Datasets:

Hendrik-a

/

arxiv

like 0

dopants of transition metal ions in ii - vi semiconductors exhibit native 2 + valency . despite this , 3 + or mixed 3+/2 + valency of iron ions in zno was reported previously . several contradictory mechanisms have been put forward for explanation of this fact so far . here , we analyze fe valency in zno by complementary theoretical and experimental studies . our calculations within the generalized gradient approximation ( gga+ ) indicate that the fe ion is a relatively shallow donor . its stable charge state is fe in ideal zno , however , the high energy of the ( + /0 ) transition level enhances the compensation of fe to fe by non - intentional acceptors in real samples . using several experimental methods like electron paramagnetic resonance , magnetometry , conductivity , excitonic magnetic circular dichroism and magneto - photoluminescence we confirm the 3 + valency of the iron ions in polycrystalline ( zn , fe)o films with the fe content attaining 0.2%.we find a predicted increase of-type conductivity upon the fe doping with the fe donor ionization energy of ev consistent with the results of theoretical considerations . moreover , our magnetooptical measurements confirm the calculated non - vanishing _ s , p - d _ exchange interaction between band carriers and localized magnetic moments of the fe ions in the zno , being so far an unsettled issue .

1603.01123

the magnetically doped zinc oxide holds a great promise for implementations in optoelectronic devices.@xcite as a member of a wide bandgap semiconductor family it is characterized by a small lattice constant , a possible large _ p - d _ hybridization , a small spin - orbit interaction , and a large exciton oscillator strength . these properties make zno based dilute magnetic semiconductors dmss highly attractive for room temperature applications.@xcite unique magnetic and magnetooptical properties have been already demonstrated , e.g. , for ( zn , co)o and ( zn , mn)o.@xcite ultra long spin coherence time ( @xmath5s ) found recently@xcite for fe@xmath2 ions in the zno indicates ( zn , fe)o as a highly promising spintronic system . systematic studies of magnetooptical properties of the zno doped with iron ions are , however , still missing . in particular , in the work of ando et al . ( ref . ) magnetic circular dichroism ( mcd ) spectrum has been shown for the ( zn , fe)o , but no dependence on the magnetic field nor the iron charge state was determined . also , there is no literature record related to properties of near - the - band - gap photoluminescence ( pl ) of the ( zn , fe)o in magnetic field . dopants of transition metal ions in ii - vi semiconductors natively exhibit the 2 + valency . in particular , 2 + valency was reported for the iron ions in cds , zns , cdse , znse , cdte and znte.@xcite in the case of zno the fe ions have been , however , observed in parallel in both valence states ( 2 + and 3+)@xcite or exclusively in 3 + state.@xcite several contradictory mechanisms have been put forward for explanation of 3 + valency of iron in zno , like promotion of fe@xmath1 into the fe@xmath2 due to a compensation induced by zn vacancies,@xcite photoionization of the fe@xmath1 centers@xcite or a direct charge transfer from the ion to the conduction band@xcite . in parallel , the long spin coherence time@xcite suggests that the fe@xmath2 ions in zno are decoupled from their environment . thus , there are at least two questions that still need to be answered : ( i ) what is the valency of fe ions in zno ? ( ii ) do the fe@xmath2 ions couple to band carriers through _ s , p - d _ exchange interaction ? in this work , we start with presenting the results of the density functional theory ( dft ) study of the electronic and magnetic properties of fe in zno . the generalized gradient approximation , together with the @xmath6 corrections considered as fitting parameters , are employed . calculations performed with @xmath0(fe ) larger than 3 ev show that the fe@xmath1 is a relatively shallow donor with the ( + /0 ) transition level in the upper part of the gap , and therefore the stable charge state of fe in ideal zno is the fe@xmath1 . the experimental ionization energy , around 0.25 ev , is reproduced with @xmath0(fe ) = 4 ev . the high energy of ( + /0 ) level relative to the valence band facilitates compensation of the fe@xmath1 to the fe@xmath2 by , e.g. , native acceptors because of the so - called fermi level effect . when @xmath0(fe ) lower than 3 ev is assumed , the donor level of the fe@xmath1 is degenerate with the conduction band , while that of the fe@xmath2 forms a gap state . in this case , a correct description of the electronic structure of fe within gga is not possible , as discussed in appendix [ appendix ] . finally , our calculations point toward non - zero values of _ s , p - d _ exchange constants in the ( zn , fe)o . in the second part of the paper we provide a set of experimental results of magnetometry , conductivity and magnetooptical measurements on the ( zn , fe)o samples with the fe content attaining 0.2% . they confirm the theoretical findings . magnetometry results point toward 3 + valency of the fe ions . conductivity measurements reveal ionization energy of the fe ions around 0.25 ev . brillouin - like dependencies found in measurements of mcd and of pl in magnetic field along with a clear curie - paramagnetic dependence of magnetization on temperature determined from magnetospectroscopy confirm presence of the fe ions in 3 + charge state and the ion - carrier _ s , p - d _ exchange interaction in the ( zn , fe)o . the paper is organized as follows . section [ theory ] presents the method and the results of the calculations , which explain the valency of iron ions in zno and predict non - zero _ s , p - d _ exchange integrals for the ( zn , fe)o . [ sec : samples ] describes the samples studied along with the results of their structural characterization . [ sec : exp : valency ] gathers experimental results obtained with non - optical methods , which testify the presence of iron ions in 3 + valency in the studied samples . [ sec : magnetospectroscopy ] reports on the results of magnetospectroscopy investigations , which confirm the findings from the sec . [ theory ] and sec . [ sec : exp : valency ] , in particular by providing evidence for the _ s , p - d _ exchange interaction in the ( zn , fe)o . theoretical description of transition metal ( tm ) impurities in semiconductors represents a demanding test for electronic structure calculations for two reasons . the first one is that the local density approximation ( lda ) and gga to the dft severely underestimate the band gap , due to which the tm levels can be incorrectly predicted to form resonances degenerate with the conduction band continuum , rather than the experimentally observed states in the band gap . this error can lead to an erroneous charge state of a tm impurity , metallic rather than insulating crystal , spurious magnetic interactions between the dopants , etc . @xcite the second problem is related to a localized nature of the 3@xmath7 wave functions , for which many body effects can play an important role , requiring usage of approaches beyond lda / gga . this holds for both orbitals of host semiconductors and @xmath7(tm ) impurity orbitals . in the case of zno , both lda and gga give a too small band gap @xmath8 of about 1.0 ev , @xcite and a too high energy of the @xmath7(zn)-derived bands relative to the valence band maximum ( vbm ) . @xcite the hybrid functionals @xcite and quasiparticle gw approaches @xcite are more accurate methods to describe the band structure . however , as recently pointed out in ref . , they improve mainly the band gap , while placing the @xmath7(zn ) band too high . consequently , lim _ et al . _ applied an on - site potential for @xmath7(zn ) states during gw calculations to shift down the @xmath7(zn)-derived bands and hence to obtain full agreement with experiment . @xcite an alternative pseudoempirical approach consists in using gga supplemented by the @xmath6 corrections , @xcite which can be calculated e.g. by the linear response method @xcite or can be treated as free parameters fitted to experimental data . the @xmath6 term applied to @xmath7(zn ) orbitals @xcite greatly improves the position of @xmath7 bands , but hardly affects the band gap . the correct gap was obtained by applying an extremely large @xmath0 term to @xmath9(zn ) states ( @xmath10 ev ) in addition to @xmath0 for @xmath7(zn ) states . @xcite another method consisted in using @xmath6 for @xmath7(zn ) supplemented by empirical non - local external potentials on @xmath9 and @xmath11 states of both zn and o atoms . @xcite to obtain the correct @xmath8 within gga+@xmath0 one should note that the upper valence band is mainly derived from the @xmath11(o ) orbitals . consequently , the @xmath0 term for the @xmath11(o ) orbitals , in addition to @xmath0(zn ) , should be included . @xcite this approach is employed in this work , as described below . the calculations are performed within dft theory with the gga for the exchange - correlation potential . @xcite the @xmath6 corrections are included according to refs . we use the pseudopotential method implemented in the quantum espresso code , @xcite with the valence atomic configuration @xmath12 for zn , @xmath13 for o and @xmath14 for fe , respectively . the plane - waves kinetic energy cutoffs of 30 ry for wave functions and 180 ry for charge density are employed . the electronic structure of zno in the wurtzite phase is examined with the @xmath15 @xmath16-point grid . analysis of a single fe impurity in zno is performed using the @xmath17 supercell with 72 atoms ( 2.8 at % of fe ) and the @xmath18 supercell with 144 atoms ( 1.4 at % of fe ) . for the density of states ( dos ) calculations , the @xmath16-space summations are performed with a @xmath19 @xmath16-point grid . ionic positions are optimized until the forces acting on ions are smaller than 0.02 ev / . methfessel - paxton @xcite smearing method with the smearing width of 0.136 ev or lower is used for partial occupancies . calculations with fixed occupation matrices are performed at the @xmath20 point only using the smaller @xmath17 supercell . the values of the @xmath0 terms for 3@xmath7(zn ) and 2@xmath11(o ) orbitals are fitted to reproduce the band structure of zno . we find that @xmath0(zn)=12.5 ev and @xmath0(o)=6.25 ev reproduce both the experimental @xmath8 of 3.3 ev , @xcite and the energy of the @xmath7(zn ) band , centered about 8 ev below the vbm . @xcite they also lead to the proper width of 6 ev of the upper valence band of mostly @xmath11(o ) character , in accordance with the experiment . @xcite our @xmath0 parameters , although relatively high , are similar to the values reported in other works . in refs and , @xmath0(zn)@xmath21 ev and @xmath0(o)@xmath22 ev were fitted to the experimental band structure . importantly , in ref . , parameters @xmath0(zn)=12.8 ev and @xmath0(o)=5.29 ev were calculated by using pseudohybrid hubbard density functional method . this is consistent with the fitted values , and thus provides a complementary justification of the used @xmath6 parameters . the relaxed crystal structure agrees well with experiment : the lattice parameters @xmath23 and @xmath24 as well as internal parameter @xmath25 are underestimated by less than 1 % in comparison with experimental values @xmath26 , @xmath27 , and @xmath25 . @xcite the value of @xmath0(fe ) is considered as a free parameter varying from 0 to 6 ev . finally , the increase of the temperature from 0 to 300 k causes a change of the measured band gap by about 0.1 ev . this change is neglected in the calculations , since it does not affect the conclusions . in ideal ii - vi semiconductors , i.e. , those without additional dopants and defects , a single fe impurity is expected to occur in the fe@xmath1 ( @xmath28 ) charge state , with 6 electrons on the @xmath7(fe)-induced levels and total spin @xmath29 . however , as it follows from the present results , due to proximity of fe levels to the zno conduction band , the description of electronic configuration of fe in zno is not straightforward . more specifically , for small values of @xmath0(fe ) , @xmath30(fe)@xmath31 ev , the calculated level of fe@xmath1 is degenerate with the continuum of the conduction band , and there are fundamental problems with the correct description of this case within lda / gga . this issue is discussed in detail in appendix a. on the other hand , those problems are absent when higher values of @xmath0(fe ) are assumed . in this case , in the ideal zno , fe is a standard impurity in the 2 + charge state . due to the proximity of the donor level to the conduction band we predict , however , an easy ionization of fe@xmath1 to the fe@xmath2 , either by thermal excitation of an electron to the conduction band , or by electron transfer to acceptors present in the host . this means that in the realistic case of imperfect samples both fe@xmath1 and fe@xmath2 should co - exist . ( color on - line ) energy bands and dos of zno with fe in : ( a ) and ( c ) @xmath32 charge state , fe@xmath2 , ( b ) and ( d ) @xmath28 charge state , fe@xmath1 . grey area and blue lines in dos indicate the total dos and the dos projected on @xmath7(fe ) orbitals , respectively . red lines denote the band gap of zno . zero energy is set at the vbm . the calculation is performed with the 144-atom supercell for @xmath0(fe ) = 4 ev . ] the band structure and dos of zno calculated assuming @xmath0(fe)@xmath33 ev is shown for the fe@xmath2 and fe@xmath1 case in fig . [ figt1](a ) and fig . [ figt1](b ) , respectively . the impurity @xmath7(fe ) gap states constitute nearly dispersionless bands , while their energies strongly depend on the fe charge state . to make the discussion transparent we begin the analysis by fe@xmath2 , and then move to fe@xmath1 . the fe ion in the @xmath32 charge state , fe@xmath2 , assumes the electronic configuration @xmath34 characterized by 5 electrons on the @xmath7(fe ) levels and total spin @xmath35 . in this case , fe introduces two spin - up states below the vbm : the @xmath36 doublet and the @xmath37 triplet , which form broad resonances within the valence band , see fig . [ figt1](c ) . the exchange spin - up@xmath38spin - down splitting is strong , @xmath39 ev , since the @xmath40 doublet is at 3.0 ev , while the @xmath41 triplet is practically degenerate with cbm . both spin down levels are empty . addition of one electron to the zno : fe@xmath2 system results in the @xmath28 charge state fe@xmath1 , with 6 electrons on the @xmath7(fe)-induced levels . in this case , the energies of fe - induced levels rise due to the increased intracenter coulomb repulsion , the @xmath36 doublet is still degenerate with the valence band , while the @xmath37 triplet is at about 0.8 ev above vbm , and is split into a @xmath42 doublet and a @xmath43 singlet . more importantly , the @xmath44 singlet occupied with one electron is 0.25 ev below cbm . ( color on - line ) energy levels of ( a ) fe@xmath2 , ( b ) `` fe@xmath1 '' ( for @xmath0(fe)@xmath31 ev , see appendix a ) and fe@xmath1 ( for @xmath0(fe)@xmath45 ev ) , and ( c ) the transition level @xmath46 calculated as a function of @xmath0(fe ) . the 72 atom supercell is used in calculations.,width=313 ] the value of @xmath0(fe ) is treated here as a parameter to be fitted to experiment . the dependence of the fe states on the @xmath0(fe ) is presented in fig . [ figt2 ] . as it is discussed in refs for transition metal impurities in gan , the @xmath0-induced shift of an impurity level depends on its occupation , and is negative ( positive ) for the occupied ( empty ) state . this feature is clearly seen in fig . [ figt2](a ) for both the occupied @xmath36 and @xmath37 , and the empty @xmath40 and @xmath41 levels of fe@xmath2 . the calculated dependencies of @xmath36 and @xmath37 on @xmath0(fe ) are non - linear because of their increasing hybridization with valence states . the case of fe@xmath1 , [ figt2](b ) , is more complex . for the @xmath0(fe ) up to 2.5 ev , the calculated energies of the spin down levels are degenerate with the conduction band , but this `` pseudo - resonant '' character of fe is largely an artefact of gga , see appendix [ appendix ] . the @xmath41 triplet is about 1 ev above @xmath40 , as for fe@xmath2 . in the interval @xmath47(fe)@xmath48 ev , an abrupt change in the order the fe - induced levels occurs . in particular , the @xmath41 triplet splits , the @xmath44 singlet derived from @xmath41 is lower in energy than the @xmath40 doublet , becoming a gap state for the @xmath0(fe ) higher than 3 ev ( see ref . ) . since @xmath44 is occupied with one electron , its energy decreases with the @xmath0(fe ) . this splitting of @xmath41 changes the atomic configuration by increasing the difference between the length of the fe - o bond along the @xmath49-axis and those of the three remaining planar bonds , see ref . . this , in turn , induces the splitting of the @xmath37 gap state for @xmath0(fe)@xmath50 ev . consequently , for the @xmath0(fe)@xmath51 ev , fe is a donor with an occupied singlet in the band gap and spin 2 . this allows for calculations of the transition level and ionization energy of fe . the change of the charge state of fe is determined by the @xmath46 transition level between @xmath32 and @xmath28 charge states . it is defined as the fermi energy relative to vbm at which formation energies of the 0 and + 1 charge states are equal : @xmath52 where @xmath53 is the total system energy . the calculations are performed along the scheme proposed in ref . . the energy of vbm , @xmath54 , is determined from the total energy difference between the pure zno crystal with and without a hole in the vbm , i.e. @xmath55 ( color on - line ) dos of pure zno and zno with fe in the @xmath32 charge state , fe@xmath2 , after potential alignment @xmath56 ev . to improve visibility , a contribution of the fe states , shown in blue , is multiplied by 10 . @xmath0(fe ) = 4 ev is used in calculations . , width=302 ] in general , the supercell results must be corrected for finite size effects by taking into account the band filling correction , and the image charge corrections together with potential alignment for charged defects . @xcite in the latter case , a compensating homogeneous background charge is assumed . this guarantees charge neutrality , i.e. , a correct value of the @xmath57 component of electrostatic potential , but it also requires an alignment of the average electrostatic potential in the supercell with and without the defect . the potential alignment energy @xmath58 is estimated by comparing the dos of pure zno and of zno : fe@xmath2 . far from the defect levels , the energy shift for charged system is @xmath59 ev ( fig . [ figt3 ] ) . image charge correction @xmath60 stems from the electrostatic coupling between charged defects in different supercells . the simplified version of the one proposed in ref . is employed , @xmath61 \frac{q^2\alpha_m}{2\varepsilon v^{1/3}},\ ] ] where @xmath62 is the madelung constant , @xmath63 is the low - frequency dielectric constant , @xmath64 is the supercell volume , and @xmath65 is the shape factor for a @xmath17 hcp supercell . this gives @xmath66 ev . thus , the potential alignment and the image charge term cancel to a good approximation . regarding the convergence with respect to the supercell size we note that supercells with about 100 atoms give well convergent results even for highly charged defects , see ref . . our results confirm this , since the values obtained with 72 and 144 atom supercells agree to better than 0.1 ev . finally , in the present case the band filling correction vanish , since the bands derived from the @xmath7(fe ) are fully occupied for both fe@xmath1 and fe@xmath2 . the calculated @xmath46 transition level is shown in fig . [ figt2](c ) as a function of @xmath0(fe ) . in agreement with fig . [ figt2](b ) , @xmath46 coincides with the cbm for @xmath0(fe)=3 ev , and decreases in energy with the increasing @xmath0(fe ) . a quantity often considered in the context of experiment is the ionization @xmath67 energy . ( here , the @xmath67 is experimentally determined from the temperature dependence of conductivity in sec . [ sec : transport ] . ) it can be calculated as the total energy difference between the final and the initial states of the system , which are the fe@xmath2 with one electron in the conduction band and fe@xmath1 , respectively . as it is pointed out in refs , the energy of the @xmath46 level counted from the vbm and the thermal ionization energy @xmath67 should add up to the @xmath8 . in practice , when the two quantities are determined in calculations small differences between them occur due to the used approximations . for instance , when calculating the total energy of the @xmath32 state , the fictitious neutralizing background is assumed . ( however , in actual calculations the background is not introduced . instead , the @xmath57 component of the electrostatic potential is assumed to vanish . @xcite ) on the other hand , in the calculations of the @xmath67 , both in the initial and the final state the supercell are electrically neutral , and introduction of the background ( or the adjustment of the @xmath68 component of the total electrostatic potential ) is not necessary . another source of discrepancies is the limitation of the brillouin zone summations to the @xmath20-point , used in the calculations of excited states with fixed non - equilibrium occupations . in our case , both approaches agree within 0.1 ev . by comparing theory with experiment we find that our experimental value @xmath69 ev is reproduced when @xmath0(fe)=4 ev , fig . [ figt2](c ) . @xcite this is close to 4.3 ev for feo , and larger than @xmath0(fe)@xmath70 ev found for bulk fe . @xcite it is worth to note that despite the almost vanishing @xmath0 for mn and fe found in the corresponding case of gan,@xcite a recent study @xcite indicates that also for mn in zno , the @xmath0(mn ) term is about 1.5 ev , i.e. , larger than in gan . previous theoretical investigations of fe in zno provided conflicting results regarding the stability of the fe@xmath1 charge state . @xcite the uncertainty is largely due to the band gap problem , mentioned in the subsection [ teo1 ] . the lda / gga calculations typically situated the @xmath40 of fe@xmath1 above the cbm , but this level ordering can result from the too low band gap . on the other hand , the gap - corrected approaches predicted the @xmath40 state below the cbm . in the former case , the instability of fe@xmath1 is expected as discussed above , but this possibility was not explicitly addressed . more specifically , in ref . the lda was used . the @xmath40 state of fe was found to pin the fermi level , and its energy was degenerate with the cbm ; this is close to the present results for small @xmath0(fe ) , see appendix [ appendix ] . such a situation was also predicted by the lda calculations of ref . . lda was also used in ref . with the @xmath6 term applied only to @xmath7(fe ) orbitals , which gave a strongly underestimated band gap , and the energy of @xmath40 above the cbm . previous theoretical approaches to fe in zno with the corrected gap included the lda with the self - interaction corrections and @xmath0(fe)@xmath71 ev ; @xcite this provided an fe - induced spin - down band at 1.8 ev above the vbm . similarly , in ref . the band structure of pure zno was fitted to experiment , and the corrections for @xmath7(fe ) , @xmath72 ev and the exchange constant @xmath73 ev ( thus effectively @xmath0(fe)=2.5 ev ) were applied . the @xmath74 transition level was found at 1.98 ev , i.e. , about 1.4 ev below the cbm . in ref . , gga@xmath6 with @xmath0(fe)@xmath75 ev and @xmath0(zn)@xmath76 ev was employed together with an empirical band gap correction . the @xmath74 transition level was predicted at 2.9 ev above the vbm . this is close to our results with @xmath0(fe)@xmath77 ev . .[table1 ] the fe magnetic moment @xmath78 ( @xmath79 ) , conduction @xmath80 ( ev ) and valence @xmath81 ( ev ) band - edge spin splittings , and exchange constants @xmath82 ( ev ) , @xmath83 ( ev ) for @xmath28 and @xmath32 charge state . results are obtained for the 72-atom supercell , @xmath84 . for @xmath85=+1 and @xmath0(fe)=0 , values of the @xmath81 and the @xmath83 can not be precisely determined , as indicated by `` @xmath86 '' ( see sec . [ teo3 ] for details ) . [ cols="<,<,^,^,^,^,^",options="header " , ] the calculated spin splitting of the conduction band @xmath87 and the valence band @xmath88 , produced by the coupling of fe with the host zno states , can be used to estimate the _ s , p - d _ coupling . within the mean - field approximation , the exchange constants are expressed by @xcite @xmath89 where @xmath90 is the concentration of the fe ions and @xmath91 is one half of the computed magnetization . the calculated values are given in table [ table1 ] . as it follows from the table , the constant @xmath82 is around @xmath92 ev for both , 2 + and 3 + , fe charge states . the obtained value is comparable to the typical one , 0.2 ev , found in ii - vi compounds . @xcite moreover , the @xmath82 depends on the @xmath0(fe ) only weakly . in contrast , the constant @xmath83 is strongly dependent on both the fe charge state and the @xmath0(fe ) . for @xmath28 and @xmath0(fe)@xmath93 ev , the @xmath83 is obtained for `` fe@xmath1 '' charge state , and thus it should be considered as an estimate . the obtained @xmath83 , @xmath94 ev , is smaller than typical iron - hole exchange integrals for ii - vi compounds , @xcite but of the same order as effective exchange integrals reported for wide gap dmss @xcite with the increasing @xmath0(fe ) , @xmath37 approaches the valence band ( see fig . [ figt2 ] ) , which increases the spin splitting of the vbm , and for the @xmath0(fe)@xmath33 ev the @xmath83 reaches 2.0 ev . on the other hand , for @xmath32 , i. e. _ , the fe@xmath2 case , evaluation of the @xmath83 is obscured by energetic proximity of the vbm and the @xmath37 level . indeed , as it follows from dos shown in fig . [ figt1](c ) , for @xmath32 the @xmath37 state is degenerate with the top of the valence band , and therefore its coupling with the host valence states is very pronounced . also a direct inspection of the wave functions of the states close to the vbm reveals the strong hybridization , which makes it impossible to clearly distinguish between band states and fe - induced states . for the @xmath0(fe)=4 ev , the @xmath37 is well below the vbm , and we obtain the @xmath95 ev . regarding the sign of the interaction , we find that conduction electrons are ferromagnetically coupled with the fe impurities , as expected for the direct exchange coupling . @xcite the fe - hole coupling has the kinetic exchange character , @xcite driven by the hybridization of the fe and host states . the sign of the coupling constant is determined by the energy of @xmath37 relative to the vbm . for @xmath28 , @xmath37 is above the vbm , and the coupling of holes with the fe is ferromagnetic . for @xmath32 the coupling changes the sign . in this case , @xmath83 is formally evaluated from the spin splitting of the highest valence states . such an approach corresponds to the interpretation of luminescence experiments , in which one determines the spin splitting of the highest valence states involved in the exciton recombination regardless of their actual orbital composition . the coupling between the fe and o first nearest neighbors induces magnetic moment on o ions . in the case of fe@xmath2 observed in epr , the calculated magnetic moments of o ions are @xmath96 , and that of the fe is @xmath97 . the corresponding total magnetic moment is @xmath98 . in the case of fe@xmath1 , the spin of the sixth @xmath7(fe ) electron is antiparallel to the remaining ones , thus both the total magnetic moment and the contributions from fe and o neighbors are lower than for the fe@xmath2 case . the results hardly depend on the @xmath0(fe ) value . ( color on - line ) contour plots of the wave functions squared of ( a ) the top of the valence band , ( b ) @xmath36 , ( c ) @xmath37 , and ( d ) the bottom of the conduction band . blue , green , and red dots denote o , zn , and fe atoms , respectively . , width=313 ] the wave functions at the @xmath20 point of the vbm , @xmath36 , @xmath37 , and the cbm are shown in fig . [ figt4 ] for @xmath28 and the @xmath0(fe)@xmath99 . as expected , the top of the valence band is composed of @xmath11(o ) and of @xmath7(fe ) , while the contribution of @xmath7(zn ) is less pronounced . the two gap states are dominated by @xmath7(fe ) , but the contribution of the host states , mostly the o nearest neighbors , is clearly visible . the hybridization is more pronounced in the case of @xmath37 . ( the structure of both @xmath36 and @xmath37 is the same as for mn in gan , @xcite while the contribution of the fe to the vbm in zno is lower than that of the mn in gan . ) the wave functions of the two fe - induced @xmath40 and @xmath41 states , degenerate with the conduction band continuum , are very similar to those of the spin up @xmath36 and @xmath37 ( see fig . [ figt4 ] ) and thus are not shown . finally , the conduction band edge , formed by @xmath9(o ) states and a small contribution of @xmath9(zn / fe ) states , is practically not perturbed by the @xmath7(fe ) . the coupling with the conduction states occurs @xmath100 the direct exchange mechanism , @xcite and does not necessitate hybridization . the calculations performed for the @xmath0(fe)= 4 ev provide comparable results . summing up the theory part of the paper , there are fundamental difficulties with a correct gga description of the @xmath28 charge state . they stem from the fact that the @xmath40 , which should be occupied by one electron in the fe@xmath1 case , is above the empty conduction band . in consequence , self - ionization should occur . however , when the electron occupies the cbm , the @xmath40 level is below cbm , which also does not correspond to the ground state configuration of the system . the paradoxal situation , analogous to that found for the isolated fe atom @xcite , is analyzed in appendix a , where it is indicated that a correct description of zno : fe is not possible within gga . this problem is absent for both the @xmath0(fe)@xmath101 ev , and for the @xmath32 charge state . in particular , the large @xmath102 ev correction induces a decrease of the @xmath44 state of fe@xmath1 below the cbm , which corresponds to a standard configuration of a relatively shallow donor with a well - defined transition level in the upper part of the band gap . this facilitates compensation by native acceptors , such as zinc vacancies @xcite thanks to the self - compensation effect ( _ i.e. _ , to the dependence of formation energy on the fermi energy ) . @xcite the fe@xmath2-induced levels are about 1.2 ev lower than those of the fe@xmath1 . this large difference in the level energies of @xmath28 and @xmath32 stems from the strong intracenter coulomb repulsion between the @xmath7(fe ) electrons caused by the localization of their wave functions . interestingly , in the case of mn in zno the intracenter coulomb repulsion is of comparable strength , and it leads to metastability of the photoexcited mn@xmath2 . @xcite the exchange coupling constant @xmath82 , about 0.4 ev , does not depend on the fe charge state . @xmath83 is non - vanishing as well , but it depends on the charge state of the fe : it amounts to about 2.0 ev for the fe@xmath1 , while for the fe@xmath2 it is predicted to be lower and of opposite sign , @xmath103 ev , since the @xmath37 level is degenerate with the vbm . studied zno layers doped with the fe ions are produced by a spray pyrolysis method on quartz or glass substrates,@xcite with a respective thicknesses of @xmath104 nm and @xmath105 nm , as determined by scanning electron microscope measurements ( not shown ) . the layers exhibit a polycrystalline structure with @xmath106 nm grain size . as determined by magnetometry measurements ( see sec . [ sec : squid ] ) , the atomic concentration of the iron ions is @xmath107 % , which corresponds to the iron ion density of @xmath108 at/@xmath109 . a layer of pure zno is also grown in the same conditions as the fe doped ones . x - ray diffraction ( xrd ) characterization shows that the samples exhibit the crystalline structure of the hexagonal zno ( see fig . [ fig : xrd ] ) . the bragg peak ( 002 ) is dominant , suggesting that the ( 002 ) plane growth rate is the fastest one . the plane ( 002 ) is arranged parallel to the substrate , i. e. , the preferential c - axis orientation is perpendicular to the film plane for all the samples . in order to determine the valence state of iron ions incorporated into the samples , we investigate magnetic and conductivity behavior of the studied layers . electron paramagnetic resonance ( epr ) measurements are performed at the room temperature using a bruker elexsys e580 cw spectrometer operating at a microwave frequency around 9.4 ghz ( x band ) with a te@xmath110 resonance cavity . to find the best measurement conditions , the spectra are measured as a function of the microwave power in the range between 0.047 mw and 150 mw , at temperature from 2 k up to the room temperature , and modulation amplitude of the magnetic field from 0.1 mt up to 0.5 mt . the signal from the quartz or glass substrate alone , _ i.e. _ , without any deposited ( zn , fe)o layer is also measured and then subtracted as the background from the total signal measured on the studied samples . no powdering of the samples is done . two relatively sharp resonance lines are observed at magnetic fields around 300 mt , superimposed on a broad band spectrum , as shown in fig . [ fig : epr ] . the fit with a derivative of a sum of two gaussian curves yields g - factor equal g@xmath111 = 2.00 @xmath112 0.01 and g * = 2.15 @xmath112 0.01 , for a resonance respectively at a higher and lower magnetic field value ( see fig . [ fig : epr ] ) . the g@xmath111 value agrees very well with the one reported in a previous study for the @xmath113 fine structure transition of isolated , substitutional fe@xmath2 ion ( s=5/2 ) in zno powders ( 2.0062).@xcite we therefore attribute it to the fe@xmath2 ions in polycrystallites with the c - axis oriented along the normal to the sample plane . we attribute the second resonance and corresponding g - factor g * to @xmath114 transition of the fe@xmath2 ions in polycrystallites , which c - axes deviate from the normal to the sample plane . a presence of such polycrystallites in the studied samples has been revealed by maxima in xrd spectrum different then 002 ( see fig . [ fig : squid ] ) . since the linewidth of the weaker @xmath115 and @xmath116 fine structure transitions is much more sensitive to variations of the polycrystallite size and position of the fe ion ( on the surface of / inside the polycrystallite ) , the corresponding resonances are smeared out and contribute only to the broad band spectrum . the smearing could be also due to a weak signal from the fe ions ( resulting from their low concentration ) and a relatively strong background signal of the substrate , which decreased a signal - to - noise ratio . the impact of spin - spin interaction is most likely negligible here , as the concentration of the iron ions is relatively low . the agreement of the measured epr spectrum with that expected for the fe@xmath2 ion in zno@xcite confirms the presence of the iron ions with the 3 + valency in the studied samples . we note that 2 + ions are not detected in the epr measurements , and in consequence their presence can not be verified using this method . however , as will be shown below , a good description of the dependence of magnetization and of the mcd on magnetic field is obtained using a brillouin function as for the fe@xmath2 ions . = 0.2 % ) sample registered at t = 300 k , acquired with the microwave power 1.5 mw , modulation amplitude 0.1 mt , and microwave frequency 9.393668 ghz . due to a very weak epr signal intensity , in order to improve signal - to - noise ratio , the spectrum has been averaged over ten acquisitions . fit with a derivative of a sum of two gaussian curves ( blue line ) yields g@xmath111 and g * values , as indicated . ] magnetization measurements are performed with the use of a squid - type magnetometer ( liquid helium cooled mpms xl device manufactured by quantum design , providing the sensitivity of @xmath117 emu ) in the temperature range 2 - 300 k and magnetic fields up to 7 t. the measured magnetic moment of each sample is a sum of paramagnetic moment of the fe@xmath2 ions , a diamagnetic contribution from the zno layer and the substrate , and a possible contribution of unintentional impurities . therefore the measured magnetization can be expressed in the form : @xmath118 where @xmath119 is a total magnetic moment of the fe@xmath2 ions in zno matrix , @xmath120 is the sum of diamagnetic susceptibility of zno layer and of the substrate ( assumed to be temperature independent in the studied temperature range ) and @xmath121 represents a contribution from possible precipitates of secondary phases . we note that in our case diamagnetic contribution dominates the others since the mass of the magnetic layer is only a tiny fraction of the total mass of the sample . in such a case a precise value of @xmath120 of the sample is crucial . a careful analysis of the data reveals the absence of ferromagnetic secondary phases ( e.g. , fe - rich aggregates ) in all of the studied samples . this is in contrast to what was reported previously for the ( zn , fe)o@xcite and other fe - doped wide band gap semiconductors , e.g. , for ( ga , fe)n.@xcite consequently , the term @xmath121 in the eq . [ eq1:magnetization ] will be neglected.@xcite having in mind the epr identification of the fe impurity as the fe@xmath2 , with spin @xmath35 and @xmath122 , we assume @xmath119 in the form : @xmath123 where @xmath124 is the brillouin function for spin s = 5/2 , g is g - factor ( see sec . [ sec : epr ] ) and a is the number of the fe centers . possible interaction between the fe@xmath2 ions is neglected here due to their expected low concentration . in order to extract @xmath119 from the total magnetization of the samples @xmath125 , the dominating diamagnetic contribution ( i.e. @xmath120 b ) is subtracted using the following procedure . at t = 300 k the brillouin function in eq . [ eq2:magnetization ] is a linear function of the magnetic field . so that fitting of the m@xmath126(b , t = 300 k ) by a straight line provides a susceptibility of the sample , which is in practice solely a diamagnetic susceptibility ( a residual paramagnetic contribution originating from the fe@xmath2 ions is negligibly small with respect to the total sample susceptibility due to a relatively high temperature and a small density of the fe@xmath2 dopant ) . the diamagnetic susceptibility evaluated in this way from the data in the field range @xmath127 ( @xmath128 ) is used for evaluation of @xmath119 according to eq . [ eq2:magnetization ] . . ] in fig . [ fig : squid ] , both the experimental data ( i.e. , @xmath129 ) and the fit ( i.e. , @xmath130 are given for an example sample grown on quartz substrate . parameter @xmath131 being the only fitting parameter provides the information about the actual molar concentration of the fe@xmath2 ions in the layers . it attains @xmath107% , that is @xmath108 at/@xmath109 , in agreement with the results of energy - dispersive x - ray ( edx ) spectroscopy ( not shown ) . this value is typical for the studied layers . we note that the magnetization measurements confirm the @xmath132 state of the iron ions , which points toward high efficiency of the autoionization and/or compensation of the ions predicted by the theory . electrical contacts on thin films are deposited using a conductive silver paint . contact ohmicity is systematically verified by i - v characteristics . dc- resistivity and hall effect measurements were performed in a van der pauw configuration in the temperature range from 90 k to 400 k and for magnetic field perpendicular to the film plane varying from 0 t to 1.6 t using a custom designed , high impedance measurement set - up . this set - up is a combination of keithley ( usa ) electrical measurement device and a temperature controller ( linkam scientific , uk ) coupled by a software written in gemac , versailles . the equipment enables measurements of highly resistant samples , up to 10 g@xmath133 . the samples exhibit a semiconducting behavior . namely , the resistivity increases with the temperature decrease . at room temperature , resistivity @xmath134 of the 0.2% ( zn , fe)o on glass substrate equals to @xmath135 cm . in conductivity _ vs _ inverse temperature plots , two different regions of the conductivity are distinguished : ( i ) for 300 k down to 200 k , where conduction due to the temperature activation to the band dominates , ( ii ) for temperatures below 200 k , where the near neighbor hopping mechanism becomes dominant . the activation energy of band conductivity has been determined to be @xmath4 ev.@xcite negative sign of hall voltage is found for all layers in hall effect measurements at room temperature . this implies that the samples , independently if deposited on glass or on fused silica , are of n - type . for pure zno electron concentration @xmath136 and mobility @xmath137 = 1.9 @xmath138/vs are found , while for the 0.2% ( zn , fe)o @xmath3 reaches @xmath139 and mobility decreases due to the presence of scattering centers down to @xmath137 = 0.5 @xmath138/vs . a natural explanation for the increase of the density of free electrons upon the iron doping is that at least some part of iron ions are donors , which are not fully compensated by the native acceptors . in fact , the increase of the electron density from @xmath140 @xmath141 to @xmath142 @xmath141 upon the fe doping is at least two orders of magnitude lower than it would result from the density of the ionized fe ions ( @xmath143 @xmath141 already at t = 2 k , see sec . [ sec : squid ] ) . thus , we state that most of the ions , which ionize from fe@xmath1 to fe@xmath2 state are compensated by native acceptors . as indicated by previous experimental theoretical @xcite and experimental @xcite works , a likely source of native acceptors in zno is a zn vacancy , however a contribution to the carrier concentration may come also from oxygen vacancies@xcite or surface states of the studied polycrystalline sample . the samples are placed at pumped helium temperature inside a cryostat equipped with a superconducting coil magnet . the emission is non - resonantly excited at 3.81 ev ( 325 nm ) using a continuous wave he - cd laser . the excitation beam is focused to a 0.1 mm spot on the sample surface . the signal is detected by a ccd camera coupled to a grating spectrometer ( 0.1 mev of overall spectral resolution of the setup ) . circular polarizations of the signal are resolved . the reflectivity and pl measurements are performed in the faraday configuration in magnetic field of up 10 t , in temperature range from 1.5 k to 50 k , with a halogen lamp serving as a source of the ultraviolet light . the mcd is determined based on the acquired reflectivity spectra as @xmath144 - r@xmath145 + r@xmath146 , where r@xmath147 and r@xmath148 represents intensity of the reflectivity spectrum in the @xmath149 and @xmath150 polarization , respectively . optical transitions of three excitons , @xmath131 , @xmath151 and @xmath121 , are present in the reflectivity spectra of the studied ( zn , fe)o layers , as expected for a wurtzite structure semiconductor ( see fig . [ fig : mcd](a ) . magnetic field - induced splittings are small with respect to the transitions linewidths , what precludes tracing of the excitonic shifts as a function of the applied magnetic field . however , a clear mcd signal related to excitons is observed in the magnetic field . the mcd spectra determined based on the reflectivity spectra at b = 9 t for temperatures of 1.5 k , 3.5 k , 10 k and 50 k are shown in fig . [ fig : mcd]b . integrated mcd intensity@xcite ( i@xmath152 ) is calculated as an integral under the mcd curve in the region of @xmath131 and @xmath151 excitonic transitions . the i@xmath152 increases with the magnetic field following a brillouin - like dependence with a saturation at around 3 t and around 7 t for 1.5 k and 3.5 k , respectively ( see fig . [ fig : mcd](c ) ) . as it is seen , the i@xmath152 is well described by the paramagnetic brillouin function with the land factor g = 2.0062 ( taken following ref . ) and spin 5/2 ( as for the fe@xmath2 ions ) , without any free fitting parameters ( saturation value of the i@xmath152 is determined once for the 1.5 k case and then kept constant in the case of fits for 3.5 k and 10 k. ) the brillouin like shape of the observed dependence proves that zeeman splitting of bands yielding the i@xmath152 results directly from the _ s - d _ and _ p - d _ exchange interactions between the iron ions spins ( s = 5/2 ) and spins of , respectively , band electrons and holes . when the magnetic moments of the iron ions become fully oriented at a sufficiently high field , the exchange interaction induced splitting , and thus the i@xmath152 saturates , as observed . in the case of a reference sample of pure zno a linear dependence of the i@xmath152 on magnetic field originating form the zeeman splitting of bands is found , as expected ( not shown ) . below the saturation , the mcd magnitude strongly decreases with the increasing temperature following a curie paramagnetic dependence , 1/t ( see inset to fig . [ fig : mcd](c ) showing the i@xmath152 as a function of the temperature at b = 1 t. ) if the exchange interaction of carriers with fe@xmath1 ions played a dominant role in the studied samples , the van vleck type paramagnetism , proper for an ion with 3(d@xmath153 ) electron configuration , thus with a degenerate ground state , would dominate.@xcite in that case the decrease of the i@xmath152 with the temperature would be only weak in the considered temperatures range . hence , the observed rapid drop of the i@xmath152 with the increasing temperature indicates that abundance of the fe@xmath1 ions in our samples is negligible and/or that the magnitude of the _ s , p - d _ interaction is much larger in the case of the fe@xmath2 than of the fe@xmath1 ions . since the theory ( see sec . [ teo3 ] ) points towards carrier - ion exchange constants comparable or larger for the fe@xmath1 than for fe@xmath2 ions , we state that a content of the fe@xmath1 ions is at least order of magnitude lower than the one of the fe@xmath2 ions . this result is consistent with the magnetometry results at t = 2 k , which are properly described without taking into account any contribution from the fe@xmath1 ions ( sec.[sec : squid ] ) . concluding this section , the mcd provides a strong confirmation for the presence of the fe@xmath2 ions , negligible density of the fe@xmath1 ions as well as for the _ s , p - d _ interaction between the fe@xmath2 ions and the band carriers in the ( zn , fe)o . figure [ fig : pl1](a ) shows pl spectra in magnetic field b = 0 t and b = 10 t recorded in two circular polarizations of the light , @xmath149 and @xmath150 , for the ( zn , fe)o sample . the contributions from a bound exciton at around 3360 mev and donor acceptor pairs ( dap ) at around 3320 mev are identified following ref . . figure [ fig : pl1](b ) shows a degree of the circular polarization @xmath154 determined from spectra in magnetic field from 0 t to 10 t for ( zn , fe)o samples . the @xmath154 is determined as : @xmath155 - i@xmath156 + i@xmath146 , where i@xmath147 and i@xmath148 represent intensity of the pl spectrum in the @xmath149 and @xmath150 polarization , respectively . a high degree of polarization @xmath154 is observed in the spectral region of the bound exciton and of the dap at energy of around 3365 mev and 3320 mev , respectively . the @xmath154 increases with the magnetic field and saturates at around b = 5 t. integrated degree of polarization i@xmath157 , defined as the area under the degree of polarization curve in the excitonic region , is plotted as a function of the magnetic field in fig . [ fig : pl1](c ) together with a paramagnetic brillouin function fit . the fitting parameters are the saturation value and the effective temperature . the best fit is obtained for factor g = 2.0062,@xcite the effective temperature t = 3.8 k and spin 5/2 as for the fe@xmath2 ions . as it is seen the dependence is well described by the brillouin function . this strongly suggests that the @xmath154 is directly proportional to the sample magnetization . we have checked that the increased temperature with respect to the temperature of experiment ( 1.8 k ) results from the sample heating with the excitation beam . a dependence on magnetic field is qualitatively the same also for the dap ( not shown ) indicating that qualitatively the same mechanism is responsible for the effects observed for the bound exciton and the dap . a lorentzian curve is fitted to the exciton transition at around 3360 mev , providing information on parameters of the transition as a function of the magnetic field . the pl intensity determined that way for the ( zn , fe)o for @xmath149 and @xmath150 polarizations is shown in the inset to fig . [ fig : pl1](c ) . the pl intensity increases with the magnetic field in both polarizations . a similar effect on the field was previously observed in dms structures and nanostructures , e.g. , involving mn . @xcite as in previous works , @xcite we attribute the emission intensity increase to the magnetic field induced reduction of efficiency of non - radiative exciton recombination assisted by auger excitation of the fe@xmath2 ion . in the absence of the magnetic field , band carriers and the fe@xmath2 ions are spin degenerate making efficiency of the auger process independent of spin . however , the magnetic field lowers the number of spin arrangements of the electron and the fe@xmath2 ion that fulfill the spin conservation rule in the process , and thus it decreases the overall efficiency of the non - radiative auger recombination . we note that the @xmath149 polarized emission increases more than the @xmath150 polarized one . similar as in the study of ( zn , mn)o ( ref . [ ] ) , this effect can be explained in terms of spin - dependent exciton formation involving free and bound exciton states , as well as @xmath131 and @xmath151 excitons relaxation involving change of the exciton symmetry . for pure zno , the intensity of the excitonic pl linearly increases for @xmath149 and decreases for @xmath150 in the magnetic field ( not shown ) due to polarization of the carriers induced by the zeeman splitting of bands . we find also that the linewidth of the bound exciton in ( the zn , fe)o decreases nonlinearly with the magnetic field in both polarizations with a saturation at around 4 t ( not shown ) . the effect is stronger for the @xmath149 than for the @xmath150 polarization . we attribute the linewidth narrowing to a reduction of the fluctuations of the magnetization of the fe@xmath2 ions induced by the magnetic field.@xcite mutually opposite shifts in the magnetic field of bound excitons originating from bands of symmetry @xmath158 and @xmath159 ( @xmath131 and @xmath151 excitons ) also contribute to the linewidth narrowing.@xcite to summarize this part , the pl results provide a strong support for the conclusions drawn from the reflectivity measurements , unequivocally confirming presence of the ion - carrier _ s , p - d _ interaction in the studied system . theoretical and experimental analysis of the electronic structure of fe ions in zno and of the magnetic properties of zno : fe was conducted . the gga@xmath6 method was employed , with the values of @xmath0 for @xmath7(zn ) and @xmath11(o ) orbitals fitted to zno experimental band structure . the @xmath0(fe ) was also considered as a free parameter . for small values of the @xmath0(fe ) , @xmath160 ev , the calculations lead to unphysical results , with fractional occupancies of the fe level degenerate with the conduction band . the calculations performed using the @xmath0(fe)@xmath51 ev indicate that the fe is a relatively shallow donor with the @xmath161 charge state stable in the ideal zno . the @xmath0(fe)@xmath33 ev allows one to reproduce the observed ionization energy @xmath162 ev . the high energy of the @xmath163 transition level facilitates compensation of the fe@xmath1 by unintentional acceptors . thus , in real imperfect samples both , the fe@xmath1 and the fe@xmath2 , should co - exist . in parallel , epr , magnetometry , reflectivity , magneto - photoluminescence , and conductivity experiments were conducted on polycrystalline zno layers with the fe content of the order of 0.1 of atomic per cent . the results of the epr measurements reveal the presence of the substitutional fe with the 3 + valency in the studied samples . the layers magnetization determined in squid measurements is well described with the paramagnetic brillouin function as for the fe@xmath2 ions , what demonstrates that the dominant fe charge state is 3 + . consistently with this finding , the magnetic field dependencies in magnetooptical response of the samples determined in the reflectivity and pl measurements are proportional to the measured magnetization . namely , they are described by the paramagnetic brillouin function determined as for the fe@xmath2 ions and they obey the paramagnetic curie law , characteristic of the fe@xmath2 in zno , rather than the paramagnetic van vleck law expected for the fe@xmath1 . according to the conductivity measurements , the samples are n - type , with the room temperature electron concentration of about 10@xmath164 @xmath141 . the measured activation energy of conductivity of @xmath4 ev is in agreement with the theoretical value of the fe ionization energy for @xmath0(fe)=4 ev . the low electron concentration indicates a high degree of compensation of the fe donors , most probably by the zinc vacancies . @xcite the presence of compensating acceptors is also reflected by the dap recombination line seen in the photoluminescence . overall , the set of experimental results indicates in a consistent way that the content of the fe@xmath1 ions in the studied samples is at least order of magnitude lower than the one of the fe@xmath2 ions . finally , the calculations indicate that the exchange interaction coupling constant @xmath82 between conduction electrons and the fe@xmath1 or the fe@xmath2 ions , is around 0.4 ev , somewhat higher than values typical for wide band gap semimagnetic semiconductors . the exchange coupling constant @xmath83 for holes strongly depends on the fe charge state : for the fe@xmath1 the @xmath83 is of the same order as reported for other wide gap dmss ( around 2 ev and ferromagnetic ) , while in the case of the fe@xmath2 it changes its sign to antiferromagnetic and is much lower ( equal to around -0.4 ev ) . the pronounced magnetooptical effects observed in the mcd and the pl unequivocally confirm a presence of the _ s , p - d _ interaction between the band carriers and the fe@xmath2 ions in the studied system . the properties resulting from the fe doping indicate the ( zn , fe)o as a promising material for implementation in functional spintronic devices . one of the authors ( jp ) acknowledges the support from preludium project nr dec-2013/11/n / st3/04062 , and two authors ( ac and pb ) the support from the project nr 2012/05/b / st3/03095 , which are financed by polish national science centre ( ncn ) . calculations were performed on icm supercomputers of university of warsaw ( grant no . g46 - 13 and g16 - 11 ) . we thank dr andrzej witowski for performing fir spectroscopy measurements . ( color on - line ) energy bands of zno with fe in : ( a ) @xmath32 charge state fe@xmath2 , and ( b ) @xmath28 charge state , `` fe@xmath1 '' . red lines denote the band gap of zno . zero energy is set at the vbm . results are obtained with the 144-atom supercell for @xmath0(fe)=0 . ] here we analyze electronic structure of fe in zno calculated assuming small @xmath0(fe ) , and show problems encountered within gga . the band structure and dos of zno doped with fe@xmath2 and fe@xmath1 is presented in fig . [ figa1](a ) and fig . [ figa1](b ) , respectively , for @xmath0(fe)@xmath99 . the impurity @xmath7(fe ) states constitute nearly flat bands , and their energies strongly depend on the fe charge state . fe@xmath2 introduces two spin - down states close to the cbm : the @xmath40 doublet is at 2.4 ev , and the @xmath41 triplet is very close to the cbm . both spin down levels are empty . addition of one electron to the zno : fe@xmath2 system should result in the @xmath28 charge state fe@xmath1 , with 6 electrons on the @xmath7(fe)-induced levels . however , this configuration can not be reached . the electronic structure for @xmath28 calculated allowing for fractional occupancies is shown in fig . [ figa1](b ) . in this case , the fe spin - down states are degenerate with the conduction band . in particular , the @xmath40 doublet is about 0.5 ev above cbm , and @xmath41 is higher by @xmath165 ev . therefore , the electron that should occupy @xmath40 autoionizes to the bottom of the conduction band . a more detailed analysis shows that autoionization is partial only . in fact , the calculated occupancy of @xmath40 is 0.7 , and the integrated occupancy of the conduction states up to the fermi energy @xmath166 is 0.3 . one should notice that since there are no empty states below @xmath166 , @xmath166 must be located within the partially occupied @xmath40 level . this electronic configuration for @xmath28 , with partial occupation of @xmath40 , denoted as `` fe@xmath1 '' , is different from both the fe@xmath2 state , and from fe@xmath1 with one electron on @xmath40 . one should observe , however , that the results for `` fe@xmath1 '' are largely artefacts which follow from the employed method of calculations . more specifically , there are two issues : the increase of eigenenergies with increasing occupation , and applicability of partial occupations approach . first , the calculated increase of the kohn - sham energy of @xmath40 with its increasing occupation is related to the spurious self - interaction obtained in the approximate versions of dft . within the exact dft , a kohn - sham energy level is independent of its fractional occupation , while an integer change of occupation induces its jump . indeed , a change of the charge state of a defect due to a capture ( emission ) of an electron implies an increased ( decreased ) coulomb repulsion between electrons situated at the relevant orbitals ( @xmath7 orbitals of a transition metal ion , broken bonds of vacancy neighbors , etc . ) . this is reflected in the increasing energies of transition levels for consecutive charge states , together with the increasing kohn - sham eigenenergies of defect - induced states . this is a genuine physical effect , which can be described by the so - called effective @xmath0 parameter , and calculated within dft @xcite for integer occupancies . second , the final result for `` fe@xmath1 '' with one electron shared between @xmath7(fe ) and cbm should not be accepted because such a partition is not physical in the present case of an isolated defect center in an insulator . in other words , the @xmath167 configuration of fe@xmath1 is not stable since @xmath40 occupied with one electron is above the ( empty ) cbm , @xcite similar to a resonant donor . however , a complete autoionization to the @xmath32 charge state can not occur because the empty @xmath40 is below the ( occupied ) cbm . neither case corresponds to the ground state , and the convergent self - consistent solution is only obtained for fractional occupation of the @xmath40 level . these problems with `` fe@xmath1 '' in zno are analogous to the difficulties encountered for isolated transition metal atoms such as fe . @xcite namely , the self - consistent solution for @xmath168 configuration leads to an empty @xmath9 state below the partially filled @xmath7 states , while in the @xmath169 configuration there are empty @xmath7 states below the @xmath9 level . the energy minimum is obtained for fractional occupations of both @xmath7 and @xmath9 , but , as indicated by janak , @xcite `` in a strict interpretation of the kohn - sham theory , this means that there is no charge density for this problem which minimizes the total energy and which is generated from a fermi distribution . '' the electronic structure of the neutral fe in zno calculated in refs is characterized by @xmath166 in the conduction band , and partially filled @xmath40 , which closely resembles the `` fe@xmath1 '' configuration . clearly , without a detailed inspection of the results it is not possible to assess the genuine resonant ( or the `` pseudo - resonant '' ) character of fe found in those works but , given the similarity to `` fe@xmath1 '' , one may expect that the above discussion may be relevant in some of those cases .

n66/ngc346 is the largest and brightest region in the small magellanic cloud and contains at least one known supernova remnant snr0057 . optical emission from the remnant is overwhelmed by the bright photoionized emission from the nebula , but the remnant has been detected by way of far ultraviolet absorption lines . here we present data from the far ultraviolet spectroscopic explorer ( _ fuse _ ) satellite showing strong and emission from a position at the edge of snr0057 . we also present high - resolution , long - slit h spectra across n66 showing high- and low - velocity emission corresponding closely to the x - ray boundaries of the supernova remnant . we use these fuv and optical data to determine the physical parameters of the shock and interaction geometry with n66 . we find that ionizing photons from the many massive cluster stars nearby likely affect the ionization balance in the post - shock gas , hindering the production of lower - ionization and neutral species . we discuss the importance and inherent difficulty of searching for supernova remnants in or near bright regions and suggest that the far ultraviolet provides a viable means to discover and study such remnants .

astro-ph0212354

the giant region n66 is the largest , brightest star forming region in the small magellanic cloud ( smc ) . the populous star cluster ngc346 embedded in the nebula is home to several dozen early type stars including a dozen of spectral types o7 and earlier . two arcminutes to the east lies the massive , luminous wn+ob system hd5980 which underwent a luminous blue variable ( lbv ) outburst in 1994 @xcite . these stars are the main source of ionization in the nebula . n66 also contains at least one known supernova remnant ( snr ) , snr0057@xmath07226 . _ einstein _ x - ray observations showed a bright source in n66 tentatively identified as a snr @xcite . radio observations by @xcite showed bright , extended , thermal emission at 408 , 843 and 5000 mhz corresponding to the region , but no clear sign of the snr . optical observations also showed only the extremely bright nebular emission but no filamentary structure associated with the snr . however , @xcite discovered faint , high - velocity h@xmath1 emission in a long - slit spectrum across n66 and suggested the existence of a snr . subsequently , @xcite subtracted smoothed h@xmath1 data from 843 mhz radio continuum observations and revealed a remnant @xmath23.2 ( @xmath255 pc at an smc distance of 59 kpc ) in diameter and confirmed the remnant snr0057@xmath07226 . x - ray spectroscopy of snr0057@xmath07226 with asca @xcite showed that the spectrum can be modeled as a hot thermal plasma . high - resolution x - ray observations of n66 with the _ chandra x - ray observatory _ @xcite show a 130@xmath3100 region of extended , center - bright , thermal emission corresponding to the snr . these high - resolution data show snr0057@xmath07226 to have relatively uniform surface brightness and no clear temperature gradient from center to rim . difference imaging techniques applied to recent radio observations @xcite at 1.42 , 2.37 , and 4.80 ghz ( 20 , 13 and 6 cm ) show snr0057@xmath07226 to be a non - thermal , limb - brightened shell with a spectral index of approximately @[email protected] ( s@xmath4 ) . the x - ray and radio emitting regions are similar in extent at a resolution of @xmath220 . hence , the @xcite estimate of angular size is probably too large , and the x - ray / radio snr is roughly 36 @xmath3 28 pc in extent . thus , snr0057@xmath07226 may be a member of the `` mixed morphology '' class of snrs @xcite , with a radio shell and filled - center x - ray emission , although the presence of hd 5980 along the sight line confuses the situation . @xcite resolve x - ray emission from hd 5980 itself for the first time , and note similarities with the galactic object @xmath5 carinae , which has diffuse surrounding x - ray emission @xcite . @xcite mention a possible association of hd 5980 and snr0057@xmath07226 , but fuv absorption studies have clearly determined that hd 5980 is behind the snr . fuv observations toward hd5980 show the expected galactic and smc absorption at @xmath60 km s@xmath7 and @xmath6@xmath8150 km s@xmath7 , respectively , as well as high - velocity absorption indicating the receding shell of an intervening snr . absorption at @xmath8300 km s@xmath7 was found in iue spectra by @xcite ; this was later confirmed by @xcite who suggested an intervening snr . @xcite detected absorption systems at @xmath8300 and @xmath8330 km s@xmath7 in , and several other ions in high - resolution _ hst _ stis spectra . they also see very weak components at @xmath820 and @xmath850 km s@xmath7 , possibly consistent with the approaching side of the remnant if indeed these features are real . @xcite also see absorption at @xmath2@xmath8300 km s@xmath7 in and in _ fuse _ observations of hd5980 and conclude that the star must lie behind the receding shell of snr0057@xmath07226 . @xcite also find that a _ fuse _ spectrum of sk80 ( av232 ) , an o7ia star @xmath21 southeast of hd5980 but also seen in projection within the x - ray emission , shows none of the high - velocity absorption seen toward hd5980 . since sk80 lies closer to the x - ray edge of the remnant , we might expect a lower line - of - sight velocity for snr gas in this sight line . however , the absorption profile near the smc systemic velocity ( @xmath2150 km s@xmath7 ) is indistinguishable from that of hd5980 . thus , sk80 must either lie within or in front of the remnant . far - ultraviolet ( fuv ) observations are useful in the study of snrs . several strong emission lines including @xmath91548,1550 , @xmath91238,1242 , and especially @xmath91032,1038 are good shock diagnostics and are produced in regions where the temperature is lower than typical x - ray producing regions and higher than regions of bright optical emission @xcite . emission , in particular , arises in gas at temperatures near @xmath10 k a condition almost never reached by photoionization and is thus a sensitive tracer of shock - heated material . in this paper we present a _ fuse _ observation of emission at the x - ray edge of snr0057@xmath07226 that shows strong emission in @xmath91032,1038 and @xmath11977 . we also present new longslit echelle observations showing high- and low - velocity material in h@xmath1 across the face of snr0057@xmath07226 . the observations are described in 2 . in 3 , we discuss the snr kinematics as revealed by the optical and fuv data . in 4 , we compare the , , and h@xmath1 measurements with the predictions of shock models to derive shock velocity , preshock density and ram pressure at one location in the snr . using the same models with the fuv absorption data of @xcite and @xcite , we derive similar values for material on the sight line toward hd5980 . we propose a physical picture for snr0057@xmath07226 in relation to n66 . finally we discuss the implications these observations have for the detection of snrs in bright nebulae and ob associations . a summary is presented in 5 . figure 1 shows images of snr0057@xmath07226 and its surroundings . the left panel shows an h@xmath1 + [ ] emission - line image taken with the curtis schmidt telescope at ctio in 1999 , december . these data are described more thoroughly by @xcite . the right panel shows a recent x - ray image of ngc346 from the _ chandra x - ray observatory _ , obtained courtesy of @xcite . these data show snr0057@xmath07226 clearly as a bright , extended source . unfortunately , the acis detector gaps partially obscure the southwestern edge of the snr and some of the core of ngc346 itself . six stars within n66/ngc346 are part of the _ fuse _ hot ism team program and are marked with small boxes in figure 1 . four of these targets are within the crowded core of the cluster itself . the cluster targets were observed with the 4@xmath320 mdrs aperture . the remaining two ( hd5980 and sk80 ) lie about two arcminutes to the east of the cluster in projection toward snr0057@xmath07226 and were observed with the 30@xmath330 lwrs aperture . _ fuse _ has three spectral apertures @xcite ; the lwrs and the mdrs apertures are centered 209 from each other . thus , while the mdrs aperture was being used to observe stars in ngc346 , the lwrs aperture fell serendipitously on other parts of the sky . in one case ( p20305 , henceforth position 1 ) , the lwrs aperture fell 75 east of hd5980 on the eastern x - ray edge of snr0057@xmath07226 . this spectrum shows bright emission lines of @xmath11977.020 and @xmath91031.926 , 1036.617 . a very faint continuum underlies the bright emission lines at a flux level of @xmath12 erg s@xmath7 @xmath13 @xmath7 . in another case ( p20302 , henceforth position 2 ) the lwrs aperture fell in a region north of hd5980 outside the boundaries of the snr s x - ray emission . this sight line shows a weak continuum spectrum similar to that at position 1 . we use this sight line to characterize the fuv background of n66 . the _ fuse _ observations are summarized in table 1 . figure 1 shows the positions of the _ fuse _ lwrs apertures for positions 1 and 2 as large boxes . the third _ fuse _ aperture , the 125@xmath320 hirs aperture , lies halfway between the other apertures . in several observations of stars in ngc346 the hirs aperture fell on x - ray bright regions of snr0057@xmath07226 . however , no flux was detected in these spectra . since the hirs aperture has less than 3% of the collecting area of lwrs for detecting diffuse emission , this negative result is consistent . _ fuse _ has four independent spectral channels ( in _ fuse _ nomenclature , lif1 , lif2 , sic1 , and sic2 ) , and because of thermal effects onboard the spacecraft , relative channel alignments shift in a complicated manner @xcite . this can cause significant complications for diffuse source observations ( cf . * ? ? ? the spacecraft guides on the lif1 channel , which also has the highest fuv throughput near 1035 . the other three channels tend to drift as much as 8 over the course of a few orbits with respect to lif1 if no realignments are made . for this reason , we use only lif1 data for stellar targets whenever possible . lif1 does not cover wavelengths below 987 . thus we use the lower - throughput sic2 channel to observe @xmath11977 . all four channels cover the 1035 region containing , so a direct comparison of and is possible . the six stellar datasets were reduced using the methods described by @xcite and references therein . spectral resolution in these data is 1520 km s@xmath7 . the lwrs data were reduced using calfuse v2.0.5 and observations were combined by channel . the region around 1032 was examined in all four channels to determine the relative shifts that exist in the wavelength solutions relative to lif1 . we increased the signal - to - noise ratio by coadding the two lif channels to create a composite spectrum . this does not significantly degrade the spectral resolution , which for an aperture - filling extended source is @xmath20.34 ( 106 km s@xmath7 at 1032 ) . flux is present in the two sic channels as well and closely resembles that in the lif channels , but with lower s / n . the two sic channels were coadded for the emission . figure 2 shows these data in comparison to position 2 which sampled a sight line outside the snr . seven longslit echelle spectra of the n66 region were obtained on the 4-m blanco telescope at the cerro tololo interamerican observatory ( ctio ) during 2000 december and 2001 december . these observations are summarized in table 2 and the slit positions are shown in figure 1 as dashed lines . for both observing runs , a 79 line mm@xmath7 echelle grating and long - focus red camera were used . the decker was opened to its maximum extent giving a useful slit length on the sky of 35 . the slit width was set to 165 . the cross - dispersing grating was replaced with a flat mirror and an order - separating filter ( @xmath14=6580 , @xmath15=150 ) was used to isolate the spectral region around the h@xmath1 line . [ ] @xmath96548,6583 also falls within the spectral range sampled . data reduction was performed with iraf using the quadproc and longslit packages . trim , bias and dark current calibration were performed first . the images were next processed to remove the tilt of the ccd axis with respect to the slit . this rendered the stellar spectra horizontal along a row . thorium - argon lamp spectral lines were traced and fit to correct for curvature in the cross - dispersion direction and provide a dispersion solution . final wavelength calibration was performed using the terrestrial airglow lines present in the observations themselves . the result is a rectified spectrum with dispersion along the x - axis and the spatial dimension along the y - axis . the calibrated data cover @xmath235 of sky at a spatial resolution of @xmath21 ( 0267 pix@xmath7 ) and spectral resolution of @xmath210 km s@xmath7 ( 0.082 pix@xmath7 = 3.7 km s@xmath7 pix@xmath7 ) . we present three of the spectra in two - dimensional format in figure 3 , where the wavelength scale has been converted to velocity . the lwrs aperture of position 1 lies 75 east of hd5980 and coincides with the edge of bright , soft x - ray emission ( figure 1 ) . strong emission is seen in both lines of the doublet as well as @xmath11977.02 ( figure 2 ) . we detect no other lines in the _ fuse _ wavelength range ( 9051187 ) besides airglow . in addition to the emission lines , a faint continuum is seen in the position 1 spectrum ( f@xmath16 erg s@xmath7 @xmath13 @xmath7 ) . the spectrum is consistent with an early - type star and shows many of the same ism absorption features seen in other _ fuse _ spectra . the unidentified star in the lwrs aperture visible in the optical image ( figure 1 ) is not the source of the fuv continuum . it does not appear in fuv uit ( @xmath17 ) images of the region @xcite and is of spectral type a or later . in contrast to position 1 , there are no and emissions detected at position 2 . this aperture fell an arcminute northeast of the edge of the snr in a field empty of bright stars , uv - bright or otherwise . however , a similar stellar continuum can be seen in this spectrum . the faint continuum observed at both positions is due to dust - scattered starlight from the many fuv - bright sources nearby . a similar effect at longer uv wavelengths was seen in iue spectra of other regions in the smc @xcite . any emission from the snr will suffer absorption from the intervening smc and galactic ism . fortunately , there is little molecular hydrogen absorption in the vicinity of 1032 ; the only ism absorption comes from smc and galactic . we use the continuum - normalized absorption profile of sk80 ( * ? ? ? * their figure 80 ) as our absorption model . a gaussian profile divided by the absorption model was convolved with a 106 km s@xmath7 aperture function . the gaussian parameters were varied and the resulting models were fit to the observed @xmath111032 line profile . the best fit parameters were : @xmath18=228 km s@xmath7 , fwhm=159 km s@xmath7 , and peak flux=2.23@xmath19 erg s@xmath7 @xmath13 @xmath7 . the total flux of the intrinsic gaussian function is 1.27@xmath19 erg s@xmath7 @xmath13 compared with an observed flux of 8.68@xmath20 erg s@xmath7 @xmath13 . both the fit to the data and the intrinsic gaussian profile are shown in the top panel of figure 4 . the @xmath111032 line should be twice the intensity of the @xmath111038 line in the absence of resonance scattering and self - absorption . however , this ratio can vary as a function of velocity within the line profile . the observed @xmath111032/@xmath111038 ratio is close to two on the blue - shifted side of the line but becomes significantly higher on the red - shifted wing . since smc and galactic absorption affect both lines in a proportional manner , we conclude that resonance scattering and self - absorption is minimal but that significant absorption from molecular hydrogen5 - 0(p1)h@xmath21 @xmath111038.156 and 5 - 0(r2)h@xmath21 @xmath111038.690affects the weaker line . we can characterize the @xmath111038 emission using the intrinsic emission profile determined from the @xmath111032 line . in this case , the absorption model was allowed to vary ; a simulated molecular absorption spectrum was calculated with the _ fuse _ spectral simulator code fsim and the sk80 @xmath111032 absorption profile ( scaled by a square root ) was multiplied in . ( since @xmath22 in the optically thin limit , @xmath23 . ) the resulting model was multiplied by the intrinsic gaussian profile determined above and convolved with the aperture function to simulate the observed @xmath111038 line . we first fit the @xmath111038 emission using two molecular hydrogen components : a galactic component @xcite ( v=0 km s@xmath7 , n(h@xmath21)=@xmath24 , b=5 km s@xmath7 , t=150 k ) and an smc component ( v=115 km s@xmath7 , n(h@xmath21)=@xmath25 , b=15 km s@xmath7 , t=400 k ) slightly modified from @xcite . however , this did not produce enough absorption on the red side of the emission . a third component , representing swept - up material from the snr at v@xmath2155 km s@xmath7 improves the fit ( figure 4 , middle panel ) , but requires an unphysically large turbulent velocity for a good fit . there could be many components at a range of velocities in the turbulent , post - shock gas which might account for the enhanced absorption . no such high - velocity components are seen by @xcite . the @xmath11977 emission is strongly affected by resonance scattering and self - absorption . the _ fuse _ spectrum shows that the galactic and smc components are saturated @xcite . therefore we do not attempt to fit the observed profile . the measured flux , 6.3@xmath20 erg s@xmath7 @xmath13 , is a lower limit to the intrinsic flux . overlaying the derived @xmath111032 profile ( figure 4 , bottom panel ) provides a first order assessment of how this line might be affected if its intrinsic line width is similar to . apart from and ( and geocoronal airglow ) , there are no other lines detected . in a spectrum of the cygnus loop , a typical middle - aged snr , the next brightest lines to and were [ ] @xmath111146 and @xmath91063,1074 at @xmath25% of the observed flux @xcite . such weak emission would not be detected in the spectrum of snr0057@xmath07226 , especially considering the lower abundances of these elements in the smc . the seven echelle observations cut across much of the n66 nebulosity in figure 1 including the core of ngc346 and the outlying regions . three spectra cross significant portions of the snr and show faint high- and low - velocity emission . the data show both h@xmath1 and [ ] emission at a velocity resolution of @xmath215 km s@xmath7 . we find the nebula itself to be surprisingly quiescent on the large scale given the violence of strong stellar winds and ongoing star formation . the bright nebular emission can be fit with a single - component gaussian with v@xmath26=157.7 km s@xmath7 , fwhm=34.7 km s@xmath7 . assuming a temperature of @xmath27 k in the photoionized gas , we derive a bulk turbulence of 23.8 km s@xmath7(b=14.3 km s@xmath7 ) . this degree of turbulence is typical of magellanic regions but much lower than that seen in giant regions such as 30 doradus and nebulae in m101 and m33 @xcite . the thermal fwhm of [ ] is narrower than that of h@xmath1 by the square root of the ratio of their ionic masses ( i.e. by @xmath28=3.7 ) . thus [ ] emission is a more sensitive probe of velocity structure than h@xmath1 . however , these lines are usually much weaker than h@xmath1 , particularly in the low - nitrogen smc ism , and are often completely undetectable in smc echelle spectra . four of the echelle spectra ( both hd5980 pointings , ngc346-wb3 + 4 and ngc346-wb6 ) have usable [ ] data . these show @xmath29 km s@xmath7 and a @xmath30=22.8 km s@xmath7 . assuming a temperature of 10@xmath31 k , we derive a bulk turbulence of 17.5 km s@xmath7 ( b=10.5 km s@xmath7 ) , also typical for ordinary magellanic regions . the two spectra which cut through ngc346 show an expansion feature bounded by the bright h@xmath1 rim visible in figure 1 surrounding the cluster core . ( ngc346-wb1 samples the same region as the latter two spectra , but has a high sky brightness masking any [ ] emission . ) the data show a maximum expansion velocity of 14 km s@xmath7 . this expansion is presumably driven by the stellar winds of the stars located within the bubble . the ngc346 expansion feature is also detected in the h@xmath1 data , but only in that a pair of gaussians are needed to fit the broad h@xmath1 profile , rather than a single component . meanwhile , the east - west hd5980 data cut through n66c , a compact region to the west of hd5980 and north east of the cluster core . the [ ] data reveal an expansion feature here with @xmath32 km s@xmath7 . high- and low - velocity h@xmath1 emission appear in the four echelle observations which cross snr0057@xmath07226 ( hd5980 ns and ew , sk80 ns and ew ) . three of these spectra are shown in figure 3 ; the fourth spectrum ( sk80 ew ) is of low signal - to - noise and intersects only the edge of the remnant . the faint , higher velocity emission in these spectra arises in shocks associated with the snr . it is weaker than the photoionized nebular emission by a factor of @xmath270 and has a very low signal to noise ratio . the snr emission is quite patchy : blobs with widths of @xmath210 ( @xmath23 pc ) are seen in the position - velocity diagrams ( figure 3 ) . we find high - velocity features all across the face of snr0057@xmath07226 with higher velocity emission typically toward the center suggesting an expanding shell . the fastest emission is at v@xmath33335 km s@xmath7 about 15 ( 4 pc ) south of hd5980 ( this is also the region of peak x - ray brightness ) . if the systemic velocity of the snr is taken to be the same as the n66 ism , this gives an expansion velocity of @xmath34 km s@xmath7 , consistent with strong production by the shock front . the h@xmath1 emission from the approaching side of the snr is also seen in figure 3 , though it is even fainter than the receding side . in the few cases where the blue - shifted material is discernible from the gaussian wings of the bright nebular emission , we see velocities as low as v@xmath35 km s@xmath7 . as with the receding side of the snr , the material with the largest velocity offsets tends to be closest to the center of the remnant . approximate spectrophotometry was performed on the echelle data using a flux - calibrated h@xmath1 + [ ] image of n66 obtained from @xcite . comparing profiles from the image with wavelength - integrated intensity along the slit gave a conversion factor of @xmath24.5@xmath36 erg s@xmath7 @xmath13 count@xmath7 . we use this value to measure fluxes and surface brightnesses for different patches of snr emission in the spectra ( figure 3 ) . a gaussian profile was fit to the main nebular emission ( v@xmath37158 km s@xmath7 ) over a range @xmath38 along the slit and divided from the data . a background level was then subtracted and the remaining counts within a wavelength ( velocity ) range @xmath39-@xmath40 were summed . the summed counts were converted into a total flux . surface brightness was determined by dividing the total flux by the area covered by the extraction region ( @xmath41 ) . typical observed surface brightnesses were @xmath42 erg s@xmath7 @xmath13 arcsec@xmath43 with a standard deviation of 50% . there is a faint patch of h@xmath1 emission at 215 km s@xmath7 at the position 1 lwrs aperture ( figure 3 , top panel ) . this corresponds reasonably well with the observed emission at 228 km s@xmath7 . we measure the surface brightness of this emission patch to be @xmath44 erg s@xmath7 @xmath13 arcsec@xmath43 over a slit range 5779 east of hd5980 . the detection of emission as well as the kinematics present in the h@xmath1 echelle data imply the presence of a strong shock in n66 associated with snr0057@xmath07226 . the snr shock model was also favored by @xcite for the high velocity absorption towards hd5980 . detection of the front side of snr0057@xmath07226 has not been clearly established because the low velocity components cited by @xcite were detected at such a marginal level . in any event , the near side of the snr is considerably more difficult to detect both in emission and absorption than the back side . optical emission from the region complicates the situation and masks the morphology of the snr . in the following discussion we adopt a working model of the physical association of various components in the n66 region . a roughly spherical snr is located on the near side of n66 . the rear side of the snr is propagating into relatively denser nebula while the near side is propagating through the more rarefied smc ism . hd5980 lies behind the remnant , embedded within n66 . the ngc346 cluster stars lie outside the snr , also embedded within n66 . we determine the shock velocity , preshock hydrogen density and ram pressure of the shock wave using the emission and absorption sight lines toward snr0057@xmath07226 . by examining the surface brightness ratio between and h@xmath1 , we measure the `` completeness '' of the shock . then we discuss the relationship between the snr and its surroundings and the implications of this analysis for observing snrs in other regions . in the n66 reference frame ( v@xmath26=@xmath8158 km s@xmath7 ) , the material at position 1 has a radial velocity of @xmath870 km s@xmath7 while toward hd5980 , @xmath45 km s@xmath7 . from x - ray and radio images , we choose the center of the remnant to be ( j2000 ) @xmath1=00:59:27 , @xmath46=@xmath072:10:15 , about 20 south of hd5980 . this position is also consistent with the location of the highest - velocity h@xmath1 emission ( figure 3 ) . using the projected distances from this center and the two radial velocity measurements , we derive a radius of curvature for the shock front of 80@xmath47 and an expansion velocity for the gas of v@xmath48 km s@xmath7 . this places the observed x - ray edge on the eastern side of the shock front inclined @xmath260 to our line of sight . the expansion velocity of 147 km s@xmath7 is not the shock speed , but rather the bulk velocity of the postshock material which should be between 0.75 and 1.0 times the shock speed . first we caculate shock models for comparison against the observations . the models were calculated using an updated version of the code described by @xcite . the main input parameters are the shock velocity , the preshock density and the elemental abundances . we ran models for shock velocities between 140 and 200 km s@xmath7 spaced by 10 km s@xmath7 . a preshock hydrogen number density of 1 @xmath49 , and smc elemental abundances @xcite were used in all the models . the calculation is followed until the the recombination zone is complete and the gas temperature has reached about 1000 k. we present the intensities of selected lines from the shock models in table 3 . the intensities are relative to h@xmath1=100 and the flux from nearby multiplet lines of an ion have been summed . the line flux produced by a shock increases sharply as a function of shock velocity in the range 150 200 km s@xmath7 . shocks below @xmath2160 km s@xmath7 do not produce much and thus this represents a lower limit to the actual shock velocity in the snr . over this same range of shock velocities , the flux remains nearly constant . the changing ratio of i()/i ( ) is due to the increasing ionization of oxygen at higher velocities . as discussed in 3.1 , the is strongly affected by absorption and scattering , and so it is not possible for us to derive an accurate assessment of the intrinsic ratio of i()/i ( ) in the spectrum of position 1 . however , we note that using the total ( corrected ) flux with the observed flux places an upper limit of f()/f()@xmath503 . ignoring the effect of differential reddening for the moment , this clearly points us toward the lower end of the velocity range where is produced . based on this ratio , the observed expansion velocity and the shock velocity requirements for production , we adopt 160 km s@xmath7 as a representative shock velocity at position 1 and use details from the model at this velocity below . we now compare the observed and calculated surface brightnesses to estimate the preshock density required to produce the observed flux . the model predicts an surface brightness ( in both lines of the doublet ) of i@xmath51 erg s@xmath7 @xmath13 into 2@xmath52 steradians ( for v@xmath53=160 km s@xmath7 , @xmath54=1 @xmath49 , he / h=0.08 ) or i@xmath55 erg s@xmath7 @xmath13 arcsec@xmath43 . in the optically thin limit , intensity will scale with density so we can write @xmath56 where @xmath57 is the reddening correction factor at 1035 , @xmath58 is the surface area of shock ( in arcseconds@xmath59 ) within the aperture and @xmath60 is the preshock density in units of @xmath49 . i@xmath61 is the total flux in both lines which we estimate by multiplying the corrected @xmath111032 flux by 1.5 , thus i@xmath62 erg s@xmath7 @xmath13 . from figure 1 , we estimate that @xmath2@xmath63 of the lwrs aperture is filled with x - ray emission . if the emission follows the same pattern and the shock front is tilted 60 from our line of sight , then the aperture sampled emission from 600 arcseconds@xmath59 of shock front . reddening is a potential source of uncertainty . @xcite found the mean e(b@xmath0v)=0.14 for the ngc346 stellar population . however , since the snr lies in front of the bulk of the cluster , this may overestimate the extinction to the snr . on the other hand , using the spectral type and observed colors of sk80 , one can derive e(b@xmath0v)=0.11 . we adopt this value and assume r@xmath64=3.1 ( typical for the diffuse ism ) and the extinction curve of @xcite ( e(1035@xmath0v)/e(b@xmath0v)=18 ) . this yields a reddening correction c@xmath65=4.5 . solving the above equation , we get n@xmath66 @xmath49 at position 1 , similar to an azimuthally - averaged electron density of 7 @xmath49 at the radius of position 1 derived by @xcite . the dynamical pressure is then @xmath67 dyne @xmath13 . if more reddening is assumed and e(b@xmath0v ) is raised to 0.14 ( c@xmath65=6.8 ) , the changes are minor : n@xmath68 @xmath49 and @xmath69 dyne @xmath13 . less reddening ( e(b@xmath0v)=0.08 , c@xmath65=3.0 ) would yield n@xmath70 @xmath49 , @xmath71 dyne @xmath13 . this pressure is higher than those seen in regions of the vela snr . @xcite derive n@xmath72 @xmath49 and p=3.7@xmath73 dyne @xmath13 for a x - ray bright , nearly face - on shock . @xcite find p=1.6@xmath74 dyne @xmath13 for an edge - on shock . @xcite find p=2 - 4@xmath73 dyne @xmath13 for another locations in vela while @xcite find p@xmath21@xmath73 dyne @xmath13 for a region of diffuse x - ray emission . this variation in pressures within the same remnant suggests local density and pressure enhancements caused by reverse shocks from blastwave - cloud interactions . we have only two widely - separated measurements for snr0057@xmath07226 and can say nothing so specific but the high pressure and density at position 1 as well as the x - ray emission morphology suggest similar shock - cloud interactions . the shock models discussed above predict the total column depths of ions produced in the post - shock flow . @xcite measured the column depths of several highly ionized species in the high - velocity component toward hd5980 ( their table 1 ) and conclude that a snr shock is the most likely production mechanism for these ions in the observed quantities . @xcite presented data on other ions not available to _ fuse_. in figure 5 , we present the observed column depths of these fuv ions and the values predicted by the models with v@xmath75=160 , 180 and 200 km s@xmath7 . the ions are ordered according to the ionization potential needed to produce them . the relative column depths of all the ions down to are well matched by the 160 km s@xmath7 model . the higher velocity models predict too much relative to the other high ionization ions . the column depths scale linearly with preshock density ( all the models were run with n@xmath76 = 1 @xmath49 ) . the offset between the 160 km s@xmath7 model and the observations is @xmath77 dex , so a preshock density of @xmath78 can reproduce the observed column depths for the high ionization lines . the shock velocity and preshock density toward hd5980 are thus in good agreement with our measurements for position 1 . the behavior of the low ionization lines is drastically different . for and , all the models predict column densities that are an order of magnitude higher than the observations ( see figure 5 ) . this dearth of low - ionization material can also be seen by comparing the observed and calculated ratio of and h@xmath1 . the shock models ( which have been calculated to the point of complete recombination ) predict i()/i(h@xmath1)=3.3 for @xmath79=160 km s@xmath7 and smc abundances . the observed ratio at position 1 is @xmath80 where @xmath81 indicates the surface brightness and @xmath82 the aperture filling factor of the relevant emission . we assume the reddening correction @xmath57=4.5 as discussed above ; reddening at h@xmath1 is moderate and we adopt @xmath83=1.24 . the filling factor @xmath82 for each aperture is more uncertain . we assume @xmath84 as discussed above . it is difficult to say where emission does and does not appear at a fine scale in the low s / n echelle data . the h@xmath1 aperture was chosen carefully to encompass only the faint emission in figure 3 , but we estimate that @xmath85 is still @xmath2@xmath86 . these values give i()/i(h@xmath1)@xmath3718 . uncertainties in the filling factors , reddening , and surface brightnesses probably make this ratio uncertain by a factor of 2 , but the ratio is still significantly higher than the predicted ratio of 3.3 . the weakness of h@xmath1 emission compared with at position 1 suggests the presence of an incomplete shock . there are many cases where incomplete shocks are observed in individual filaments or small portions of snrs @xcite ; however , these regions are all at sub - parsec scales . the lwrs aperture at position 1 is sampling a region of snr0057@xmath07226 roughly 8 pc across . furthermore , the observed h@xmath1 surface brightness does not vary strongly between different portions of the snr implying that the conditions at position 1 are typical of the remnant as a whole . in figure 3 , we see weak snr h@xmath1 emission or none at all . this is a large object with moderate shock velocities implying that it is a middle - aged remnant like the cygnus loop or the vela snr ; most of the optical emission from these remnants comes from complete , radiative shocks and these remnants are bright in h@xmath1 . if some portions of snr0057@xmath07226 were recombinationally complete , we would expect to see h@xmath1 emission @xmath210 times brighter than is observed . thus , it appears that there is a global effect preventing the cooler parts of the recombination flow from forming . we argue that the gas at the back end of the post - shock flow is being prevented from recombining by the ionizing flux from the bright stars in n66 . there are eleven stars of spectral type o6.5 or earlier located 2 ( 35 pc , in projection ) to the west of snr0057@xmath07226 . the combined ionizing flux from ngc346 is 4.0@xmath310@xmath87 erg s@xmath7 @xcite or @xmath21.5@xmath310@xmath88 ionizing photons s@xmath7 . hd5980 itself contributes @xmath89 ionizing photons s@xmath7 . using basic ionization / recombination equations from @xcite we find that this photon flux in a medium with density @xmath25 @xmath49 will yield an ionized sphere over 100 pc in radius . indeed , @xcite find that n66 is density bounded and that 45% of the ionizing photons from cluster stars escape to ionize the general smc ism . this zone of ionization could encompass snr0057@xmath07226 and affect the ionization balance in the recombining post - shock gas . the dynamical time scale of the shock is of order a few hundred years . the photoionization time scales are much shorter ( @xmath2100 days ) , while the recombination times are longer ( of order 10,000 yr ) . thus the ionizing flux will suppress h@xmath1 emission . the strength and spectral energy distribution of the local ionizing flux will affect the column densities of the low and moderately ionized species . in theory , much of the discrepancy between the modeled and observed and columns towards hd5980 ( figure 5 ) could be due to iron and silicon being ionized to higher ionization stages . if the best fit model is scaled up to n@xmath90 @xmath49 , the column is underpredicted by about 0.3 dex . this is clearly not enough to explain the 1.9 dex discrepancy in the column , but the stellar spectrum extends out to @xmath250 ev @xcite and the local flux could be producing . however , no such discrepancies are seen in the and columns . a more likely scenario is that the sight line to hd5980 is passing through a small region where the swept up column is lower than average for snr0057@xmath07226 . while the zone is complete , the zone would be incomplete regardless of ionizing radiation . a self - consistent shock model calculation that includes a photoionizing field is necessary to predict the column densities and the h@xmath1 flux accurately . while such modeling is beyond the scope of this paper , we conclude that the local ionizing field must have a significant effect on the observed characteristics of snr0057@xmath07226 as a whole . the emission and absorption data discussed here lead to a schematic structure for the n66snr region shown in figure 6 . snr0057@xmath07226 lies on the near side of n66 and is encountering the denser nebular material on its back side . position 1 is located at the edge of the x - ray emission ( figure 1 ) . the observed radial velocity of and h@xmath1 emission at position 1 is @xmath2225 km s@xmath7 . this is lower than the observed velocity of absorption ( @xmath8300 km s@xmath7 ) toward hd5980 which lies projected closer to the center of the remnant . also it is @xmath275 km s@xmath7 higher than the systemic velocity of n66 ( @xmath8158 km s@xmath7 ) , which we would expect for material viewed tangentially at the edge of the expanding remnant . either the systemic velocity of the snr and n66 differ by @xmath270 km s@xmath7 , or the observed x - ray edge ( a portion of which is position 1 ) does not represent the actual limb of the snr shock . the latter is assumed in figure 6 . we propose that the x - ray emission arises over a spherical cap where the shock has encountered and heated denser material . the remnant is not yet even half - way submerged in the n66 material , and thus the observed radius of x - ray emission is smaller than the actual blast wave radius . the observed x - ray edge at position 1 is inclined to our line of sight by @xmath260 . hd5980 and the other uv - bright stars are located within n66 and have ionized a region of surrounding material . the combined stellar winds and thermal pressure have formed a slowly expanding envelope around the cluster core . since high - velocity absorption is not seen in the spectrum of sk80 @xcite , it must lie within or in front of the snr . hd5980 lies behind the remnant and affects the ionization balance in the swept - up gas , preventing the formation of , , , and other low - excitation species . the bright nebular emission seen surrounding the uv bright stars of n66 arises in dense material behind an ionized layer . the approaching snr material is apparent in h@xmath1 in figure 3 , but has not been conclusively identified in fuv spectra . if the far side of the remnant is seen near @xmath2300 km s@xmath7 and if the snr is centered at the smc systemic velocity of @xmath2155 km s@xmath7 , we expect the nearside absorption to be seen near v=10 km s@xmath7 . this region is essentially masked by galactic absorption making it difficult to detect the snr . @xcite compared the spectra of sk80 and hd5980 near zero velocity and found no difference . however , both of these sight lines are toward the remnant and may contain nearly identical foreground columns . the four _ fuse _ targets in ngc346 lie less than 2 from hd5980 and are outside the snr sight line . a comparison of the averaged , normalized profiles from these four targets and the two sight lines projected within the snr is shown in figure 7 . we find a small but systematic excess in absorption near v=0 km s@xmath7 in the two remnant profiles . the difference represents an extra column density of about @xmath91 @xmath13 or about @xmath92 that of the backside value reported by @xcite . we also find that the velocity centroid of the hd5980/sk80 zero velocity absorption is shifted by about @xmath810 km s@xmath7 with respect to the ngc346 galactic absorption . this is not too different from the velocity found by @xcite for and absorption features that they attribute to front - side snr material . recent work by @xcite on galactic halo absorption has shown significant variations in column over as little as 3 . the variation in n ( ) between the four ngc346 targets ( mean separation 19 ) is 9% ( the maximum variation between two measurements is 12% ) while the variation between the two snr sight lines ( separation @xmath931 ) is only 8% , comparable to the uncertainties due to continuum placement , noise , and limits of integration . however , the cluster and snr stars show a relative difference of 53% in the zero - velocity column over @xmath22 . because of the potential galactic halo variation on small scales , the matter of frontside absorption remains inconclusive . however , it is plausible we are seeing front side absorption near v=@xmath810 km s@xmath7 in the hd5980/sk80 data . if true , then sk80 would be inside snr0057@xmath07226 , as shown in figure 6 . even if it is not , we can place an upper limit of n()@xmath94 @xmath13 for the frontside column density , much less than is observed from the back side of snr0057@xmath07226 . in a recent paper , @xcite model the collision of a snr shock with a wind - blown bubble around an evolved star and suggest that the far side of snr0057@xmath07226 is interacting with the stellar wind of hd5980 . the x - ray image of the system @xcite does , in fact , show some resemblance to the models and there is reasonable agreement between predicted and observed x - ray fluxes . our observations allow us to further test this wind - snr model . the velzquez et al . wind - blown bubble has a radius of @xmath210 pc while our position 1 ( at the rim of the snr ) is at least twice that distance ( in projection ) from hd5980 ; thus , position 1 is sampling the ism , not stellar wind . in 4.1 we derived a preshock density of @xmath26 @xmath49 for the material at both position 1 and in the shock toward hd5980 , at least two orders of magnitude higher than the interior of the modeled wind bubbles ( see figures 1 and 2 of * ? ? ? * ) . furthermore , the @xmath8300 km s@xmath7 component exhibits an ionization structure expected for a single snr shock @xcite . if this component arose in a turbulent region between snr and stellar wind shock fronts , as suggested by @xcite , we might not expect to see the kinematics structure observed by @xcite . clearly , higher resolution spectral observations and detailed shock models are required . core - collapse supernovae arise from massive stars often formed in clusters and associations the same massive stars which photoionize their surrounding gas and create bright regions . massive sn progenitors can not migrate far from their birth places in their short lifetimes and thus snrs associated with h ii regions should occur frequently . to various degrees , this is not seen in optical snr surveys of nearby galaxies . the primary optical method for detecting snrs , especially in galaxies beyond the magellanic clouds , is to compare h@xmath1 to [ ] emission , and pick out objects with [ ] /h@xmath1@xmath95 0.4 ( cf . * ? ? ? * ; * ? ? ? * and references therein ) . in recent years , as the sensitivity and resolution of radio and x - ray observations has become competitive with optical ground - based observations , independent searches at these wavelengths have been undertaken and compared against the optical snr lists @xcite . these multiwavelength surveys and comparisons provide a more complete survey method , but in general do not find many of the same snrs . discussions of selection effects usually point toward `` confusion '' as being responsible for incompleteness of optical snr surveys in crowded and complex regions of emission . clearly this is plausible at some level , but the denser conditions in regions might also be expected to cause brighter optical ( radiative ) snrs since the shock emissivity increases as the density squared . scanning through the optical snr catalogs of nearby galaxies ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * to name a few ) , one can find a number of stunning counter examples to the `` confusion '' hypothesis . there are often bright optical snrs found in confused regions and adjacent to bright region emission . the mechanism we propose in 4.2 seems to be a more plausible explanation for why the optical snr searches miss some snrs and not others in confused regions . if a supernova occurs close enough to ionizing stars , the potential bright optical snr emission in h@xmath1 and is prevented from forming and no optical snr can be detected , especially against bright nebular emission . our investigation of snr0057@xmath07226 , however , demonstrates that bright fuv emission lines are still present in such a situation and may provide an alternate way of not only detecting but studying such snrs in detail . detecting snrs with higher ions circumvents the problems of truncated recombination and bright background emission . stellar photoionization will not prevent the production of high ions nor will they be contaminated significantly by surrounding nebular emission . our observations have shown that in a bright region , the detection of localized emission from high - ionization species such as can provide a clear signature of snr emission . position 1 lies on the edge of the snr and shows strong emission in and . position 2 lies outside the remnant but still within bright optical emission ; yet no line emission is present . in conjunction with radio and x - ray data , such a detection is conclusive . we note however that the use of the fuv is restricted to lines of sight where extinction is low . nearby extragalactic targets such as the magellanic clouds and local group galaxies will be the easiest to investigate with this technique . we see strong emission in and at a position on the eastern x - ray edge of snr0057@xmath07226 . absorption in these same ions is seen at @xmath2300 km s@xmath7 toward the bright lbv system hd5980 which lies behind the remnant . another position away from the remnant but within the bright nebular extent of n66 shows no fuv line emission . the remnant lies on the near side of n66 and the is formed in shocks as the blast wave encounters the denser material of the nebula . seven longslit echelle spectra in h@xmath1 and [ ] show that n66 has turbulence similar to other magellanic regions . two slowly expanding features are seen : the compact region n66c and a bright shell around the core of ngc346 . three of the echelle spectra show faint patches of h@xmath1 emission at high and low velocities which appear as a rough expansion feature corresponding spatially and kinematically to snr0057@xmath07226 . observed and flux compared with shock models allows us to estimate the shock velocity and determine the ambient density and ram pressure at position 1 . we use observed column depths from @xcite and @xcite to derive the shock velocity and density of the snr shock toward hd5980 . these values are in good agreement with each other and with other values in the literature . weak h@xmath1 emission as well as low column densities of low - ionization species such as show that the ionization balance in the post - shock gas is affected by ionizing stellar photons . the ionizing flux from the many hot stars in ngc346 is sufficient to keep swept - up material out of ionization equilibriuim . this highlights the importance and inherent difficulty of looking for snrs in bright regions . direct optical identification is complicated by the bright nebular emission . fuv emission and absorption , along with x - ray and radio observations , are good ways to circumvent these difficulties . the authors wish to acknowledge useful discussion with charles hoopes , miroslav filipovi , lister staveley - smith and alex fullerton . mike corcoran provided the x - ray images . chris smith supplied the interference filters used in the schmidt imagery . bryan dunne helped with the echelle observations . robert kennicutt supplied the flux - calibrated image of n66 . john raymond made available the shock code used in calculating the models . we would also like to recognize ctio for providing the excellent observing facilities and travel support . this work contains data obtained for the guaranteed time team by the nasa - cnes - csa _ fuse _ mission operated by the johns hopkins university . financial support has been provided by nasa contract nas5 - 32985 . n66snr barba , r. h. , niemela , v. s. , baume , g. , & vazquez , r. a. 1995 , , 446 , l23 blair , w. p. , et al . 1991 , , 379 , l33 blair , w. p. , & long , k. s. 1997 , , 108 , 261 blair , w. p. , raymond , j. c. , danziger , j. , matteucci , f. 1989 , , 338 , 812 blair , w. p. , sankrit , r. , & tulin , s. m. 2002 , , 140 , 367 chu , y - h . , & kennicutt , r. c. 1988 , , 95 , 1111 chu , y - h . , & kennicutt , r. c. 1994 , , 216 , 253 danforth , c. w. , blair , w. p. , & raymond , j. c. , 2001 , , 122 , 938 danforth , c. w. , howk , j. c. , fullerton , a. w. , blair , w. p. , & sembach , k. r. 2002 , , 139 , 81 de boer , k. s. , & savage , b. d. 1980 , , 238 , 86 fesen , r. a. , blair , w. p. , & kirshner , r. p. 1982 , , 262 , 171 filipovi , m. , staveley - smith , l. , ried , w. , & danforth , c. w. , 2003 , in prep . fitzpatrick , e. l. & savage , b. d. 1983 , , 267 , 93 gordon , s. m. , duric , n. , kirshner , r. p. , goss , w. m. , & viallefond , f. 1999 , , 120 , 247 gordon , s. m. , kirshner , r. p. , long , k. s. , blair , w. p. , duric , n. , & smith , r. c. 1998 , , 117 , 89 hoopes , c. g. , sembach , k. r. , howk , j. c. , & blair , w. p. 2001 , , 558 , l35 hutchings , j. b. , & giasson , j. 2001 , , 113 , 1205 howk , j. c. , savage , b. , sembach , k. r. , & hoopes , c. g. 2002 , , 572 , 264 inoue , h. , koyama , k. , & tanaka , y. 1983 , iaus , 101 , 535 jenkins , e. b. , & wallerstein , g. 1995 , , 440 , 227 kahn , s. m. , gorenstein , p. , harnden , f. r. , & seward , f. d. 1985 , , 299 , 821 kennicutt , r. c. 1984 , , 287 , 116 kennicutt , r. c. , & hodge , p. 1986 , , 306 , 130 koenigsberger , g. , guinan , e. , auer , l. , & georgiev , l. 1995 , , 452 , l107 koenigsberger , g. , et al . 2001 , , 121 , 267 lacey , c. k. , duric , n. , & goss , w. m. 1997 , , 108 , 417 long , k. s. , blair , w. p. , kirshner , r. p. , & winkler , p. f. 1990 , , 72 , 61 massey , p. , parker , j. w. , & garmany , c. d. 1989 , , 98 , 1305 matonick , d. m. , & fesen , r.a . 1997 , , 112 , 49 mills , b. y. , little , a. g. , durdin , j. m. , & kesteven , m. j. 1982 , , 200 , 1007 moos , h. w. , et al . 2000 , , 538 , l1 naz , y. , et al . 2002 , , in press osterbrock , d. e. 1989 , `` astrophysics of gaseous nebulae and active galactic nuclei '' , university science books , mill valley , ca pannuti , t. g. , duric , n. , lacey , c. k. , goss , w. m. , hoopes , c. g. , walterbos , r. a. m. , & magnor , m. a. 2000 , , 544 , 780 pannuti , t. g. , duric , n. , lacey , c. k. , ferguson , a. n. m. , magnor , m. a. , & mendelowitz , c. 2002 , , 565 , 966 parker , j. w. , et al . 1998 , , 116 , 180 raymond , j. c. 1979 , , 39 , 1 raymond , j. c. , blair , w. p. , long , k. s. , vancura , o. , edgar , r. j. , morse , j. , hartigan , p. , & sanders , w. t. 1997 , , 482 , 881 rela~ no , m. , peimbert , m. , & beckman , j. 2002 , , 564 , 704 rho , j. , & petre , r. 1998 , , 503 , l167 russell , s. c. , & dopita , m. a. 1990 , , 74 , 93 sahnow , d. j. , et al . 2000 , , 538 , l7 sankrit , r. , shelton , r. l. , blair , w. p. , sembach , k. r. , & jenkins , e. b. 2001 , , 549 , 416 seward , f. d. , & mitchell , m. 1981 , , 243 , 736 shull , j. m. , et al . 2000 , , 538 , l73 smith , m. g. , & weedman , d. w. 1970 , , 161 , 33 smith , m. g. , & weedman , d. w. 1971 , , 169 , 271 smith , m. g. , & weedman , d. w. 1973 , , 179 , 461 sutherland , r. s. , & dopita , m. a. 1993 , , 88 , 253 tumlinson , j. , et al . 2002 , , 566 , 857 velzquez , p. f. , koenigsberger , g. , & raga , a. c. 2003 , , 584 , in press wang , q. , & wu , x. 1992 , , 78 , 391 ye , t. , turtle , a. j. , & kennicutt , r. c. 1991 , , 249 , 722 yokogawa , j. , imanishi , k. , tsujimoto , m. , nishiuchi , m. , koyama , k. , nagase , f. , & corbet , r. h. d. 2000 , , 128 , 491 llllllcrl ngc346-wb6 & 00 58 57.74 & @xmath072 10 33.6 & o4v((f ) ) & p20305&mdrs & 5&10992&2001-sep-25 + ngc346-wb4 & 00 59 00.39 & @xmath072 10 37.9 & o5 - 6v & p20304&mdrs & 6&11853&2001-aug-25 + ngc346-wb3 & 00 59 01.09 & @xmath072 10 28.2 & o2iii(f * ) & p20303&mdrs & 3 & 8482&2001-aug-25 + ngc346-wb1 & 00 59 04.81 & @xmath072 10 24.8 & o4iii(n)(f ) & p20302&mdrs & 3 & 4625&2001-aug-25 + hd5980 & 00 59 26.55 & @xmath072 09 53.8 & wn var & p10301 & lwrs & 4 & 5734 & 2000-jul-02 + sk80 & 00 59 31.95 & @xmath072 10 45.8 & o7 iaf+ & p10302 & lwrs & 4&11699 & 2000-jul-02 + n66-pos1 & 00 59 42 & @xmath072 09 49 & & p20305 & lwrs & 5&10992 & 2001-sep-25 + n66-pos2 & 00 59 37 & @xmath072 07 57 & & p20302 & lwrs & 3 & 4625 & 2001-aug-25 + lllrrc hd5980 ns & 00 59 26.55 & -72 09 53.8 & 0 & 1200 & 2001-dec-24 + hd5980 ew & 00 59 26.55 & -72 09 53.8 & 90 & 1200 & 2001-dec-25 + sk80 ns & 00 59 31.95 & -72 10 45.8 & 0 & 1200 & 2001-dec-26 + sk80 ew & 00 59 31.95 & -72 10 45.8 & 90 & 600 & 2000-dec-10 + ngc346-wb6 & 00 58 57.74 & -72 10 33.6 & 90 & 600 & 2000-dec-10 + ngc346-wb3 + 4 & 00 59 01.09 & -72 10 28.2 & 197 & 600 & 2000-dec-10 + ngc346-wb1 & 00 59 04.81 & -72 10 24.8 & 0 & 600 & 2001-dec-26 + lrrrrrrr h@xmath1 6563 & 100 & 100 & 100 & 100 & 100 & 100 & 100 + s vi 937 & 12 & 19 & 18 & 20 & 23 & 27 & 30 + c iii 977 & 352 & 304 & 275 & 258 & 247 & 239 & 232 + n iii 991 & 17 & 13 & 12 & 11 & 10 & 10 & 10 + ne vi 1006 & 0 & 1 & 4 & 12 & 27 & 45 & 58 + o vi 1034 & 9 & 90 & 334 & 708 & 996 & 1156 & 1236 + s iv 1070 & 11 & 8 & 8 & 7 & 7 & 7 & 7 + ne v 1146 & 3 & 8 & 13 & 18 & 24 & 27 & 28 + n v 1240 & 17 & 27 & 23 & 23 & 24 & 24 & 24 + si iv 1397 & 31 & 28 & 26 & 25 & 24 & 24 & 23 + o iv ] 1402 & 100 & 112 & 108 & 101 & 97 & 94 & 92 + n iv ] 1490 & 9 & 7 & 6 & 5 & 5 & 5 & 5 + c iv 1549 & 211 & 174 & 156 & 148 & 145 & 142 & 140 + o iii ] 1662 & 59 & 52 & 47 & 44 & 42 & 41 & 40 + n iii ] 1748 & 6 & 5 & 4 & 4 & 4 & 4 & 4 + i(h@xmath1 ) & 1.268 & 1.551 & 1.819 & 2.048 & 2.249 & 2.431 & 2.622 +

the fundamental rovibrational band of co near 4.7 m is a sensitive tracer of the presence and location of molecular gas in the planet - forming region of protoplanetary disks at 0.0110au . we present a new analysis of a high - resolution spectral survey ( r,000 , or ) of co rovibrational lines from protoplanetary disks spanning a wide range of stellar masses and of evolutionary properties . we find that the co emission originates in two distinct velocity components . line widths of both components correlate strongly with disk inclination , as expected for gas in keplerian rotation . by measuring the line flux ratios between vibrational transitions , we find that the two velocity components are clearly distinct in excitation . the broad component ( ) probes the disk region near the magnetospheric accretion radius atau , where the gas is hot (k ) . the narrow component ( ) probes the disk at larger radii of 0.110au , where the gas is typically colder ( 200700k ) . co excitation temperatures and orbital radii define an empirical temperature - radius relation as a power law with index between 0.053au . the broad co component , co - spatial with the observed orbital distribution of hot jupiters , is rarely detected in transitional and herbig ae disks , providing evidence for an early dissipation of the innermost disk . an inversion in the temperature profile beyond 3au is interpreted as a tracer of a regime dominated by uv pumping in largely devoid inner disks , and may be a signature of the last stage before the disk enters the gas - poor debris phase . _ accepted by the astrophysical journal _

1507.06650

the inner @xmath9au of protoplanetary disks is thought to be the region in which most exoplanets form @xcite . initially , the inner disk likely has a smooth radial distribution of gas and dust that is truncated by the stellar magnetic field at the co - rotation radius , within 0.1au . from this point , disk material accretes onto the central star through magnetospheric accretion ( e.g. * ? ? ? however , over time , inner disks evolve , and gaps and inner holes frequently develop within a few myr @xcite . some of these gaps may be carved by photo - evaporation @xcite , and others by dynamical clearing by planets coupled with pressure - driven dust migration @xcite . the evolutionary window available for planet formation is therefore limited by , and linked to , the time scale of inner disk dispersal of a few myr @xcite . when observing the inner regions of protoplanetary disks and the process of planet formation , a central challenge is that the angular size of these regions is small ( @xmath10 ) , even for the closest star - forming clouds . further , the gas temperatures in the planet - forming region typically spans @xmath11k , pushing the main observational line tracers into the near- to far - infrared wavelength range . this leads to the common use of infrared molecular spectroscopy to investigate the properties of inner disk gas . the co molecule is a ubiquitous and sensitive tracer of gas in protoplanetary disks . while pure rotational co lines observed by millimeter interferometers trace cold gas extending to the outermost disk regions , co lines from rovibrational transitions at infrared wavelengths are typically excited closer to the central star , where the disk is hotter and exposed to strong infrared ( ir ) and ultraviolet ( uv ) radiation fields . surveys of rovibrational co emission from the fundamental ( @xmath12 ) band near 4.7@xmath0 m have demonstrated that the lines probe the properties of inner disks between the magnetospheric accretion radius and the inner 10 au ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? co line profiles , when spatially- and velocity - resolved , have shown a wealth of co isotopologues and complex line profiles , as indicative of a range in excitation conditions and of physically distinct emission components @xcite . emission from @xmath13co @xmath14=1 - 0 and @xmath14=2 - 1 lines is strong enough to be detected in the majority of gas - rich disks , demonstrating that co is often vibrationally excited . in the innermost disk , co can be efficiently excited by ir photons from the stellar radiation and the local @xmath0m - size hot dust through ir pumping ( e.g. * ? ? ? detection of higher vibrational lines in some disks implies vibrational temperatures that are several times larger than the rotational temperatures , providing evidence for the radiative cascade from pumping of directly uv - irradiated co gas @xcite . detection of rarer isotopologues , including @xmath15co that is commonly seen in its @xmath14=1 - 0 transitions , has been usually attributed to large column densities of co gas found in young disks . recent analyses of velocity - resolved co rovibrational emission in one protoplanetary disk found that the @xmath14=1 - 0 lines are a superposition of a broad ( fwhm @xmath16kms@xmath17 ) and a narrow ( fwhm @xmath18kms@xmath17 ) components @xcite . @xcite found evidence for similarly broad and narrow co components in @xmath19 more embedded disks , and found that , while the two components are blended in @xmath13co @xmath14=1 - 0 lines , only the broad component contributes to the @xmath13co @xmath14=2 - 1 lines , while the @xmath15co lines are dominated by the narrow component . in this paper , we systematically apply a spectral decomposition analysis to high resolution ( @xmath20 ) spectra of a sample of @xmath21 protoplanetary disks , to test the hypothesis that co rovibrational emission in protoplanetary disks may generally be described by two distinct components in velocity and excitation : specifically , a _ narrow component _ , or nc , and a distinctly _ broad component _ , or bc . the decomposition allows us to estimate both the location and excitation temperature of the emitting co gas . while most disks in the sample do show two distinct co components that get colder at larger disk radii , some show only narrow emission lines and signatures of having cleared - out inner disk regions . together , this dataset reveals a sequence of inner disk dispersal across a wide range of stellar masses , from 0.3 to 3@xmath22 . the disks sample includes 37 disks with 4.7@xmath0 m spectra obtained with the cryogenic infrared echelle spectrometer ( crires , * ? ? ? * ) on the very large telescope ( vlt ) of the european southern observatory ( eso ) , as part of the eso large program 179.c-0151 @xcite . further , one spectrum was added from program 093.c-0432 , providing an additional epoch of data for a disk around the variable star exlup @xcite . the original survey included disks mostly around young solar - mass stars ( m@xmath23 m@xmath24 , t tauri stars ) as well as some around intermediate - mass stars ( @xmath25 m@xmath26 m@xmath24 , mostly herbig ae / be stars ) . the disks were also selected to cover a range of evolutionary phases , from the early stages when the disks are still embedded in diffuse envelopes to the clearing stages of transitional disks . for this paper , we selected those disks with spectra that show co lines in emission , indicative of the presence of warm co gas in the inner part of the disk . the sample includes 30 t tauri disks and 7 herbig disks ; 2 disks have been observed to have diffuse envelopes , and 7 disks are known to be transitional ( see table [ tab : incl ] ) . all spectra were reduced using procedures described in @xcite , and were already presented in several papers , providing general overviews as well as more detailed studies of various subsamples @xcite . the observations were carried out between 2007 and 2010 , but consistent instrumental parameters were maintained to ensure homogeneity of the data . the slit width was @xmath27 , providing a spectral resolution of @xmath28kms@xmath17 , which is generally confirmed by measured widths of narrow atmospheric absorption lines . early - type telluric standard stars were observed at similar airmasses to the science targets ( typically within 0.1 difference ) to allow an accurate telluric correction . the observations were generally obtained using the crires adaptive optics correction . the four crires detectors provide a combined instantaneous spectral coverage of @xmath29@xmath0 m . the observing strategy targeted the @xmath13co rovibrational p - branch lines in several non - contiguous settings , with at least one centered between 4.65 and 4.77@xmath0 m ( see figure [ fig : 5um spectrum ] ) . additionally , a spectral setting covering wavelengths up to 5.0@xmath0 m was obtained for most targets . the crires spectra do not provide continuous coverage due to gaps from detector chips and the telluric absorption from atmospheric co lines doppler - shifted from the targeted lines depending on the month of observation . in some cases , the spectra were repeated by applying a small shift to the central wavelength or taken in different months of the year to compensate for those effects and provide a more complete spectral sampling ( see * ? ? ? this strategy was adopted to obtain complete velocity - resolved co line profiles . in this section we describe the methodology applied to the analysis of crires spectra , while comparison of our results to previous work and discussion are provided in section [ sec : disc ] . we apply a spectral decomposition algorithm , which essentially consists of the following three basic steps . _ step 1 _ : : high signal - to - noise profiles of @xmath13co @xmath14=1 - 0 and @xmath14=2 - 1 emission are built by stacking multiple lines from different rotational levels . _ step 2 _ : : the stacked line profiles are split into two ( broad ; bc and narrow ; nc ) components by scaling the @xmath13co @xmath14=2 - 1 profile to match the wings of the @xmath13co @xmath14=1 - 0 profile , and then subtracting it . this step reveals if the emission is made of a single ( sc ) rather than of two components ( bc and nc , figure [ fig : multi - comp analysis ] ) . _ step 3 _ : : the line widths and the vibrational @xmath3 ratios are then measured for each detected component . this procedure is applied to each co spectrum in our sample . the first two steps are described in sections [ sec : ana1 ] and [ sec : ana2 ] . the third step provides the data used to find two empirical relations that probe the properties of inner disks ( sections [ sec : ana3 ] and [ sec : ana4 ] ) . we build high signal - to - noise co line profiles for each science target by stacking several transitions observed in the 4.7@xmath0 m spectra . nominally , the maximum spectral coverage obtained in this survey includes ro - vibrational co transitions from p(1)p(32 ) for @xmath13co @xmath14=1 - 0 , p(1)p(26 ) and r(0)r(6 ) for @xmath13co @xmath14=2 - 1 , and p(1)p(23 ) and r(0)r(15 ) for @xmath15co @xmath14=1 - 0 . in practice , the number of transitions available for each disk depends on the spectral coverage obtained for individual targets and on the gaps caused by detector chips and telluric absorption , which in turn depends on the time of year of a given observation . for stacking and averaging line profiles , we use the lines that are most commonly available , which have similar integrated flux , and which are not significantly blended with other transitions . these criteria preserve both the line flux and shape in the stacked profiles , and enforce homogeneity in the analysis method for the entire disk sample . note that , within the relatively narrow spectral range @xmath30 m , lines out of a given vibrational level have very similar fluxes and excitation energies ( within 10 - 30% ) and line profiles depend only very weakly on the rotational quantum number ( e.g. * ? ? ? * ; * ? ? * ; * ? ? ? * ) . for high rotational quantum numbers , this may no longer be true ( @xmath31 ) . given these constraints , the @xmath13co @xmath14=1 - 0 lines that are most suitable for stacking are those between p(7 ) and p(12 ) , while the @xmath13co @xmath14=2 - 1 lines are those between p(3 ) and p(6 ) . as for the @xmath15co lines , several can typically be used over the entire spectrum , where not blended with @xmath13co . each line is normalized to its local continuum , re - sampled on a common velocity grid , and the stacked line profiles are built by weighted average of the pixel values from the individual transitions . while the stacked @xmath14=2 - 1 line profiles typically have continuous spectral sampling , @xmath14=1 - 0 lines often have spectral gaps on one side due to the counterpart telluric absorption lines at the doppler velocity of the target at the time of observation ( see for instance the gap on the red side of the @xmath14=1 - 0 line in figure [ fig : multi - comp analysis ] ) . in some disks ( cw tau , doar 24e s , ht lup , vssg 1 ) , an absorption component is present on top of @xmath13co emission in the @xmath14=1 - 0 lines . this is typically due to absorption from unrelated foreground gas @xcite . in these targets , we exclude pixels affected by the absorption component from the stacked profiles and further analysis . l l l l c l c l c l aatau & s & tt & 0.85 & _ m1 _ & - & - & 71(1 ) & _ i1 _ + as205 n & d & tt & 1.1 & _ m2 _ & 20(5 ) & _ i2 _ & 25 & _ i3 _ + as209 & d & tt & 1.4 & _ m3 _ & - & - & 3840 & _ i3 _ + cra - irs2 & d & tt , em & 1.0 & _ m4 _ & 17(11 ) & _ this work _ & - & - + cvcha & d & tt & 2.1 & _ m3 _ & 30(12 ) & _ this work _ & - & - + cwtau & d & tt & 1.2 & _ m2 _ & 28(10 ) & _ this work _ & - & - & @xmath32 agrees with @xmath33 from _ i4 _ + dftau & d & tt & 0.5 & _ m5 _ & 65(15 ) & _ this work _ & - & - & @xmath32 agrees with @xmath34 from _ i5 _ + doar24e s & d & tt & 0.7 & _ m2 _ & 20(5 ) & _ i2 _ & - & - + doar44 & s & tt , tr & 1.3 & _ m3 _ & - & - & 40(20 ) & _ i3 _ & + drtau & d & tt & 1.0 & _ m2 _ & 9(5 ) & _ i2 _ & 37(3 ) & _ i6 _ & + ec82 & s & tt & 1.0 & _ m4 _ & 51(18 ) & _ this work _ & - & - & + exlup 08 & d & tt & 0.6 & _ m6 _ & 44(8 ) & _ this work _ & - & - & ex lup observation from august 2008 + exlup 14 & d & tt & 0.6 & _ m6 _ & 48(17 ) & _ this work _ & - & - & ex lup observation from april 2014 + fntau & s & tt & 0.35 & _ m7 _ & - & - & 20(10 ) & _ i7 _ + fztau & s & tt & 0.7 & _ m3 _ & 38(15 ) & _ this work _ & - & - & + gqlup & d & tt & 0.8 & _ m2 _ & 65(10 ) & _ i2 _ & - & + hd135344b & s & h , tr & 1.6 & _ m2 _ & 14(4 ) & _ i2 _ & 23(0.3 ) & _ i8 _ + hd142527 & s & h , tr & 3.5 & _ m2 _ & 20(2 ) & _ i2 _ & 23(8 ) & _ i9 _ + hd144432s & s & h & 1.7 & _ m2 _ & 27(3 ) & _ i2 _ & - & - + hh100 & d & tt , em & 0.4 & _ m3 _ & 26(4 ) & _ this work _ & - & - & + htlup & d & tt & 2.5 & _ m3 _ & 28(8 ) & _ this work _ & - & - & + imlup & d & tt & 1.0 & _ m4 _ & - & - & 49(4 ) & _ i10 _ + irs48 & s & h , tr & 2.0 & _ m3 _ & 42(6 ) & _ i11 _ & 48(8 ) & _ i12 _ + lkha330 & s & h , tr & 2.5 & _ m2 _ & 12(2 ) & _ i2 _ & 42 & _ i13 _ + rno90 & d & tt & 1.5 & _ m2 _ & 37(4 ) & _ i2 _ & - & - + rulup & d & tt & 0.7 & _ m2 _ & 35(5 ) & _ i2 _ & - & - + scra s & d & tt & 0.6 & _ m3 _ & 26(4 ) & _ this work _ & - & - + scra n & d & tt & 1.5 & _ m2 _ & 10(5 ) & _ i2 _ & - & - + sr21 & s & h , tr & 2.2 & _ m2 _ & 15(1 ) & _ i2 _ & 15(0.4 ) & _ i8 _ & + ttau n & d & tt & 2.1 & _ m3 _ & 20(13 ) & _ this work _ & - & - & @xmath35 agrees with @xmath36 from _ i14 _ + twcha & s & tt & 1.0 & _ m8 _ & 67(23 ) & _ this work _ & - & - + twhya & s & tt , tr & 0.7 & _ m2 _ & 4(1 ) & _ i2 _ & 12 & _ i6 _ & + vssg1 & d & tt & 0.5 & _ m3 _ & - & - & 53 & _ i3 _ & + vvser & s & h & 3.0 & _ m2 _ & 65(5 ) & _ i2 _ & 70(5 ) & _ i15 _ & + vwcha & d & tt & 0.6 & _ m3 _ & 44(17 ) & _ this work _ & - & - + vzcha & d & tt & 0.8 & _ m3 _ & 25(10 ) & _ this work _ & - & - + waoph6 & d & tt & 0.9 & _ m9 _ & - & & 39 & _ i3 _ + wxcha & d & tt & 0.6 & _ m8 _ & 87(31 ) & _ this work _ & - & - l c c c c c c[ht ] as205 n & 52 & 0.38 & 0.17 & 970 & 13 & 2.3 + as209 & 160 & 0.91 & 0.07 & 1054 & 33 & 1.7 + cra - irs2 & 53 & 0.37 & 0.11 & 960 & 17 & 1.1 + cvcha & 102 & 0.25 & 0.18 & 790 & 47 & 0.9 + cwtau & 54 & 0.32 & 0.32 & 890 & 45 & 0.5 + dftau & 89 & 0.43 & 0.20 & 1060 & 53 & 0.6 + doar24e s & 81 & 0.30 & 0.04 & 860 & 29 & 0.4 + drtau & 37 & 0.39 & 0.06 & 990 & 13 & 0.5 + exlup 08 & 155 & 0.61 & 0.06 & 1420 & 32 & 1.3 + exlup 14 & 150 & 0.51 & 0.07 & 1220 & 40 & 1.0 + gqlup & 107 & 0.28 & 0.20 & 830 & 62 & 0.6 + hh100 & 80 & 1.02 & 0.04 & 1200 & 12 & 1.9 + htlup & 190 & 0.65 & 0.05 & 1510 & 35 & 1.6 + imlup & 200 & 1.16 & 0.05 & 1400 & 40 & 1.3 + rno90 & 92 & 0.21 & 0.23 & 730 & 52 & 0.7 + rulup & 111 & 0.50 & 0.07 & 1200 & 18 & 2.5 + scra s & 101 & 0.40 & 0.04 & 1000 & 15 & 1.8 + scra n & 57 & 0.30 & 0.05 & 860 & 12 & 1.1 + ttau n & 110 & 0.57 & 0.07 & 1340 & 28 & 1.1 + vssg1 & 66 & 0.36 & 0.27 & 940 & 35 & 1.0 + vwcha & 80 & 0.32 & 0.16 & 890 & 33 & 0.9 + vzcha & 53 & 0.44 & 0.18 & 1080 & 24 & 0.9 + waoph6 & 133 & 0.33 & 0.07 & 900 & 32 & 1.2 + wxcha & 140 & 0.45 & 0.11 & 1100 & 67 & 0.5 l c c c c[ht ] aatau & 115 & 0.14 & 0.2 & 630 + doar44 & 72 & 0.11 & 0.4 & 590 + ec82 & 49 & 0.03 & 0.9 & 420 + fntau & 9 & 0.01 & 1.8 & 330 + fztau & 31 & 0.20 & 1.0 & 720 + hd135344b & 14 & 0.05 & 1.7 & 420 + hd142527 & 25 & 0.04 & 2.3 & 460 + hd144432s & 40 & 0.09 & 0.8 & 560 + lkha330 & 9 & 0.09 & 4.7 & 420 + twcha & 65 & 0.16 & 0.7 & 660 + twhya & 6 & 0.09 & 0.3 & 500 + vvser & 55 & 0.16 & 2.9 & 660 + + irs48 & 14 & 0.38 & 16 . & 970 + sr21 & 9 & 0.64 & 6.5 & 1480 + hd100546 @xmath37 & 16 & 0.27 & 15 . & 810 + hd97048 @xmath37 & 17 & 0.33 & 14 . & 900 + hd179218 @xmath37 & 18 & 0.37 & 21 . & 950 + hd141569 @xmath37 & 16 & 0.56 & 17 . & 1300 + hd190073 @xmath37 & 16 & 0.27 & 6.0 & 810 + hd98922 @xmath37 & 21 & 0.29 & 9.3 & 840 the presence of separate broad and narrow components is apparent by visual inspection of many disk spectra ( e.g. figure [ fig : 5um spectrum ] ) . in practice , the difference is formalized by comparison of the @xmath14=1 - 0 and @xmath14=2 - 1 stacked line profiles . we scale the @xmath14=1 - 0 to match the broad wings of the @xmath14=2 - 1 profile , and then subtract one from the other . the best scaling factor is found by @xmath38 minimization between the scaled @xmath14=1 - 0 line profile and the @xmath14=2 - 1 line profile , excluding the central region of narrow emission , if present . the nc is typically much narrower than bc , such that most of the line profile can be used for the @xmath38 minimization . the choice of the wavelength exclusion region is determined iteratively , starting by using the full line profile . if the residuals after subtraction of the two line profiles are consistent with zero over the entire spectral range of the @xmath14=2 - 1 line , the nc is not detected and the procedure stops there . in the contrary case , the fit is iterated after exclusion of increasingly large portions at the center of the line profile , until the residuals are composed of a flat continuum and of the nc emission line . figure [ fig : multi - comp analysis ] shows a demonstrative sketch of the outcome , while spectral decomposition plots for all disks are shown in appendix [ app : co_decomp ] . two basic parameters are measured from the decomposition procedure : the line width and the vibrational flux ratio @xmath39 . line widths are measured at 50% of the line peak ( the full width at half maximum ; fwhm ) from fits to the line profiles , allowing up to two gaussian functions to account for double - peaked profiles . the vibrational flux ratio is measured from the scaling factor that matches the @xmath14=1 - 0 to the @xmath14=2 - 1 lines as explained above , and is exactly equivalent to the integrated line intensity ratio between the @xmath14=2 - 1 and the @xmath14=1 - 0 lines in each spectrum ( further details are provided in the appendix ) . it is possible to measure vibrational ratios in all single - component disks and for the bc in all double - component disks . for the nc , @xmath3 can be measured in only three disks : as 205 n , dr tau , and s cra n. in these disks , the nc can also be identified in the @xmath14=2 - 1 lines through fits of multiple gaussians to the line profile , one for each of the broad and narrow components ( plots are shown in the appendix ) . the non - detection of the nc in the @xmath14=2 - 1 lines of all other double - component disks is generally consistent with an upper limit on the vibrational flux ratio of @xmath40 . tables [ tab:2co comp ] and [ tab:1co comp ] report line widths and vibrational ratios for the entire disk sample . by implementing this decomposition procedure , we find that all disks fall into two phenomenological classes : the single - component " disks ( 14 of 37 ) , where the @xmath14=1 - 0 line profile , once scaled down , is consistent with that of the @xmath14=2 - 1 line , and the double - component " disks ( 23 of 37 ) , where the @xmath14=2 - 1 profile matches only the wings of the @xmath14=1 - 0 profile , revealing an additional narrower component of the @xmath14=1 - 0 profile . in the double - component disks , the @xmath15co lines are typically significantly narrower than the bc , but match the profile of the nc ( figure [ fig : multi - comp analysis ] ) . only as 209 and i m lup have @xmath15co lines detected in the bc only , rather than in the nc , while a few disks tentatively show both the bc and nc in @xmath15co ( cra - irs 1 , as 205 n , and ru lup ) . in double - component disks , the bc profile shows properties distinct from those of the single - component disks : broader lines ( @xmath41 50200kms@xmath17 ) , a clear separation in terms of line flux ratios with a boundary at @xmath42 , and weak or undetected @xmath15co lines . conversely , the nc shares the same line widths ( @xmath41 1070kms@xmath17 ) , same vibrational ratios @xmath43 , and same @xmath15co detection rate as for co in 12/14 single - component disks ( sc ) of our sample . when using line widths to infer radial properties of keplerian disks , it is necessary to correct for inclination effects . consequently , we collect values of disk inclinations from the literature , where available , for the disks in our sample . we carefully searched for the most reliable estimates obtained using spatially - resolved disk images , either from millimeter interferometry , optical / near - infrared ( nir ) scattered - light imaging , or spectro - astrometry from crires . note that for spectro - astrometry the derived disk inclination is degenerate with an assumed stellar mass . for consistency , we therefore adopt both the stellar masses and the disk inclinations reported by @xcite . table [ tab : incl ] shows the adopted disk inclinations , with notes on the technique used to measure the inclination for each disk . using ten double - component disks with known inclinations ( seven from spectro - astrometry , and three from interferometry ) , we find strong correlation between the co line width and disk inclination , with pearson coefficients of 0.7 for the bc and of 0.9 for the nc . strong correlation between nir co line widths and disk inclination has been reported previously by @xcite from a sample of five herbig disks . correlation between these parameters is expected when the gas is in keplerian rotation around the star , a condition that is suggested by the high detection rate of double - peaked line profiles in our disk sample ( @xmath44% ) and that has been proposed since the first surveys of nir co emission in disks ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? note that the two - component decomposition significantly decreases the scatter in this relation relative to what is seen when the line profiles are not fully velocity - resolved as with crires , or if the line profiles were all modeled with a single gaussian profile to derive average line widths ( i.e. without separating bc from nc ) . gas in keplerian rotation at a disk radius @xmath45 produces emission at a maximal velocity @xmath46 from the rest frequency of a given co line following kepler s law as @xmath47 , where @xmath48 is the disk inclination , @xmath49 is the gravitational constant and @xmath50 is the stellar mass . using this relation , it is possible to derive an emitting radius for the gas in a rotating disk from the observed line widths . in general , emission from higher velocities originate from smaller orbital radii . these velocities , usually taken at the 50% or 10% of the line peak value , therefore provide estimates of the disk radii where co is emitting . we refer to the innermost emitting radius of co , @xmath51 , when derived from @xmath52hwhm as proposed by @xcite . in comparison , the velocity at @xmath53hwhm provides an estimate of the disk radius @xmath54 that emits the peak line flux ( in agreement with the emitting radii derived from spectro - astrometry by * ? ? ? * ; * ? ? ? figure [ fig : fwhm_incl ] shows the measured co line widths divided by @xmath55 . removing the stellar mass dependence on the inclination relation leaves only the dependence on the keplerian radius @xmath45 , which can be compared to a simple model of gas emission in keplerian rotation around a @xmath56 star ( shown in figure [ fig : fwhm_incl ] at three different disk radii ) . in these terms , the bc has @xmath57 between 0.01 and 0.1 au . this is consistent with the disk region of the corotation and magnetospheric accretion radii for solar - mass stars ( e.g. * ? ? ? * ) , as found by previous studies of rovibrational co emission from disks ( e.g. * ? ? ? * ; * ? ? ? the nc , instead , emits from larger disk radii , with @xmath57 between 0.1 and 1 au . the single - component disks show co emission at radii similar to , or larger than , those of the nc . in figure [ fig : fwhm_incl ] we show linear fits to the bc and nc in disks with known inclinations , obtained using the bayesian method by @xcite . the regression takes into account the uncertainties on both variables , as well as the intrinsic scatter in the regression relation due to physical quantities not explicitly accounted for . typical errors on the line widths are @xmath58 for the bc and @xmath59 for the nc , the stellar masses are assumed to be known to 20% and the disk inclinations are typically known to accuracies better than 10% . the best - fit parameters and their errors are taken as the median and standard deviation of the posterior distributions of the regression parameters . the median intercept and slope are used to plot the lines in figure [ fig : fwhm_incl ] , while the posterior distributions are plotted in the appendix . the single linear fits shown in figure [ fig : fwhm_incl ] assume that differences in @xmath57 due to the dependence on stellar and disk luminosity for either the bc and the nc is weak relative to the inclination dependence . this is supported by the bayesian regression of line widths versus disk inclinations , which confirms strong correlation ( @xmath60 ) for both the bc and nc , although the uncertainties in the regression parameters for bc are 34 times larger than for nc . this is also consistent with previous results for disks around solar - mass stars where @xmath57 was found to vary only within a factor of a few @xcite , and with the theoretical expectation that the inner radius increases only with @xmath61 . the residual intrinsic scatter on the linear relation for the bc is four times larger than for the nc , which may point to a remaining dependence on the stellar and ( variable ) accretion luminosity . given the empirical line width - inclination relation , it is possible to estimate inclinations for disks with high resolution co spectroscopy even when resolved imaging is not yet available . in doing so , it is assumed that the two relations found for the bc and nc represent a general property of rovibrational co emission in the double - component disks . the advantage of having two relations , one for each velocity component , is that for a given double - component disk there are two constraints to derive one parameter . we take the measured values of the bc and nc line widths in each disk and use them to derive the disk inclination by using the regression coefficients @xmath62 from the empirical linear relations : @xmath63 where @xmath64 and @xmath65 for the bc , and @xmath66 and @xmath67 for the nc . we derive two inclination values , one for the bc and one for the nc , and adopt the mean of the 1-@xmath68 range in common between the two . among the single - component disks , only three have unknown inclination : ec 82 , fz tau , tw cha . for these disks , we adopt the nc relation , since single - component disks with known inclinations span disk radii similar to those of the nc of double - component disks ( figure [ fig : fwhm_incl ] ) . following this procedure , we estimate an inclination for a total of fifteen new disks . these values are included in table [ tab : incl ] . using the inclinations described in the previous section , we derive the co emitting radii for all disks in the sample . in figure [ fig : vibr_rin ] we show a diagram made with the vibrational line flux ratios plotted against the co emitting radii . the diagram at the left of the figure shows a strong logarithmic anti - correlation between @xmath69 and @xmath3 , with a pearson coefficient of -0.9 . in addition to spanning spatially separated @xmath54 ranges , as already shown in figure [ fig : fwhm_incl ] , the bc and nc clearly span distinct values in @xmath3 . here too , the single - component disks have similar values to those of the nc . two obvious outliers are sr 21 and irs 48 , both transitional disks with large inner gaps , which have vibrational excitation temperatures as high as the bc at much smaller disk radii ( they will be discussed in section [ sec : disc ] ) . excluding them , figure [ fig : vibr_rin ] demonstrates that the vibrational flux ratio decreases with disk radius as : @xmath70 the vibrational line flux ratios measure how much the vibrational level @xmath14=2 is populated as compared to @xmath14=1 , which in turn is a measure of the vibrational excitation temperature @xmath71 ( e.g. * ? ? ? the relation between @xmath3 and @xmath54 can thus be converted into a radial profile for the co vibrational excitation temperature in inner disks . the conversion of the vibrational flux ratio to a temperature depends on the co column density , so the model - independent relation is that using the vibrational flux ratio ( eqn . 2 ) . to estimate the corresponding temperature relation , we use a model of a slab of gas in thermal equilibrium , where the level populations are defined by a single temperature following the boltzmann distribution . other analyses of co rovibrational emission from disks consistently find a relatively narrow range of column densities . consequently , we adopt a single @xmath13co column density of @xmath72@xmath73 , as an order - of - magnitude value representative of the range found in previous studies ( @xmath74@xmath75@xmath73 , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the details of the conversion are described in the appendix [ app : vibr_convers ] . the resulting temperature radial profile is : @xmath76 if disks without previously known inclinations are excluded , the exponent becomes @xmath77 , within the error , indicating that the result is not biased by our empirical derivation of disk inclinations . the excitation temperatures estimated for each disk are reported in tables [ tab:2co comp ] and [ tab:1co comp ] . the nominal uncertainty on the temperatures , once the column density is fixed , is less than 50 k as due to the uncertainty on the vibrational ratios ( which is typically @xmath78 ) . @xcite proposed that the non - detection of @xmath15co lines in the bc , combined with the curvature seen in rotation diagrams , constrained the bc column density to within @xmath79@xmath72@xmath73 . on the other hand , rotation diagrams of @xmath15co lines were found to require column densities of @xmath80@xmath75@xmath73 for the nc . if we relax our assumption and allow the bc to have columns as low as @xmath74@xmath73 , and nc as large as @xmath75@xmath73 , then the temperatures will decrease / increase for the nc / bc , respectively , and the overall temperature profile may exhibit a steeper index up to @xmath81 ( see appendix [ app : vibr_convers ] ) . the vibrational excitation temperatures of the bc match those previously found from rotational analyses of @xmath13co lines ( 7001700k , e.g. * ? ? ? * ) , while the co in the nc and sc disks typically have lower temperatures , corresponding to those determined for @xmath15co line fluxes ( 200600k , e.g. * ? ? ? this is consistent with our finding that the @xmath15co line profiles match those of the nc , as indicative of probing a similar disk region . in turn , this suggests that the nc is formed in an optically thick layer , leading to high @xmath82 ratios , while the bc must be more optically thin . a similar conclusion was reached for disks in embedded young stars @xcite . the radii derived from the bc span the range found previously for disks around solar - mass stars ( 0.010.1au ) while the nc and sc span the range found for herbig and transitional disks ( 0.110 au ) . this is consistent with the finding that double - component disks ( where bc is detected ) are primarily disks around low - mass stars , while the single - component disks are primarily herbig and transitional disks . the nc and part of sc share similar properties in terms of line widths , orbital radii , and vibrational ratios ( section [ sec : ana ] and table [ tab : co summary ] ) , and they may trace the same gas component , regardless of the presence of a bc . l c c c c c broad ( bc ) & 50200 & 0.20.6 & 0.040.3 & 8001500 & 3/23 + narrow ( nc ) & 1070 & @xmath83 & 0.23 & @xmath84 & 18/23 + single i ( sc ) & 670 & 0.010.2 & 0.25 & 300 - 700 & 9/12 + single ii ( sc ) & 920 & 0.30.6 & 620 & 800 - 1500 & 8/8 while a disk origin for the bc is supported by the common double - peak profiles , as well as by a keplerian signal seen in spectro - astrometry @xcite , the origin of the nc is less clear . a disk wind has been proposed for single - peak @xmath13co @xmath14=1 - 0 line profiles ( those dominated by the nc , e.g. as205 n ) to reproduce the sub - keplerian motions of the line peak , observed with spectro - astrometry @xcite . in one case , a corresponding large - scale disk wind has possibly been seen with alma @xcite . an observational characteristic of a disk wind launched at small radii , and preserving angular momentum as it expands , is that it the gas contributing to the line becomes increasingly sub - keplerian closer to the line center . specifically , the line wings are exactly keplerian , while the inner few kms@xmath17 are strongly sub - keplerian , tracing radii significantly smaller than what their velocity would suggest . this could explain why observables constructed from the line wings of both the broad and narrow components , as we do here , yield results consistent with a nearly keplerian disk . broad and narrow components that may be similar to those of the rovibrational co lines have been found in the 6300[oi ] line by @xcite . these authors suggest a disk origin for the broad component , and an unbound cool disk wind as the origin of the narrow component . it is not clear whether the narrow components of [ oi ] and co are generally tracing the same disk wind , but they share close resemblance in at least a few disks . for the disks in common between this work and @xcite , the fwhm of nc in co and [ oi ] agrees to within @xmath85kms@xmath17 in dr tau , vz cha , ru lup , tw hya , and cw tau ; in dr tau , the same is true also for the bc of co and [ oi ] . in section [ sec : ana4 ] we found a strong logarithmic anti - correlation between the co vibrational excitation temperature and emitting radius . this relation corresponds to an empirical temperature profile of a typical inner disk surface . however , it was noted that two disks from the sample with large co emitting radii , the transitional disks sr 21 and irs 48 , clearly depart from the relation , showing significantly higher vibrational ratios at their respective co emitting radii than the relation predicts . to explore this further , we add 6 disks from the crires survey of 12 disks around intermediate - mass ( @xmath86 ) stars from @xcite . this survey presents 4.7@xmath0 m spectroscopy with a similar experimental design and wavelength coverage as our survey . specifically , we include the full subset of disks for which lines are detected from both @xmath13co @xmath14=1 - 0 and @xmath14=2 - 1 transitions in a wavelength range similar to that of our survey . following the same procedure as for other disks , we derive @xmath54 and @xmath71 for the disks around hd100546 , hd179218 , hd190073 , hd97048 , hd141569 , hd98922 ( included in table [ tab:1co comp ] ) . their locations in the temperature - radius diagram are included in figure [ fig : tempprofs_compar ] , which shows that disks around intermediate - mass stars have vibrational temperatures that are typically higher at a given radius than the relation defined by the lower - mass stars at smaller disk radii . it is clear that sr 21 and irs 48 belong to this vibrationally hot class of disks . the combination of crires surveys of disks spanning a wide range of stellar masses therefore demonstrates the existence of two trends : a vibrational temperature decrease out to radii of @xmath87au followed by a temperature increase at radii @xmath88au . we note that the remaining six disks from van der plas et al . ( 2015 ) lack @xmath14=2 - 1 measurements due either to poor telluric corrections and low signal - to - noise , or the line widths and disk inclinations put them in the ir regime of our diagram ( @xmath85au ) , where low @xmath14=2 - 1 line fluxes are expected . .,scaledwidth=50.0% ] analyses by @xcite , @xcite , and @xcite included two possible excitation regimes for co rovibrational emission : in the inner disk , ir pumping due to the local dust continuum radiation , and at larger disk radii , where the dust temperature is too low to generate sufficient amounts of 4.7@xmath0 m photons , pumping by uv photons from the star and/or accretion shock . the specificity of uv pumping is to populate very high vibrational levels ( @xmath89 ) by decay of an excited electronic state @xcite , while ir pumping drives the co level population to equilibrate to the local color temperature of the dust . at dust temperatures of 500 - 1000k , this will only populate the @xmath14=1 and @xmath14=2 levels . indeed , figure [ fig : tempprofs_compar ] can be interpreted as evidence for a transition between the ir and uv pumping regimes near 3au . the ir regime " , at small disk radii , and the uv regime " , at larger disk radii , are correspondingly marked in the figure . further evidence is provided by comparison to rotational temperatures t@xmath90 derived in the same disks from previous studies @xcite . figure [ fig : tempprofs_compar ] shows that @xmath91 in the ir regime , indicating that the co rovibrational population is thermalized to the local dust temperature in the inner disk @xcite . conversely , in the uv regime , @xmath92 @xcite . this is consistent with the finding of others , for instance @xcite found @xmath93k and @xmath94k in the disk of hd141569 . note that , while the vibrational temperatures of the @xmath14=2 - 1 transitions are higher than the rotational temperature in the uv regime , they are typically much lower than the vibrational temperature of even higher - lying levels ( @xmath14=6 - 5 can be as high as 6000k ) . therefore , while we use the @xmath14=2 - 1 temperature since it is easy to measure in most disks , our results are consistent with previous work citing very high temperatures of high - lying vibrational levels . + in figure [ fig : final ] we illustrate how the temperature - radius diagram of rovibrational co emission resembles key aspects of the current theoretical picture of protoplanetary disk dispersal . in the phenomenological class of the double - component disks , the bc is observed from a hot disk region extending to the magnetospheric accretion radius at @xmath5au @xcite , demonstrating the presence of abundant gas at the smallest radii dynamically permissible for keplerian orbits . these disks are likely to be in a primordial phase , before giant planets have fully formed ( sketch 1 ) . dr tau is an example of a primordial disk as based on its co emission . the next stage in the co sequence leads to the suppression and eventual disappearence of bc ( sketch 2 ) , strongly suggesting the opening of partly devoid inner gaps or an inner region depleted in gas compared to the primordial disks . in fact , it has been observed by monitoring co spectra of a strong episodic accretor , ex lup , that the bc gets significantly weaker relative to nc after an accretion outburst that depleted the gas mass in the inner disk by an order of magnitude @xcite . in this stage , double- and single - component disks overlap in the diagram . nc is detected from fractions of an au up to @xmath87au . at similar radii , and even larger , co is observed in single - component disks , where the bc is not detected . tw hya is an example of this second class of disks , which probably has significant overlap with the `` pre - transitional '' disks as defined from dust emission @xcite . indeed , for tw hya , an inner gap in the optically thick dust disk has been detected @xcite . note that a non - detection of the bc does not directly imply the complete absence of dust or gas at @xmath950.1 au . residual gas and dust in the innermost region , which absorbs most of the direct uv radiation from the star , may likely explain why the temperature profile in some single - component disks continues along an ir - pumped power law . photo - evaporation probably plays a role in opening disk gaps , and it is interesting to note that uv photo - evaporation is expected to take over at a few au @xcite , close to where the transition between the ir and uv regimes is found in the co diagram . other clearing mechanisms may dominate the disk dissipation at smaller disk radii , including disk accretion and dust evolution to growing planetesimals and protoplanets ( e.g. * ? ? ? the transition to a uv - dominated regime of co rovibrational excitation defines a third class of disks , as discussed in section [ sec : disc2 ] . in order for uv radiation to excite the gas to high vibrational temperatures beyond a few au , the inner disk region must be largely devoid of shielding dust . we identify these disks as having large inner gaps , with very little amounts of residual material in the inner disk ( sketch 3 ) . this interpretation is consistent with that proposed by @xcite , who also suggested that uv - pumped co emission may be a distinct signature of devoid inner gaps , and @xcite , who imaged large inner dust gaps in two herbig disks in the uv branch of our diagram , one of which was previously unknown ( in hd97048 ) . in total , large inner gaps in the dusty disks have been detected in up to 6 out of 8 disks in the uv branch so far , and for the remaining two ( hd98922 and hd190073 ) the current lack of resolved imaging may simply be due to their large distance ( @xmath96 pc ) . potentially , our analysis suggests that all disks that show a single rovibrational co component are developing or have developed gaps in their inner 10 au region , making them prime targets for future imaging detections . a stellar - mass dependence of @xmath54 may be implied by figure [ fig : final ] , where higher - mass stars tend to have larger @xmath54 relative to lower - mass stars . co gas from @xmath97au is rarely detected in disks around stars of @xmath98 . this can not be explained by stellar magnetospheric truncation of the disk , which is expected to happen at radii @xmath99au across the stellar mass spectrum @xcite , and an inner disk radius set by dust sublimation may be marginally consistent only in a few disks , and only at @xmath100au given their bolometric luminosities @xcite . alternatively , a stellar - mass dependent inner disk lifetime would be consistent with the co temperature - radius diagram , with disks around intermediate - mass stars dispersing faster than those around low- and solar - mass stars ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? that is , we detect fewer broad co components in disks with short dispersal time scales . in this scenario , a mass - dependence in figure [ fig : final ] would be directly linked to the timescale for gas dispersal in inner disks , and therefore to planet - formation processes and timescales . in summary , we suggest that the co temperature - radius diagram shows a sequence of gap opening in protoplanetary disks , where formation of smaller gaps ( interpreted as an early transition , in evolutionary terms ) is marked by the disappearance of the co broad component , while opening of larger gaps ( or a more evolved phase ) is marked by an inversion in the trend of the temperature profile . the uv pumping regime may represent the very last step before the primordial gas in the disk is dispersed , setting the beginning of the debris disks phase ( see the beta pic point in figure [ fig : final ] , from rovibrational co @xmath14=1 - 0 lines detected by * ? ? ? * ) . a high - resolution spectrum of co emission at 4.7@xmath0 m may therefore be a powerful discriminator between primordial , pre - transitional and transitional disk stages at radii between 0.110au . observations of the relative radial profiles of co gas and of dust grains of different sizes in inner disks may provide a powerful tool to distinguish between competing gap formation mechanisms . the lifetime of the planet - forming region of protoplanetary disks is linked to the observed orbital distribution of exoplanets . in particular , the semi - major axis distribution of giant exoplanets is known to contain structure indicating the presence of multiple populations , including the `` hot jupiters '' , orbiting very close to the star at a small fraction of an au @xcite , and a deficit of giant planets out to 0.5au @xcite . it is generally accepted that hot jupiters migrated inwards from a natal orbit at a few au , either due to angular momentum exchange with protoplanetary gas @xcite during the lifetime of the disk , or due to a later dynamical process , such as the kozai mechanism @xcite . ( from the database of exoplanets.org ) . the exoplanet histogram is arbitrarily scaled for comparison to the co histogram . _ bottom _ : same as plot above , but separated into two stellar mass bins . here we show the innermost co gas in each disk , therefore excluding the nc in double - component disks.,scaledwidth=50.0% ] to explore this in the context of our observed inner disk radii , we show the distribution of @xmath51 compared to the semi - major axes of known exoplanets with @xmath101 in figure [ fig : corad_compar ] . the bc radius distribution matches the orbital distribution of hot jupiters ( 0.010.1 au ) , as expected in the type ii migration scenario . that is , we observe that gas is indeed often available at the right disk radii within 12myr to allow hot jupiters to migrate inwards to the inner gas disk truncation radius . however , hot jupiters probably comprise a small fraction of giant exoplanets @xcite . if the inner disk dissipates rapidly , most planets may not have time to migrate , and the orbital distribution of `` cold jupiters '' will be determined by the inner gas radius of transitional disks , as traced by the narrow co component . indeed , figure [ fig : corad_compar ] shows that the nc distribution peaks slightly inward of the bulk of the exo - jupiter distribution ( 110 au ) . a stellar - mass dependence of the inner disk gas radius may also be reflected in the exoplanet population . it is known that intermediate - mass stars have very few hot jupiters compared to lower - mass stars @xcite , although a possible overestimation of the intermediate - mass stellar masses is currently under debate @xcite . @xcite proposed that the paucity of hot jupiters around intermediate - mass stars may be due to a stellar - mass dependent disk lifetime , where inner disks around intermediate - mass stars dissipate before any giant planets have time to migrate to the disk corotation radius . in the lower panel of figure [ fig : corad_compar ] , we compare the distributions of innermost gas disk radii separated into two stellar mass bins at @xmath102 . this distribution suggests that the bc component is rarely found in disks around intermediate - mass stars , relative to disks around low- and solar - mass disks . that is , the paucity of the bc within 0.1au in disks around intermediate - mass stars is reflected by a similar paucity of hot jupiters in the same stellar mass bin . a likely interpretation of our co observations is that the gaseous disk at 0.01 - 0.5au around stars with @xmath103 indeed dissipates faster than the planet migration time scale . larger samples of co rovibrational spectra of disks and exoplanets around intermediate - mass stars are needed to confirm this link between the natal disk properties and planetary architectures . we presented a new analysis of vlt - crires high - resolution ( r@xmath196,000 ) 4.7@xmath0 m spectra of rovibrational co emission from protoplanetary disks . we applied an empirical decomposition of co line profiles to the sample of co - emitting disks , finding that rovibrational line profiles consist of two distinct velocity components , a broad ( bc ) and a narrow ( nc ) component , primarily in disks around solar - mass stars , and of a single co component in transitional and herbig disks . when splitting the observed line profiles into two components , we find an empirical relation between the co line width and the disk inclination as expected when the gas is in keplerian rotation in a disk . this allowed us to estimate the inclination for 15 disks in the sample for which it was previously unknown . using the profile decomposition to separate blended information from different disk radii , we define an empirical diagram of the keplerian co emitting radius and the vibrational excitation temperature , leading to the definition of a universal excitation temperature profile . between 0.05 and 3au , we find a power law @xmath104 . beyond 3au , the vibrational temperature turns around and begins to increase with radius . we interpret this as the presence of two different excitation regimes , with pumping by infrared photons dominating in the inner disk , and fluorescence excitation by uv photons beyond 3au . we discuss the co temperature - radius diagram in the context of the theoretical picture of inside - out protoplanetary disk evolution , where the innermost disk is dissipated first . we find evidence that the co rovibrational line profiles are tracers of the evolutionary stage of inner protoplanetary disks , potentially providing a gaseous equivalent to the traditional definition of classical , pre - transitional , and transitional disks as based on the dust continuum emission . specifically , we propose that disks exhibiting only a single component in rovibrational co should be in more advanced stages of gap opening and inner disk dissipation than the double - component disks , where the disappearance of bc marks an earlier stage and the uv pumping regime a final stage before the debris disks phase . rovibrational co emission is sensitive to the innermost location of abundant gas , including for transitional disks , which may have more than one inner region due to the presence of gaps . consequently , we compared the observed distribution of inner disk molecular gas to the semi - major axis distribution of giant exoplanets . we find that the distribution of co emitting radii for the broad component matches that of the semi - major axis distribution of hot jupiters . this is consistent with the majority of hot jupiters migrating inward due to type ii migration , and being halted by the inner truncation of the gaseous disk during an early evolutionary phase . the radial distribution of the narrow co component peaks at larger radii , close to the `` period deficit '' between the hot jupiter population and the bulk of the giant exoplanet distribution . finally , we find that disks around intermediate - mass stars rarely have gas in the hot jupiter region . this matches an observed deficit of hot jupiters around intermediate - mass stars , lending support to the idea that their inner disks may evolve too fast for giant planets to migrate inwards to the disk corotation radius . this analysis ultimately supports the potential of future spatially - resolved imaging of rovibrational co to move significant steps forward in our understanding of disk evolution and planet formation in protoplanetary disks . herbig disks can already be resolved with adaptive - optics assisted observations , for instance with crires , but disks around solar - mass stars require either spectro - astrometric techniques , infrared interferometric imaging ( e.g. , vlt - matisse ) , or full aperture spectro - imaging at 4.7@xmath0 m with the future generation of extremely large telescopes ( e.g. , e - elt - metis ) . the authors acknowledge helpful discussions with c. salyk , e. van dishoeck , and g. blake , and constructive comments from an anonymous referee . this research has made use of the exoplanet orbit database and the exoplanet data explorer at exoplanets.org . acknowledges financial support by a nasa origins of the solar system grant no . oss 11-oss11 - 0120 , a nasa planetary geology and geophysics program under grant nag 5 - 10201 . this work is based on observations made with eso telescopes at the paranal observatory under programs 179.c-0151 and 093.c-0432 . the vibrational ratio @xmath105 is the flux ratio @xmath3 between the @xmath106 lines and the @xmath107 lines with same rotational quantum numbers in the two vibrational levels . in this work , @xmath105 values are measured from stacked lines built by using multiple rotational lines and thus provide an average estimate of the vibrational ratio between @xmath108 and @xmath109 . previously , vibrational ratios of co emission were measured by @xcite for 25 disks that are included in our sample , using the same crires spectra we use here . they published the median of five values measured in each disk , as obtained from ratios of the corresponding rotational lines in the two vibrational levels . figure [ fig : vibr_plev ] ( panel @xmath110 ) compares the values from @xcite to those measured in this work . the values measured in the single - component disks agree to a 1:1 relation , showing that our estimates are consistent . however , the vibrational ratios in previous work have been underestimated in the double - component disks , because the different contributions from bc and nc in the @xmath14=1 - 0 lines were not separated . the vibrational ratio @xmath105 depends on the temperature and column density of the gas , as well as on the excitation process . in this work we use a model of a slab of gas in thermal equilibrium that accounts for the line opacity , where the level populations are determined by the boltzmann distribution ( the model is explicitly described in * ? ? ? the molecular parameters for co rovibrational lines are taken from the 2012 hitran and the hitemp databases @xcite . we produced a grid of models to convert the measured @xmath105 into an estimate of the vibrational excitation temperature @xmath71 of the gas . we used three values for the co column density n@xmath111 to span the range of values proposed by previous studies of nir co emission from protoplanetary disks ( @xmath74@xmath75@xmath73 , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? similarly , for the temperature we use eleven values between 300 and 5000 k. for each set of @xmath71 and @xmath112 a co model spectrum is produced , assuming a line fwhm of @xmath113kms@xmath17 ( the median value for the bc ) and using a pixel sampling equivalent to the crires spectra . we then re - sample the model spectrum using a typical value for the rms noise as measured in the observed spectra , and then measure vibrational ratios following three methods : 1 ) using directly the individual line fluxes before convolution with the observed fwhm ( the true " value ) , 2 ) taking the median of seven ratios of the line fluxes measured in the convolved spectrum ( equivalent to the method used by * ? ? ? * ) , and 3 ) using the method adopted in this work . figure [ fig : vibr_plev ] ( panel @xmath114 ) shows one model as an example , for illustration of the outcomes of this procedure . it is important to highlight that the true " ratio is not a single value . in the example , the vibrational ratio varies between 0.2 and 0.6 depending on the line used to measure it . moreover , vibrational ratios measured from the individual line fluxes as observed in the convolved spectrum tend to underestimate the true value . this is due to blending of the @xmath14=1 - 0 lines with higher vibrational lines , which contribute increasingly to the line flux as @xmath71 , @xmath112 , and the line fwhm increase . the averaged ratios estimated in this work in general agree well with the range of true values for each model ( panel @xmath115 in figure [ fig : vibr_plev ] ) . this is true in particular in the case of a co column density of @xmath72@xmath73 , which is a good approximation for most disks in our sample ( e.g. * ? ? ? * ) . for higher values of @xmath116@xmath73 , the vibrational ratios measured here may overestimate the true values by up to 50% . we convert the vibrational ratios measured in this work into temperatures using the curves shown in panel @xmath115 of figure [ fig : vibr_plev ] . in this regard , it is not important that our measured @xmath105 may overestimate the true value , as long as we are self - consistent in the conversion , i.e. as long as we use the same method to measure @xmath105 in the data and in the model . in tables [ tab:2co comp ] and [ tab:1co comp ] we put the temperature values estimated by using a co column of @xmath72@xmath73 . in the case of as 209 , hh 100 , and i m lup , the measured vibrational ratios are too high to be modeled with a co column of @xmath72@xmath73 , which never produces values higher than @xmath10.8 . therefore , for these disks we use a co column of @xmath75@xmath73 , which can reproduce @xmath105 as high as 1.2 and would be appropriate if the emission is more optically thick ( our analysis indeed finds that the bc in as 209 and i m lup is more optically thick than in other double - component disks , as the bc is detected also in @xmath15co , see figure [ fig : multi - comp analysis ] ) . for each one of the three values adopted for the co column density , we obtain a different empirical relation between the temperature and the radius of the co emission . this is shown in panel @xmath117 of figure [ fig : vibr_plev ] , where we plot the linear fits obtained as in figure [ fig : vibr_rin ] . the index @xmath118 of the temperature profile @xmath119 varies between 0.20 ( for @xmath112 = @xmath75@xmath73 ) to 0.33 ( for @xmath120@xmath73 ) . if we allow a radial gradient in @xmath112 such to have higher values of @xmath121@xmath73 as representative of the more optically - thick nc , and lower values of @xmath122@xmath73 as representative of the bc ( as found by * ? ? ? * ) , the index may become as steep as @xmath123 . in figure [ fig : decomp figure ] we show the outcome of the co decomposition procedure as applied to all disk spectra in our sample . we implement the bayesian method by @xcite in the linear fits to the empirical relations presented in sections [ sec : ana3 ] and [ sec : ana4 ] . @xcite provided a bayesian method for linear regression that accounts for errors on both the dependent and independent variables as well as an intrinsic scatter due to physical properties that are not explicitly included . a weighted sum of gaussian distributions is used to obtain an accurate approximation of the true probability distribution of the independent variable . to sample the posterior distribution we use the markov chain monte carlo method from the metropolis - hastings algorithm , which is more efficient than the gibbs sampler in case of small sample sizes . a thorough discussion of the method and of its comparison to other statistical methods is provided in @xcite . the method has been made publicly available through the idl routine ` linmix_err.pro ` , that we use in our analysis . figure [ fig : post_distr ] shows the posterior distributions we obtained for the linear regression parameters presented in section [ sec : ana ] .

distance covariance and distance correlation are scalar coefficients that characterize independence of random vectors in arbitrary dimension . properties , extensions , and applications of distance correlation have been discussed in the recent literature , but the problem of defining the partial distance correlation has remained an open question of considerable interest . the problem of partial distance correlation is more complex than partial correlation partly because the squared distance covariance is not an inner product in the usual linear space . for the definition of partial distance correlation we introduce a new hilbert space where the squared distance covariance is the inner product . we define the partial distance correlation statistics with the help of this hilbert space , and develop and implement a test for zero partial distance correlation . our intermediate results provide an unbiased estimator of squared distance covariance , and a neat solution to the problem of distance correlation for dissimilarities rather than distances .

1310.2926

distance covariance ( dcov ) and distance correlation characterize multivariate independence for random vectors in arbitrary , not necessarily equal dimension . in this work we focus on the open problem of _ partial distance correlation_. our intermediate results include methods for applying distance correlation to dissimilarity matrices . the distance covariance , denoted @xmath0 , of two random vectors @xmath1 and @xmath2 characterizes independence ; that is , @xmath3 with equality to zero if and only if @xmath1 and @xmath2 are independent . this coefficient is defined by a weighted @xmath4 norm measuring the distance between the joint characteristic function ( c.f . ) @xmath5 of @xmath1 and @xmath2 , and the product @xmath6 of the marginal c.f.s of @xmath1 and @xmath2 . if @xmath1 and @xmath2 take values in @xmath7 and @xmath8 , respectively , @xmath0 is the non - negative square root of @xmath9 where @xmath10 . the integral exists provided that @xmath1 and @xmath2 have finite first moments . note that feuerverger ( 1993 ) @xcite proposed a bivariate test based on this idea and applied the same weight function @xmath11 , where it may have first appeared . the following identity is established in szkely and rizzo ( * ? ? ? * theorem 8 , p. 1250 ) . let @xmath12 , @xmath13 , and @xmath14 be independent and identically distributed ( iid ) , each with joint distribution @xmath12 . then @xmath15 provided that @xmath1 and @xmath2 have finite first moments . in section [ s4 ] an alternate version of ( [ edcov ] ) is defined for @xmath1 and @xmath2 taking values in a separable hilbert space . that definition and intermediate results lead to the definition of partial distance covariance . we summarize a few key properties and computing formulas below for easy reference . the distance correlation ( dcor ) @xmath16 is a standardized coefficient , @xmath17 , that also characterizes independence : @xmath18 for more details see @xcite and @xcite . properties , extensions , and applications of distance correlation have been discussed in the recent literature ; see e.g. dueck et al . @xcite , lyons @xcite , kong et al . @xcite , and li , zhong , and zhu @xcite . however , there is considerable interest among researchers and statisticians on the open problem of a suitable definition and supporting theory of partial distance correlation . among the many potential application areas of partial distance correlation are variable selection ( see example [ ex7 ] ) and graphical models ; see e.g. wermuth and cox @xcite for an example of work that motivated the question in that context . in this work we introduce the definition of partial distance covariance ( pdcov ) and partial distance correlation ( pdcor ) statistics and population coefficients . first , let us see why it is not straightforward to generalize distance correlation to partial distance correlation in a meaningful way that preserves the essential properties one would require , and allows for interpretation and inference . one could try to follow the definitions of the classical partial covariance and partial correlation that are based on orthogonal projections in a euclidean space , but there is a serious difficulty . orthogonality in case of partial distance covariance and partial distance correlation means independence , but when we compute the orthogonal projection of a random variable onto the condition variable , the `` remainder '' in the difference is typically not independent of the condition . alternately , the form of sample distance covariance ( definition [ defvn ] ) may suggest an inner product , so one might think of working in the hilbert space of double centered distance matrices ( [ dcenter ] ) , where the inner product is the squared distance covariance statistic ( [ e : anxy ] ) . here we are facing another problem : what would the projections represent ? the difference @xmath19 of double centered distance matrices is typically not a double centered distance matrix of any sample . this does not affect formal computations , but if we can not interpret our formulas in terms of samples then inference becomes impossible . to overcome these difficulties while preserving the essential properties of distance covariance , we finally arrived at an elegant solution which starts with defining an alternate type of double centering called `` @xmath20-centering '' ( see definition [ def : tildea ] and proposition [ p : unbiased ] below ) . the corresponding inner product is an unbiased estimator of squared population distance covariance . in the hilbert space of `` @xmath20-centered '' matrices , all linear combinations , and in particular projections , are zero diagonal @xmath20-centered matrices . we prove a representation theorem that connects the orthogonal projections to random samples in euclidean space . methods for inference are outlined and implemented , including methods for non - euclidean dissimilarities . definitions and background for dcov and dcor are summarized in section 2 . the partial distance correlation statistic is introduced in section 3 , and section 4 covers the population partial distance covariance and inference for a test of the hypothesis of zero partial distance correlation . examples and applications are given in section 5 , followed by a summary in section 6 . appendix a contains proofs of statements . in this section we summarize the definitions for distance covariance and distance correlation statistics . random vectors are denoted with upper case letters and their sample realizations with lowercase letters . in certain contexts we need to work with a data matrix , which is denoted in boldface , such as @xmath21 or @xmath22 . for example , the sample distance correlation of a realization from the joint distribution of @xmath23 is denoted @xmath24 . the norm @xmath25 denotes the euclidean norm when its argument is in @xmath26 , and we omit the subscript if the meaning is clear in context . a primed random variable denotes an independent and identically distributed ( iid ) copy of the unprimed symbol ; so @xmath27 are iid , etc . the pair @xmath28 denotes two realizations of the random variable @xmath1 . the transpose of @xmath29 is denoted by @xmath30 . other notation will be introduced within the section where it is needed . in this paper a _ random sample _ refers to independent and identically distributed ( iid ) realizations from the underlying joint or marginal distribution . the distance covariance and distance correlation statistics are functions of the double centered distance matrices of the samples . for an observed random sample @xmath31 from the joint distribution of random vectors @xmath32 and @xmath33 , compute the euclidean distance matrices @xmath34 and @xmath35 . define @xmath36 where @xmath37 similarly , define @xmath38 , for @xmath39 . in previous work , we denoted the double centered distance matrix of a sample by a plain capital letter , so that @xmath40 and @xmath41 above were denoted by @xmath42 and @xmath43 . in this paper we work with two different methods of centering , so we have introduced notation to clearly distinguish the two methods . in the first centering above the centered versions @xmath44 have the property that all rows and columns sum to zero . another type of centering , which we will call unbiased or @xmath20-centering and denote by @xmath45 in definition [ def : tildea ] below , has the additional property that all expectations are zero : that is , @xmath46 = 0 $ ] for all @xmath47 . although this centering has a somewhat more complicated formula ( [ au ] ) , we will see in this paper the advantages of the new centering . [ defvn ] the sample distance covariance @xmath48 and sample distance correlation @xmath49 are defined by @xmath50 and @xmath51 respectively , where the squared sample distance variance is defined by @xmath52 although it is obvious , it is perhaps worth noting that @xmath53 and @xmath49 are rigid motion invariant , and they can be defined entirely in terms of the distance matrices , or equivalently in terms of the double centered distance matrices . this fact is important for what follows , because as we will see , one can compute partial distance covariance by operating on certain transformations of the distance matrices , without reference to the original data that generated the matrices . ( these definitions are not invariant to monotone transformation of coordinates , such as ranks . ) if @xmath1 and @xmath2 have finite first moments , the population distance covariance coefficient @xmath54 exists and equals zero if and only if the random vectors @xmath1 and @xmath2 are independent . some of the properties of distance covariance and distance correlation include : 1 . @xmath53 and @xmath24 converge almost surely to @xmath0 and @xmath16 , as @xmath55 . 2 . @xmath56 and @xmath57 if and only if every sample observation is identical . [ th4 - 1 ] @xmath58 . 4 . [ th4 - 2 ] if @xmath59 then there exists a vector @xmath60 , a non - zero real number b and an orthogonal matrix @xmath61 such that @xmath62 , for the data matrices @xmath63 and @xmath64 . a consistent test of multivariate independence is based on the sample distance covariance : large values of @xmath65 support the alternative hypothesis that @xmath1 and @xmath2 are dependent ( see @xcite ) . a high - dimensional distance correlation @xmath66-test of independence was introduced by szkely and rizzo @xcite ; the tests are implemented as ` dcov.test ` and ` dcor.ttest ` , respectively , in the _ energy _ package ( rizzo and szkely , @xcite ) for r ( r core team @xcite ) . the definitions of sample distance covariance and sample distance correlation can be extended to random samples taking values in any , possibly different , metric spaces . for defining the population coefficient we need to suppose more ( see lyons @xcite ) . if @xmath1 and @xmath2 take values in possibly different separable hilbert spaces and both @xmath1 and @xmath2 have finite expectations then it remains true that @xmath67 , and equals zero if and only if @xmath1 and @xmath2 are independent . this implies that the results of this paper can be extended to separable hilbert space valued variables . extensions and population coefficients pdcov and pdcor are discussed in section [ s4 ] . extending lyons @xcite , sejdinovic et al . @xcite discuss the equivalence of distance based and rkhs - based statistics for testing dependence . for theory , background , and further properties of the population and sample coefficients , see szkely , et al . @xcite , szkely and rizzo @xcite , and lyons @xcite , and the references therein . on the weight function applied see also feuerverger @xcite and szkely and rizzo @xcite . for an overview of recent methods for measuring dependence and testing independence of random vectors , including distance covariance , readers are referred to josse and holmes @xcite . we introduce partial distance correlation by first considering the sample coefficient . [ def : tildea ] let @xmath68 be a symmetric , real valued @xmath69 matrix with zero diagonal , @xmath70 . define the @xmath20-centered matrix @xmath71 as follows . let the @xmath72-th entry of @xmath71 be defined by @xmath73 here `` @xmath20-centered '' is so named because as shown below , the corresponding inner product ( [ vu ] ) defines an unbiased estimator of squared distance covariance . [ p : unbiased ] let @xmath74 denote a sample of observations from the joint distribution @xmath75 of random vectors @xmath1 and @xmath2 . let @xmath68 be the euclidean distance matrix of the sample @xmath76 @xmath77 from the distribution of @xmath1 , and @xmath78 be the euclidean distance matrix of the sample @xmath79 from the distribution of @xmath2 . then if @xmath80 , for @xmath81 , @xmath82 is an unbiased estimator of squared population distance covariance @xmath54 . proposition [ p : unbiased ] is proved in appendix ( [ prfunbiased ] ) . it is obvious that @xmath83 if all of the sample observations are identical . more generally , @xmath84 if and only if the @xmath85 sample observations are equally distant or at least @xmath86 of the @xmath85 sample observations are identical . for a fixed @xmath87 , we define a hilbert space generated by euclidean distance matrices of arbitrary sets ( samples ) of @xmath85 points in a euclidean space @xmath7 , @xmath88 . consider the linear span @xmath89 of all @xmath69 distance matrices of samples @xmath90 . let @xmath91 be an arbitrary element in @xmath89 . then @xmath42 is a real valued , symmetric matrix with zero diagonal . let @xmath92 and for each pair of elements @xmath93 , @xmath94 in the linear span of @xmath95 define their inner product @xmath96 if @xmath97 then @xmath98 for any @xmath99 . below in theorem [ thmmds ] it is shown that every matrix @xmath100 is the @xmath20-centered distance matrix of a configuration of @xmath85 points in a euclidean space @xmath7 , where @xmath101 . [ t : hilbert ] the linear span of all @xmath69 matrices @xmath102 is a hilbert space with inner product defined by ( [ vu ] ) . let @xmath103 denote the linear span of @xmath104 . if @xmath105 and @xmath106 , then @xmath107 so @xmath108 . it is also true that @xmath100 implies that @xmath109 , and the zero element is the @xmath69 zero matrix . then since @xmath110 for the inner product , we only need to prove that for @xmath111 and real constants @xmath112 , the following statements hold : 1 . @xmath114 only if @xmath115 . 3 . @xmath116 4 . @xmath117 . statements ( i ) and ( ii ) hold because @xmath118 is proportional to a sum of squares . statements ( iii ) and ( iv ) follow easily from the definition of @xmath103 , @xmath71 , and ( [ vu ] ) . the space @xmath103 is finite dimensional because it is a subspace of the space of all symmetric , zero - diagonal @xmath69 matrices . in what follows , @xmath103 denotes the hilbert space of theorem [ t : hilbert ] with inner product ( [ vu ] ) , and @xmath119 is the norm of @xmath71 . from theorem [ t : hilbert ] a projection operator ( [ proj ] ) can be defined in the hilbert space @xmath103 , @xmath87 , and applied to define partial distance covariance and partial distance correlation for random vectors in euclidean spaces . let @xmath120 and @xmath121 be elements of @xmath103 corresponding to samples @xmath122 and @xmath123 , respectively , and let @xmath124 denote the orthogonal projection of @xmath125 onto @xmath126 and the orthogonal projection of @xmath127 onto @xmath126 , respectively . in case @xmath128 the projections are defined @xmath129 and @xmath130 . clearly @xmath131 and @xmath132 are elements of @xmath103 , their dot product is defined by ( [ vu ] ) , and we can define an estimator of pdcov(@xmath133 ) via projections . [ def.pdcov ] let @xmath134 be a random sample observed from the joint distribution of @xmath135 . the sample partial distance covariance ( pdcov ) is defined by @xmath136 where @xmath131 , and @xmath132 are defined by ( [ proj ] ) , and @xmath137 let @xmath134 be a random sample observed from the joint distribution of @xmath135 . then sample partial distance correlation is defined as the cosine of the angle @xmath138 between the ` vectors ' @xmath131 and @xmath132 in the hilbert space @xmath103 : @xmath139 and otherwise @xmath140 . sample pdcov and pdcor have been defined via projections in the hilbert space generated by @xmath20-centered distance matrices . in this section , we start to consider the interpretation of sample pdcov and sample pdcor . since pdcov is defined as the inner product ( [ pdcov ] ) of two @xmath20-centered matrices , and ( unbiased squared ) distance covariance ( [ vudcov ] ) is computed as inner product , a natural question is the following . are matrices @xmath131 and @xmath132 the @xmath20-centered _ euclidean _ distance matrices of samples of points @xmath141 and @xmath142 , respectively ? if so , then the sample partial distance covariance ( [ pdcov ] ) is distance covariance of @xmath21 and @xmath22 , as defined by ( [ vudcov ] ) . for every sample of points @xmath143 $ ] , @xmath144 , there is a @xmath20-centered matrix @xmath145 in @xmath103 . conversely , given an arbitrary element @xmath146 of @xmath103 , does there exist a configuration of points @xmath147 $ ] in some euclidean space @xmath8 , for some @xmath148 , such that the @xmath20-centered euclidean distance matrix of sample @xmath141 is exactly equal to the matrix @xmath146 ? in this section we prove that the answer is yes : @xmath131 , @xmath132 of ( [ pdcov ] ) , and in general every element in @xmath103 , is the @xmath20-centered distance matrix of some sample of @xmath85 points in a euclidean space . first a few properties of centered distance matrices are established in the following lemma . [ lemmau ] let @xmath71 be a @xmath20-centered distance matrix . then 1 . rows and columns of @xmath71 sum to zero . 2 . that is , if @xmath43 is the matrix obtained by @xmath20-centering an element @xmath150 , @xmath151 . @xmath71 is invariant to double centering . that is , if @xmath43 is the matrix obtained by double centering the matrix @xmath71 , then @xmath152 . [ l4 ] if @xmath112 is a constant and @xmath43 denotes the matrix obtained by adding @xmath112 to the off - diagonal elements of @xmath71 , then @xmath153 . see appendix [ prflemmau ] for proof of lemma [ lemmau ] . as lemma [ lemmau](iv ) is essential for our results , it becomes clear that we can not apply double centering as in the original ( biased ) definition of distance covariance here . the invariance with respect to the constant @xmath112 in ( iv ) holds for @xmath20-centered matrices but it does not hold for double centered matrices . an @xmath69 matrix @xmath19 is called _ euclidean _ if there exist points @xmath154 in a euclidean space such that their euclidean distance matrix is exactly @xmath19 ; that is , @xmath155 , @xmath39 . necessary and sufficient conditions that @xmath19 is euclidean are well known results of classical ( metric ) mds . with the help of lemma [ lemmau ] , and certain results from the theory of mds , we can find for each element @xmath156 a configuration of points @xmath154 in euclidean space such that their @xmath20-centered distance matrix is exactly equal to @xmath146 . a solution to the underlying system of equations to solve for the points @xmath157 is found in schoenberg @xcite and young and householder @xcite . it is a classical result that is well known in ( metric ) multidimensional scaling . mardia , kent , and bibby @xcite summarize the result in theorem 14.2.1 , and provide a proof . for an overview of the methodology see also cox and cox @xcite , gower @xcite , mardia @xcite , and torgerson @xcite . we apply the converse ( b ) of theorem 14.2.1 as stated in mardia , kent , and bibby @xcite , summarized below . let @xmath158 be a dissimilarity matrix and define @xmath159 . form the double centered matrix @xmath160 . if @xmath40 is positive semi - definite ( p.s.d . ) of rank @xmath161 , then a configuration of points corresponding to @xmath162 can be constructed as follows . let @xmath163 be the positive eigenvalues of @xmath40 , with corresponding normalized eigenvectors @xmath164 , such that @xmath165 , @xmath166 . then if @xmath167 is the @xmath168 matrix of eigenvectors , the rows of @xmath167 are @xmath161-dimensional vectors that have interpoint distances equal to @xmath158 , and @xmath169 is the inner product matrix of this set of points . the solution is constrained such that the centroid of the points is the origin . there is at least one zero eigenvalue so @xmath170 . when the matrix @xmath40 is not positive semi - definite , this leads us to the _ additive constant problem _ , which refers to the problem of finding a constant @xmath112 such that by adding the constant to all off - diagonal entries of @xmath158 to obtain a dissimilarity matrix @xmath171 , the resulting double centered matrix is p.s.d . let @xmath172 denote the double centered matrix obtained by double centering @xmath173 , where @xmath174 is the kronecker delta . let @xmath175 denote the matrix obtained by double centering @xmath176 . the smallest value of @xmath112 that makes @xmath172 p.s.d . is @xmath177 , where @xmath178 is the smallest eigenvalue of @xmath179 . then @xmath172 is p.s.d . for every @xmath180 . the number of positive eigenvalues of the p.s.d . double centered matrix @xmath172 determines the dimension required for the representation in euclidean space . however , we require a constant to be added to the elements @xmath181 rather than @xmath182 . that is , we require a constant @xmath112 , such that the dissimilarities @xmath183 are euclidean . the solution by cailliez @xcite is @xmath184 , where @xmath184 is the largest eigenvalue of a @xmath185 block matrix @xmath186 where @xmath187 is the zero matrix and @xmath188 is the identity matrix of size @xmath85 ( see cailliez @xcite or cox and cox ( * ? ? ? 2.2.8 ) for details ) . this result guarantees that there exists a constant @xmath184 such that the adjusted dissimilarities @xmath189 are euclidean . in this case at most @xmath190 dimensions are required ( cailliez ( * ? ? ? * theorem 1 ) ) . finally , given an arbitrary element @xmath146 of @xmath103 , the problem is to find a configuration of points @xmath191 $ ] such that the @xmath20-centered distance matrix of @xmath22 is exactly equal to the element @xmath146 . thus , if @xmath192 we are able to find points @xmath22 such that the euclidean distance matrix of @xmath22 equals @xmath193 , and we need @xmath194 . now since @xmath195 we can apply lemma [ lemmau ] to @xmath146 , and @xmath196 follows from lemma [ lemmau](ii ) and lemma [ lemmau](iv ) . hence , by applying classical mds with the additive constant theorem , and lemma [ lemmau ] ( ii ) and ( iv ) , we obtain the configuration of points @xmath22 such that their @xmath20-centered distances are exactly equal to the element @xmath156 . lemma [ lemmau](iv ) also shows that the inner product is invariant to the constant @xmath112 . this establishes our theorem on representation in euclidean space . [ thmmds ] let @xmath146 be an arbitrary element of the hilbert space @xmath103 of @xmath20-centered distance matrices . then there exists a sample @xmath154 in a euclidean space of dimension at most @xmath190 , such that the @xmath20-centered distance matrix of @xmath154 is exactly equal to @xmath146 . the above details also serve to illustrate why a hilbert space of double centered matrices ( as applied in the original , biased statistic @xmath197 ) is not applicable for a meaningful definition of partial distance covariance . the diagonals of double centered distance matrices are not zero , so we can not get an exact solution via mds , and the inner product would depend on @xmath112 . another problem is that while @xmath197 is always non - negative , the inner product of projections could easily be negative . [ [ methods - for - dissimilarities ] ] methods for dissimilarities + + + + + + + + + + + + + + + + + + + + + + + + + + + in community ecology and other fields of application , it is often the case that only the ( non - euclidean ) dissimilarity matrices are available ( see e.g. the genetic distances of table [ t : maize7 ] ) . suppose that the dissimilarity matrices are symmetric with zero diagonal . an application of theorem [ thmmds ] provides methods for this class of non - euclidean dissimilarities . in this case , theorem [ thmmds ] provides that there exists samples in euclidean space such that their @xmath20-centered euclidean distance matrices are equal to the dissimilarity matrices . thus , to apply distance correlation methods to this type of problem , one only needs to obtain the euclidean representation . existing software implementations of classical mds can be applied to obtain the representation derived above . for example , classical mds based on the method outlined in mardia @xcite is implemented in the r function ` cmdscale ` , which is in the _ stats _ package for r. the ` cmdscale ` function includes options to apply the additive constant of cailliez @xcite and to specify the dimension . the matrix of points @xmath22 is returned in the component ` points ` . for an exact representation , we can specify the dimension argument equal to @xmath190 . [ ex0 ] to illustrate application of theorem [ thmmds ] for non - euclidean dissimilarities , we computed the bray - curtis dissimilarity matrix of the ` iris ` _ setosa _ data , a four - dimensional data set available in r. the bray - curtis dissimilarity defined in cox and cox ( * ? ? ? * table 1.1 ) as @xmath198 is not a distance since it does not satisfy the triangle inequality . a bray - curtis method is available in the ` distance ` function of the _ ecodist _ package for r @xcite . we find a configuration of 50 points in @xmath199 that have @xmath20-centered distances equal to the @xmath20-centered dissimilarities . the mds computations are handled by the r function ` cmdscale ` . function ` ucenter ` , which implements @xmath20-centering , is in the r package _ pdcor _ @xcite . .... > x < - iris[1:50 , 1:4 ] > iris.d < - distance(x , method="bray - curtis " ) > au < - ucenter(iris.d ) > v < - cmdscale(as.dist(au ) , k=48 , add = true)$points .... the points ` v ` are a @xmath200 data matrix , of 50 points in @xmath199 . next we compare the @xmath20-centered distance matrix of the points ` v ` with the original object ` au ` from @xmath103 : .... > vu < - ucenter(v ) > all.equal(au , vu ) [ 1 ] true .... the last line of output shows that the points ` v ` returned by ` cmdscale ` have @xmath20-centered distance matrix ` vu ` equal to our original element ` au ` of the hilbert space . @xmath201 example [ ex0 ] shows that the sample distance covariance can be defined for dissimilarities via the inner product in @xmath103 . alternately one can compute @xmath202 , where @xmath21 , @xmath22 are the euclidean representations corresponding to the two @xmath20-centered dissimilarity matrices that exist by theorem [ thmmds ] . using the corresponding definitions of distance variance , sample distance correlation for dissimilarities is well defined by ( [ rcorrected ] ) or @xmath203 . similarly one can define pdcov and pdcor when one or more of the dissimilarity matrices of the samples is not euclidean distance . however , as in the case of euclidean distance , we need to define the corresponding population coefficients , and develop a test of independence . for the population definitions see section [ s4 ] . let @xmath204 where @xmath205 are the @xmath20-centered distance matrices of the samples @xmath29 and @xmath206 , and @xmath207 . the statistics @xmath208 and @xmath209 take values in @xmath210 $ ] , but they are measured in units comparable to the squared distance correlation @xmath211 . [ p : pdcor2 ] if @xmath212 , a computing formula for @xmath213 in definition ( [ pdcor2 ] ) is @xmath214 see appendix [ prfrxy ] for a proof . equation ( [ r.alt ] ) provides a simple and familiar form of computing formula for the partial distance correlation . the computational algorithm is easily implemented , as summarized below . _ algorithm to compute partial distance correlation @xmath215 from euclidean distance matrices @xmath216 , @xmath217 , and @xmath218 _ : 1 . compute the @xmath20-centered matrices @xmath71 , @xmath219 , and @xmath121 , using @xmath220 and @xmath221 . 2 . compute inner products and norms using @xmath222 and @xmath208 , @xmath223 , and @xmath224 using @xmath225 3 . if @xmath226 and @xmath227 @xmath228 otherwise apply the definition ( [ pdcor2 ] ) . note that it is not necessary to explicitly compute the projections , when ( [ r.alt ] ) is applied . the above algorithm has a straightforward translation to code ; see e.g. the _ pdcor _ package @xcite for an implementation in r. the population distance covariance has been defined in terms of the joint and marginal characteristic functions of the random vectors . here we give an equivalent definition following lyons @xcite , who generalizes distance correlation to separable hilbert spaces . instead of starting with the distance matrices @xmath34 and @xmath229 the starting point of the population definition are the bivariate distance functions @xmath230 and @xmath231 where @xmath28 are realizations of the random variables @xmath1 and @xmath232 are realizations of the random variable @xmath2 . we can also consider the random versions . let @xmath233 and @xmath234 be random variables with finite expectations . the random distance functions are @xmath235 and @xmath236 . here the primed random variable @xmath1 denotes an independent and identically distributed ( iid ) copy of the variable @xmath1 , and similarly @xmath237 are iid . the population operations of double centering involves expected values with respect to the underlying population random variable . for a given random variable @xmath1 with cdf @xmath238 , we define the corresponding _ double centering function with respect to _ @xmath1 as @xmath239 provided the integrals exist . here @xmath240 is a real valued function of two realizations of @xmath1 , and the subscript @xmath1 references the underlying random variable . similarly for @xmath241 iid with cdf @xmath238 , we define the random variable @xmath242 as an abbreviation for @xmath243 , which is a random function of @xmath244 . similarly we define @xmath245 and the random function @xmath246 . now for @xmath247 iid , and @xmath237 iid , such that @xmath248 have finite expectations , the population distance covariance @xmath249 is defined by @xmath250.\ ] ] the definition ( [ e : dcov2 ] ) of @xmath54 is equivalent to the original definition ( [ e : aw ] ) . however , as we will see in the next sections , ( [ e : dcov2 ] ) is an appropriate starting point to develop the corresponding definition of pdcov and pdcor population coefficients . more generally , we can consider _ dissimilarity functions _ @xmath251 . in this paper , a dissimilarity function is a symmetric function @xmath252 with @xmath253 . the corresponding random dissimilarity functions @xmath254 are random variables such that @xmath255 , and @xmath256 . double centered dissimilarities are formally defined by the same equations as double centered distances in ( [ e : hatax ] ) . the following lemma establishes that linear combinations of double - centered dissimilarities are double - centered dissimilarities . [ lemma3 ] suppose that @xmath233 , @xmath234 , @xmath257 is a dissimilarity on @xmath258 , and @xmath259 is a dissimilarity on @xmath260 . let @xmath261 and @xmath262 denote the dissimilarity obtained by double - centering @xmath257 and @xmath259 , respectively . then if @xmath263 and @xmath264 are real scalars , @xmath265 where @xmath266 \in \mathbb r^p \times \mathbb r^q$ ] , and @xmath267 is the result of double - centering @xmath268 . see appendix [ prflemma3 ] for the proof . the linear span of double - centered distance functions @xmath261 is a subspace of the space of double - centered dissimilarity functions with the property @xmath269 . in this case all integrals in ( [ e : hatax ] ) and @xmath270 are finite if @xmath1 has finite expectation . the linear span of the random functions @xmath270 for random vectors @xmath1 with @xmath271 is clearly a linear space , such that the linear extension of ( [ e : dcov2 ] ) @xmath272 = \mathcal v^2(x , y)\ ] ] to the linear span is an inner product space or pre - hilbert space ; its completion with respect to the metric arising from its inner product @xmath273\ ] ] ( and norm ) is a hilbert space which we denote by @xmath274 . [ defpdcov ] introduce the scalar coefficients @xmath275 if @xmath276 define @xmath277 . the double - centered projections of @xmath270 and @xmath278 onto the orthogonal complement of @xmath279 in hilbert space @xmath274 are defined @xmath280 or in short @xmath281 and @xmath282 , where @xmath279 denotes double - centered with respect to the random variable @xmath283 . the population partial distance covariance is defined by the inner product @xmath284.\ ] ] [ defpdcor ] population partial distance correlation is defined @xmath285 where @xmath286 if @xmath287 we define @xmath288 . note that if @xmath277 , we have @xmath289 = \mathcal v^2(x , y ) , $ ] and @xmath290 . the population coefficient of partial distance correlation can be evaluated in terms of the pairwise distance correlations using formula ( [ pdcor.pop ] ) below . theorem [ t : pdcor.pop ] establishes that ( [ pdcor.pop ] ) is equivalent to definition [ defpdcor ] , and therefore serves as an alternate definition of population pdcor . [ t : pdcor.pop ] the following definition of population partial distance correlation is equivalent to definition [ defpdcor ] . @xmath291 where @xmath16 denotes the population distance correlation . the proof of theorem [ t : pdcor.pop ] is given in appendix [ prfthm3 ] partial distance correlation is a scalar quantity that captures dependence , while conditional inference ( conditional distance correlation ) is a more complex notion as it is a function of the condition . although one might hope that partial distance correlation @xmath292 if and only if @xmath1 , and @xmath2 are conditionally independent given @xmath283 , this is , however , not the case . both notions capture _ overlapping _ aspects of dependence , but mathematically they are not equivalent . the following examples illustrate that @xmath293 is not equivalent to conditional independence of @xmath1 and @xmath2 given @xmath283 . suppose that @xmath294 are iid standard normal variables , @xmath295 , @xmath296 , and @xmath297 . then @xmath298 , @xmath299 , @xmath300 , and @xmath1 and @xmath2 are conditionally independent of @xmath283 . one can evaluate @xmath301 by applying ( [ pdcor.pop ] ) and the following result . suppose that @xmath75 are jointly bivariate normal with correlation @xmath302 . then by ( * ? ? ? * theorem 7(ii ) ) , @xmath303 in this example @xmath304 , @xmath305 , and @xmath306 . in the other direction we can construct a trivariate normal @xmath135 such that @xmath307 but @xmath308 . according to baba , shibata and sibuya @xcite , if @xmath309 are trivariate normal , then pcor@xmath310 if and only if @xmath1 and @xmath2 are conditionally independent given @xmath283 . a specific numerical example is as follows . by inverting equation ( [ cor2dcor ] ) , given any @xmath311 , one can solve for @xmath312 in @xmath313 $ ] . when @xmath314 , and @xmath315 , the nonnegative solutions are @xmath316 , @xmath317 . the corresponding @xmath318 correlation matrix is positive definite . let @xmath309 be trivariate normal with zero means , unit variances , and correlations @xmath316 , @xmath317 . then pcor@xmath319 @xmath320 and therefore conditional independence of @xmath1 and @xmath2 given @xmath283 does not hold , while @xmath321 is exactly zero . in the non - gaussian case , it is not true that zero partial correlation is equivalent to conditional independence , while partial distance correlation has the advantages that it can capture non - linear dependence , and is applicable to vector valued random variables . we have defined partial distance correlation , which is bias corrected and has an informal interpretation , but we also want to determine whether @xmath322 is significantly different from zero , which is not a trivial problem . although theory and implementation for a consistent test of multivariate independence based on distance covariance is available @xcite , clearly the implementation of the test by randomization is not directly applicable to the problem of testing for zero partial distance correlation . since @xmath323 if and only if @xmath324 , we develop a test for @xmath325 vs @xmath326 based on the inner product @xmath327 . unlike the problem of testing independence of @xmath1 and @xmath2 , however , we do not have the distance matrices or samples corresponding to the projections @xmath131 and @xmath132 . for a test we need to consider that @xmath131 and @xmath132 are arbitrary elements of @xmath103 . we have proved ( theorem [ thmmds ] ) that @xmath328 can be represented as the @xmath20-centered distance matrix of some configuration of @xmath85 points @xmath21 in a euclidean space @xmath7 , @xmath101 , and that @xmath329 has such a representation @xmath22 in @xmath8 , @xmath330 . hence , a test for @xmath324 can be defined by applying the distance covariance test statistic @xmath202 . the resulting test procedure is practical to apply , with similar computational complexity as the original dcov test of independence . one can apply the test of independence implemented by permutation bootstrap in the ` dcov.test ` function of the energy package @xcite for r. alternately a test based on the inner product ( [ pdcov ] ) is implemented in the _ pdcor _ @xcite package . in this section we summarize simulation results for tests of the null hypothesis of zero partial distance correlation and two applications . we compared our simulation results with two types of tests based on linear correlation . methods for dissimilarities are illustrated using data on genetic distances for samples of maize in example [ ex6 ] , and variable selection is discussed in example [ ex7 ] . the test for zero partial distance correlation is a test of whether the inner product @xmath331 of the projections is zero . we obtain the corresponding samples @xmath332 such that the @xmath20-centered distances of @xmath21 and @xmath22 are identical to @xmath328 and @xmath329 , respectively , using classical ( metric ) mds . we used the ` cmdscale ` function in r to obtain the samples @xmath21 and @xmath22 in both cases . the _ pdcov _ test applies the inner product of the double centered distance matrices of @xmath21 and @xmath22 as the test statistic . alternately , one could apply the original _ dcov _ test , to the joint sample @xmath333 . these two test statistics are essentially equivalent , but the _ pdcov _ test applies an unbiased statistic defined by ( [ vu ] ) , while the _ dcov _ test applies a biased statistic , @xmath334 ( [ e : anxy ] ) . both tests are implemented as permutation tests , where for a test of @xmath335 the sample indices of the @xmath1 sample are randomized for each replicate to obtain the sampling distribution of the test statistic under the null hypothesis . the term ` permutation test ' is sometimes restricted to refer to the exact test , which is implemented by generating all possible permutations . except for very small samples , it is not computationally feasible to generate all permutations , so a large number of randomly generated permutations are used . this approach , which we have implemented , is sometimes called a randomization test to distinguish it from an exact permutation test . in our simulations the _ pdcov _ and _ dcov _ tests for zero partial distance correlation were equivalent under null or alternative hypotheses in the sense that the type 1 error rates and estimated power agreed to within one standard error . in power comparisons , therefore , we reported only the _ pdcov _ result . the linear partial correlation @xmath336 measures the partial correlation between one dimensional data vectors @xmath29 and @xmath206 with @xmath123 removed ( or controlling for @xmath123 ) . the sample partial correlation coefficient is @xmath337 where @xmath338 denotes the linear ( pearson ) sample correlation . the partial correlation test is usually implemented as a @xmath66 test ( see e.g. legendre @xcite ) . in examples where @xmath29 , @xmath206 , @xmath123 are one dimensional , we have included the partial correlation @xmath66-test in comparisons . however , for small samples and some non - gaussian data , the type 1 error rate is inflated . in cases where type 1 error rate of _ pcor _ was not controlled we did not report power . the partial mantel test is a test of the hypothesis that there is a linear association between the distances of @xmath1 and @xmath2 , controlling for @xmath283 . this extension of the mantel test @xcite was proposed by smouse , et al . @xcite for a partial correlation analysis on three distance matrices . the mantel and partial mantel tests are commonly applied in community ecology ( see e.g. legendre and legendre @xcite ) , population genetics , sociology , etc . let @xmath339 denote the upper triangles of the @xmath69 distance matrices of the @xmath1 , @xmath2 and @xmath283 samples , respectively . then the partial mantel test statistic is the sample linear partial correlation between the @xmath340 elements of @xmath341 and @xmath342 controlling for @xmath343 . let @xmath344 be the corresponding data vectors obtained by representing @xmath339 as vectors . the partial mantel statistic is @xmath345 , computed using formula ( [ pcor.stat ] ) . since @xmath346 are not iid samples , the usual @xmath66 test is not applicable , so the partial mantel test is usually implemented as a permutation ( randomization ) test . see legendre @xcite for a detailed algorithm and simulation study comparing different methods of computing a partial mantel test statistic . based on the results reported by legendre , we implemented the method of permutation of the raw data . the algorithm is given in detail on page 44 by legendre @xcite , and it is very similar to the algorithm we have applied for the _ energy _ tests ( _ pdcov _ and _ dcov _ tests for zero pdcor ) . this method ( permutation of raw data ) for the partial mantel test is implemented in the _ ecodist _ package @xcite and also the _ vegan _ package @xcite for r. for simulations , the ` mantel ` function in the _ ecodist _ package , which is implemented mainly in compiled code , is much faster than the ` mantel ` or ` mantel.partial ` functions in the _ vegan _ package , which are implemented in r. [ covdist ] both the _ pdcov _ and partial mantel ( _ mantel _ ) tests are based on distances . one may ask `` is distance covariance different or more general than covariance of distances ? '' the answer is yes ; it can be shown that @xmath347 where @xmath348 and @xmath349 are iid . the _ dcov _ tests are tests of independence of @xmath1 and @xmath2 ( @xmath350 ) , while the mantel test is a test of the hypothesis @xmath351 . an example of dependent data such that their distances are uncorrelated but @xmath352 is given e.g. by lyons @xcite . thus , distance covariance tests are more general than mantel tests , in the sense that distance covariance measures all types of departures from independence . in each permutation test @xmath353 replicates are generated and the estimated @xmath161-value is computed as @xmath354 where @xmath355 is the indicator function , @xmath356 is the observed value of the test statistic , and @xmath357 is the statistic for the @xmath358-th sample . the test is rejected at significance level @xmath359 if @xmath360 . the partial correlation test ( _ pcor _ ) is also included for comparison it is implemented as a t - test @xcite . type 1 error rates and estimated power are determined by a simulation size of 10,000 tests in each case ; for @xmath361 the number of tests is 100,000 . the standard error is at most 0.005 ( 0.0016 for @xmath361 ) . [ ex1a ] this example is a comparison of type 1 error on uncorrelated standard normal data . the vectors @xmath1 , @xmath2 , and @xmath283 are each iid standard normal . results summarized in table [ t : ex1a ] show that type 1 error rates for _ pdcov _ , _ dcov _ , and _ partial mantel _ tests are within two se of the nominal significance level , while type 1 error rates for pcor are inflated for @xmath362 . .example [ ex1a ] : type 1 error rates at nominal significance level @xmath359 for uncorrelated standard trivariate normal data.[t : ex1a ] [ cols=">,>,>,>,>,>,>,>,>,>,>",options="header " , ] here the goal was to fit a linear model to predict the response _ lpsa _ given one or more of the predictor variables above . the train / test set indicator is in the last column of the data set . for comparison with the discussion and analysis in hastie et al . @xcite , we standardized each variable , and used the training set of 67 observations for variable selection . feature screening by distance correlation has been investigated by li et al . @xcite . in this example we introduce a partial distance correlation criterion for variable selection . for simplicity , we implement a simple variant of forward selection . this criterion , when applied for a linear model , can help to identify possible important variables that have strong non - linear association with the response , and thus help researchers to improve a linear model by transforming variables or improve prediction by extending to a nonlinear model . in the initial step of pdcor forward selection , the first variable to enter the model is the variable @xmath363 for which distance correlation @xmath364 with response @xmath206 is largest . after the initial step , we have a model with one predictor @xmath363 , and we compute pdcor@xmath365 , for the variables @xmath366 not in the model , then select the variable @xmath367 for which pdcor@xmath365 is largest . then continue , at each step computing @xmath368 for every variable @xmath363 not yet in the model , where @xmath11 is the vector of predictors currently in the model . the variable to enter next is the one that maximizes @xmath369 . according to the pdcor criterion , the variables selected enter the model in the order : _ lcavol , lweight , svi , gleason , lbph , pgg45_. if we set a stopping rule at 5% significance for the pdcor coefficient , then we stop after adding _ lbph _ ( or possibly after adding _ pgg45 _ ) . at the step where _ gleason _ and _ lbph _ enter , the @xmath161-values for significance of pdcor are 0.016 and 0.001 , respectively . the @xmath161-value for _ pgg45 _ is approximately 0.05 . the models selected by pdcor , best subset method , and lasso ( hastie et al . * table 3.3 , p. 63 ) ) are : @xmath370 the order of selection for ordinary forward stepwise selection ( cp ) is _ lcavol , lweight , svi , lbph , pgg45 , age , lcp , gleason_. comparing pdcor forward selection with lasso and forward stepwise selection , we see that the results are similar , but _ gleason _ is not in the lasso model and enters last in the forward stepwise selection , while it is the fourth variable to enter the pdcor selected model . the raw gleason score is an integer from 2 to 10 which is used to measure how aggressive is the tumor , based on a prostate biopsy . plotting the data ( see figure [ f : prostate1 ] ) we can observe that there is a strong _ non - linear _ relation between _ gleason _ and _ lpsa_. this example illustrates that partial distance correlation has practical use in variable selection and in model checking . if we were using pdcor only to check the model selected by another procedure , it would show in this case that there is some nonlinear dependence remaining between the response and the predictors excluded from the lasso model or the best subsets model . using the pdcor selection approach , we also learn which of the remaining predictors may be important . finally , it is useful to recall that pdcor has more flexibility to handle predictors that are multi - dimensional . one may want groups of variables to enter or leave the model as a set . it is often the case that when dimension of the feature space is high , many of the predictor variables are highly correlated . in this case , methods such as partial least squares are sometimes applied , where a small set of derived predictors ( linear combinations of features ) become the predictor variables . using methods of partial distance correlation , one could evaluate the subsets ( as predictor sets ) without reducing the multivariate observation to a real number via linear combination . the pdcor coefficient can be computed for multivariate predictors ( and for multivariate response ) . partial distance covariance and partial distance correlation coefficients are measures of dependence of two random vectors @xmath1 , @xmath2 , controlling for a third random vector @xmath283 , where @xmath1 , @xmath2 , and @xmath283 are in arbitrary , not necessarily equal dimensions . these definitions are based on an unbiased distance covariance statistic , replacing the original double centered distance matrices of the original formula with @xmath20-centered distance matrices . the @xmath20-centered distance covariance is the inner product in the hilbert space @xmath103 of @xmath20-centered distance matrices for samples of size @xmath85 , and it is unbiased for the squared population distance covariance . each element in the hilbert @xmath103 is shown to be the @xmath20-centered distance matrix of some configuration of @xmath85 points in a euclidean space @xmath7 , @xmath101 . the proof and the solution are obtained through application of theory of classical mds . this allows one to define dcor and dcov for dissimilarity matrices , and provides a statistically consistent test of independence based on the inner product statistic . this methodology also provides a similar test of the hypothesis of zero partial distance correlation based on the inner product . all pdcor and pdcov methods have been implemented and simulation studies illustrate that the tests control the type 1 error rate at its nominal level . power performance was superior compared with power of partial correlation and partial mantel tests for association . methods have been implemented and illustrated for non - euclidean dissimilarity matrices . a ` pdcor forward selection ' method was applied to select variables for a linear model , with a significance test as the stopping rule . more sophisticated selection methods will be investigated in future work . with the flexibility to handle multivariate response and/or multivariate predictor , pdcor offers a new method to extend the variable selection toolbox . software is available in the _ energy _ @xcite package for r , and the _ pdcor _ @xcite package for r. proposition [ p : unbiased ] asserts that @xmath371 is an unbiased estimator of the population coefficient @xmath54 . when the terms of @xmath371 are expanded , we have a linear combination of terms @xmath372 . the expected values of these terms differ according to the number of subscripts that agree . define @xmath373=e[|x - x'| ] , \qquad \beta : = e[b_{kl}]=e[|y - y'| ] , \qquad k \neq l,\\ \delta & : = e[a_{kl}b_{kj } ] = e[|x - x'||y - y''| ] , \qquad j , k , l \ ; \mathrm { distinct},\\ \gamma & : = e[a_{kl}b_{kl}]=e[|x - x'||y - y'| ] , \qquad k \neq l,\end{aligned}\ ] ] where @xmath348 @xmath349 are iid . due to symmetry , the expected value of each term in the expanded expression @xmath371 is proportional to one of @xmath374 , @xmath375 , or @xmath376 , so the expected value of @xmath371 can be written as a linear combination of @xmath377 , @xmath375 , and @xmath376 . the population coefficient can be written as ( see ( * ? ? ? * theorem 7 ) ) @xmath378 + e[|x - x'|]\,e[|y\!-\!y'| ] - 2e[|x - x'||y\!-\!y''| ] \\ & = \gamma + \alpha \beta - 2 \delta.\end{aligned}\ ] ] let us adopt the notation @xmath379 , @xmath380 , and @xmath381 , where @xmath382 , @xmath383 , and @xmath384 . similarly define @xmath385 , and @xmath386 . then @xmath387 @xmath388 let @xmath389 then @xmath390 it is easy to see that @xmath391=n(n-1)\gamma$ ] . by expanding the terms of @xmath392 and @xmath393 , and combining terms that have equal expected values , one can obtain @xmath394 = n(n-2)\{(n-2)(n-3)\alpha \beta + 2 \gamma + 4(n-2)\delta\},\ ] ] and @xmath395 = n(n-1)\{(n-2)\delta + \gamma\}.\ ] ] then @xmath396 & = \frac{1}{n(n-3 ) } e\left [ t_1 - \frac{t_2}{(n-1)(n-2)^2 } -\frac{2t_3}{n-2}\right ] \\ & = \frac{1}{n(n-3 ) } \left\ { \frac{n^3 - 5n^2 + 6n}{n-2 } \gamma + n(n-3)\alpha \beta + ( 6n-2n^2)\delta \right \ } \\ & = \gamma + \alpha \beta - 2 \delta = \mathcal v^2(x , y).\end{aligned}\ ] ] here the inner product is ( [ vu ] ) and the ` vectors ' are @xmath20-centered elements of the hilbert space @xmath397 . equation ( [ r.alt ] ) can be derived from ( [ pdcor2 ] ) using similar algebra with inner products as used to obtain the representation @xmath398 for the linear correlation @xmath399 ( see e.g. huber @xcite ) . the details for @xmath400 are as follows . if either @xmath401 or @xmath402 then @xmath403 so by definition @xmath404 , @xmath405 , and ( [ r.alt ] ) is also zero . if @xmath406 but @xmath407 , then @xmath408 . in this case @xmath409 and @xmath410 , so that @xmath411 which equals expression ( [ r.alt ] ) since @xmath408 . suppose that none of @xmath412 are zero . then @xmath413 similarly , in the denominator of ( [ pdcor2 ] ) we have @xmath414 thus , if the denominator of ( [ pdcor2 ] ) is not zero , the factor @xmath415 cancels from the numerator and denominator and we obtain ( [ r.alt ] ) . the sum of the first row of @xmath71 is @xmath416 similarly each of the rows of @xmath71 sum to zero , and by symmetry the columns also sum to zero , which proves statement ( i ) . statements ( ii ) and ( iii ) follow immediately from ( i ) . in ( iv ) let @xmath417 be the @xmath418-th element of @xmath43 . then @xmath419 , and for @xmath420 , @xmath421 . hence @xmath422 , @xmath420 and @xmath423 . therefore @xmath424 which proves ( iv ) . it is clear that @xmath425 is identical to the double - centered dissimilarity @xmath426 . it remains to show that the sum of two arbitrary elements @xmath427 is a double - centered dissimilarity function . let @xmath266 \in \mathbb r^p \times \mathbb r^q$ ] . consider the dissimilarity function @xmath428 , where @xmath429 $ ] and @xmath430 $ ] . then @xmath431df_t(t ' ) \\ & \qquad - \int [ a(x , x')+b(y , y')]df_t(t ) \\ & \qquad + \iint [ a(x , x')+b(y , y')]df_t(t')df_t(t ) \\ & = a(x , x ' ) + b(y , y ' ) -e[a(x , x')+b(y , y ' ) ] \\ & \qquad - e[a(x , x')+b(y , y ' ) ] + e[a(x , x')+b(y , y ' ) ] \\&= a_x(x , x ' ) + b_y(y , y').\end{aligned}\ ] ] it is straightforward to check the first three special cases . case ( i ) : if @xmath283 is constant a.s . then @xmath432 , and @xmath433 . in this case both @xmath321 and ( [ pdcor.pop ] ) simplify to @xmath311 . case ( ii ) : if @xmath1 or @xmath2 is constant a.s . and @xmath283 is not a.s . constant , then we have zero in both definition [ defpdcor ] and ( [ pdcor.pop ] ) . case ( iii ) : if none of the variables @xmath434 are a.s . constant , but @xmath435 or @xmath436 , then @xmath437 by definition and @xmath438 or @xmath439 . thus ( [ pdcor.pop ] ) is also zero by definition . case ( iv ) : in this case none of the variables @xmath434 are a.s . constant , and @xmath440 . thus @xmath441 = ( a_x-\alpha c_z , b_y-\beta c_z ) \\ \notag & = ( a_x , b_y ) - \alpha ( b_y , c_z ) - \beta ( a_x , c_z ) + \alpha \beta ( c_z , c_z ) \\ & = \notag \mathcal v^2(x , y ) - \frac{\mathcal v^2(x , z)\mathcal v^2(y , z)}{\mathcal v^2(z , z ) } \\ & \notag \qquad - \frac{\mathcal v^2(x , z)\mathcal v^2(y , z)}{\mathcal v^2(z , z ) } + \frac{\mathcal v^2(x , z)\mathcal v^2(y , z ) \mathcal v^2(z , z ) } { \mathcal v^2(z , z ) \mathcal v^2(z , z ) } \\ \notag & = \mathcal v^2(x , y ) - \frac{\mathcal v^2(x , z)\mathcal v^2(y , z)}{\mathcal v^2(z , z)}.\end{aligned}\ ] ] similarly @xmath442 and @xmath443 hence , using ( [ e : iv-1])([e : iv-3 ] ) @xmath444 thus , in all cases definition [ defpdcor ] and ( [ pdcor.pop ] ) coincide . the authors thank russell lyons for several helpful suggestions and comments on a preliminary draft of this paper . baba , k. , shibata , r. and sibuya , m. ( 2004 ) . partial correlation and conditional correlation as measures of conditional independence . _ australian and new zealand journal of statistics _ * 46 * ( 4 ) : 657664 . kong , j. , klein , b. e. k. , klein , r. , lee , k. , and wahba , g. ( 2012 ) . using distance correlation and ss - anova to assess associations of familial relationships , lifestyle factors , diseases , and mortality , _ proc . of the national acad . of sciences _ , * 109 * ( 50 ) , 2035220357 . doi:10.1073/pnas.1217269109 jari oksanen , f. guillaume blanchet , roeland kindt , pierre legendre , peter r. minchin , r. b. ohara , gavin l. simpson , peter solymos , m. henry h. stevens and helene wagner ( 2013 ) . vegan : community ecology package . r package version 2.0 - 7 . http://cran.r-project.org/package=vegan reif , j. c. , a. e. melchinger , x. c. xia , m. l. warburton , d. a. hoisington , s. k. vasal , g. srinivasan , m. bohn , and m. frisch ( 2003 ) . genetic distance based on simple sequence repeats and heterosis in tropical maize populations , _ crop science _ * 43 * , no . 4 , 12751282 . schoenberg , i. j. ( 1935 ) . remarks to maurice frchet s article `` sur la definition axiomatique dune classe despace distancis vectoriellement applicable sur lespace de hilbert . '' _ ann . _ , * 36 * , 724732 . stamey , t. a. , kabalin , j. n. , mcneal , j. e. , johnstone , i. m. , freiha , f. , redwine , e. a. and yang , n. ( 1989 ) . prostate specific antigen in the diagnosis and treatment of adenocarcinoma of the prostate : ii . radical prostatectomy treated patients , _ journal of urology _ * 141*(5 ) , 10761083 . wermuth , n. and cox , d. r. ( 2013 ) . concepts and a case study for a flexible class of graphical markov models . in becker , c. , fried , r. and kuhnt , s. ( eds . ) _ robustness and complex data structures . festschrift in honour of ursula gather . _ springer , heidelberg , 331350 .

we analyze archival _ hst_/stis / fuv - mama imaging and spectroscopy of 13 compact star clusters within the circumnuclear starburst region of m83 , the closest such example . we compare the observed spectra with semi - empirical models , which are based on an empirical library of galactic o and b stars observed with _ iue _ , and with theoretical models , which are based on a new theoretical uv library of hot massive stars computed with wm - basic . the models were generated with starburst99 for metallicities of and , and for stellar imfs with upper mass limits of 10 , 30 , 50 , and 100m . we estimate the ages and masses of the clusters from the best fit model spectra , and find that the ages derived from the semi - empirical and theoretical models agree within a factor of 1.2 on average . a comparison of the spectroscopic age estimates with values derived from _ multi - band photometry shows a similar level of agreement for all but one cluster . the clusters have a range of ages from about 3 to 20myr , and do not appear to have an age gradient along m83 s starburst . clusters with strong p - cygni profiles have masses of a fewm , seem to have formed stars more massive thanm , and are consistent with a kroupa imf fromm . field regions in the starburst lack p - cygni profiles and are dominated by b stars .

1011.4449

in the context of disk galaxies , a nuclear or circumnuclear ring is a region of large gas surface density and enhanced star formation , located within 2 kpc of the galactic nucleus . @xcite distinguish between nuclear disks and nuclear rings , by setting the maximum width of the ring to half the ring radius . the most compact rings ( ultra - compact nuclear rings , ucnrs ) have radii smaller than 200 pc @xcite . the atlas of images of nuclear rings @xcite is the most complete atlas of galactic nuclear rings to date and includes 101 barred and unbarred disk galaxies with rings . the dust nuclear rings around elliptical and s0 galaxies constitute a different class of galactic ring with no associated star formation . the present work is not concerned with the latter rings . circumnuclear rings in disk galaxies give rise to starbursts @xcite , which observationally are ring- or arc - shaped . note that the term `` starburst '' does not have a general definition @xcite . a `` circumnuclear starburst '' is a region composed of star condensations with individual masses ranging from a few@xmath6 to greater than @xmath7m@xmath2 , bright in the ultraviolet ( uv ) because they have recently emerged from their birth clouds ( the nuclear galactic ring ) , and contain hot and massive o and/or b stars . proposed scenarios for the origin of star - forming galactic rings are galaxy collisions or mergers , accretion of intergalactic gas , and resonances caused by the actions of rotating bars or other non - axisymmetric disturbances on the motions of disk gas clouds @xcite . note that in their analysis of h@xmath8 and continuum images of 327 local disk galaxies , @xcite found no significant increase in the central concentration of star formation as a result of the presence of a close companion . on the other hand , there is strong observational evidence for a connection between bars and circumnuclear starbursts @xcite . in addition , non - starburst barred galaxies may eventually become starburst barred galaxies @xcite . the location of nuclear rings in barred galaxies may be set by inner lindblad resonances ( ilrs ) , which can prevent gas from flowing further in . such resonances can cause gas to accumulate in circumnuclear rings where clusters can form @xcite . alternative models for the location of starbursts in barred galaxies can be found in @xcite and @xcite , while @xcite have a model for the migration of circumnuclear star clusters and nuclear rings . circumnuclear starbursts in barred galaxies are perhaps the most numerous class of nearby starburst regions @xcite . their detailed study can provide a stepping stone for studies of starbursts at higher redshifts . indeed , local starburst galaxies ( including circumnuclear starbursts ) , show star formation rate densities @xcite , uv colors @xcite , and spectral morphologies @xcite , similar to those of high redshift lyman - break galaxies ( also see @xcite ) . determining the ages , masses , and stellar initial mass functions ( imfs ) of individual star clusters within circumnuclear starbursts is important for understanding the relations between ( 1 ) galaxy dynamics and interactions and the origin of circumnuclear star clusters ; ( 2 ) massive star feedback and the fueling of active galactic nuclei ( agn , @xcite ) ; and ( 3 ) young massive star clusters and globular clusters @xcite . in this study , we determine the ages and masses , and constrain the imfs of 13 star clusters in messier 83 ( m83 , ngc 5236 ) , which hosts the nearest example of a circumnuclear starburst in a barred galaxy . m83 is a nearby ( [email protected] , @xcite ) , southern ( dec[j2000]@xmath10 ) , nearly face - on ( @xmath11 , @xcite ) , sab(s)c grand - design spiral galaxy @xcite , with an optical disk spanning @xmath12 ( @xmath13kpc@xmath14kpc ) . @xcite have estimated that at the distance of m83 , a central black hole of moderate size and luminosity would be detectable . however , the galaxy shows no indication from radio or x - ray observations of hosting an agn @xcite . m83 has a close dynamical companion in the dwarf irregular galaxy ngc 5253 @xcite , which contains a starburst nucleus of @xmath15pc in size ( @xcite , @xcite , @xcite ) . m83 s arc - shaped circumnuclear starburst has been observed at wavelengths ranging from the radio to the x - rays ( see @xcite for references ) . it spans about 200pc in length and 35pc in thickness , and it is located between two rings of dusty gaseous material which may be coincident with two inner lindblad resonances @xcite . the starburst qualifies as an ucnr ( ring radius of @xmath16 pc ) , and as shown in this work , it breaks up into about 20 compact fuv bright star clusters . two proposed scenarios for its origin are , the interaction of m83 with its neighbor ngc 5253 about 1gyr ago @xcite , and/or the merging of m83 with a dwarf satellite in the past . the former scenario is supported by the `` finger '' of h i gas projecting outward from the halo of m83 toward ngc 5253 @xcite , and by the presence of off - disk h i clouds in m83 @xcite . the merger scenario is based on kinematic studies using r band @xcite , near - infrared @xcite , and submillimeter @xcite spectroscopy , which have revealed the presence of a controversial second nucleus , hidden by dust , more massive than the optical nucleus , and located @xmath17 northwest of the latter . recent n - body simulations by @xcite , which account for this second nucleus , predict the disruption of the circumnuclear starburst in less than an orbital time . however @xcite have concluded that there is no second nucleus in m83 , since they can not confirm the local maximum in the stellar velocity dispersion at the location of the `` second nucleus '' found by @xcite . a third scenario for the origin of the starburst could be internal galaxy dynamics associated with the bar . the ages of circumnuclear star clusters in m83 have been previously estimated . based on the analysis of co absorption and br@xmath18 emission - line data @xcite found that the northwestern end of m83 s starburst is the youngest , while the southeastern end is the oldest . @xcite compared _ hst_/wfpc2 photometry with theoretical population synthesis models and found that most of m83 s circumnuclear star clusters are less than 10 myr old , with clusters younger than 5 myr preferentially located along the northwestern end . the statistical analysis of the photometry of @xcite by @xcite confirms the age gradient along the starburst . in their analysis , @xcite divided the @xmath19 starburst region into 25@xmath20 angular sectors , and considered the mean cluster age as well as the oldest age within each sector , finding 5myr for the oldest cluster in the northwest and 25myr for the oldest cluster in the southeast . the above age gradient suggests that star formation occurs in an ordered manner , and supports the idea that the starburst is fed by the inflow of bar - driven material . if real , this gradient has profound implications for the formation and evolution of circumnuclear starbursts . for this reason , in this work , we age - date 13 clusters within the northern portion of m83 s starburst , using three different techniques , i.e. , by comparing archival _ fuv spectroscopy of the clusters with semi - empirical and new fully theoretical model spectra , and by deriving the ages from high spatial resolution _ hst_/wfc3 multi - band optical photometry . although our study does not cover the full extent of the starburst s arc , this does not affect our results . in addition , we analyze the massive star population in regions between the clusters . another goal of this work is to constrain the upper imfs of our clusters . this is particularly interesting in the metal rich circumnuclear environment of m83 . m83 s circumnuclear h ii regions have an oxygen to hydrogen number density ratio of 12+log(o / h)[email protected] @xcite , and a corresponding fraction of all metals to all elements by mass ( hereafter , metallicity ) that is intermediate between z=0.020 and z=0.040 , and supersolar ( 12+log[o / h]@[email protected] , z@xmath2=0.013 , @xcite ) . we estimate the cluster masses from spectroscopy and photometry , and check which clusters have formed stars more massive than @xmath22m@xmath2 . the random sampling of the stellar imf may prevent this from happening in low mass clusters @xcite . this paper is organized as follows : [ observations ] describes the observations and data reduction , [ models ] presents the models used in the estimate of the spectroscopic ages and masses of the clusters , [ analysis ] explains our procedure for determining the spectroscopic and photometric properties of the clusters , in [ discussion ] we present and discuss our results , finally , we conclude in [ conclusion ] . we analyzed archival fuv imaging and spectroscopy of compact star clusters and field regions located between about @xmath23 ( @xmath24110pc ) and @xmath25 ( @xmath24198pc ) of m83 s optical center , and spanning an arc of @xmath26200pc in size . the data were taken with the stis onboard the _ hst _ as part of programs gto-7577 and gto-8785 ( pi : sara heap ) . we complemented these data with optical images of the same clusters taken with _ s wfc3 as part of program go / dd-11360 ( pi : oconnell ) . @xcite discuss properties derived from optical photometry of star clusters within the larger field of view covered by the wfc3 images . our main focus is on the fuv data , but we also briefly discuss the optical data . figure [ fig_composite ] shows a false - color logarithmic - scale image of m83 s arc - shaped circumnuclear starburst . the image is a composite of the stis image taken with the f25qtz ( fuv ) filter , and wfc3 images taken with the f657n ( h@xmath8 ) and f555w ( v ) filters . north is at the top and east is to the left . the optical nucleus of m83 is the large yellow - green dot in the upper left . the starburst starts northwest of the nucleus , wraps about the nucleus towards the southeast , and is punctuated by about 20 uv bright clusters . the fuv bright clusters appear white , while fainter ones appear blue . the location of sn1968l , which was identified by @xcite , is indicated . at m83 s distance , 1 represents about 22 pc . the two dusty rings discovered by @xcite and embracing the starburst are clearly seen . the clusters with available stis fuv spectroscopy are labeled . note that although clusters 3 , 4 and 7 seem to be composed of several bright knots clustered together , one knot always dominates the brightness . figure [ fig_composite ] was kindly produced for us by w. p. blair . the fuv data set consists of a direct image and four two - dimensional ( 2-d ) spectrograms . the direct image is projected onto the mama detector , which has a field of view of 25x25 and a plate scale of 0@xmath27024 pixel@xmath28 . the spectrograms , one long slit ( 25x2 ) and three multi - object ( slitless ) , were taken with the g140l grating , which has an average dispersion of 0.584 pixel@xmath28 . table [ tab_data ] gives the observation date ( col . 2 ) , exposure time ( col . 3 ) , aperture position angle in degrees east of north ( col . 4 ) , filter name ( col . 5 ) , wavelength coverage for the detector@xmath29filter combination ( col . 6 ) , and extension of the most processed data file available from the hubble data archive of the fuv observations . the data were partially processed through the on - the - fly reprocessing ( otfr ) pipeline of the archive , at retrieval time . this included data linearization , dark subtraction , flat fielding , and summing of the repeated observations . in addition , the pipeline corrected the direct image for geometric distortion . after processing through the pipeline , in order to increase the signal - to - noise ( s / n ) , we added together exposures taken with the same instrument configuration , i.e. , the two slitless exposures taken with the f25sfr2 aperture , which we refer to as the f25sfr2 exposure . because the spectrograms were taken in either a slitless or a wide slit mode with the g140l grating , the spectral traces have large offsets in the dispersion direction . therefore , the otfr pipeline did not wavelength calibrate or convert these spectra to absolute flux . in order to extract wavelength - calibrated one - dimensional ( 1-d ) spectra from the 2-d spectra , we first located the position of each spectral trace along the cross dispersion direction , and then used the x1d module ( mcgrath , busko , & hodge , stis isr 99 - 03 ) available in the image reduction and analysis facility ( iraf , @xcite ) for this task . several iterations were required in order to find the optimal value of xoffset , which is the parameter used by x1d for shifting a spectrum along the dispersion direction . we used the absorption feature c ii @xmath30 ( hereafter c ii 1335 ) , for finding the optimal value of xoffset . in our spectra , this feature is unresolved and is a blend of milky way interstellar and m83 stellar components . for the width of the extraction box we used the value recommended in the stis manual for point sources , i.e. , 11 pixels @xcite . background spectra extracted from outer portions of the 2-d spectrograms were assumed to represent purely geocoronal emission ( air glow ) from h i lyman-@xmath8 ( ly@xmath8 ) at @xmath31 and/or , [ o i ] @xmath32 , depending on the aperture , and were subtracted from the cluster spectra . we did not subtract the underlying diffuse light because of its large variations as a function of position . the direct image and the 2-d spectrograms from the stis are shown in the four panels of fig . [ fig_fuvset ] . panel ( a ) features the 25x25 direct image in logarithmic scale , with the location of m83 s optical center marked with a cross , and the spatial scale marked with a vector . panels ( b ) and ( c ) show the long - slit and the multi - object spectrograms respectively , both taken with the same aperture ( 25mama ) . the vertical band in panel ( b ) is due to ly@xmath8 airglow filling the 2-wide slit ( it is seen more clearly in the electronic version ) . the ly@xmath8 airglow contaminates much of the slitless spectrogram shown in panel ( c ) . finally , panel ( d ) features the sum of the two slitless spectrograms taken with the f25sfr2 aperture , which rejects the ly@xmath8 airglow . no contamination from the [ o i ] air glow is apparent in panel ( d ) because the emission brightness geocoronal [ o i ] is typically less than @xmath33 of that of geocoronal ly@xmath8 , depending on the time of observation and the position of the target relative to the earth @xcite . we used the slitless exposures for extracting spectra of those clusters falling outside the slit , as well as for studying the effect of rejecting the ly@xmath8 air glow on our results . the spectral resolution , depends on the spatial extent of the star clusters . in order to estimate the spectral resolution of our spectra , we determined the full width at half maximum ( fwhm ) of the cross dispersion profiles of the spectral traces . for our point - like representative , gd 153 , the fwhm is 0.09 . due to blending with the profiles of close neighbours , we were only able to determine the fwhm of four clusters by our method . for clusters 1 , 4 , 10 , and 11 + 12 we obtained fwhm of 0.17 , 0.30 , 0.19 , and 0.23 , respectively , which suggests that the clusters are partially resolved . since these are neither point sources nor fully extended sources , we expected the spectral resolution of our long - slit spectra to be intermediate between the values given in the stis instrument handbook , i.e. , 0.9 ( point source ) and 48 ( extended source ) . multi - component gaussian fitting of the c ii feature at @xmath30 , carried out with the peak analysis code ( available at ftp://ftp.ncnr.nist.gov/pub/staff/dimeo/pan.zip ) reveals that the fwhm of this feature ranges from 2.5 - 7 . our field spectra are expected to have the resolution of a fully extended source . we corrected the cluster spectra for reddening due to the presence of dust in the galaxy and in m83 . for the galaxy s contribution to the reddening , we adopt @xmath34 @xcite , and we assume the galactic extinction curve of @xcite . for m83 s contribution , we ( i ) fitted a power law of the form @xmath35 to the continuum in the range @xmath36 , ( ii ) assumed that any deviation of @xmath37 from @xmath38 , which is the expected value for a dust free starburst , is due to extinction by local dust , and ( iii ) dereddened the observed spectrum until @xmath38 was reached , using the starburst obscuration law of @xcite . in order to match each cluster with its fuv spectral trace , we started by directly comparing the positions of the clusters in the stis direct image , with the positions of the spectral traces in the 2-d spectrograms . for this , we rotated the image and vertically shifted the spectrograms until the orientations matched , as shown in fig . [ fig_fuvmatch ] , which shows enlarged portions of the stis image , side by side with two spectrograms . note that the slitless spectrograms all have the same orientation , while the orientation of the long - slit spectrogram is different . therefore , in fig . [ fig_fuvmatch ] , we show only the slitless spectrogram taken with the f25sfr2 aperture . in panel ( a ) of fig . [ fig_fuvmatch ] , we indicate the position of the long - slit , the locations where we extracted the sky and field region spectra , and the correspondence between clusters within the slit and their spectral traces . panel ( c ) , which shows the same image as panel ( a ) , but with a slightly different orientation , shows only one sky region due to the fact that the second falls outside of the field displayed in the panel . unfortunately , and as can be seen in fig . [ fig_fuvmatch ] , in spite of the different orientations of the spectrograms , the spectral traces of the two brightest clusters , 11 and 12 , always overlap . furthermore , as seen in panel ( d ) , the faint spectral traces of clusters 4 and 5 also overlap , as do the traces of cluster 14 and a source not in our sample . fortunately , we can use the long - slit spectra for clusters 4 and 14 . however , cluster 5 was rejected from our analysis due to its poor spectral trace . the spatial correspondence between star clusters and their spectral traces can be checked in fig . [ fig_fuvprof ] , where we compare the cluster radial profiles derived from the stis direct image ( bottom curves in left and right panels ) , with the cross dispersion profiles derived from the long - slit ( top curve , left panel ) and the slitless ( top curve , right panel ) 25mama spectrograms . also shown in fig . [ fig_fuvprof ] is the cross - dispersion profile of a point - like source , represented by white dwarf gd 153 ( top dotted curve in left and right panels ) , observed as part of program 7097 , in the same configuration as the long - slit spectrogram . for convenience , we divided each curve in fig . [ fig_fuvprof ] by the amplitude of the tallest peak , which for the clusters is the amplitude of target 11@xmath3912 ( which overlap ) , and for the white dwarf is its own amplitude . for clarity , the top curves in fig . [ fig_fuvprof ] ( cross dispersion profiles ) were all shifted by 0.2 . the clusters and regions used for the background subtraction are labeled . for the radial profiles , we first rotated the image to match the orientation of each spectrogram , and we then obtained the count rate at a given vertical position by integrating over the horizontal direction . the radial profiles in the left and right panels look slightly different because of the different angles by which we rotated the image . we note that although the peaks for clusters in our sample are dominated by emission from the clusters themselves , diffuse light and fainter sources also contribute to their shape and amplitude . this however , does not affect the identification of the spectrum of each cluster . for the spectral profiles , the count rate at a given position in the cross dispersion direction was obtained by integrating over the dispersion axis . note the higher background of the top right curve with respect to the top left curve , which is due to contamination with ly@xmath8 air glow over a larger portion of the detector in the slitless spectrogram ( right panel of fig . [ fig_fuvprof ] ) . circular aperture photometry was performed on the clusters using the iraf task phot , using an aperture radius of 3 pixels and a background annulus with radii of 10 and 13 pixels . aperture corrections , based on the estimated size of each cluster , were applied to convert the fixed aperture magnitudes to total magnitudes . the instrumental magnitudes in the f336w , f438w , f555w , f658n , and f814w filters were converted to the vegamag photometric system by applying zeropoints available at the following url : http://www.stsci.edu/hst/wfc3/phot_zp_lbn . we refer to the broad - band filters as the `` @xmath40 '' , `` @xmath41 '' , `` @xmath42 '' , and `` @xmath43 '' magnitudes for simplicity , although they were not transformed to the johnson - cousins system . for more details , refer to @xcite . we compared our observed reddening corrected fuv spectra with synthetic dust free spectra generated with the windows version @xcite of the widely used package starburst99 ( s99 , @xcite ) . the main input parameters in s99 are 1 ) the star - formation law , 2 ) the imf , 3 ) the metallicity and stellar evolution tracks , and 4 ) the stellar spectral library . we tried the following parameters . 1 ) a single burst of star formation , since @xcite ruled out the possibility that star formation within individual clusters proceeded continuously . indeed , in m83 s circumnuclear region , the he i 5876 absorption line is detected in h ii regions ionized by the youngest clusters , but not in those ionized by clusters older than about @xmath44myr . if star formation was continuous , the line strength ratio of he i 5876 to h@xmath37 would be the equilibrium value in the older clusters . 2 ) imfs with high mass limits of 10 , 30 , 50 , and 100@xmath45 . 3 ) the stellar evolution tracks for non - rotating stars of @xcite for masses @xmath46 , and of @xcite for masses @xmath47 , at metallicities @xmath0 and @xmath1 . note that from h ii region measurements , @xcite found that the metallicity of m83 s nuclear region is @xmath48 , which is in between the above two metallicities . 4 ) an empirical stellar library of galactic o and b stars , as well as a new uv high resolution theoretical library for massive stars of metallicities @xmath49 , and @xmath50 . the difference in metallicity between the stellar evolutionary tracks and the theoretical stellar library is due to the use of different solar compositions . indeed , the wm - basic models use the revised solar composition of asplund et al . ( 2005 ) , which has a lesser content in metals than the solar composition of the stellar evolution tracks . note that there are no empirical uv stellar libraries for @xmath51 . we describe the empirical and the new fully theoretical stellar spectral libraries in [ libraries ] . we discuss the advantages and disadvantages of using the theoretical library in [ theo_lib ] . hereafter , we refer to single stellar population ( ssp ) spectra generated with the empirical stellar library as the semi - empirical models , and to ssp spectra computed with the theoretical stellar library , as the theoretical models . we compare theoretical models at @xmath0 and @xmath1 in [ solar_vs_supersolar ] . we compare theoretical models corresponding to different high mass limits of the imf in [ upimf ] . finally , we compare the two sets of models in [ old_vs_new ] . the semi - empirical models were built from spectral libraries of galactic o and b stars observed in the @xmath52 range at 0.12 resolution with the far ultraviolet spectroscopic explorer @xcite , and in the @xmath53 range at 0.75 resolution with the international ultraviolet explorer @xcite . comparisons of the semi - empirical models with observations of local and distant galaxies generally yield quite good agreement @xcite . the theoretical models were built from a spectral library of hot massive stars ( @xmath54k , @xmath55 ) that was computed as described in leitherer et al . ( 2010 ) , using the stellar atmosphere code wm - basic @xcite , following the approach pioneered by @xcite . the models span the spectral range @xmath56 at high ( 0.5 ) resolution . this resolution is sufficient to resolve the main uv stellar wind lines , which have widths in excess of 1000kms@xmath28 . the strengths and weaknesses of the theoretical stellar library are the following . 1 ) the two main publicly available codes optimized for modeling the uv spectra of ob stars with winds are cmfgen @xcite and wm - basic . wm - basic , which we utilized , solves the radiative transfer and the hydrodynamics self - consistently to derive the density structure of the stellar wind , while cmfgen assumes an analytical approximation for the wind structure a priori . 2 ) wolf - rayet ( w - r ) stars , whose stellar evolution is quite uncertain , are modeled using wm - basic atmospheres . this ensures that their continuum fluxes are accounted for , but w - r - specific lines are not reproduced correctly . fortunately , w - r stars in the local universe do not contribute significantly to the uv spectrum of ob stellar populations , except for he ii @xmath571640 , which should be viewed with care . in individual w - r stars , the equivalent width ( ew ) of the he ii 1640 emission component ranges from 3 to 100 @xcite , with a preference for smaller values ( 3 - 30 or so ) for late - type wn stars , which tend to dominate observed statistics due to their high absolute visual magnitude . in starburst clusters , the typical ew of the 1640 bump is about 2 - 3 @xcite . if ew=15 is the typical value for late - type wn stars , this suggests a continuum diluting about 80% of the light at 1640 , and w - r stars can not contribute more than 20% to the uv continuum . the dilution is caused by the strong continuum of o main sequence and o supergiant stars . 3 ) we omit the geometrical effect of clumps in the stellar wind , but the associated x - ray emission from shocks propagating in the wind is included . clumping affects the lines from trace ions , e.g. , o vi @xmath571035 and o v @xmath571371 @xcite . in addition , the strength of the wind lines depends on the assumed ionization balance . while photo - ionization is the primary mechanism determining the ionization state of the wind , x - rays arising from shocked material can be important as well ( e.g. , @xcite ) . 4 ) we omit stark broadening of the hydrogen lowest lyman lines , which is computationally expensive . a pure doppler profile of hydrogen is very similar to one with stark broadening included , as long as the upper principal quantum of the transition is small @xcite . this is the result of the stark width s dependence on the fourth power of the upper principal quantum number . in practice , this means stark broadening becomes negligible in hydrogen lines at higher energies and originating at lower densities . lyman-@xmath8 , owing to its large optical depth , is formed far out in the wind even at relatively low mass - loss rates . therefore , stark broadening is not important at ages younger than 10 myr . 5 ) the evolution of massive stars in binary systems and of rotating massive stars requires further study before it can be included in the models @xcite . 6 ) the models can be used for interpreting the rest - frame uv spectra of objects with metallicities other than near - solar or those of the large ( @xmath58 ) and small ( @xmath59 ) magellanic clouds @xcite . because there are no empirical uv spectral stellar libraries at @xmath51 , we have to rely on fully theoretical star cluster spectra , at @xmath1 . here we discuss differences in the theoretical spectra of single stellar populations that were computed with stellar evolution tracks corresponding to metallicities of @xmath0 and @xmath1 . line - driven wind theory predicts lower mass - loss rates and lower wind terminal velocities for o stars of lower metallicity @xcite , and therefore , weaker wind features in the spectra of these stars . observations of ms o stars in the magellanic clouds and the galaxy support the theory . for instance , see @xcite , where empirical stellar wind results for the milky way , the lmc and the smc are compared with the ob star theoretical wind models of @xcite , which are an extension of line driven wind theory . wm - basic models do a good job of reproducing lower wind terminal velocities at lower metallicity @xcite , as can be seen in fig . [ fig_solar_vs_supersolar ] , which shows a comparison between the @xmath0 and the @xmath1 models , for ages from one to 20 myr ( as labeled on the right ) , in the wavelength range @xmath62 , at 0.75 resolution . for instance , compare the absorption components of c iv 1550 at @xmath0 and @xmath1 . in addition , the photospheric lines are deeper at the higher metallicity , as expected . figure [ fig_upimf ] shows semi - empirical and theoretical models for ssps with different high mass limits of the stellar imf . the figures show ages 2 , 3 , 5 , and 10myr , and focus on c iv 1550 and si iv 1400 , whose profiles are sensitive to the distributions of o main sequence and o supergiant stars , respectively . all models in fig . [ fig_upimf ] have a lower mass limit of @xmath63m@xmath2 . the thin black curves use a kroupa imf from 0.1 to 100m@xmath2 , while the thick grey curves ( red in the electronic version ) use imf high mass limits ( @xmath64 ) of 10 , 30 , and 50m@xmath2 . lowering @xmath64 , i.e. , gradually removing the o and the most luminous b stars , has the following effects . ( i ) as @xmath64 decreases , the p - cygni profiles of si iv 1400 and of c iv 1550 become weaker . in particular , the emission component of these profiles is non - existent at @xmath65 for the semi - empirical models , and drops in strength more gradually in the theoretical models . also the strong absorption component of these profiles is purely photospheric at @xmath66 , for both models . ( ii ) for the theoretical models at @xmath67 , the absorption lines are dominated by photospheric lines from b stars . here we discuss differences between the semi - empirical and the theoretical models at @xmath0 . figure [ fig_old_vs_new ] shows the theoretical and semi - empirical models , using the same format of fig . [ fig_solar_vs_supersolar ] . there are several noteworthy features . 1 ) in the region below 1240 the semi - empirical spectra suffer from ly@xmath8 absorption in the galactic plane . 2 ) the p - cygni profiles of the resonant doublets of n v at 1238.8 + 1242.8 ( hereafter n v 1240 ) and of c iv at 1548.2 + 1550.8 ( hereafter , c iv 1550 ) , which originate in the stellar wind , decrease in strength as the o stars evolve and disappear . 3 ) the resonant doublet of si iv at 1393.8 + 1402.8 ( hereafter , si iv 1400 ) is absent in main sequence ( ms ) o stars , and develops p - cygni profiles in giant and supergiant o stars . this is why the p - cygni profile of si iv 1400 is strong at ages @xmath68myr , when o supergiants are present . note that the lack of strong n v 1240 , c iv 1550 , or si iv 1400 p - cygni profiles , in the spectrum of a star cluster , indicates either that the cluster did not form very massive o stars , or that the cluster is too old to show the signatures of such stars . 4 ) at 3 myr , the semi - empirical model has somewhat stronger n v , si iv , and c iv lines than the theoretical model . at this particular time step , stars with stronger winds make a stronger contribution to the semi - empirical than to the theoretical spectrum . this difference serves as a reminder of a crucial assumption made when linking model spectra to stellar evolution models . in the case of the theoretical spectra , one simply adopts the effective temperature of the atmospheres for matching the positions of the evolutionary tracks . in contrast , for the empirical spectra , we obtain the effective temperatures from spectral types via a temperature scale . starburst99 assumes the relation between spectral type and effective temperature of o stars derived by @xcite . we note that switching between the relations of martins et al . and the earlier work of @xcite or @xcite modifies the peak of the si iv emission by almost 10% . careful comparisons with observations should provide guidance for choosing the preferred approach in the synthesis models . in addition , the wm - basic models use the revised solar composition of asplund et al . ( 2005 ) , which has a lesser content in metals than the solar composition of the stellar evolution tracks . while the revised solar abundances have a small effect on the evolutionary tracks of massive stars , they do affect the computed uv spectra via changed opacities and wind properties . 5 ) many of the narrow absorption lines in the semi - empirical models are of interstellar origin . for the first four myr , the absorption line si ii @xmath571264 is stronger in the semi - empirical models because of the interstellar contribution . 6 ) o v @xmath571371 , which is strong at ages @xmath69myr , i.e. , in the hottest stars , is stronger in the theoretical models due to the neglect of wind clumping @xcite . table [ tab_observed ] allows an assessment of the uncertainties in the reddened flux at 1500 , @xmath70 , and in the power law index of the reddened fuv continuum , @xmath71 , for the 13 clusters in our final sample . for each spectral exposure and each cluster , columns ( 2)-(12 ) in the table give ( 1 ) the cluster i d ; ( 2 ) and ( 3 ) the right ascension and declination measured from the stis direct image ; ( 4)-(6 ) the median signal to noise ( s / n ) of the reddened spectrum ; ( 7)-(9 ) @xmath70 ; and ( 10)-(12 ) @xmath71 . the faint clusters 4 and 14 have the lowest s / n values . the rest of the clusters have reasonably good s / n ( @xmath72 ) . the @xmath70 fluxes from the slitless exposures taken with the 25mama and f25sfr2 apertures agree within @xmath33 . therefore , the ly@xmath8 airglow has little effect on @xmath70 . the slit blocks the light from nearby sources and lowers @xmath70 by @xmath73 . the value of @xmath71 ( i ) increases by @xmath74 when the wavelength range for fitting the continuum is extended ( compare the two values in column 10 ) ; ( ii ) varies by at most @xmath33 , due to geocoronal ly@xmath8 contamination ( compare columns 11 and 12 ) ; and ( iii ) increases by at most @xmath33 when the slit is removed ( compare columns 10 and 11 or 12 ) . the properties of clusters @xmath75 derived from the slitless spectra are less reliable than those derived from the long - slit spectra , and are only shown for comparison . since the two slitless exposures yield similar values in tab . [ tab_observed ] , we decided to discard the exposure spanning the shortest wavelength range , i.e. , the one taken with the f25sfr2 aperture . for the nine clusters within the slit , i.e. , 1 , 2 , 3 , 4 , 10 , 11 , 12 , 13 , and 14 , the ` good ' data span the wavelength range @xmath76 , while for the rest of clusters , i.e. , 6 , 7 , 8 , and 9 , they span the range @xmath77 . we adopted the algorithm developed by @xcite for deriving the age and mass of each cluster from the best fit model to the data . the algorithm takes as input the de - reddened cluster spectrum and a grid of starburst99 models spanning ages from 1 to 20myr in step sizes of 0.1myr . the goodness of the fit is characterized by @xmath78 , where @xmath79 , and where @xmath80 represents the observed data for the @xmath81th pixel , @xmath82 the model data , @xmath83 the error in the observed spectrum , and @xmath84 the assigned weight . the weighting scheme eliminates from consideration the interstellar lines , and gives the stellar wind lines a weight 1.5 times greater than that of the continuum . prior to fitting , the resolution of the models was degraded to match the approximate resolution of the observed spectra . the spectroscopic masses were derived by ( i ) fitting power laws to the continua of the observed and the model spectra of each cluster , where the model corresponds to the best fit age and a stellar mass of 10@xmath85m@xmath2 ; ( ii ) computing the ratio of the observed to the model power law at each wavelength point ; ( iii ) taking the average of the sum of the ratios obtained in ( ii ) ; and ( iv ) multiplying the result by 10@xmath85m@xmath2 . the metallicity of m83 s starburst is @xmath48 with a large error bar ( see @xcite ) . we considered theoretical models with metallicities of @xmath0 and @xmath1 . @xcite derived the age and extinction for each cluster by finding the minimum @xmath78 when comparing observed ubvih@xmath8 magnitudes with those predicted by the charlot & bruzual ( 2009 , private communication ) stellar population synthesis models , assuming a milky way - type extinction curve @xcite , metallicities of @xmath86 and @xmath87 , and a chabrier imf from @xmath5m@xmath2 @xcite . here , the narrow band filter contains both stellar continuum and nebular line emission , i.e. no continuum subtraction was performed for the narrow band image , and the predictions were modeled as line plus continuum . the mass of each cluster was estimated from its extinction - corrected @xmath42 band luminosity and the age - dependent mass - to - light ratio from the stellar population models , assuming a chabrier stellar imf and a distance of 4.5 mpc . the ages and masses derived using this technique are compared to the spectroscopic values below . on average , the theoretical spectra fit the data better for @xmath1 than for @xmath0 ( on average , the reduced @xmath78 is 1.3 times larger at @xmath0 than at @xmath1 , with a standard deviation of 0.3 ) . note that some of the differences between the data and the best - fit models are not due to the metallicity choice , but to the quality of the data , which suffers from crowding , as well as from worse spectral resolution than when a narrow slit is used . in any case , on average , the spectroscopic ages and masses at the two metallicities are identical . hereafter , we only consider spectroscopic results at @xmath0 , which is the metallicity of the semi - empirical models . cluster 14 is the youngest ( @xmath263 myr , average of spectroscopic and photometric ages ) and the least massive ( photometric mass of about @xmath88m@xmath2 ) in our sample . according to @xcite cluster 14 should have a maximum stellar mass of about 80 m@xmath2 ( i.e. below 100 m@xmath2 ) . unfortunately , the spectrum of this cluster is too noisy to reliably pin down its maximum stellar mass . cluster 1 is the next best cluster in our sample to test the upper end of the stellar imf , and has a mean age of @xmath264myr and a photometric mass of @xmath263@xmath6m@xmath2 . as shown in fig . [ fig_1and10 ] , left panel , the strong si iv 1400 and c iv 1550 p - cygni profiles of cluster 1 can not be reproduced with models having maximum stellar masses of 30m@xmath2 or less . models with maximum stellar masses of 50 and 100m@xmath2 fit cluster 1 equally well . note that @xcite found optical signatures of the presence of wn stars in the region of our clusters @xmath75 . the p - cygni profile of he ii at 1640 , which is characteristic of wn stars , is barely detected in the fuv spectra of clusters @xmath75 , but given the point that we make in [ theo_lib ] , this is not surprising . the rest of clusters with strong p - cygni profiles of n v , si iv and c iv are consistent with having formed stars more massive than 30m@xmath2 . the right panel of fig . [ fig_1and10 ] shows that the upper mass limit of the imf is harder to constrain for cluster 10 , which is without p - cygni profiles . clusters with strong absorption profiles of si iv and c iv are consistent with having formed at least some b stars , if we assume that the absorptions are not of interstellar origin . figures [ fig_fits_1 - 6 ] to [ fig_fits_12 - 14 ] show the background and reddening corrected rest - frame spectra of the clusters in our sample , along with best fit semi - empirical and theoretical models corresponding to a metallicity of @xmath0 and a kroupa imf from @xmath5m@xmath2 . we used the latter models to determine the spectroscopic ages and masses of all clusters , including those without strong p - cygni profiles of n v , si iv , and c iv in their spectra . therefore , we are assuming that clusters without strong p - cygni profiles formed massive o stars but are older than @xmath2612 myr . another possibility for the latter clusters is that they are younger than @xmath2612 myr and did not form very massive o stars . here , we compare our ages derived using spectroscopic and photometric techniques . we also compare our ages with estimates from the literature . our spectroscopic and photometric ages are compiled in columns ( 1)-(9 ) of tab . [ tab_age ] , which give ( 1 ) the cluster i d ; ( 2 ) and ( 4 ) , the spectroscopic ages from the semi - empirical and theoretical models ; ( 3 ) and ( 5 ) , the reduced @xmath78 value of the best fits to the observed spectra ; ( 6 ) and ( 7 ) , the photometric ages from the @xmath86 and @xmath87 models ; ( 8)-(13 ) ratios of the age estimates derived from the different techniques . the last four rows of tab . [ tab_age ] give the minimum , maximum , mean , and standard deviation values for relevant columns . figure [ fig_photsed ] shows the photometric age determinations for cluster 1 ( mean photometric age @xmath264myr ) and cluster 10 ( mean photometric age @xmath2619myr ) . they were derived using the technique described in @xcite . figure [ fig_photsed ] also shows the measured v - b and v - i colors for clusters 1 and 10 . considering the mean of the spectroscopic and photometric ages , clusters 1 and 10 are representative of young age ( @xmath89myr ) and older age ( @xmath90myr ) clusters , respectively . the ages estimated from the four different techniques are in excellent agreement . ages derived by comparing uv spectroscopy with semi - empirical and theoretical predictions are within a factor of 1.5 . this is similar to the @xmath91 uncertainties of @xmath261myr found previously for young clusters from uv spectroscopy @xcite . ages determined by comparing the _ hst_/wfc3 optical colors of the clusters with predictions from stellar population synthesis models at the two metallicities agree within a factor of 2 . finally , the spectroscopic and photometric ages are in excellent agreement ( within a factor of 1.2 on average ) . table [ tab_age ] shows that most clusters are quite young , with ages @xmath92myr . clusters 6 , 7 , and 10 , which have the oldest spectroscopic ages , also have the oldest photometric ages . the worst agreement between the spectroscopic and photometric ages is for cluster 8 ( difference of a factor of 4 ) . the fuv spectrum of this cluster lacks the strong p - cygni profiles found in young star clusters , yet the cluster has very blue colors , resulting in a young ( @xmath264myr ) photometric age . comparing with previous work , our ages agree with the photometric ages derived by @xcite based on _ hst_/wfpc2 photometry , except for clusters 6 , 7 , and 10 , which are older than 6myr in our case . we found 3.6 , 3.4 , and 4.1myr , for the ages of clusters 1 , 2 , and 3 respectively ( means of spectroscopic and photometric ages ) , which is in excellent agreement with the ages found by @xcite for region a ( all @xmath264myr ) , using the same spectroscopic data as here . on the other hand , @xcite determined an age of 7myr for their region b , from the equivalent width of h@xmath37 . region b corresponds to the location of our clusters 7 - 9 . we found mean ages of 15.5myr for cluster 7 and 3.4myr for cluster 9 . as explained in the previous paragraph , we were unable to pin down the age of cluster 8 , which is also in region b. finally , region d of @xcite corresponds to a cluster that we did not include in our sample , due to its truncated spectral trace . we conclude that our spectroscopic and photometric age estimates are within the expected uncertainties for 12 out of 13 of the clusters . the clusters are @xmath263 - 20myr old and were not all formed at the same time . in addition , we do not find any compelling evidence for an age gradient along m83 s arc - shaped starburst , but rather see clumping of clusters of similar age with very young clusters dominating the northern - most and southern - most portion of the starburst that we studied , and somewhat older clusters found between . this is in disagreement with @xcite and @xcite , who have suggested that there is an age gradient along the starburst , as described in the introduction . here , we compare the masses estimated for our clusters using spectroscopic and photometric techniques . these are compiled in columns ( 1)-(7 ) of tab . [ tab_mass ] , which give ( 1 ) the cluster i d ; ( 2 ) and ( 3 ) , the masses derived by fitting the semi - empirical and the theoretical models to the fuv spectroscopy ; ( 4 ) and ( 5 ) the masses derived from the optical photometry assuming metallicities of @xmath49 and @xmath86 ; and ( 6)-(11 ) ratios of masses obtained from the different techniques . column ( 6 ) of tab . [ tab_mass ] shows that the spectroscopic masses from the semi - empirical models are in very good agreement with those from the theoretical models , and that the semi - empirical masses are systematically larger by a factor of 1.5 on average than the theoretical masses . the latter is because the semi - empirical models have systematically larger values of the mean ratio described in ( ii ) of the last paragraph of [ ana_spec ] . column ( 7 ) of tab . [ tab_mass ] shows that the photometric masses are not significantly affected by metallicity . finally , columns ( 8) to ( 11 ) show that the spectroscopic masses are systematically larger than the photometric masses by a factor of 4 - 6 on average . we suggest that this is the result of the high background contamination of our wide slit and slitless crowded spectra . optical photometry provides more leverage for determining the cluster mass . this is because photometry captures light from low and intermediate mass stars , which contribute more by mass to the imf than massive stars . in addition , the total light from the cluster is estimated better from the photometry . therefore , we consider our photometric masses ( m@xmath93 ) as more reliable than our spectroscopic masses . our most massive cluster has m@xmath93 between 2 and @xmath94m@xmath2 , comparable to the virial mass of the ionizing cluster of 30 doradus , ngc 2070 ( @xmath95m@xmath2 , @xcite ) , in the large magellanic cloud . according to @xcite , young star clusters with masses larger than 10@xmath96m@xmath2 can last an age comparable to or exceeding the age of the universe . therefore , this massive cluster in m83 could be a globular cluster progenitor , while two other clusters , which have m@xmath97m@xmath2 , may also survive . cluster 14 has m@xmath98m@xmath2 and appears to be the most compact in size in the fuv image . finally , the rest of the clusters have m@xmath93 of a few@xmath3m@xmath2 . note that our older clusters tend to be more massive than the younger ones . this is probably an observational effect . older clusters at lower masses are harder to detect in the fuv . columns ( 1)-(5 ) of tab . [ tab_reddening ] give ( 1 ) the cluster s i d , ( 2 ) @xmath99 , the power - law index of the continuum over the wavelength range @xmath77 , corrected for a foreground milky way reddening of @xmath34 ; ( 3 ) the intrinsic reddening of the cluster derived from the fuv spectra ; ( 4 ) the intrinsic reddening derived from the optical photometry ; and ( 5 ) @xmath100 , the intrinsic luminosity of the cluster at 1500 , derived by adopting a distance of 4.5 mpc to m83 . overall , the agreement between the reddening values derived from the spectroscopic and the photometric analysis is fair . for clusters 3 and 14 , which have dusty local environments ( see fig . [ fig_composite ] ) , the agreement is good . indeed , both clusters have large intrinsic spectroscopic and photometric reddenings . the agreement is also good for clusters 11 and 12 , which appear to be the least reddened by both measures of the intrinsic extinction . however , for cluster 1 , e(b - v)=0.18 from the spectroscopy , while e(b - v)@xmath1010.06 from the photometry . cluster 3 has the largest intrinsic value of @xmath100 , while clusters 4 , 13 , and 14 , which appear faint in the optical images , have the lowest values of @xmath100 . in addition to compact star clusters , there is uv emission from the diffuse field star population . we obtained the spectrum representing the diffuse stellar field by summing spectra extracted from the long slit exposure , at the locations indicated by the dashed lines in fig . [ fig_fuvmatch ] . we performed no background or reddening correction for the field spectrum . in addition , we did not shift the spectral traces in the dispersion direction , due to the lack of narrow interstellar lines that could be used for reference . figure [ fig_fieldsp ] shows the result . the spectrum lacks the p - cygni profiles of n v , si iv , and c iv , which are signatures of o stars , but it shows strong absorption components in these features , which indicates the presence of b stars . this is consistent with a picture where ( massive ) o stars form mostly in clusters but not in the field , and where the field is dominated by b stars from clusters which disrupt rapidly , on timescales of @xmath2610myr @xcite . we estimated the ages , reddenings , and masses of 13 clusters within the circumnuclear starburst of m83 , the nearest such example , based on _ hst_/stis / fuv - mama imaging and spectroscopy , and _ hst_/wfc3/uvis photometry . we derived four ages for each cluster . the first two ages were obtained by fitting the uv spectra with semi - empirical @xcite and theoretical @xcite models corresponding to a metallicity of @xmath0 , giving the most weight to the spectral features n v 1240 , si iv 1400 , and c iv 1500 , which are expected to be prominent in the observed spectra of clusters with ages of less than 20myr . note that theoretical models with @xmath1 yield , on average , ages and masses identical to those at @xmath0 . we compared our spectroscopic ages with the photometric ages derived by @xcite using charlot and bruzual stellar population synthesis models corresponding to metallicities of @xmath86 and @xmath87 . the spectroscopic ages derived with the semi - empirical and the theoretical predictions are within a factor of 1.2 on average . the spectroscopic and photometric ages agree at a similar level . our ages agree with those derived from _ hst_/wfpc2 photometry by @xcite , except for clusters 6 , 7 , and 10 , which are older than 6myr in our case . our ages for clusters 1 - 3 agree with the ages of region a derived from stis fuv spectroscopy by @xcite . the clusters are @xmath263 - 20myr old and were not all formed at the same time . we found no age gradient along m83 s starburst , in disagreement with @xcite and @xcite . our clusters with strong p - cygni profiles have photometric masses of at least a few@xmath3m@xmath2 , seem to have formed stars more massive than @xmath4m@xmath2 , and are consistent with a kroupa imf from @xmath102m@xmath2 . the rest of the clusters are consistent with having formed at least some b stars . b stars dominate the field population . this work was supported by nasa grant n1317 . r. c. is grateful for support from nsf through career award 0847467 . we would like to thank william p. blair for producing fig . [ fig_composite ] . in addition , we would like to thank brad whitmore and our referee for comments that greatly improved the quality of this paper . , l. , jogee , s. , barazza , f. d. , marinova , i. , & shen , j. 2009 , in astronomical society of the pacific conference series , vol . 419 , astronomical society of the pacific conference series , ed . s. jogee , i. marinova , l. hao , & g. a. blanc , 402+ , d. 1986 , in presented at the society of photo - optical instrumentation engineers ( spie ) conference , vol . 627 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . d. l. crawford , 733+ , t. , lundgren - andersson , a. , & olofsson , h. 2004 , in iau symposium , vol . 222 , the interplay among black holes , stars and ism in galactic nuclei , ed . t. storchi - bergmann , l. c. ho , & h. r. schmitt , 219220 lccccccc slitless spectra & 1998 - 05 - 02 & 42 & -24.9 & f25sfr2 & 1275 - 1700 & sfl + slitless spectra & 1998 - 05 - 03 & 45 & -24.9 & f25sfr2 & 1275 - 1700 & sfl + slitless spectra & 1998 - 05 - 03 & 39 & -24.9 & 25mama & 1150 - 1700 & sfl + direct image & 1999 - 05 - 19 & 23 & 32.9 & f25qtz & 1475 - 1700 & x2d + long slit spectra & 2000 - 04 - 23 & 42 & -9.5 & 25mama & 1150 - 1700 & flt [ tab_data ] lccccccccccc 1 & 13:37:00.49 & -29:51:54.93 & 18 & 17 & 27 & 6.0e-15 & 8.9e-15 & 8.0e-15 & -1.3 ( -1.1 ) & -0.7 & -0.5 + 2 & 13:37:00.45 & -29:51:55.40 & 16 & 14 & 24 & 4.8e-15 & 6.8e-15 & 6.4e-15 & -1.5 ( -1.2 ) & -0.9 & -0.9 + 3 & 13:37:00.54 & -29:51:55.61 & 14 & 10 & 19 & 3.8e-15 & 4.4e-15 & 4.4e-15 & -0.6 ( -0.4 ) & 0.2 & 0.8 + 4 & 13:37:00.42 & -29:51:57.52 & 6 & & & 9.2e-16 & & & -0.6 ( -0.7 ) & & + 5 & 13:37:00.34 & -29:51:58.31 & & & & & & & & & + 6 & 13:37:00.37 & -29:51:59.48 & & 16 & 23 & & 5.3e-15 & 5.4e-15 & & -1.4 & -1.5 + 7 & 13:37:00.42 & -29:51:59.83 & & 24 & 35 & & 1.1e-14 & 1.1e-14 & & -1.4 & -1.5 + 8 & 13:37:00.44 & -29:51:59.99 & & 24 & 36 & & 1.2e-14 & 1.2e-14 & & -1.6 & -1.6 + 9 & 13:37:00.43 & -29:52:00.20 & & 22 & 33 & & 9.5e-15 & 1.0e-14 & & -1.6 & -1.7 + 10 & 13:37:01.00 & -29:52:00.71 & 21 & 22 & 34 & 7.7e-15 & 1.1e-14 & 1.2e-14 & -1.5 ( -1.3 ) & -1.4 & -1.3 + 11 & 13:37:01.00 & -29:52:01.45 & 34 & 35 & 50 & 2.1e-14 & 2.4e-14 & 2.6e-14 & -2.2 ( -1.9 ) & -2.1 & -2.2 + 12 & 13:37:00.54 & -29:52:01.62 & 34 & 35 & 50 & 2.1e-14 & 2.5e-14 & 2.6e-14 & -2.3 ( -2.0 ) & -2.1 & -2.2 + 13 & 13:37:00.56 & -29:52:01.91 & 22 & 23 & 35 & 9.0e-15 & 1.2e-14 & 1.2e-14 & -2.3 ( -2.0 ) & -2.0 & -2.1 + 14 & 13:37:00.53 & -29:52:02.62 & 7 & & & 1.5e-15 & & & -0.7 ( -0.7 ) & & [ tab_observed ] lcccccccccccc 1 & 3.9 & 3.3 & 2.6 & 5.4 & 4.2 & 3.8 & 1.5 & 1.1 & 0.9 & 0.6 & 1.0 & 0.7 + 2 & 3.4 & 4.2 & 2.4 & 7.4 & 4.2 & 3.8 & 1.4 & 1.1 & 0.8 & 0.6 & 0.9 & 0.6 + 3 & 3.5 & 4.8 & 3.5 & 4.0 & 4.0 & 5.5 & 1.0 & 0.7 & 0.9 & 0.9 & 0.6 & 0.6 + 4 & 3.5 & 1.5 & 2.7 & 1.9 & 4.2 & 5.0 & 1.3 & 0.8 & 0.8 & 0.7 & 0.7 & 0.5 + 6 & 19.8 & 1.9 & 16.8 & 1.7 & 10.0 & 24.0 & 1.2 & 0.4 & 2.0 & 1.7 & 0.8 & 0.7 + 7 & 14.2 & 2.3 & 14.3 & 2.9 & 9.5 & 24.0 & 1.0 & 0.4 & 1.5 & 1.5 & 0.6 & 0.6 + 8 & 15.6 & 2.4 & 14.3 & 2.4 & 4.0 & 3.6 & 1.1 & 1.1 & 3.9 & 3.6 & 4.3 & 3.9 + 9 & 4.5 & 5.3 & 3.7 & 2.3 & 3.3 & 2.2 & 1.2 & 1.5 & 1.4 & 1.1 & 2.1 & 1.7 + 10 & 11.8 & 2.9 & 14.1 & 4.7 & 13.2 & 24.0 & 0.8 & 0.5 & 0.9 & 1.1 & 0.5 & 0.6 + 11 & 3.4 & 6.2 & 2.6 & 8.8 & 6.3 & 4.6 & 1.3 & 1.4 & 0.5 & 0.4 & 0.7 & 0.6 + 12 & 3.4 & 7.1 & 2.6 & 10.4 & 6.3 & 4.4 & 1.3 & 1.4 & 0.5 & 0.4 & 0.8 & 0.6 + 13 & 4.2 & 2.6 & 2.8 & 4.9 & 9.1 & 5.0 & 1.5 & 1.8 & 0.5 & 0.3 & 0.8 & 0.6 + 14 & 2.2 & 2.4 & 2.6 & 4.1 & 3.3 & 2.2 & 0.8 & 1.5 & 0.7 & 0.8 & 1.0 & 1.2 + min & 2.2 & 1.5 & 2.4 & 1.7 & 3.3 & 2.2 & 0.8 & 0.4 & 0.5 & 0.3 & 0.5 & 0.5 + max & 19.8 & 7.1 & 16.8 & 10.4 & 13.2 & 24.0 & 1.5 & 1.8 & 3.9 & 3.6 & 4.3 & 3.9 + mean & 7.2 & 3.6 & 6.5 & 4.7 & 6.3 & 8.6 & 1.2 & 1.1 & 1.2 & 1.0 & 1.1 & 1.0 + stdev & 5.9 & 1.8 & 5.8 & 2.7 & 3.2 & 8.8 & 0.2 & 0.5 & 0.9 & 0.9 & 1.0 & 0.9 [ tab_age ] lcccccccccc 1 & 1.1e+05 & 6.9e+04 & 2.5e+04 & 2.7e+04 & 1.5 & 0.9 & 4.3 & 2.8 & 4.0 & 2.6 + 2 & 6.0e+04 & 5.2e+04 & 1.9e+04 & 2.1e+04 & 1.2 & 0.9 & 3.1 & 2.7 & 2.9 & 2.5 + 3 & 1.4e+05 & 1.2e+05 & 4.4e+04 & 4.7e+04 & 1.2 & 0.9 & 3.2 & 2.8 & 3.0 & 2.6 + 4 & 3.1e+04 & 2.3e+04 & 1.7e+04 & 1.8e+04 & 1.3 & 0.9 & 1.9 & 1.4 & 1.7 & 1.3 + 6 & 7.4e+05 & 3.8e+05 & 6.4e+04 & 9.9e+04 & 2.0 & 0.6 & 11.6 & 5.9 & 7.5 & 3.8 + 7 & 1.0e+06 & 6.5e+05 & 1.9e+05 & 3.1e+05 & 1.6 & 0.6 & 5.5 & 3.4 & 3.4 & 2.1 + 8 & 9.3e+05 & 5.2e+05 & 4.1e+04 & 3.6e+04 & 1.8 & 1.1 & 22.8 & 12.7 & 25.6 & 14.2 + 9 & 1.2e+05 & 7.7e+04 & 1.8e+04 & 2.0e+04 & 1.5 & 0.9 & 6.4 & 4.2 & 6.0 & 3.9 + 10 & 5.7e+05 & 4.5e+05 & 6.8e+04 & 1.3e+05 & 1.3 & 0.5 & 8.3 & 6.6 & 4.4 & 3.5 + 11 & 9.1e+04 & 7.3e+04 & 5.0e+04 & 5.8e+04 & 1.3 & 0.9 & 1.8 & 1.4 & 1.6 & 1.2 + 12 & 7.9e+04 & 6.3e+04 & 5.5e+04 & 4.8e+04 & 1.3 & 1.1 & 1.4 & 1.1 & 1.6 & 1.3 + 13 & 4.3e+04 & 2.3e+04 & 2.6e+04 & 1.6e+04 & 1.8 & 1.6 & 1.7 & 0.9 & 2.7 & 1.4 + 14 & 3.0e+04 & 2.8e+04 & 3.9e+03 & 4.2e+03 & 1.1 & 0.9 & 7.6 & 7.2 & 7.1 & 6.7 + min & 3.1e+04 & 2.3e+04 & 1.7e+04 & 1.6e+04 & 1.2 & 0.5 & 1.4 & 0.9 & 1.6 & 1.2 + max & 1.0e+06 & 6.5e+05 & 1.9e+05 & 3.1e+05 & 2.0 & 1.6 & 22.8 & 12.7 & 25.6 & 14.2 + mean & 3.3e+05 & 2.1e+05 & 5.1e+04 & 6.9e+04 & 1.5 & 0.9 & 6.1 & 4.1 & 5.5 & 3.6 + stdev & 3.8e+05 & 2.2e+05 & 4.7e+04 & 8.3e+04 & 0.3 & 0.3 & 5.9 & 3.3 & 6.4 & 3.5 + [ tab_mass ] lccccc 1 & -1.7 & 0.18 & 0.02 & 0.06 & 1.4e+38 + 2 & -1.9 & 0.16 & 0.06 & 0.10 & 9.4e+37 + 3 & -0.9 & 0.28 & 0.28 & 0.20 & 2.0e+38 + 4 & -1.0 & 0.23 & 0.24 & 0.20 & 5.0e+37 + 6 & -1.8 & 0.13 & 0.20 & 0.04 & 8.8e+37 + 7 & -1.7 & 0.13 & 0.22 & 0.06 & 1.9e+38 + 8 & -1.9 & 0.10 & 0.16 & 0.16 & 1.5e+38 + 9 & -1.9 & 0.10 & 0.14 & 0.18 & 1.3e+38 + 10 & -1.9 & 0.14 & 0.02 & 0.00 & 1.3e+38 + 11 & -2.5 & 0.05 & 0.00 & 0.08 & 1.5e+38 + 12 & -2.6 & 0.04 & 0.00 & 0.00 & 1.3e+38 + 13 & -2.6 & 0.03 & 0.00 & 0.02 & 4.8e+37 + 14 & -1.1 & 0.23 & 0.18 & 0.22 & 5.7e+37 [ tab_reddening ]

we derive an asymptotic expansion for the excess risk ( regret ) of a weighted nearest - neighbour classifier . this allows us to find the asymptotically optimal vector of non - negative weights , which has a rather simple form . we show that the ratio of the regret of this classifier to that of an unweighted-nearest neighbour classifier depends asymptotically only on the dimension of the feature vectors , and not on the underlying populations . the improvement is greatest when , but thereafter decreases as . the popular bagged nearest neighbour classifier can also be regarded as a weighted nearest neighbour classifier , and we show that its corresponding weights are somewhat suboptimal when is small ( in particular , worse than those of the unweighted-nearest neighbour classifier when ) , but are close to optimal when is large . finally , we argue that improvements in the rate of convergence are possible under stronger smoothness assumptions , provided we allow negative weights . * key words : * bagging , classification , nearest neighbours , weighted nearest neighbour classifiers .

1101.5783

supervised classification , also known as pattern recognition , is a fundamental problem in statistics , as it represents an abstraction of the decision - making problem faced by many applied practitioners . examples include a doctor making a medical diagnosis , a handwriting expert performing an authorship analysis , or an email filter deciding whether or not a message is genuine . classifiers based on nearest neighbours are perhaps the simplest and most intuitively appealing of all nonparametric classifiers . the @xmath0-nearest neighbour classifier was originally studied in the seminal works of @xcite ( later republished as @xcite ) and @xcite , but it retains its popularity today . surprisingly , it is only recently that detailed understanding of the nature of the error probabilities has emerged @xcite . arguably the most obvious defect with the @xmath0-nearest neighbour classifier is that it places equal weight on the class labels of each of the @xmath0 nearest neighbours to the point @xmath5 being classified . intuitively , one would expect improvements in terms of the misclassification rate to be possible by putting decreasing weights on the class labels of the successively more distant neighbours . the first purpose of this paper is to describe the asymptotic structure of the difference between the misclassification rate ( risk ) of a weighted nearest neighbour classifier and that of the optimal bayes classifier for classification problems with feature vectors in @xmath6 . theorem [ thm : main ] in section [ sec : main ] below shows that , subject to certain regularity conditions on the underlying distributions of each class and the weights , this excess risk ( or _ regret _ ) asymptotically decomposes as a sum of two dominant terms , one representing bias and the other representing variance . for simplicity of exposition , we will deal initially with binary classification problems , though we also indicate the appropriate extension to general multicategory problems . our second contribution , following on from the first , is to derive the vector of non - negative weights that is asymptotically optimal in the sense of minimising the misclassification rate ( cf . theorem [ thm : optweights ] ) . in fact this asymptotically optimal weight vector has a relatively simple form : let @xmath7 denote the sample size and let @xmath8 denote the weight assigned to the @xmath9th nearest neighbour ( normalised so that @xmath10 . then the optimal choice is to set @xmath11 ( an explicit expression for @xmath12 is given in ( [ eq : kstar ] ) below ) and then let @xmath13 & \mbox{for $ i=1,\ldots , k^*$ } \\ 0 & \mbox{for $ i = k^*+1,\ldots , n$. } \end{array } \right.\ ] ] thus , in the asymptotically optimal weighting scheme , only a proportion @xmath14 of the weights are positive . the maximal weight is almost @xmath15 times the average positive weight , and the discrete distribution on @xmath16 defined by the asymptotically optimal weights decreases in a concave fashion when @xmath4 , in a linear fashion when @xmath17 and in a convex fashion when @xmath18 ; see figure [ fig : decweights ] . when @xmath1 is large , about @xmath19 of the weights are above the average positive weight . [ t][t]0 [ t][t]0.002 [ t][t]0.004 [ t][t]0.006 [ t][t]0.008 [ t][t]0.01 [ t][t]0 [ t][t]20 [ t][t]40 [ t][t]60 [ t][t]80 [ t][t]100 [ t][t]neighbour order [ t][t]optimal weight another consequence of theorem [ thm : optweights ] is that @xmath20 is bigger by a factor of @xmath21 than the asymptotically optimal choice of @xmath0 for traditional , unweighted @xmath0-nearest neighbour classification . it is notable that this factor , which is around 1.27 when @xmath4 and increases towards 2 for large @xmath1 , does not depend on the underlying populations . this means that there is a natural correspondence between any unweighted @xmath0-nearest neighbour classifier and one of optimally weighted form , obtained by multiplying @xmath0 by this dimension - dependent factor to obtain the number @xmath22 of positive weights for the weighted classifier , and then using the weights given in ( [ eq : optweights ] ) with @xmath22 replacing @xmath20 . in corollary 3 we describe the asymptotic improvement in the excess risk that is attainable using the procedure described in the previous paragraph . since the rate of convergence to zero of the excess risk is @xmath23 in both cases , the improvement is in the leading constant , and again it is notable that the asymptotic improvement does not depend on the underlying populations . the improvement is relatively modest , which goes some way to explaining the continued popularity of the ( unweighted ) @xmath0-nearest neighbour classifier . nevertheless , for @xmath24 , the improvement in regret is at least 5% , though it is negligible as @xmath3 ; the greatest improvement occurs when @xmath2 , and here it is just over 8% . see figure [ fig : asympimp ] . [ t][t]0.90 [ t][t]0.92 [ t][t]0.94 [ t][t]0.96 [ t][t]0.98 [ t][t]1.00 [ t][t]1.02 [ t][t]0 [ t][t]10 [ t][t]20 [ t][t]30 [ t][t]40 [ t][t]50 [ t][t]@xmath1 [ t][t]regret ratio another popular way of improving the performance of a classifier is by _ bagging _ @xcite . short for ` bootstrap aggregating ' , bagging involves combining the results of many empirically simulated predictions . empirical analyses , e.g. @xcite , have reported that bagging can result in improvements over unweighted @xmath0-nearest neighbour classification . moreover , as explained by @xcite , understanding the properties of the bagged nearest neighbour classifier is also of interest because they provide insight into _ random forests _ @xcite . random forest algorithms have been some of the most successful ensemble methods for regression and classification problems , but their theoretical properties remain relatively poorly understood . when bagging the nearest neighbour classifier , we can draw resamples from the data either with- or without - replacement . we treat the ` infinite simulation ' case , where both versions take the form of a weighted nearest neighbour classifier with weights decaying approximately exponentially on successively more distant observations from the point being classified @xcite . the crucial choice is that of the resample size , or equivalently the sampling fraction , i.e. the ratio of the resample size to the original sample size . in section [ sec : bagged ] , we describe the asymptotically optimal resample fraction ( showing in particular that it is the same for both with- and without - replacement sampling ) and compare its regret with those of the weighted and unweighted @xmath0-nearest neighbour classifiers . finally , in section [ sec : negweights ] , we consider the problem of choosing optimal weights without the restriction that they should be non - negative . the situation here is somewhat analogous to the use of higher order kernels for classifiers based on kernel density estimates of each of the population densities . in particular , subject to additional smoothness assumptions on the population densities , we find that powers of @xmath7 arbitrarily close to the ` parametric rate ' of @xmath25 for the excess risk are attainable . all proofs are deferred to the appendix . classification has been the subject of several book - length treatments , including @xcite , @xcite and @xcite . in particular , classifiers based on nearest neighbours form a central theme of @xcite . the review paper by @xcite contains 243 references and provides a thorough survey of the classification literature up to 2005 . more recently , @xcite have discussed the relative merits of _ plug - in _ classifiers ( a family to which weighted nearest neighbour classifiers belong ) and classifiers based on _ empirical risk minimisation _ , such as support vector machines @xcite . weighted nearest neighbour classifiers were first studied by @xcite ; see also @xcite . @xcite proved that if @xmath26 as @xmath27 and @xmath28 for some @xmath29 with @xmath30 as @xmath27 , then risk of the weighted nearest neighbour classifier converges to the risk of the bayes classifier ; see also @xcite . as mentioned above , this work attempts to study the difference between these risks more closely . weighted nearest neighbour classifiers are also related to classifiers based on kernel estimates of each of the class densities ; see for example the review by @xcite , as well as @xcite . the @xmath23 rates of convergence obtained in this paper for non - negative weights are the same as those obtained by @xcite under similar twice - differentiable conditions with second - order kernel estimators of the class densities . @xcite proved that in a certain sense this is the minimax optimal rate , though his assumptions and context are slightly different from what is studied here . further related work includes the literature on _ highest density region _ or _ level set _ estimation @xcite . @xcite and @xcite proved an analogous result for the bagged nearest neighbour classifier to the @xcite result described in the previous paragraph . more precisely , if the resample size @xmath31 used for the bagging diverges to infinity , and @xmath32 as @xmath27 , then the risk of the bagged nearest neighbour classifier converges to the bayes risk . note that this result does not depend on whether the resamples are taken with or without replacement from the training data . @xcite have recently proved a striking rate of convergence result for the bagged nearest neighbour estimate ; this is described in greater detail in section [ sec : bagged ] . let @xmath33 be independent and identically distributed pairs taking values in @xmath34 . we suppose that @xmath35 for some @xmath36 and that @xmath37 for @xmath38 , where @xmath39 is a probability measure on @xmath6 . we write @xmath40 for the marginal distribution of @xmath41 and let @xmath42 denote the corresponding regression function . a _ classifier _ @xmath43 is a borel measurable function from @xmath6 to @xmath44 , with the interpretation that the point @xmath45 is classified as belonging to class @xmath46 . the _ misclassification rate _ , or _ risk _ of @xmath43 over a borel measurable set @xmath47 is defined to be @xmath48.\ ] ] the classifier which minimises the risk over @xmath49 is the bayes classifier , given by @xmath50 its risk is @xmath51 for each @xmath52 , let @xmath53 denote a vector of weights , normalised so that @xmath54 . fix @xmath55 and an arbitrary norm @xmath56 on @xmath6 , and let @xmath57 denote a permutation of the training sample @xmath58 such that @xmath59 . we define the _ weighted nearest neighbour classifier _ to be @xmath60 we also write @xmath61 where it is necessary to emphasise the weight vector , for example when comparing different weighted nearest neighbour classifiers . our initial goal is to study the asymptotic behaviour of @xmath62,\ ] ] where the probability is taken over the joint distribution of @xmath63 and @xmath64 . it will be convenient to define a little notation : for a smooth function @xmath65 , we write @xmath66 for its derivative at @xmath5 , and @xmath67 for its @xmath68th partial derivative at @xmath5 . analogously , we write @xmath69 for the second derivative of @xmath70 at @xmath5 , and @xmath71 for the @xmath72th element of the corresponding hessian matrix at @xmath5 . we let @xmath73 denote the closed ball of radius @xmath74 centered at @xmath5 in the norm @xmath56 , and let @xmath75 denote the @xmath1-dimensional lebesgue measure of the unit ball @xmath76 . we will make use of the following assumptions for our theoretical results : ( a.1 ) : : the set @xmath47 is a compact @xmath1-dimensional manifold with boundary @xmath77 . ( a.2 ) : : the set @xmath78 is non - empty . there exists an open subset @xmath79 of @xmath6 that contains @xmath80 and such that the following properties hold : firstly , @xmath81 is bounded away from zero for @xmath82 , where @xmath83 is an open set containing @xmath49 ; secondly the restrictions of @xmath84 and @xmath85 to @xmath79 are absolutely continuous with respect to lebesgue measure , with twice continuously differentiable radon nikodym derivatives @xmath86 and @xmath87 respectively . ( a.3 ) : : there exists @xmath88 such that @xmath89 . moreover , for sufficiently small @xmath90 , the ratio @xmath91 is bounded away from zero , uniformly for @xmath55 . ( a.4 ) : : for all @xmath92 , we have @xmath93 , and for all @xmath94 , we have @xmath95 , where @xmath96 denotes the restriction of @xmath97 to @xmath77 . the introduction of the compact set @xmath49 finesses the problem of performing classification in the tails of the feature vector distributions . see for example ( * ? ? ? * section 3 ) for further discussion of this point and related results , as well as @xcite . @xcite and @xcite impose similar compactness assumptions for their results . the set @xmath49 may be arbitrarily large , though the larger it is , the stronger are the requirements in ( * a.2 * ) . although as stated , the assumptions on @xmath49 are quite general , little is lost by thinking of @xmath49 as a large closed euclidean ball . its role in the asymptotic expansion of theorem [ thm : optweights ] below is that it is involved in the definition of the set @xmath80 , which represents the decision boundary of the bayes classifier . we will see that the behaviour of @xmath86 and @xmath87 on the set @xmath80 is crucial for determining the asymptotic behaviour of weighted nearest neighbour classifiers . the second part of ( * a.3 * ) asks that the ratio of the @xmath98-measure of small balls to the corresponding @xmath1-dimensional lebesgue measure is bounded away from zero . this requirement is satisfied , for instance , if @xmath84 and @xmath85 are absolutely continuous with respect to lebesgue measure , with radon nikodym derivatives that are bounded away from zero on the open set @xmath83 . the assumption in ( * a.4 * ) that @xmath93 for @xmath92 asks that @xmath86 and @xmath87 , weighted by the respective prior probabilities of each class , should cut at a non - zero angle along @xmath80 . in the language of differential topology , this means that 1/2 is a _ regular value _ of the function @xmath97 , and the second part of ( * a.4 * ) asks for 1/2 to be a regular value of the restriction of @xmath97 to @xmath77 . together , these two requirements ensure that @xmath80 is a @xmath99-dimensional submanifold with boundary of @xmath6 , and the boundary of @xmath80 is @xmath100 @xcite . the requirement in ( * a.4 * ) that @xmath93 for @xmath92 is related to the well - known _ margin condition _ of , e.g. @xcite and @xcite ; when it holds ( and in the presence of the other conditions ) , there exist @xmath101 such that @xmath102 for sufficiently small @xmath103 ; see ( * ? ? ? * proposition 1 ) . a proof of this fact , which uses weyl s tube formula @xcite , is given after the proof of theorem [ thm : main ] in the appendix . in this sense , we work in the setting of a margin condition with the power parameter equal to 1 . we now introduce some notation needed for theorem [ thm : main ] below . for @xmath104 , let @xmath105 denote the set of all sequences of non - negative deterministic weight vectors @xmath53 satisfying * @xmath106 ; * @xmath107 , where @xmath108 ; note that this latter expression appears in ( [ eq : optweights ] ) ; * @xmath109 , where @xmath110 ; * @xmath111 ; * @xmath112 . observe that @xmath113 for @xmath114 . the first and last conditions ensure that the weights are not too concentrated on a small number of points ; the second amounts to a mild moment condition on the probability distribution on @xmath16 defined by the weights . the next two conditions ensure that not too much weight ( or squared weight in the case of the latter condition ) is assigned to observations that are too far from the point being classified . although there are many requirements on the weight vectors , they are rather mild conditions when @xmath115 is small , as can be seen by considering the limiting case @xmath116 . for instance , for the unweighted @xmath0-nearest neighbour classifier with weights @xmath53 given by @xmath117 , we have that @xmath118 for small @xmath104 provided that @xmath119 . thus for the vector of @xmath0-nearest neighbour weights to belong to @xmath105 for all large @xmath7 , it is necessary that the usual conditions @xmath120 and @xmath30 for consistency are satisfied , and these conditions are almost sufficient when @xmath104 is small . the situation is similar for the bagged nearest neighbour classifier see section [ sec : bagged ] below . the fact that the weights are assumed to be deterministic means that they depend only on the ordering of the distances , not the raw distances themselves ( as would be the case for a classifier based on kernel density estimates of the population densities ) . such classifiers are not necessarily straightforward to implement , however : @xcite showed that even in the simple situation where @xmath4 and @xmath121 and @xmath122 cross at a single point @xmath123 , the optimal order of the bandwidth for the kernel depends on the sign of @xmath124 . continuing with our notational definitions , let @xmath125 , and let @xmath126 where @xmath127 denotes the laplacian of @xmath97 at @xmath5 . define @xmath128 where @xmath129 denotes the natural @xmath99-dimensional volume measure that @xmath80 inherits as a subset of @xmath6 . note that @xmath130 , and @xmath131 , with equality if and only if @xmath132 is identically zero on @xmath80 . although the definitions of @xmath133 and @xmath134 are complicated , we will see after the statement of theorem [ thm : main ] below that they are comprised of terms that have natural interpretations . [ thm : main ] assume ( * a.1 * ) , ( * a.2 * ) , ( * a.3 * ) and ( * a.4 * ) . then for each @xmath104 , @xmath135 as @xmath27 , uniformly for @xmath118 , where @xmath136 theorem [ thm : main ] tells us that , asymptotically , the dominant contribution to the excess risk , or _ regret _ , of the weighted nearest neighbour classifier over @xmath49 , can be decomposed as a sum of two terms . the two terms , constant multiples of @xmath137 and @xmath138 respectively , represent variance and squared bias contributions to the regret . it is interesting to observe that , although the 0 - 1 classification loss function is quite different from the squared error loss often used in regression problems , we nevertheless obtain such an asymptotic decomposition . the constant multiples of the dominant variance and squared bias terms depend only on the behaviour of @xmath86 and @xmath87 ( and their first and second derivatives ) on @xmath80 , as seen from ( [ eq : c1c2 ] ) . moreover , we can see from the expression for @xmath133 in ( [ eq : c1c2 ] ) that the contribution to the dominant variance term in the regret will tend to be large in the following three situations : firstly , when @xmath139 is large on @xmath80 ; secondly when the @xmath129 measure of @xmath80 is large ; and thirdly when @xmath140 is small on @xmath80 . in the first two of these situations , the probability is relatively high that a point to be classified will be close to the bayes decision boundary @xmath80 , where classification is difficult . in the latter case , the regression function @xmath97 moves away from @xmath141 only slowly as we move away from @xmath80 , meaning that there is a relatively large region of points near @xmath80 where classification is difficult . from the expression for @xmath134 in ( [ eq : c1c2 ] ) , we see that the dominant squared bias term is also large in these situations , and also when @xmath142 is large on @xmath80 . from the proof of theorem [ thm : main ] , it is apparent that @xmath143 is the dominant bias term for @xmath144 as an estimator of @xmath145 . indeed , by a taylor expansion , @xmath146 the two summands in the definition of @xmath147 represent asymptotic approximations to the respective summands in this approximation . consider now problem of optimising the choice of weight vectors . let @xmath148 and then define the weights @xmath149 as in ( [ eq : optweights ] ) . the first part of theorem [ thm : optweights ] below can be regarded as saying that the weights @xmath150 are asymptotically optimal . [ thm : optweights ] assume * ( a.1)*(*a.4 * ) and assume also that @xmath151 . for any @xmath104 and any sequence @xmath152 , we have @xmath153 moreover , the ratio in ( [ eq : regratio ] ) above converges to 1 if and only if both @xmath154 and @xmath155 . equivalently , this occurs if and only if both @xmath156 finally , @xmath157 now write @xmath158 for the traditional , unweighted @xmath0-nearest neighbour classifier ( or equivalently , the weighted nearest neighbour classifier with @xmath159 for @xmath160 and @xmath161 otherwise ) . another consequence of theorem [ thm : main ] is that , provided * ( a.1)*(*a.4 * ) hold and @xmath151 , the quantity @xmath20 defined in ( [ eq : kstar ] ) is larger by a factor of @xmath21 ( up to an unimportant rounding error ) than the asymptotically optimal choice of @xmath162 for @xmath158 ; see also @xcite . we can therefore compare the performance of @xmath163 with that of @xmath164 . [ cor : regretratio ] assume * ( a.1)*(*a.4 * ) and assume also that @xmath151 . then @xmath165 as @xmath27 . since the limit in ( [ eq : regretratio ] ) does not depend on the underlying populations , we can plot it as a function of @xmath1 ; cf . figure [ fig : asympimp ] . in fact , corollary [ cor : regretratio ] suggests a natural correspondence between any unweighted @xmath0-nearest neighbour classifier @xmath158 and the weighted nearest neighbour classifier which we denote by @xmath166 whose weights are of the optimal form ( [ eq : optweights ] ) , but with @xmath20 replaced with @xmath167 under the conditions of corollary [ cor : regretratio ] , we can compare @xmath158 and @xmath166 , concluding that for each @xmath168 , @xmath169 as @xmath27 , uniformly for @xmath170 . the fact that the convergence in ( [ eq : regretratio2 ] ) is uniform for @xmath0 in this range means that the ratio on the left - hand side of ( [ eq : regretratio2 ] ) has the same limit if we replace @xmath0 by an estimator @xmath171 constructed from the training data @xmath172 , provided that @xmath171 lies in this range with probability tending to 1 . finally in this section , we note that the theory presented above can be extended in a natural way to multicategory classification problems , where the class labels take values in the set @xmath173 . writing @xmath174 , let @xmath175 for distinct indices @xmath176 . in addition to * ( a.1 ) * and the obvious analogues of the conditions * ( a.2 ) * , * ( a.3 ) * and ( * a.4 * ) , we require : ( a.5 ) : : for each @xmath177 , the submanifolds @xmath178 and @xmath179 of @xmath6 are transversal . condition ( * a.5 * ) ensures that @xmath180 is either empty or a @xmath181-dimensional submanifold of @xmath6 @xcite . under these conditions , the conclusion of theorem [ thm : main ] holds , provided that the constants @xmath133 and @xmath134 are replaced with @xmath182 and @xmath183 respectively , where each term @xmath184 and @xmath185 is an integral over @xmath178 . apart from the obvious notational changes involved in converting @xmath133 and @xmath134 to @xmath184 and @xmath185 , the only other change required is to replace the constant factor @xmath186 in the definition of @xmath133 with @xmath187 where @xmath188 denotes the common value that @xmath189 and @xmath190 take at @xmath191 . this change accounts for the fact that @xmath188 is not necessarily equal to @xmath141 on @xmath178 . it follows ( provided also that @xmath192 ) that the asymptotically optimal weights are still of the form ( [ eq : optweights ] ) , but with the ratio @xmath193 in the expression for @xmath20 in ( [ eq : kstar ] ) replaced with @xmath194 . moreover , the conclusion of corollary [ cor : regretratio ] and the subsequent discussion also remain true . traditionally , the bagged nearest neighbour classifier is obtained by applying the 1-nearest neighbour classifier to many resamples from the training data . the final classification is made by a majority vote on the classifications obtained from the resamples . in the most common version of bagging where the resamples are drawn with replacement , and the resample size is the same as the original sample size , bagging the nearest neighbour classifier gives no improvement over the 1-nearest neighbour classifier @xcite . this is because the nearest neighbour occurs in more than half ( in fact , roughly a proportion @xmath195 ) of the resamples . nevertheless , if a smaller resample size is used , then substantial improvements over the nearest neighbour classifier are possible , as has been verified empirically by @xcite . in fact , if the resample size is @xmath196 , then the ` infinite simulation ' versions of the bagged nearest neighbour classifier in the with- and without - replacement resampling cases are weighted nearest neighbour classifiers with respective weights @xmath197 and @xmath198 of course , the observations above render the resampling redundant , and we regard the weighted nearest neighbour classifiers with the weights above as defining the two versions of the bagged nearest neighbour classifier . it is convenient to let @xmath199 denote the resampling fraction . intuitively , for large @xmath7 , both versions of the bagged nearest neighbour classifier behave like the weighted nearest neighbour classifier with weights @xmath200 which place a @xmath201 distribution ( conditioned on being in the set @xmath16 ) on the weights : @xmath202 the reason for this is that , in order for the @xmath9th nearest neighbour of the training data to be the nearest neighbour of the resample , the nearest @xmath203 neighbours must not appear in the resample , while the @xmath9th nearest neighbour must appear , and these events are almost independent when @xmath7 is large ; see @xcite . naturally , the parameter @xmath204 plays a crucial role in the performance of the bagged nearest neighbour classifier , and for small @xmath104 , the three vectors of weights given in ( [ eq : bagged1 ] ) , ( [ eq : bagged2 ] ) and ( [ eq : bagged3 ] ) belong to @xmath105 for all large @xmath7 if @xmath205 . in the following corollary of theorem [ thm : main ] , we write @xmath206 to denote either of the bagged nearest neighbour classifiers with weights ( [ eq : bagged1 ] ) , ( [ eq : bagged2 ] ) or their approximation with weights ( [ eq : bagged3 ] ) . [ cor : baggednnexp ] assume ( * a.1*)(*a.4 * ) . for every @xmath168 , @xmath207 uniformly for @xmath208 , where @xmath209 this result is somewhat related to corollary 10 of @xcite . in that paper , the authors study the bagged nearest neighbour estimate @xmath210 of the regression function @xmath97 . they prove in particular that under regularity conditions ( including a lipschitz assumption on @xmath97 ) and for a suitable choice of resample size , @xmath211 = o(n^{-2/(d+2)})\ ] ] for @xmath18 . it is known , e.g. @xcite that this is the minimax optimal rate for their problem . corollary [ cor : baggednnexp ] may also be applied to deduce that the asymptotically optimal choice of @xmath204 in all three cases is @xmath212 thus , in an analogous fashion to section [ sec : main ] , we can consider the performance of @xmath213 relative to that of @xmath163 . [ cor : bagged ] assume * ( a.1)*(*a.4 * ) and assume also that @xmath151 . then @xmath214 as @xmath27 . the limiting ratio in ( [ eq : regretratio3 ] ) is plotted as a function of @xmath1 in figure [ fig : regretratio2 ] . the ratio is about 1.18 when @xmath4 , showing that the bagged nearest neighbour classifier has asymptotically worse performance than the @xmath0-nearest neighbour classifier in this case . the ratio is equal to 1 when @xmath17 , and is less than 1 for @xmath18 . the facts that the asymptotically optimal weights decay as illustrated in figure [ fig : decweights ] and that the bagged nearest neighbour weights decay approximately geometrically explain why the bagged nearest neighbour classifier has almost optimal performance among weighted nearest neighbour classifiers when @xmath1 is large . similar to the discussion following corollary [ cor : regretratio ] , based on the expressions for @xmath162 and @xmath215 , there is a natural correspondence between the unweighted @xmath0-nearest neighbour classifier @xmath216 with data driven @xmath171 , and the bagged nearest neighbour classifier @xmath217 , where @xmath218 the same limit ( [ eq : regretratio3 ] ) holds for the regret ratio of these classifiers , again provided there exists @xmath168 such that @xmath219 . [ t][t]0.90 [ t][t]0.92 [ t][t]1.00 [ t][t]1.05 [ t][t]1.10 [ t][t]1.15 [ t][t]1.20 [ t][t]0 [ t][t]10 [ t][t]20 [ t][t]30 [ t][t]40 [ t][t]50 [ t][t]@xmath1 [ t][t]regret ratio if we allow negative weights , it is possible to choose weights satisfying @xmath220 . this means that we can eradicate the dominant squared bias term in the asymptotic expansion of theorem [ thm : main ] . it follows that , subject to additional smoothness conditions , we can achieve faster rates of convergence with weighted nearest neighbour classifiers , as we now describe . the appropriate variant of condition ( * a.2 * ) , which we denote by * ( a.2)(r ) * , is as follows : ( a.2)(r ) : : the set @xmath78 is non - empty . there exists an open subset @xmath79 of @xmath6 that contains @xmath80 and such that the following properties hold : firstly , @xmath81 is bounded away from zero for @xmath82 , where @xmath83 is an open set containing @xmath49 ; secondly the restrictions of @xmath84 and @xmath85 to @xmath79 are absolutely continuous with respect to lebesgue measure , with @xmath221-times continuously differentiable radon nikodym derivatives @xmath86 and @xmath87 respectively . thus condition * ( a.2)(1 ) * is identical to * ( a.2)*. note that we are still in the setting of a margin condition with power parameter equal to 1 . for non - negative integers @xmath222 and @xmath223 and @xmath224 , let @xmath225 let @xmath226 be a multi - index ( i.e. a @xmath1-tuple of non - negative integers ) . we write @xmath227 , and for @xmath228 , we write @xmath229 . now , for @xmath230 , let @xmath231 where we evaluate the integral by transforming to spherical coordinates . it is convenient here to use multi - index notation for derivatives , so we write @xmath232 . as non - standard multi - index notation , it is also convenient for @xmath230 to write @xmath233 . now let @xmath234 so that @xmath235 . let @xmath236 for @xmath237 , define @xmath238 . we consider restrictions on the set of weight vectors analogous to those imposed on @xmath239th order kernels in kernel density estimation . specifically , we let @xmath240 denote the set of deterministic weight vectors @xmath53 satisfying * @xmath54 , @xmath241 for @xmath242 ; * @xmath106 ; * @xmath243 ; * there exists @xmath244 such that @xmath245 and such that @xmath246 ; * @xmath111 ; * @xmath247 . finally , we are in a position to state the analogue of theorem [ thm : main ] for weight vectors in @xmath240 . [ thm : main2 ] assume ( * a.1 * ) , ( * a.2*)(*r * ) , ( * a.3 * ) and ( * a.4 * ) . then for each @xmath104 , @xmath248 as @xmath27 , uniformly for @xmath249 , where @xmath250 a consequence of theorem [ thm : main2 ] is that we can construct weighted nearest neighbour classifiers which , under conditions ( * a.1 * ) , ( * a.2*)(*r * ) , ( * a.3 * ) and ( * a.4 * ) , and provided @xmath251 , achieve the rate of convergence @xmath252 for the regret . to illustrate this , set @xmath253 , and in order to satisfy the restrictions on the allowable weights , consider weight vectors with @xmath161 for @xmath254 . then , by mimicking the proof of theorem [ thm : optweights ] and seeking to minimise ( [ eq : gamma ] ) subject to the constraints @xmath255 and @xmath256 for @xmath242 , we obtain minimising weights of the form @xmath257 the equations @xmath258 and @xmath259 for @xmath260 for weight vectors of the form ( [ eq : negweightvec ] ) yield @xmath239 linear equations in the @xmath261 unknowns @xmath262 . although these equations can be solved directly in terms of @xmath263 say , simpler expressions are obtained by solving asymptotic approximations to these equations . in particular , since it is an elementary fact that for non - negative integers @xmath264 and @xmath265 , @xmath266 as @xmath120 , we can just deal with the dominant terms . as examples , when @xmath267 , we find @xmath268 and when @xmath269 , we should take @xmath270 under the conditions of theorem [ thm : main2 ] , and provided @xmath251 , these weighted nearest neighbour classifiers achieve the @xmath252 convergence rate . the choice of @xmath263 involves a trade - off between the desire to keep the remaining squared bias term @xmath271 small , and the need for it to be large enough to remain the dominant bias term . this reflects the fact that the asymptotic results of this section should be applied with some caution . besides the discomfort many practitioners might feel in using negative weights , one would anticipate that rather large sample sizes would be needed for the leading terms in the asymptotic expansion ( [ eq : higherorderexp ] ) to dominate the error terms . this is also the reason why we do not pursue here methods such as lepski s method @xcite that adapt to an unknown smoothness level around @xmath80 . of theorem [ thm : main ] the proof is rather lengthy , so we briefly outline the main ideas here . write @xmath272 and observe that @xmath273 \ , dp_1(x ) \nonumber \\ & \hspace{1cm}+ \int_{\mathcal{r } } ( 1-\pi)\bigl[\mathbb{p}\{\hat{c}_n^{\mathrm{wnn}}(x ) = 1\ } - \mathbbm{1}_{\{c^{\mathrm{bayes}}(x)=1\}}\bigr ] \ , dp_2(x ) \nonumber \\ & = \int_{\mathcal{r } } \biggl\{\mathbb{p}\biggl(\sum_{i=1}^n w_{ni } \mathbbm{1}_{\{y_{(i)}=1\ } } < \frac{1}{2}\biggr ) - \mathbbm{1}_{\{\eta(x ) < 1/2\}}\biggr\ } \ , dp^\circ(x).\end{aligned}\ ] ] for @xmath103 , let @xmath274 where @xmath275 . moreover , let @xmath276 the dominant contribution to the integral in ( [ eq : mainarg ] ) comes from @xmath277 , where @xmath278 . since the unit vector @xmath279 is orthogonal to the tangent space of @xmath80 at @xmath123 , we can decompose the integral over @xmath277 as an integral along @xmath80 and an integral in the perpendicular direction . we then apply a normal approximation to the integrand to deduce the result . this normal approximation requires asymptotic expansions to the mean and variance of the sum of independent random variables in ( [ eq : mainarg ] ) , and these are developed in * step 1 * and * step 2 * below respectively . in order to retain the flow of the main argument , we concentrate on the dominant terms in the first five steps of the argument , simply labelling the many remainder terms as @xmath280 . the sizes of these remainder terms are controlled in * step 6 * , where we also present an additional side calculation . by a taylor expansion , @xmath285 where we show in * step 6 * that @xmath286 uniformly for @xmath118 . writing @xmath287 , we also show in * step 6 * that for @xmath288 and @xmath289 , the restriction of the distribution of @xmath290 to a sufficiently small ball about the origin is absolutely continuous with respect of lebesgue measure , with radon nikodym derivative given at @xmath291 by @xmath292 let @xmath293 . by examining the argument leading to ( [ eq : aic ] ) , we see that we can replace @xmath74 there with @xmath294 , to conclude that for all @xmath295 , @xmath296 it follows that @xmath297 uniformly for @xmath288 and @xmath298 . similarly @xmath299 uniformly for @xmath288 and @xmath298 . let @xmath300 , and let @xmath301 with @xmath302 ( where we introduce the comma here for clarity ) . by a taylor expansion , we have @xmath303\mathbb{p}\{\mathrm{bin}(n-1,p_{\|u\| } ) = i-1\ } \ , du \nonumber \\ & = \{1+o(1)\}\sum_{i = k_1}^{k_2 } n \delta w_{ni}\sum_{j=1}^d \int_{\|u\| \leq \delta_n } \bigl\{\eta_j(x)u_j^2\bar{f}_j(x ) + \tfrac{1}{2}\eta_{jj}(x)u_j^2 \bar{f}(x)\bigr\}\mathbb{p}\{\mathrm{bin}(n-1,p_{\|u\| } ) < i\ } \ , du,\end{aligned}\ ] ] uniformly for @xmath288 and @xmath118 . now , @xmath304 is decreasing in @xmath305 and is close to 1 when @xmath305 is small and close to zero when @xmath305 is large . to analyse this more precisely , note that @xmath306 as @xmath307 , uniformly for @xmath288 , so it is convenient to let @xmath308 and set @xmath309 . then there exists @xmath310 such that for @xmath311 , we have for all @xmath288 , all @xmath312 $ ] and all @xmath313 that @xmath314 thus by bernstein s inequality @xcite , for each @xmath315 and for @xmath311 , @xmath316 } \sup_{k_1 \leq i \leq k_2 } \bigl[1 - \mathbb{p}\{\mathrm{bin}(n-1,p_{\|v\|/b_n } ) < i\}\bigr ] \leq \exp\biggl(-\frac{k_1}{3\log^2 n}\biggr ) = o(n^{-m}).\ ] ] similarly , for @xmath311 , @xmath317 } \sup_{k_1 \leq i \leq k_2 } \mathbb{p}\{\mathrm{bin}(n-1,p_{\|v\|/b_n } ) < i\ } \leq \exp\biggl(-\frac{k_1}{3\log^2 n}\biggr ) = o(n^{-m}).\ ] ] we deduce from ( [ eq : firstderiv ] ) , ( [ eq : secondderiv ] ) , ( [ eq : delta ] ) , ( [ eq : lower ] ) and ( [ eq : upper ] ) that @xmath318 uniformly for @xmath288 and @xmath118 . combining ( [ eq : taylor ] ) , ( [ eq : r1 ] ) and ( [ eq : biascompletion ] ) , this completes * step 1*. * step 2 * : let @xmath319 and let @xmath320 . we claim that @xmath321 uniformly for @xmath118 . to see this , note that @xmath322 + \sum_{i=1}^n w_{ni}^2 \mathrm{var } \ \eta(x_{(i ) } ) \\ & = \sum_{i=1}^n w_{ni}^2 \bigl[\mathbb{e}\eta(x_{(i ) } ) - \{\mathbb{e}\eta(x_{(i)})\}^2\bigr ] . \end{aligned}\ ] ] but by a simplified version of the argument in * step 1 * , we have @xmath323 it follows that @xmath324 uniformly for @xmath118 . similarly , @xmath325 uniformly for @xmath118 . this completes * step 2*. * step 3 * for @xmath326 and @xmath327 , we write @xmath328 for brevity . moreover , we write @xmath329 for the radon nikodym derivative with respect to lebesgue measure of the restriction of @xmath330 to @xmath331 for large @xmath7 . we show that @xmath332 \ , d(p_1 - p_2)(x ) \nonumber \\ & = \int_{\mathcal{s } } \int_{-\epsilon_n}^{\epsilon_n } \psi(x_0^t)\bigl[\mathbb{p}\{s_n(x_0^t ) < 1/2\ } - \mathbbm{1}_{\{t < 0\}}\bigr ] \ , dt \ , d\mathrm{vol}^{d-1}(x_0)\{1 + o(1)\},\end{aligned}\ ] ] uniformly for @xmath118 . recalling the definition of @xmath333 in ( [ eq : sepseps ] ) , note that for large @xmath7 , the map @xmath334 is a diffeomorphism from @xmath335 onto @xmath331 @xcite . observe that @xmath336 moreover , for large @xmath7 and @xmath337 , we have @xmath338 . the pullback of the @xmath1-form @xmath339 is given at @xmath340 by @xmath341 where the error term is uniform in @xmath340 for @xmath326 and @xmath337 . it follows from the theory of integration on manifolds , as described in @xcite and ( * ? ? ? * theorems 3.15 and 4.7 ) ( see also @xcite ) , that @xmath342 \ , dp^\circ(x ) \nonumber \\ & = \int_{\mathcal{s}^{\epsilon_n \epsilon_n } } \int_{-\epsilon_n}^{\epsilon_n } \psi(x_0^t)\bigl[\mathbb{p}\{s_n(x_0^t ) < 1/2\ } - \mathbbm{1}_{\{t < 0\}}\bigr ] \ , dt \ , d\mathrm{vol}^{d-1}(x_0)\{1 + o(1)\},\end{aligned}\ ] ] uniformly for @xmath118 . but @xmath343 , and this latter set has volume @xmath344 by weyl s tube formula ( * ? ? ? * theorem 4.8 ) . thus the integral over @xmath331 in ( [ eq : fubini ] ) may be replaced with an integral over @xmath277 and , similarly , the integral over @xmath333 may be replaced with an integral over @xmath80 , without changing the order of the error term in ( [ eq : fubini ] ) . thus ( [ eq : rfubini ] ) holds , and this completes * step 3*. * step 4 * : we now return to the main argument to bound the contribution to the risk ( [ eq : mainarg ] ) from @xmath345 . in particular , we show that @xmath346 \ , dp^\circ(x ) = o(n^{-m}),\ ] ] for all @xmath295 . to see this , recall that @xmath81 is assumed to be bounded away from zero on the set @xmath347 ( for fixed @xmath103 ) , and @xmath348 is bounded away from zero for @xmath326 . hence , by ( [ eq : containment ] ) in * step 3 * , there exists @xmath349 such that , for sufficiently small @xmath103 , @xmath350 we also claim that @xmath282 is similarly bounded away from @xmath141 uniformly for @xmath351 . in fact , we have by hoeffding s inequality that @xmath352 it follows that @xmath353 for sufficiently large @xmath7 . similarly , @xmath354 for large @xmath7 . now we may apply hoeffding s inequality again , this time to @xmath355 , to deduce that @xmath356 for each @xmath295 , using ( [ eq : upper ] ) and ( [ eq : lower ] ) and the fact that @xmath357 for @xmath118 . this completes * step 4*. * step 5 * we now show that @xmath358 \ , dt \ , d\mathrm{vol}^{d-1}(x_0 ) = b_1 s_n^2 + b_2t_n^2 + o(s_n^2 + t_n^2),\ ] ] uniformly for @xmath118 , where @xmath133 and @xmath134 were defined in ( [ eq : c1c2 ] ) . when combined with ( [ eq : mainarg ] ) and the results of * step 3 * and * step 4 * ( in particular , ( [ eq : rfubini ] ) and ( [ eq : outer ] ) ) , this will complete the proof of theorem [ thm : main ] . first observe that @xmath359 \ , dt \ , d\mathrm{vol}^{d-1}(x_0 ) \nonumber \\ & = \int_{\mathcal{s } } \int_{-\epsilon_n}^{\epsilon_n } t\|\dot{\psi}(x_0)\|\bigl[\mathbb{p}\{s_n(x_0^t ) < 1/2\ } - \mathbbm{1}_{\{t < 0\}}\bigr ] \ , dt \ , d\mathrm{vol}^{d-1}(x_0)\{1+o(1)\}.\end{aligned}\ ] ] now , @xmath355 is a sum of independent , bounded random variables , so by the non - uniform version of the berry esseen theorem , there exists @xmath360 such that @xmath361 } \biggl|\mathbb{p}\biggl(\frac{s_n(x_0^t ) - \mu_n(x_0^t)}{\sigma_n(x_0^t ) } \leq y\biggr ) - \phi(y)\biggr| \leq \frac{c_1}{n^{1/2}(1+|y|^3)},\ ] ] where @xmath362 denotes the standard normal distribution function . thus @xmath363 where we show in * step 6 * that latexmath:[\[\label{eq : r2 } @xmath118 . finally , we can make the substitution @xmath367 to conclude that @xmath368 \ , dt \ , d\mathrm{vol}^{d-1}(x_0 ) \nonumber \\ & = \frac{s_n^2}{4 } \int_{\mathcal{s } } \int_{-\infty}^\infty u\|\dot{\psi}(x_0)\|\biggl\{\phi\biggl(-u\|\dot{\eta}(x_0)\| - \frac{2t_n}{s_n}a(x_0)\biggr ) - \mathbbm{1}_{\{u < 0\}}\biggr\ } \ , du \ , d\mathrm{vol}^{d-1}(x_0 ) + r_4 \nonumber \\ & = b_1 s_n^2 + b_2t_n^2 + r_4,\end{aligned}\ ] ] where @xmath133 and @xmath134 were defined in ( [ eq : c1c2 ] ) . we have used the fact that @xmath369 for @xmath326 in the final step of this calculation . once we have shown in * step 6 * that latexmath:[\[\label{eq : r4 } * step 6 * _ to show ( [ eq : density ] ) , which gives the radon nikodym derivative of the restriction of the distribution of @xmath290 to a small ball about the origin _ : recall that @xmath371 , and that @xmath372 denotes @xmath1-dimensional lebesgue measure . for a borel subset @xmath373 of @xmath6 , let @xmath374 . it follows from the hypothesis * ( a.2 ) * that for @xmath288 with @xmath7 sufficiently large , and for @xmath289 , the restriction of the distribution of @xmath290 to a small ball about the origin is absolutely continuous with respect to @xmath372 . thus for @xmath288 with @xmath7 sufficiently large , for @xmath289 , for @xmath375 with @xmath305 sufficiently small , and for @xmath376 , @xmath377 as @xmath378 , by the lebesgue differentiation theorem . for the other bound , write @xmath379 and observe that @xmath380 \\ & \hspace{1cm}= \frac{1}{a_d \delta^d}\biggl[n\biggl\{\int_{b_\delta(x+u ) } \bar{f}(v ) \ , dv\biggr\}\binom{n-1}{i-1}p_{\|u\|-\delta}^{i-1}(1-p_{\|u\|+\delta})^{n - i } + o(\delta^d)\biggr ] \\ & \hspace{1cm}\rightarrow n\bar{f}(x+u)\binom{n-1}{i-1}p_{\|u\|}^{i-1}(1-p_{\|u\|})^{n - i}\end{aligned}\ ] ] as @xmath378 . the result therefore follows by ( * ? ? ? * theorem 3.22 ) . _ to show ( [ eq : r1 ] ) , which bounds @xmath381 _ : we have @xmath382 \\ & \hspace{0.5cm}+ \sum_{i=1}^{k_2 } w_{ni}\biggl[\mathbb{e}\{\eta(x_{(i)})\ } - \eta(x ) - \mathbb{e}\{(x_{(i)}-x)^t\dot{\eta}(x)\ } - \frac{1}{2 } \mathbb{e}\{(x_{(i)}-x)^t\ddot{\eta}(x)(x_{(i)}-x)\}\biggr ] \\ & \equiv r_{11 } + r_{12},\end{aligned}\ ] ] say . now @xmath383 , so @xmath384 to handle @xmath385 , observe that by a taylor expansion , given @xmath103 , we can find @xmath90 such that for all sufficiently large @xmath7 , all @xmath288 and all @xmath386 , we have @xmath387 for @xmath298 , let @xmath388 . further , let @xmath389 and let @xmath390 , where @xmath391 denotes the largest eigenvalue of a matrix . then for large @xmath7 and @xmath288 , latexmath:[\[\label{eq : r13 } we can apply a very similar argument to that employed in * step 1 * to deduce that uniformly for @xmath298 , @xmath393 now @xmath394 for @xmath90 sufficiently small , there exists @xmath395 such that @xmath396 . so by hoeffding s inequality , for any @xmath397 , @xmath398 \leq ( 1+t_0)e^{-2c_2 ^ 2n\delta^{2d } } = o(n^{-m}),\ ] ] for every @xmath295 . moreover , using the moment bound in ( * a.3 * ) , @xmath399 as @xmath400 , uniformly for @xmath288 . therefore we can apply bennett s inequality @xcite to show that there exist @xmath401 such that for sufficiently large @xmath7 and @xmath402 and all @xmath403 , @xmath404 \leq ( 1+c_4t^{\rho/2})^{-c_3 n/2}.\ ] ] we deduce from ( [ eq : esq ] ) , ( [ eq : tleqt0 ] ) and ( [ eq : bennett ] ) that @xmath405 for all @xmath295 . this result , combined with ( [ eq : r13dom ] ) and markov s inequality applied to the two central terms in ( [ eq : r13 ] ) , proves ( [ eq : r1 ] ) as required . _ to show ( [ eq : r2 ] ) , which bounds @xmath406 _ : observe that by * step 1 * and * step 2 * , there exist constants @xmath407 such that @xmath408 uniformly for @xmath118 . hence , @xmath409 uniformly for @xmath118 , as required . _ to show ( [ eq : r3 ] ) , which bounds @xmath410_. let @xmath411 using the results of * step 1 * and * step 2 * , given @xmath412 sufficiently small , for large @xmath7 we have that for all @xmath118 , all @xmath326 and all @xmath413 $ ] that @xmath414 it follows that for large @xmath7 , @xmath415 we deduce that for large @xmath7 , @xmath416 this allows us to conclude ( [ eq : r3 ] ) . of the fact that conditions ( * a.1*)(*a.4 * ) imply the margin condition ( [ eq : margin ] ) for the upper bound , recall from ( [ eq : infeta ] ) that by the mean value theorem , for sufficiently small @xmath103 , @xmath420 where we may take @xmath421 , which is positive . by shrinking @xmath79 if necessary , we may assume that @xmath422 , and it follows that for small @xmath103 , @xmath423 where @xmath424 , using weyl s tube formula @xcite . for the lower bound , we construct a tube similar to @xmath425 , but contained in @xmath49 . to do this , let @xmath426 and let @xmath427 further , let @xmath428 , which is finite . again by the mean value theorem , for sufficiently small @xmath103 , @xmath429 thus , letting @xmath430 , for sufficiently small @xmath103 , @xmath431 where @xmath432 , again using weyl s tube formula . @xmath419 of theorem [ thm : optweights ] consider any vector of non - negative weights @xmath433 that minimises the function @xmath434 defined in the statement of theorem [ thm : main ] . since @xmath435 is symmetric in @xmath436 , while @xmath437 is increasing in @xmath9 , we see that @xmath438 is decreasing in @xmath9 . we let @xmath439 . now form the lagrangian @xmath440 then for some @xmath441 , @xmath442 for @xmath443 . by summing ( [ eq : lagderiv ] ) from @xmath443 , and then multiplying ( [ eq : lagderiv ] ) by @xmath437 and again summing from @xmath443 , we obtain two linear equations in @xmath444 and @xmath441 , which can be solved and substituted back into ( [ eq : lagderiv ] ) to yield @xmath445 for @xmath443 . in particular , @xmath446 is the unique minimiser of @xmath434 . the weight vector @xmath150 is asymptotically equivalent to @xmath446 in the following sense : elementary calculations reveal that @xmath447 , and moreover that @xmath448 it follows immediately that @xmath449 and therefore that ( [ eq : regratio ] ) holds . arguing similarly to the above , we see that the conditions @xmath154 and @xmath155 , or equivalently ( [ eq : equivcond ] ) , are sufficient for ( [ eq : regratio ] ) to hold . to see the necessity of these conditions , suppose for now that for some small @xmath104 , the weight vector @xmath118 satisfies @xmath450 then , by almost the same lagrangian calculation as that above , we have @xmath451 where @xmath452 is given by @xmath453 and where @xmath454 . it follows that for small @xmath104 , and for any @xmath118 satisfying @xmath455 , we have @xmath456 a very similar argument shows that if @xmath457 then the conclusion of ( [ eq : liminf ] ) also holds . but if ( [ eq : regratio ] ) holds and it is not the case that both @xmath154 and @xmath155 , then either ( [ eq : lowlimit ] ) or ( [ eq : lowlimit2 ] ) would have to hold on a subsequence . but then we see from ( [ eq : liminf ] ) that ( [ eq : regratio ] ) can not hold , and this contradiction means that the conditions ( [ eq : equivcond ] ) are necessary for ( [ eq : regratio ] ) . of theorem [ thm : main2 ] let @xmath459 . we only need to show that @xmath460 uniformly for @xmath249 , because the rest of the proof is virtually identical to that of theorem [ thm : main ] . the appropriate analogue of ( [ eq : delta ] ) is @xmath461\mathbb{p}\{\mathrm{bin}(n-1,p_{\|u\| } ) = i-1\ } \ , du \nonumber \\ & = \{1+o(1)\}\sum_{i = k_1}^{k_2 } \frac{n \delta w_{ni}}{(2r-2)!}\sum_{j=1}^d \sum_{s:|s|=r-1 } \int_{\|u\| \leq \delta_n } \bigl\{\eta_j(x)u_j^2 u^s\bar{f}_{s , j}(x ) + \tfrac{1}{2}\eta_{jj}(x)u_j^2 u^s \bar{f}_s(x)\bigr\ } \nonumber \\ & \hspace{9cm}\times \mathbb{p}\{\mathrm{bin}(n-1,p_{\|u\| } ) < i\ } \ , du,\end{aligned}\ ] ] uniformly for @xmath288 and @xmath249 . combining ( [ eq : firstderiv ] ) , ( [ eq : secondderiv ] ) , ( [ eq : delta2 ] ) , ( [ eq : lower ] ) and ( [ eq : upper ] ) , we have @xmath462 uniformly for @xmath288 and @xmath249 . combining ( [ eq : taylor ] ) , the analogue of ( [ eq : r1 ] ) and ( [ eq : biascompletion2 ] ) , this proves ( [ eq : bias2 ] ) . @xmath419 * acknowledgements * : i am very grateful to peter hall for introducing me to this topic , and to the anonymous referees and associate editor for their constructive comments , which significantly helped to improve the paper . i thank the statistical and applied mathematical sciences institute ( samsi ) in north carolina for kind hospitality when i attended the ` program on analysis of object data ' from september to november 2010 , during which time part of this research was carried out . finally , i gratefully acknowledge the support of a leverhulme research fellowship . biau , g. and devroye , l. ( 2010 ) on the layered nearest neighbour estimate , the bagged nearest neighbour estimate and the random forest method in regression and classification . _ j. mult . _ , * 101 * , 24992518 . fix , e. and hodges , j. l. ( 1951 ) discriminatory analysis nonparametric discrimination : consistency properties . tech . rep . 4 , project no . 21 - 29 - 004 , usaf school of aviation medicine , randolph field , texas .

a recently introduced numerical scheme for calculating self - diffusion coefficients of solid objects embedded in lipid bilayer membranes is extended to enable calculation of hydrodynamic interactions between multiple objects . the method is used to validate recent analytical predictions by oppenheimer and diamant [ biophys . j. , 96 , 3041 , 2009 ] related to the coupled diffusion of membrane embedded proteins and is shown to converge to known near - field lubrication results as objects closely approach one another , however the present methodology also applies outside of the limiting regimes where analytical results are available . multiple different examples involving pairs of disk - like objects with various constraints imposed on their relative motions demonstrate the importance of hydrodynamic interactions in the dynamics of proteins and lipid domains on membrane surfaces . it is demonstrated that the relative change in self - diffusion of a membrane embedded object upon perturbation by a similar proximal solid object displays a maximum for object sizes comparable to the saffman - delbrck length of the membrane .

1410.0721

lateral motion of membrane components is required for proper physiological functioning in cellular biology @xcite and serves as fertile grounds for biophysical studies involving lipid bilayer model systems @xcite . a key aspect of dynamics taking place at the membrane surface is hydrodynamic flow , both within the membrane itself and in the fluids surrounding the membrane . the membrane environment is thus distinct from traditional three dimensional ( 3d ) or two dimensional ( 2d ) hydrodynamic systems , incorporating aspects of both 2d and 3d flow ; the membrane components move in a two dimensional space defined by the membrane surface , but are coupled to fluid motions in a full 3d space . for this reason , it has been suggested that membrane hydrodynamics might best be described as quasi-2d " @xcite . saffman and delbrck @xcite ( sd ) introduced the standard hydrodynamic model for lipid bilayer membranes . in the sd model , the membrane is treated as a flat thin incompressible fluid sheet with surface viscosity @xmath0 , surrounded by an infinite incompressible aqueous bulk of viscosity @xmath1 . this model can be solved in the creeping flow limit to yield analytical predictions for the diffusion coefficient of cylindrical bodies embedded within the membrane via no - slip boundary conditions @xcite , which are in good agreement with experiments recording the diffusion of membrane proteins and lipid domains @xcite . ( however , some experimental studies claim deficiencies in the sd model for membrane protein diffusion @xcite . ) the sd model also correctly predicts the dynamics of lipid domain boundary fluctuations @xcite and the dynamics of phase separation in ternary model membrane systems @xcite . perhaps the most concise way to express the physics contained within the sd model is through the green s function formulation of the hydrodynamic problem . the membrane s velocity response to in - plane forcing is given by @xmath2 here , @xmath3 is the @xmath4 velocity of the membrane at position @xmath5 . ( we assume the membrane plane is coincident with the @xmath4 plane of our coordinate system . ) @xmath6 is an in - plane force / area applied to the membrane and the green s function tensor @xmath7 captures the hydrodynamic response to forcing in the sd model . explicitly @xcite , @xmath8 \right .\delta_{ij } \nonumber \\ & & \left . - \left [ h_0 ( \tilde{r } ) - \frac{2 h_1 ( \tilde{r})}{\tilde{r } } + y_2(\tilde{r } ) + \frac{4}{\pi \tilde{r}^2 } \right ] \frac{\tilde{r}_i \tilde{r}_j}{\tilde{r}^2 } \right \}\end{aligned}\ ] ] where @xmath9 and @xmath10 are bessel functions of the second kind and struve functions , respectively . the dimensionless distance @xmath11 , where @xmath12 is the saffman - delbruck length . @xmath13 defines a crossover between 2d and 3d - like hydrodynamics in the quasi-2d sd model ; for separations well below @xmath13 the green s function is that of a 2d fluid with surface viscosity @xmath0 , whereas at separations larger than @xmath13 the green s function is similar to that of a 3d system of viscosity @xmath1 @xcite . for typical lipid bilayer systems @xmath13 is on the order of a micron @xcite . in principle , eq . [ eq : v - field ] is restricted to fluid regions and would not immediately seem to be of use in solving problems involving particles suspended within the fluid ( e.g. prediction of diffusion coefficients ) . however , within creeping flow , there is no physical or mathematical distinction between a solid particle embedded within the fluid via no - slip boundary conditions and a fluid " region that occupies the same physical space as the embedded particle with the supplemental restriction to undergo only rigid body motions . computational strategies that supplement eq . [ eq : v - field ] ( typically the 3d version of it ) with the constraint of rigid - body motion over particle associated regions within the fluid have the capability to predict particulate dynamics . well known schemes that take advantage of this strategy to calculate diffusion coefficients and related properties associated with particulate flows include the kirkwood approximation @xcite , shell method @xcite and related techniques @xcite . ( see ref . and the references within for a review of these approaches . ) though the majority of work in this direction has been aimed at studying various solid bodies immersed in homogeneous 3d hydrodynamic environments , the quasi-2d membrane case has previously been considered with a method similar to the kirkwood approximation by levine liverpool and mackintosh @xcite . ( it should also be mentioned that the green s function approach naturally leads to diffusion coefficients for circular fluid domains in the membrane geometry @xcite . ) though the kirkwood approximation is simple to implement and often gives rise to qualitatively correct predictions , the method is known to be inexact and , in extreme cases , can yield unphysical results @xcite . in 3d geometries , the shell method @xcite resolves these shortcomings , however this method is not easily applied to the sd problem as it requires use of the rotne - prager tensor @xcite ; a quasi-2d analog to the rotne - prager tensor has not been derived . an appealing alternative to the shell method is the method of regularized stokeslets ( rs ) @xcite . ( section [ sec : rs - method ] provides an introduction of the rs method . ) this technique is known to reproduce the correct velocity fields and particle diffusion coefficients for model problems in 3d @xcite and is immediately extended to the sd hydrodynamic geometry without difficulty . in recent work @xcite , two of us presented this extension and verified that rs calculations quantitatively reproduce the known analytical results for single - particle mobility problems ( i.e. prediction of diffusion coefficients ) in the quasi-2d environment . while single - particle mobility / diffusion problems present a convenient test of computational methodologies , the methods discussed above ( rs included ) are applicable to a broader class of problems involving the motions of multiple embedded bodies and the hydrodynamic interactions between them . the study of many body dynamics in fluid environments has received considerable theoretical / numerical attention @xcite , but primarily in the context of 3d systems . the few prior studies in the membrane geometry are restricted to limiting regimes and circular particles @xcite , due to the focus of these works on analytical calculations . the complicated form of eq . [ eq : mem_green ] hinders analytical progress without invoking some manner of approximation . a versatile numerical tool to quantitatively study the dynamics of multiple membrane - embedded bodies has yet to be developed , despite the growing experimental interest in the dynamics of solid body suspensions in membrane systems and related quasi-2d environments @xcite . in this work , we extend the membrane rs method to consider many - body systems . a general framework is presented , which is applied in detail to the the study of two circular disks within the membrane . we also briefly revisit our earlier calculations @xcite involving the diffusion of diamond shaped lipid domains and tethered protein assemblies , as motivated by recent experiments @xcite . this work is structured as follows : in sec . [ sec : dif - tensor ] we discuss and present the diffusion matrix for a many - body system . in sec . [ sec : rs - method ] we generalize the quasi-2d rs method to calculations involving multiple bodies embedded in the membrane . this allows for the calculation of the full many - body diffusion matrix . though both sections [ sec : dif - tensor ] and [ sec : rs - method ] focus primarily on the two - body case for concreteness , the many - body generalization is straightforward and is discussed as well . in sec . [ sec : res ] we present detailed results for the case of two identical disks embedded within the membrane . in sections [ sec : dimers ] and [ sec : diamonds ] we consider the case of tethered two - disk dimers and correlations between diamond shaped domains , respectively . finally , sec . [ sec : summary ] discusses our results and concludes . the velocities and angular velocities of rigid bodies embedded within a fluid are related to the forces and torques applied to these bodies through the diffusion matrix @xmath14 @xcite : @xmath15 here , @xmath16 , @xmath17 , @xmath18 and @xmath19 contain the vector components of force , torque , velocity and angular velocity for all the bodies present in the fluid . @xmath20 is the boltzmann constant ant @xmath21 is temperature . this formulation , which describes an instantaneous transmission of force through an otherwise quiescent fluid , assumes an overdamped creeping flow " hydrodynamic regime neglecting inertial effects . ( the mobility matrix @xmath22 notation is preferred by some authors ; the relation between mobility and diffusion is assured by the einstein relation @xcite . ) the blocks @xmath23 and @xmath24 in eq . are associated with translational and rotational motion , respectively . the blocks @xmath25 and @xmath26 couple translation and rotation . subblocks of @xmath23 , @xmath24 , @xmath25 and @xmath26 associated with specific pairs of particles ( e.g. @xmath27 or @xmath28 ) are tensors . note that @xmath29 where @xmath30 " denotes the transpose @xcite . the inverse of @xmath14 defines the resistance matrix . @xmath31 the natural physical interpretation of the resistance matrix relates the _ hydrodynamic forces / torques _ ( i.e. frictions or drags ) imparted to the bodies as they move through the fluid with prescribed motions @xmath32 however , the hydrodynamic drags exactly compensate the applied forces(torques ) in creeping flow and it is mathematically convenient to express this relationship in terms of the applied forces and torques . this is the formulation obtained by inverting eq . . @xmath33 with @xmath34 @xcite . throughout this work we will be concerned with pairs of objects residing in flat membranes spanning the @xmath4 plane . consequently , the two vectors appearing in eqs . and may be ordered as @xmath35 and @xmath36 with the @xmath37 indices specifying particle identity . we will refer to these two vectors as the f - vector and v - vector , respectively . explicitly writing the full @xmath38 resistance matrix yields ( the diffusion matrix follows similarly ) @xmath39 where the subscript @xmath40 denotes elements associated with the rotational motion . the lines in eq . divides the friction tensor into @xmath41 , @xmath42 , @xmath43 and @xmath44 blocks , see eq . . schematic of the two disk system . ( a ) side view : the two disks reside in a saffman - delbrck membrane , which treats the bilayer as a thin structureless sheet with surface viscosity @xmath45 surrounded by a bulk fluid with viscosity @xmath46 . ( b ) top view : the disks have radius @xmath47 and center - to - center distance @xmath48 ; equivalently , the closest approach separation is @xmath49 . ] most of the calculations in this work involve two identical disks of radius @xmath47 , separated by a center - to - center distance @xmath48 ( see fig . [ fig : topside ] ) . this situation introduces significant symmetry relative to the general case and the diffusion and resistance matrices simplify considerably @xcite . without loss of generality , we assume the disk centers lie on the @xmath50 axis , which yields @xmath51 and @xmath52 here , we have introduced the subscripts @xmath53 and @xmath21 to denote motions longitudinal or transverse to the disk separation axis and the superscripts @xmath54 and @xmath55 to denote self and coupling components to the matrices . in our numerical calculations for the friction and diffusion matrices ( see below ) , we do not impose any of the above symmetries and calculate all 36 elements independently . however , our results display the above symmetries to very high numerical precision ( e.g. the zero " elements in eqs . [ eq : circ - r - mat ] and [ eq : circ - dif - mat ] are calculated to be @xmath56 to @xmath57 times smaller than the nonzero " elements in these matrices . ) , which provides a welcome check of our numerical algorithms . the elements of the diffusion matrix predict the thermal fluctuations in position and orientation of the disks during a short time interval @xmath58 in the absence of any applied forcing . generically , we express this as @xmath59 with @xmath60 representing any of the coordinates or angular coordinates in our system and it is understood that @xmath58 must be short enough to ensure that the observed @xmath61 fluctuations are too small to lead to significant changes in the elements of @xmath14 . more specifically : ( @xmath62 ) @xmath63 this provides a possible experimental route toward measuring the elements of @xmath14 that does not involve directly perturbing the system @xcite . in previous work @xcite , two of us introduced a regularized stokeslets @xcite ( rs ) numerical scheme for calculating the diffusion and resistance matrices for a single solid body embedded within a membrane . the present paper extends this method to calculations involving multiple solid bodies . this section briefly reviews the membrane rs method and elaborates on those aspects of the calculation necessary to consider multiple bodies . for details on the numerical implementation , the reader is referred to our original treatment @xcite . the starting point of the rs method is a discretized version of eq . [ eq : v - field ] @xmath64 = \sum_{n=1}^n \mathbf{t}(\mathbf{r}_{m}-\mathbf{r}_{n};\epsilon ) \,\mathbf{g}[\mathbf{r}_{n}]\,.\ ] ] this equation results from inserting @xmath65 \phi_{\epsilon}(\mathbf{r } ' - \mathbf{r}_{n})\ ] ] into eq . [ eq : v - field ] and defining the regularized stokeslet " as @xmath66 the blob function " @xmath67 represents a localized envelope over which forces @xmath68 $ ] , at discrete spatial positions , @xmath69 , are transmitted to the fluid . the blob function must be centered at zero , integrate to 1 and , as @xmath70 , must limit to the dirac delta function @xmath71 . we employ a gaussian blob function @xcite , @xmath72 in all of our calculations , but other choices are certainly possible . the numerical evaluation of @xmath73 for our quasi-2d saffman - delbrck systems is described in detail in ref . . it should be clear that eq . [ eq : rs ] is a simple matrix equation . if we restrict ourselves to only calculating velocities over the same set of positions where we impose forces , i.e. @xmath74 , the relevant matrix is square . we choose the blob positions , @xmath75 , to tile all particles embedded in the fluid ( see fig . [ fig : circle - blob ] ) and solve eq . [ eq : rs ] to determine the forces , @xmath76\}$ ] , necessary to effect a specified set of velocities , @xmath77 \}$ ] . _ the velocities are always chosen to describe rigid body motions within each particle _ so that @xmath78=\mathbf{v}_i+\mathbf{\omega}_i\times ( \mathbf{r}_{m_i } - \mathbf{r}_{c_i})$ ] for each particle @xmath79 . here , the subscript @xmath80 is used to index the subset of blobs associated with particle @xmath79 and @xmath81 is the hydrodynamic center of particle @xmath79 . we note that blobs are chosen to cover the entire disk , rather than just the perimeter . this is because , as discussed in ref.@xcite mandating rigid body motion of the perimeter does not ensure rigid body motion in the disk interior . in practice , the generalized minimal residual method algorithm ( gmres ) @xcite is used to solve eq . [ eq : rs ] for the blob forces , @xmath76\}$ ] . the blob forces combine to yield the total forces and torques on the individual particles , @xmath82\ , \nonumber \\ \mathbf{\tau}_i & = & \sum_{m_i } ( \mathbf{r}_{m_i}- \mathbf{r}_{c_i})\times \mathbf{g}[\mathbf{r}_{m_i}]\,.\end{aligned}\ ] ] therefore , the rs method provides a route to determining the f - vector corresponding to a given v - vector , using the language of the preceding section . the numerical calculations involve a large number of blobs ( @xmath83 ) to discretize the particles , but the f - vector and v - vector reflect forces / torques and velocities / angular velocities on the particles and are modestly sized ( six elements each for wo particle calculations ) . as a practical matter , it is most convenient to calculate the resistance matrix by choosing the v - vector input to the rs calculation to be composed entirely of zeros , except for a single element . the f - vector corresponding to this input then corresponds to a complete column of the resistance matrix . this process is repeated for each element of the v - vector to generate the full resistance matrix , column by column . the diffusion matrix is then obtained by inversion of the resistance matrix . for concreteness and to emphasize numerical details we provide a step - by - step procedure for calculating the diffusion matrix corresponding to the case of two identical disks of radius @xmath47 ( see fig . [ fig : topside ] ) embedded within the membrane . cartoon of the blob distribution on a pair of disks . in the actual calculations , thousands of blobs are distributed on the disks . ] 1 . discretize each disk with @xmath84 blobs , see fig . [ fig : circle - blob ] . similar to ref . , the inter - blob spacing , @xmath54 , is chosen to be between @xmath85 . @xmath86 is set to be half of the spacing . all the elements of the v - vector are zeroed except for @xmath87 , _ i.e. _ the disk 1 moves along the x - axis without rotation and the disk 2 is stationary . this v - vector is used as input for the rs routine to obtain the forces and torques on the disks ( f - vector ) . the obtained f - vector is divided by @xmath88 to obtain the first column of the resistance matrix . repeat steps 2 and 3 , keeping a single non - zero element of the v - vector each time to obtain the remaining columns of the resistance matrix . for example , in the fifth run , we set all the elements of the v - vector to zero except @xmath89 , _ i.e. _ disk 1 only rotates and the disk 2 is stationary . the resulting f - vector determines the fifth column of the friction tensor . calculate the diffusion tensor by numerically inverting the friction tensor , see eq . . repeat steps 1 - 5 for different values of the spacing @xmath54 . we find that the elements of the diffusion matrix change linearly in @xmath54 and this allows us to extrapolate to the infinite resolution limit @xmath90 . the reported results are obtained by this extrapolation . step six of the above procedure provides a natural criterion for establishing the credibility of our numerics . the mean squared error of the linear fit in step 6 is readily determined for a given physical situation ( i.e. specification of @xmath91 and @xmath92 ) by considering @xmath93 different resolutions : @xmath94 . if @xmath95 is the numerical value of an element of the diffusion matrix for the resolution @xmath79 and @xmath96 is the best - fit line to the points @xmath97 , the root mean square error reads @xmath98 ^ 2}\,.\ ] ] the rmse for all the nonzero " elements of the diffusion matrices which are reported in this work are two to three orders of magnitude smaller than the extrapolated elements themselves . however , we point out that the data reported in this paper are restricted to separations @xmath99 . at smaller separations the rmse starts to become large and the linear fit over the data taken at various resolutions is clearly poor ; slight adjustments to the resolution within the range indicated above lead to wild vacillations in the predictions . this is a clear indication that the rs scheme at computationally practical resolutions is breaking down . so , although the rs scheme is not able to directly probe the small separation limit , we have a clear numerical marker that this is the case . the numbers reported in this work are meaningful and we have avoided reporting on any regimes that are problematic . we do note that the problematic regime is readily captured via lubrication theory results and the rs numerics and lubrication results share a window of common validity ( see app . [ app : lubrication ] ) . the preceding sections have detailed the mechanics for calculating the resistance and diffusion matrices for multiple solid bodies embedded in a membrane . here , we apply these methods to the calculation of these matrices for the specific case of two identical disks embedded within the bilayer . as mentioned previously , this case has the advantage of maximal symmetry and allows us to explore some consequences of the hydrodynamic calculations while keeping the set of physical and geometric parameters to a minimum . as we will see , even this simplest of systems is rather complex to describe . the translational coupling elements of the diffusion matrix quantify correlations in the thermal brownian motions of the two disks ( see eq . [ eq : fluctuations ] ) . in the limit of point - like particles , @xmath100 and @xmath101 , follow immediately from the green s function of eq . [ eq : mem_green ] . specifically , if the disks are separated by distance @xmath48 along the @xmath50 axis as diagrammed in fig . [ fig : topside ] , the limiting forms are provided by @xmath102\,,\ ] ] and @xmath103 \ , . \end{gathered}\ ] ] direct use of these two expressions within the diffusion matrix , would be the membrane analog to the kirkwood approximation for these elements . recently @xcite , oppenheimer and diamant have derived a finite size correction to eqs . [ eq : mu - c - l ] and [ eq : mu - c - t ] in the regime where particle separation is well below the saffman delbruck length ( i.e. @xmath104 ) . for disks of radius @xmath105 they find @xmath106\,,\ ] ] to leading order in @xmath107 , where the upper ( lower ) signs correspond to the longitudinal ( transverse ) diffusion and @xmath108 is euler s constant . ( we mention that the corresponding formula in ref.@xcite contains a sign error . a brief derivation of eq . [ eq : mu - size - corr ] can be found in app . [ app : corr ] to justify the result provided here . ) in the limit that @xmath109 , the last term of eq . 18 vanishes and the remaining result is the small @xmath110 limit of eqs . [ eq : mu - c - l ] and [ eq : mu - c - t ] . it is important to emphasize the limitation of this formula to the @xmath104 regime . we are unaware of any comparable analytical formula(e ) that extend the finite size results beyond this regime . comparison of the rs data for the coupled - diffusions @xmath100 and @xmath101 with the related analytical expressions , for the saffman - delbrck length * ( a ) * @xmath111 and * ( b ) * @xmath112 : the transverse coupled - diffusion @xmath101 scaled by @xmath113 versus the ratio of the surface - to - surface distance @xmath114 and the radius of the disks @xmath47 . the circles are the rs data for the transverse coupled - diffusion . the blue and dashed lines represent eqs . and , respectively . as expected , the dashed line , which includes corrections to account for finite disk size , improves upon the point particle approximation ( blue line ) . inset : the longitudinal coupled - diffusion @xmath100 scaled by @xmath113 versus the ratio of the surface - to - surface distance @xmath114 and the disks radius @xmath47 . the squares are the rs data for the longitudinal coupled - diffusion . the red lines represent eq . . ] comparison of the rs data for the coupled - diffusions @xmath100 and @xmath101 with the related analytical expressions , for the saffman - delbrck length * ( a ) * @xmath111 and * ( b ) * @xmath112 : the transverse coupled - diffusion @xmath101 scaled by @xmath113 versus the ratio of the surface - to - surface distance @xmath114 and the radius of the disks @xmath47 . the circles are the rs data for the transverse coupled - diffusion . the blue and dashed lines represent eqs . and , respectively . as expected , the dashed line , which includes corrections to account for finite disk size , improves upon the point particle approximation ( blue line ) . inset : the longitudinal coupled - diffusion @xmath100 scaled by @xmath113 versus the ratio of the surface - to - surface distance @xmath114 and the disks radius @xmath47 . the squares are the rs data for the longitudinal coupled - diffusion . the red lines represent eq . . ] results from rs numerical calculations for @xmath100 and @xmath101 are displayed in fig . [ fig : haimshort ] . we display two different cases corresponding to the behavior of small bodies ( lipids or proteins , @xmath115 ) and large bodies ( ten micron scale lipid domains , @xmath116 ) in the bilayer . in both cases it is observed that the point - like - particles approximation performs well until the disks come within several radii of one another . for separations of @xmath117 and larger , eqs . and are indistinguishable from the full numerical calculations . there are clear deviations from the point - particle results for small inter - particle separations . in the @xmath118 regime these differences are nicely explained by the theory of oppenheimer and diamant ( eq . [ eq : mu - size - corr ] ) , though the theory is not quantitatively accurate at very small separations ( as expected ) . in the regime of particles large compared to @xmath13 we lack an analytical prediction for finite size corrections to eqs . and . though finite - particle - size effects make negligible contributions to @xmath100 for the @xmath116 , case , there are large effects seen in @xmath101 . in fact , for two identical circular domains with @xmath119 microns on a membrane with @xmath120 micron , at @xmath121 micron separation the prediction of eq . for the transverse mobility is off by more than 800% from the numerical results . although it is perhaps unsurprising that finite particle size should play a major role at such separations , it is interesting to see just how big the effect is , especially in contrast to the related ( lack of ) effect on the longitudinal motion . the self - diffusion matrix elements , @xmath122 , @xmath123 , are affected by the presence of other proximal solid particles @xcite , though this influence requires a finite particle size and is a shorter ranged hydrodynamic effect as compared to that observed in the coupling diffusion elements . indeed , many common theoretical / simulation tools including the kirkwood approximation @xcite , brownian dynamics with hydrodynamic interactions @xcite and related methodologies neglect all influence of neighboring particles on self - diffusivity . in this subsection , we present rs predictions for how disk self diffusion is impacted by the presence of a second nearby disk . to this end , we compare the self - diffusion of a disk in the two disk system with the self - diffusion of an isolated single disk control . we consider two different scenarios : ( 1 ) both tracked disk and perturbing disk are freely moving , and ( 2 ) the tracked disk is free , but the perturber is fixed in place . we report the relative change in the self - diffusions compared to the single disk case as @xmath124 where @xmath125 is the ( control ) self - diffusion of an isolated single disk , and @xmath126 is the self - diffusion of a disk in the two disk system . l , t and r , for the longitudinal , transverse and rotational mobilities . the relative change in longitudinal self - diffusion as a function of @xmath127 is depicted in fig . [ fig : self - both - moving]a . the plots display results for three different disk geometries corresponding to @xmath128 swept over a wide range of saffman - delbrck lengths . for a given choice of @xmath129 , any particular choice of @xmath47 ( or @xmath114 ) yields an identical two - disk geometry upon proper scaling and it is most useful to regard changes along horizontal axis ( @xmath127 ) in this figure as resulting from altering the membrane @xmath13 while maintaining a given two - disk geometry . the first point to note about fig . [ fig : self - both - moving]a is that the cases plotted cover a significantly abbreviated range of values for @xmath129 relative to the cases considered in the context of coupled diffusion . the passive effect of the second disk on the motion of the first is a considerably shorter ranged phenomenon than the direct transmission of force from disk one to disk two at play in the coupling diffusion coefficients . at edge - to - edge separations equal to one disk radius , the strongest observed influence on self diffusion is only on the order of 5% and shrinks still further for larger separations . only for very closely placed disks does the effect become appreciable . we also note that the extreme values of @xmath13 occurring at the edges of the plot place the system in either a limiting 3d hydrodynamic regime independent of @xmath13 ( at small @xmath13 ) or the regularized 2d limit of the saffman - delbrck model ( at large @xmath13 ) with weak ( logarithmic @xcite , see eq . [ eq : mu - size - corr ] ) dependence on @xmath13 . the relative change in the transverse self - diffusion @xmath130 remains very small at all separations , although it displays the same trends as in fig . [ fig : self - both - moving]a . the maximum relative change in the transverse self - diffusion for the separation ratio @xmath131 is less than 5% and we have not explicitly included @xmath130 plots . the most striking feature of fig . [ fig : self - both - moving ] is the non - monotonic behavior for the longitudinal translation case , with a maximum for disks of radius comparable to the saffman - delbrck length , _ i.e. _ @xmath132 . although this feature draws the eye , we do not believe it to be as interesting as might first be believed . we recall that the quantities plotted in fig . [ fig : self - both - moving ] are _ relative _ differences in self - diffusion , with the difference itself plotted in ratio to the corresponding self - diffusion of an isolated single - disk control . both the numerator and denominator of the ratio display strong , but clearly monotonic dependence on @xmath133 in the vicinity of @xmath133 ( see fig . [ fig : why - pick ] ) . the observed maximum simply results from the product of two functions , one sharply increasing and the other sharply decreasing . the appearance of a maximum can be qualitatively reproduced using less elaborate numerical models , as is explicitly demonstrated in app . [ app : dim - mon ] using a kirkwood approximation for a simplified system . two disks free to move and rotate : the relative change in the ( a ) longitudinal and ( b ) rotational self - diffusion versus the ratio of the radius of the disks @xmath47 and the saffman - delbrck length @xmath13 . in both figures ( a ) and ( b ) the squares , triangles and circles represent the separations @xmath134 and 1 , respectively . the percentage changes to @xmath130 are all less than 5% for the cases considered here and are not explicitly plotted . ] two disks free to move and rotate : the relative change in the ( a ) longitudinal and ( b ) rotational self - diffusion versus the ratio of the radius of the disks @xmath47 and the saffman - delbrck length @xmath13 . in both figures ( a ) and ( b ) the squares , triangles and circles represent the separations @xmath134 and 1 , respectively . the percentage changes to @xmath130 are all less than 5% for the cases considered here and are not explicitly plotted . ] two disks free to move and rotate : diamonds show the dimensionless difference between the longitudinal self - diffusions of a disk in the presence of another one @xmath135 and a single disk @xmath136 versus the ratio of the radius of the disks and the saffman - delbrck length @xmath127 . down triangles represent the dimensionless inverse of the longitudinal self - diffusion of the single disk @xmath137 versus @xmath127 . @xmath138 is the longitudinal self - diffusion coefficient of the single disk for @xmath139 . inset : short and intermediate distance close - up of @xmath136 on the linear scale . ] the relative changes in rotational self - diffusion are presented in fig . [ fig : self - both - moving]b . the trends of @xmath140 are completely different than the @xmath141 case , as might have been anticipated due to the shorter ranged hydrodynamic flows associated with rotational motion relative to translational motion . here , the larger effects are seen in the 2d hydrodynamic regime relative to the 3d regime and the crossover is completely monotonic . unlike translational diffusion , rotational diffusion does not suffer from a stokes paradox in 2d and requires no regularization by a surrounding subphase to obtain finite results @xcite ; the large @xmath13 limit for rotational diffusion saturates to a true 2d limit without any residual logarithmic corrections . in reconciling the somewhat backward trends between panes ( a ) and ( b ) of fig . [ fig : self - both - moving ] it is helpful to recognize that the effects described in this figure result from the perturbation to single disk motion caused by constraining a nearby region of the membrane to undergo only rigid - body motion . we expect that the largest changes to self diffusion will occur when the unperturbed flows around a moving isolated disk 1 deviate most strongly from rigid body flows at the location of disk 2 . in the limit of large @xmath13 in fig . [ fig : self - both - moving]a the velocity field proximal to a single translating disk is very nearly constant . the constraint of rigid body motion over the envelope of disk 2 is hardly a constraint at all and has minimal effects on the translational self diffusion . rotational motions , on the other hand , generate considerably shorter ranged flows which vary appreciably proximal to the rotating disk and a strong effect is seen in the large @xmath13 limit of fig . [ fig : self - both - moving]b . in the small @xmath13 ( 3d ) regime the translational flows are not as flat as in the opposite regime , which explains the relatively larger changes on the right of fig . [ fig : self - both - moving ] a. the rotational flows are also more rapidly varying in the 3d regime relative to the 2d regime , and actually die off so rapidly away from the rotating disk that they are quite small in the vicinity of the second disk , explaining the small effects on the right . of fig . [ fig : self - both - moving ] b. in biological contexts , it is common for proteins and larger assemblies to be immobilized on the membrane surface due to interactions with cytoskeletal filaments and related cellular structures @xcite . it is thus natural to consider the influence of immobilized perturbers on particle self diffusion in addition to the case considered in the previous section . the self - diffusion of a disk adjacent to a completely immobile ( i.e. no translation and no rotation ) neighboring disk is a distinct hydrodynamic problem from that considered in the previous section . indeed , constraining @xmath142 in a region proximal to the diffuser is clearly a significantly more severe restriction than only requiring rigid - body motion over the same region . within the rs framework , immobility of disk 2 is enforced by setting @xmath143 in eq . [ eq:2-obj - fric - tensor - general ] . further , we only require the @xmath144 , @xmath145 and @xmath146 components of the f - vector to determine the mobilities of disk 1 as the forces on the immobilized disk are irrelevant to us . eq.[eq:2-obj - fric - tensor - general ] is thus reduced to an effective single particle resistance problem @xmath147 the first of these equations corresponds to the case of two general objects with object 2 immobilized ; the second equation is specific to the symmetries present in the two disk case with the second disk immobilized . we stress that the individual elements of the above matrices are identical to those appearing in eqs . [ eq:2-obj - fric - tensor - general ] and [ eq : circ - r - mat ] and need to be calculated via the full 2 body rs scheme described in section [ sec : rs - method ] . no new rs calculations are needed here , beyond those made to calculate the elements of the general two - body resistance matrix . the effect of second body immobilization on the self - diffusion of body one is captured through the different matrix inversion required in this case relative to the case considered in the previous section and as indicated in eq . [ eq : diff - fric - tensor ] . the self - diffusivities of particle 1 in the present case are obtained by inverting the truncated @xmath148 resistance matrix of eq . [ eq:1-obj - fric - tensor - general ] and not the full @xmath149 resistance matrix . as expected , compared to a mobile adjacent disk , a stationary disk exerts a considerably stronger influence on translational self - diffusions , @xmath150 ( see fig . [ fig : self - one - stat]a ) . the effect is particularly strong in the 2d regime of large @xmath13 . although enforcing rigid body motion over the perturbing disk has little effect due to the nearly homogeneous velocity field around the diffuser at large @xmath13 , forcing complete immobility of the perturbing disk has a huge effect and can slow diffusion by nearly a factor of two . this is true for both longitudinal and transverse motions . in the 3d regime , flows die off more rapidly away from the diffuser and the perturbation by disk 2 is smaller than in the 2d regime . however , the effect of an immobile perturber is always stronger than a mobile one ; forcing immobility on the fluid flow in a region in space is a stronger perturbation than just requiring rigid body motion over the same region . the relative change in the rotational self - diffusion , @xmath140 , is presented in fig . [ fig : self - one - stat]b . comparing figs . [ fig : self - both - moving]b and [ fig : self - one - stat]b , the rotational self - diffusion is affected by the immobile disk between 5 % to 10% more strongly than the free disks case , but the trends are similar . an immobile and a free disk ( the free disk is free to move and rotate ) : the relative change in the ( a ) transverse and ( b ) rotational self - mobilities versus the ratio of the radius of the disks @xmath47 and the saffman - delbrck length @xmath13 . in both figures ( a ) and ( b ) the squares , triangles and circles represent the separations @xmath134 and 1 , respectively . inset : the relative change in the longitudinal self - diffusion . ] an immobile and a free disk ( the free disk is free to move and rotate ) : the relative change in the ( a ) transverse and ( b ) rotational self - mobilities versus the ratio of the radius of the disks @xmath47 and the saffman - delbrck length @xmath13 . in both figures ( a ) and ( b ) the squares , triangles and circles represent the separations @xmath134 and 1 , respectively . inset : the relative change in the longitudinal self - diffusion . ] inspired by the recent experiments of knight and coworkers @xcite , we consider the motion of two - disk dimers ( fig . [ fig : dimerss ] ) on the membrane surface . the experimental systems involve constructs of two lipid associated proteins bound together by molecular ( peptide ) linkers of various lengths . ( trimers were also studied in the experiments , but are not considered here . ) in a previous study @xcite , we treated these dimers as a completely rigid body and modeled the translational diffusion for direct comparison to experiment . we did not consider rotational diffusion in that work because rotation was not tracked experimentally and we had not developed the theoretical methodology to treat any motions beyond rigid body translations and rotations . given the flexible nature of the peptide linkers used in experiment , the assumption of a completely rigid dimer is questionable and might naively be expected to be particularly problematic for rotational motions . requiring the individual monomers of the dimer to rotate with the same angular velocity as the whole assembly is a significant constraint . the purpose of this section is to evaluate how relaxing this constraint affects overall dimer rotation . as we have seen in sec . [ sec : fixed_disk ] , the calculation of particle mobilities / diffusivities under the constraint that certain motions are disallowed is straightforward . once the elements of the resistance matrix have been calculated for the general case absent constraints , the hard work has been done . imposing the constraint(s ) simply means choosing input v - vectors consistent with the constraint(s ) and only tracking those component(s ) of the f - vector orthogonal to constrained motions . the reduced resistance matrix obtained in this manner may then be inverted to obtain the diffusion matrix for the unconstrained motions . this procedure was particularly simple for the example of a completely immobile monomer studied in sec . [ sec : fixed_disk ] , since our starting basis involved single disk motions and the reduced basis was obtained simply by zeroing out half of the v - vector corresponding to motions of disk 2 . similarly , the forces / torque associated with disk 2 were ignored , which was also completely natural within the single disk basis . the present case involving dimer motions is slightly more complicated because our resistance matrix elements have been calculated in the single disk basis , but we need to impose constraints that involve both disks . these constraints are handled much more naturally in a basis of center - of - mass and relative motions of the two disks . to this end , we consider a v - vector @xmath151 defined by the transformation matrix @xmath152 two disk dimers are derived from the general two disk systems of the previous sections under the constraint of constant separation between the disks . rigid dimers " are further constrained to treat the dimer assembly as a single rigid body . `` dimers with rotating monomers '' allow the monomers to freely spin with angular velocities that are not slaved to the overall rotation of the dimer about its center of mass . ] @xmath153 , @xmath154 and @xmath155 are the translational and angular velocities of the dimer relative to its center of mass . @xmath156 is the relative velocity of the monomers along the axis joining them and @xmath157 are the angular velocities of the individual disks measured relative to @xmath155 . the dimers we consider will always have a fixed separation and @xmath158 . additionally , rigid dimers " are defined by @xmath159 , whereas dimers with freely rotating monomers impose no such constraint ; see fig . [ fig : dimerss ] . if we desire to preserve a similar expression of the fluctuation - dissipation relationship in our new basis as seen in the original two - particle basis ( i.e. eq . [ eq : fluctuations ] ) , it is essential to define the elements of our transformed f - vector as @xmath160 with the inverse of the transpose of @xmath161 given explicitly by @xmath162 it then follows that the resistance and mobility matrices in our new basis are related to their counterparts in the two - particle basis as @xmath163 further , the matrices obey the expected inverse relation to one another , @xmath164 and the elements of @xmath165 are simply related to the transformed coordinate displacements in a short time interval in analogy to eq . [ eq : fluctuations ] : @xmath166 . we calculate the diffusion coefficients for the rigid dimer by considering only the diagonal upper left @xmath167 corner of the @xmath168 matrix defined by eq . [ eq : transformed_matrices ] and inverting this reduced resistance matrix to obtain a diagonal @xmath167 diffusion matrix . the resulting diffusion coefficients @xmath169 ( longitudinal translational motion along the dimer axis ) , @xmath170 ( transverse translational motion perpendicular to the dimer axis ) and @xmath171 ( rotational motion of the dimer ) have the explicit forms @xmath172 it was verified that results obtained in this manner are numerically identical to direct rs calculations treating the entire dimer as a single rigid body ( as in ref . the corresponding quantities for the dimer with rotating monomers are calculated by considering the upper left @xmath173 corner of @xmath168 and performing the @xmath173 inversion to obtain the diffusion matrix . this matrix is not diagonal , however the first three elements along the diagonal still correspond to mean square displacements and rotation of the dimer about the center of mass . the results are @xmath174 our interest is in the comparison between rotational diffusion for the two different dimer assemblies . to this end , we compare the rotational diffusion of a rigid dimer with fixed monomers @xmath175 and a dimer with rotating monomers @xmath176 via the relative difference in the rotational self - diffusions as @xmath177 relative difference between rotational self - diffusions for dimers with and without freely rotating monomers ( calculated by eq . ) versus the ratio of the distance of the closest approach between the monomers @xmath114 and the radius of the monomers @xmath47 . the upper and lower plots are for the saffman - delbrck lengths @xmath178 and 0.1 , respectively . the triangles correspond to the full calculation including all hydrodynamic interactions between the monomers . the dashed curves represent the results obtained by neglecting monomer - monomer hydrodynamic interactions ( see text ) . ] figure [ fig : dimer ] plots @xmath179 as a function of the dimer separation for two saffman - delbrck lengths : @xmath180 ( protein - sized monomers ) and 0.1 ( domain - sized monomers ) . the results for the two regimes are rather similar , with only modest differences between the two models . however , it is interesting to note that this similarity between the two regimes as well as the small magnitudes seen in the plots both result from particle - particle hydrodynamic interactions damping otherwise more significant effects . the dashed lines in fig . [ fig : dimer ] result from approximating the full calculation by turning off all hydrodynamic interactions between the monomers ( i.e. by zeroing all off - diagonal contributions to the resistance matrix and calculating all diagonal elements in the absence of the second disk in eq . [ eq : circ - r - mat ] prior to transforming to @xmath168 ) . in this limit , eq . [ eq : rel - dif - drot ] takes the simple form @xmath181 , where the @xmath182 " superscript indicates the self resistances are calculated for an isolated single particle . it is expected that @xmath179 is a positive quantity . constraining the monomers to rotate with the overall rotation of the dimer can only lead to additional dissipation beyond what the unconstrained system would experience under a prescribed angular velocity . as the dashed lines in fig . [ fig : dimer ] indicate , this effect is potentially large ( especially for protein sized monomers ) when calculated in the absence of direct monomer - monomer hydrodynamic coupling . it is expensive to force the monomers to rotate . however , this cost is substantially reduced when the full calculation is performed . with hydrodynamic coupling present , the monomers naturally spin in the same direction as the dimer rotations , so forcing the completely in - phase motion between dimer and monomers is less costly . the effect is especially prominent for dimers in the 2d hydrodynamic regime , where the hydrodynamic interactions are strong . though the preceding examples have been limited to circular disks to maximize symmetry and simplify analysis , a strength of the rs methodology is the ability to consider bodies of arbitrary shape . as a specific example , we consider the diamond - shaped lipid domains studied in ref . previously , we used the rs technique to study single diamond - shaped domains @xcite . this section considers the diffusion of diamond domains in proximity to other diamond shaped domains , which corresponds to the actual experimental conditions of ref . @xcite . in these experiments , micron - scale diamond - shaped ( with aspect ratio of 1.42 , see fig . [ fig : diamonds - blob ] ) solid lipid domains in a background fluid lipid phase ( with saffman - delbrck length @xmath183 ) were tracked in time to measure the translational and rotational diffusion coefficients of the domains . though precise area coverages of the diamonds are not reported for the experiments , the provided images clearly display diamonds in proximity to one another and it is natural to ask how the diffusive characteristics of a tracked domain are expected to be altered by a neighbors . the results of sec . [ sec : both_moving_self ] for disks suggest that that since the diamond size is close to @xmath13 in these experiments , the effect should be nearly as strong as is possible . to determine the impact of hydrodynamic interactions on the self - diffusion of the diamonds , rs calculations were performed for diamond pairs on a membrane with the saffman - delbrck length of 1 micron ( see the discussion in ref . @xcite regarding the estimation of @xmath13 in the diamonds experiment ) . cartoon of the blob distribution for the ( a ) side - to - side and ( b ) tip - to - tip diamonds . in the calculations , thousands of blobs tile each diamond . ] in accordance with typical experimental numbers @xcite , the short axis of the diamonds is chosen to be 2.00@xmath184 and the long axis 2.84@xmath184 , corresponding to the aspect ratio of 1.42 mentioned above . this coincides with an effective radius @xmath185 , for direct comparison to the data in ref . ( the effective radius is the radius of a hypothetical disk that shares the same area as the diamond . ) two identical diamonds free to move and rotate : the relative change in the longitudinal and rotational ( inset ) self - diffusions versus the separation between the diamonds , for @xmath186 . the aspect ratio of the diamonds is 1.42 and the shorter diagonal is 2@xmath184 ( equivalent to the effective radius of @xmath185 ) . @xmath48 is the separation between diamond centers . the black diamonds and blue squares respectively represent the data for the side - to - side and tip - to - tip geometries , see fig . [ fig : diamonds - blob ] . the circles show the data for the disks with the radius @xmath187 on the same membrane . note that the shortest separations are excluded for the tip - to - tip geometry as they correspond to overlapping configurations . ] figure [ fig : diamond ] plots the relative change ( as defined in eq . [ eq : rel - err ] ) in the longitudinal and rotational self - diffusions for side - to - side and tip - to - tip diamond geometries in addition to the corresponding quantities obtained by approximating the diamonds as effective circles with areas identical to the diamond areas . the relative change depends both on the separation and orientation of the diamonds . for separations in the range @xmath188 the relative change is between 5% to 25% and dies off rapidly with increasing separation ( especially for rotational motion ) . although closely spaced neighboring domains can quantitatively affect the measured diffusion , the effect is not large and is not strikingly different from what could be estimated by approximating the domains as circles . in particular , domain - domain hydrodynamic interactions probably can not explain the apparent experimental inconsistencies between rotational and translational domain motion discussed at some length in ref . the rs approach @xcite is a versatile and powerful tool for studying particulate flows in traditional 3d hydrodynamic geometries and , more recently , in the quasi-2d geometries associated with lipid bilayer membranes @xcite . this work has further validated the membrane rs methodology of ref . @xcite and demonstrated its utility in application to problems involving multiple membrane - embedded objects . the applications considered herein involve comparisons to analytical theory and experimental measurements , but one of the most promising future applications of this work will be its use in parameterizing and validating approximate simulation schemes for the dynamics of protein laden membranes and related inhomogeneous biomembrane systems . standard techniques for simulating particulate flows in 3d @xcite start with a pairwise approximation to the diffusion matrix for the full many body system . typically , the rotne - prager approximation @xcite is employed . in the case of membrane systems , there is no rotne - prager analog available . one possibility for developing membrane simulations analogous to brownian dynamics with hydrodynamic interactions " @xcite is to numerically tabulate the coupling elements of the diffusion matrix as a function of particle separation to build up a pairwise approximation to the diffusion matrix that can be implemented in simulations , as has been suggested for 3d systems @xcite . the rs calculations presented in this work are a natural method to use for such a tabulation and could , perhaps , be supplemented with the lubrication results of ref . @xcite to capture near - field effects , as in stokesian dynamics @xcite simulations in 3d . an alternate strategy for studying protein laden membranes based on an immersed boundary ( ib ) scheme @xcite has recently been introduced @xcite . this method does not require input of a diffusion matrix , but rather tracks fluid flow explicitly and derives hydrodynamic interactions from the fluid velocity field . however , the treatment of particles in this scheme is approximate ( as in any ib simulation ) and has only been verified to agree with the far - field predictions related to those of oppenheimer and diamant @xcite . comparison between rs results and the membrane ib simulations will be an important future test of this simulation methodology and may suggest future improvements to this ( or other ) biomembrane simulation scheme incorporating hydrodynamic interactions . work along these lines is currently being pursued . thanks haim diamant and gilad haran for helpful conversations . this research was supported in part by a grant from the bsf ( grant no . 2012084 to fb ) and the nsf ( grant no . che-1153096 to fb ) . this appendix compares rs numerical results with the limiting small - separation lubrication ( lub ) analytical expressions of bussell , koch and hammer @xcite . the lub results for each element of the friction tensor are presented in table [ tab : lub ] . rs calculations as described in the main text were used to compute the friction tensor for two identical proteins with radii @xmath189 nm and the surface - to - surface separation @xmath114 ranging from 0.1 nm to 10 nm embedded in a membrane with the saffman - delbrck length 1 micron . the ratio of each element of the rs friction tensor to the corresponding lub friction tensor element is plotted in fig . [ fig : lubr ] . .[tab : lub ] lubrication predictions for two disks of radius @xmath47 with closest separation distance @xmath114 @xcite . these are the limiting results in the regime of small @xmath129 , where both lengths are also much smaller than @xmath13 . the result @xmath190 " indicates saturation to a constant at small separations in contrast to the divergent results expected for the remaining elements . [ cols="^,^,^,^",options="header " , ] panes [ fig : lubr ] ( a ) and ( b ) show that rs numerics and lub predictions for the longitudinal resistance tensor elements fully converge to one another by @xmath191 nm . figures [ fig : lubr ] ( c)-(g ) show near convergence between rs and lub results by @xmath191 nm for other elements of the friction tensor . as discussed in the main text , smaller @xmath114 geometries begin to show numerical artifacts in the rs computations for computationally feasible blob resolutions , which is why calculations were halted at @xmath191 nm . the coupled rotational friction element ( pane [ fig : lubr ] ( h ) ) lacks a limiting lub result to compare against , and we have plotted the friction element itself in this case . the results are in good agreement with the numerics of ref . in this appendix , we outline the derivation of finite - size corrections to the coupled - diffusions @xmath100 and @xmath101 , as originally presented in ref . @xcite . the derivation proceeds as follows : 1 . two disks of radii @xmath192 and @xmath193 , separated by the distance @xmath48 considered . it is assumed that @xmath194 and , without loss of generality , that the position of disk 1 is the origin and disk 2 lies on the x axis . 2 . since both disk radii and separation are all far smaller than @xmath13 , the flow profile generated around a forced disk 1 in the vicinity of disk 2 , but ignoring the presence of disk 2 , can be approximated by the formula for the flow around a forced disk in 2d : @xmath195f_{1,j}\ , . \end{aligned}\ ] ] here , @xmath196 is the euler - mascheroni constant , @xmath197 are the cartesian components of force on disk 1 and the einstein summation convention is assumed . the membrane faxn law derived by oppenheimer and diamant @xcite relates the components of force applied to disk 2 , @xmath198 , the components of velocity of disk 2 , @xmath199 and the ambient membrane velocity field were particle 2 not present _ assuming the 2d hydrodynamic regime specified in point 1_. this field is provided by @xmath200 from eq . [ eq : near - field ] in the case of a flow driven by forcing on particle 1 . the result is @xmath201 under the assumption that particle 2 is unforced , this implies @xmath202 where @xmath200 and its laplacian are evaluated at the center of disk 2 , i.e. at @xmath203 . carrying out the indicated derivatives and retaining terms only up to order second order in @xmath204 and @xmath205 yields @xmath206 f_{1,x } \nonumber \\ u_{2,y } & = & \frac{1}{4 \pi \eta_{m } } \bigg [ -\left\{\ln ( r/2\ell_0 ) + \gamma_e + 1/2 \right\ } + \frac{r_1 ^ 2 + r_2 ^ 2}{2d^2 } \bigg ] f_{1,y}. \nonumber\end{aligned}\ ] ] 4 . given the geometry specified in point 1 and comparing to eq . [ eq : circ - r - mat ] we have @xmath207 and @xmath208 . this yields the result quoted in eq . [ eq : mu - size - corr ] . the non - monotonic behavior observed in fig . [ fig : self - both - moving ] can be qualitatively captured with a simple 3-bead kirkwood type model . the perturbation of a nearby solid disk on the self - diffusivity of a tracked disk is due to the rigid - body constraint on fluid flow over the spatial envelope of the perturber . for longitudinal motions , we can mimic this full rigid body constraint by a simpler rigid - body constraint between two beads . we consider the situation pictured in fig . [ fig : dim - mon ] : a diffusing bead is proximal to a rigid dimer of beads and all three beads are co - linear . all beads are identical with radius @xmath47 and the dimer separation is @xmath209 . the distance between the dimer beads and the monomer are @xmath210 and @xmath211 with @xmath212 . @xmath114 , the point of closest approach between dimer and monomer is given by @xmath213 . within the kirkwood approximation @xcite the diffusion tensor of the three - bead system for motions along the x - axis reads where @xmath215 is the diffusion coefficient for an isolated single disk of radius @xmath47 and @xmath216 is the longitudinal coupled - diffusion in the point - particle approximation given by eq . . @xmath215 is given by the hughes , pailthorpe , white solution @xcite , which we implement numerically via the empirical fit introduced by petrov and schwille @xcite . we assume that a force @xmath217 is applied to the monomer along the x - axis and the dimer is force - free . an undetermined lagrange multiplier @xmath218 is introduced to enforce the constant separation of beads 1 and 2 in the dimer . the resulting mobility problem is thus @xmath219 where @xmath88 , @xmath220 and @xmath221 are the velocities of the beads in response to @xmath217 . using the rigid body constraint on the dimer @xmath222 , we find @xmath223 } \right \ } f_{3x}\,.\ ] ] inserting eq . into eq . and using @xmath224 yields the longitudinal self - diffusion coefficient for bead 3 in the presence of the rigid dimer . the relative change associated with adding the dimer as defined in eq . [ eq : rel - err ] is then given by @xmath225 ^ 2}{2d^\mathrm{0}\big[d^\mathrm{0}-d_\mathrm{l}^\mathrm{c}(d_{12}/\ell_0)\big]}\,.\ ] ] the above equation is plotted in fig . [ fig : dimon ] for three separations . a maximum , similar to that appearing in eq . [ fig : self - both - moving ] , is clearly apparent . 74ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop ph.d . thesis , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop _ _ , ed . ( , , ) @noop * * , ( )

we present a new time - dependent density functional approach to study the relaxational dynamics of an assembly of interacting particles subject to thermal noise . starting from the langevin stochastic equations of motion for the velocities of the particles we are able by means of an approximated closure to derive a self - consistent deterministic equation for the temporal evolution of the average particle density . the closure is equivalent to assuming that the equal - time two - point correlation function out of equilibrium has the same properties as its equilibrium version . the changes in time of the density depend on the functional derivatives of the grand canonical free energy functional $ ] of the system . in particular the static solutions of the equation for the density correspond to the exact equilibrium profiles provided one is able to determine the exact form of $ ] . in order to assess the validity of our approach we performed a comparison between the langevin dynamics and the dynamic density functional method for a one - dimensional hard - rod system in three relevant cases and found remarkable agreement , with some interesting exceptions , which are discussed and explained . in addition , we consider the case where one is forced to use an approximate form of $ ] . finally we compare the present method with the stochastic equation for the density proposed by other authors [ kawasaki , kirkpatrick etc . ] and discuss the role of the thermal fluctuations .

cond-mat9810403

in recent years the off - equilibrium properties of extended systems have represented a very active field of research . in fact , while the present understanding of systems in thermodynamic equilibrium is rather satisfactory and is based on well established theoretical methods , the comprehension of their dynamical aspects is far from complete , in spite of massive experimental and theoretical investigations @xcite . in the present paper we shall focus attention on the density functional method , which represents a powerful and widely used tool to investigate the static properties of many particles systems @xcite and consider the possibility of extending this approach to off - equilibrium situations . some authors have already employed similar approaches on a purely phenomenological basis by analogy with the popular ginzburg - landau time dependent equation and the cahn - hilliard equation , but these methods are not applicable to the highly structured density profiles that one observes at the onset of crystallization . the density functional ( df ) formalism , with advanced models for the non - local functional dependence of the helmholtz free energy on the density distribution , has provided a good framework to study the solid - liquid transition and other highly structured systems . there have been previous attempts to derive a dynamic density functional ( ddf ) theory from the microscopic equations of motions @xcite- @xcite , so that the density functional approximations developed for systems at equilibrium might be extended to the dynamics of these systems . however , none of these proposals is fully satisfactory as we shall demonstrate , while the derivation that we consider makes direct contact with the equilibrium df formulation and is therefore consistent with thermodynamic requirements . the theoretical foundations of the density functional methods are based on the concept that the intrinsic helmholtz free energy of a fluid that exhibits a spatially varying equilibrium density @xmath1 , is a unique functional @xmath0 $ ] and is independent of the applied external fields for a given intermolecular potential . an exact knowledge of @xmath0 $ ] allows obtaining in a self - consistent fashion the profile @xmath1 and all the n - point correlations via functional differentiation . when dealing with non - equilibrium situations , caused by some changes of the external constraints , such as the temperature , the pressure , or electric field , it would be extremely useful to have similar methods at our disposal . mode coupling theories @xcite provide a kinetic approach to the dynamic of supercooled fluids and structural glass transitions , but fail to predict the crystallization process . in in other cases the phase ordering dynamics of liquids has been based on schematic model hamiltonians of the ginzburg - landau type , which neglect the microscopic structure or on heuristic approximations for the free energy @xcite . in principle the density is not the only relevant variable in a dynamical description . the velocity distribution and correlation are crucial in the understanding of hydrodynamic modes , while they do not appear in the equilibrium df for classical fluids . however , one could argue following cohen and de schepper @xcite that when the density is large , the momentum and the energy flow quickly through the system via collisions , while the density variable decays slowly . the hydrodynamic modes should become irrelevant for the dynamics of dense and strongly structured systems ; the only relevant variable should be the density distribution , as in the equilibrium case , and the use of the equilibrium @xmath0 $ ] is a promising starting point to include the effects of the density correlations . in the present study we restrict ourselves to systems with such `` relaxational dynamics '' , in which the velocity distribution plays no relevant role . instead of starting directly with the newtonian dynamics of the particles , we force the irrelevance of the velocity distribution at microscopic level and begin from the stochastic brownian equations of motion of a system of @xmath2 particles interacting via two - body forces . in the equilibrium limit , as a very long time average of the dynamic evolution , the newtonian and the brownian equations of motion should give the same results and be equal to those of the equilibrium statistical ensemble . the relaxational dynamics of dense systems has also to be similar for the two types of microscopic dynamics : the rapid flow of momentum and energy due to particle - particle collisions in the newtonian dynamics , is given ( most efficiently ) by the bath in the brownian dynamics . of course , there are cases ( including some of the examples analyzed in this work ) for which the newtonian and the brownian dynamics have very different results . in those cases our proposed df approach to the dynamics would still be useful for systems following the brownian equation of motion , like colloidal particles in a bath , but it would not be appropriate for systems in which the microscopic dynamics is newtonian and the velocity distribution becomes relevant . the paper is organized as follows : in section 2 we derive the dynamic density functional theory starting from the stochastic equations of motion of the particles and discuss the main features of the resulting ddf approach . in section 3 we apply the method to few systems of hard molecules in one dimension , for which the exact equilibrium free energy density functional is known ; the comparison between the ddf and the averages over the langevin simulations gives a clear view of the validity of our proposal . in section 4 we explore the same systems but using now approximate forms for @xmath0 $ ] , of the same type as those developed for realistic systems in three dimensions ; in this way we analyze the effects of the approximations used for @xmath0 $ ] , which would be unavoidable in the practical use of the ddf formalism . in 5 we draw the conclusions , discuss previous approaches and present the future perspectives . one considers an assembly of @xmath2 brownian particles of coordinates @xmath3 interacting via an arbitrary pair potential @xmath4 and experiencing an external field @xmath5 . neglecting the inertial term and the hydrodynamic interaction their motion can be described by the following set of coupled stochastic equations : @xmath6 + { \bf\eta}_i(t ) \label{eq : uno}\ ] ] where the term @xmath7 represents the influence of the thermal bath and has the properties : @xmath8 and @xmath9 where the average is over the gaussian noise distribution and @xmath10 run over @xmath11 . the constants @xmath12 and @xmath13 give the mobility and the diffusion coefficient of the particles , respectively . the einstein relation gives @xmath14 , and from here on we take @xmath15 to fix the unit of time and have @xmath16 . the evolution law drives the system towards the equilibrium situation which is described by the canonical gibbs probability measure . instead of considering all the trajectories generated from eq . ( [ eq : uno ] ) we shall consider the evolution of the density of particles . in order to render the paper self - contained we rederive briefly the transformation @xcite , using the rules of the ito stochastic calculus . in order to do so , we recall that if @xmath17 is an arbitrary function of @xmath18 given by the process : @xmath19 with @xmath20 its evolution is given by the following ito prescription for the change of variables ( see ref . @xcite ) : @xmath21 thus employing eqs . ( [ eq : gard2 ] ) and ( [ eq : uno ] ) we obtain : @xmath22 \nabla_i f({\bf r_i})+\nabla^2_i f({\bf r_i } ) + { \bf \nabla_i } f({\bf r_i } ) { \bf\eta_i(t ) } \label{eq : cinque}\ ] ] after inserting the identity @xmath23 and using the arbitrariness of @xmath17 , we obtain the equation for the partial density operator , @xmath24 : @xmath25 + { \bf \nabla } \eta_i({\bf r},t ) \hat\rho_i({\bf r},t ) \label{eq : sei}\ ] ] where , after some manipulation , the instantaneous global density operator , @xmath26 , can be shown to obey the following multiplicative noise equation : @xmath27 . \label{eq : sette}\ ] ] eq . ( [ eq : sei ] ) is a mere rewriting of the original eqs ( [ eq : uno ] ) and as it stands does not represent an improvement . also eq . ( [ eq : sette ] ) is a stochastic equation for the total instantaneous density @xmath28 and is not an average , thus it is a sum of @xmath2 spikes located at the positions of the particles at the positions of the particles at time @xmath29 . the connection with the df formalism was made by dean @xcite in the following terms : the helmholtz free energy density functional , @xmath30= t \int d { \bf r } \rho({\bf r } ) [ \log(\rho({\bf r } ) ) -1 ] + \int d r \rho({\bf r } ) \ v_{ext}({\bf r } ) + \delta f[\rho ] , \label{eq : f}\ ] ] contains the exact ideal gas entropy and the external potential contribution in the first two terms , while the third one includes the effects of interactions and correlations between the particles and its exact form is known only for very few systems . the first two terms in the bracket of ( [ eq : sette ] ) correspond precisely to the contributions of the ideal gas and the external potential to @xmath31 , and within an implicit mean field approximation it was observed that the third term in the bracket of ( [ eq : sette ] ) can also be cast in terms of the functional derivative of @xmath32 $ ] . in a slightly different language the analysis by kawasaki @xcite led to the same type of proposal : a dynamic density functional equation for @xmath33 in term of the functional derivative of @xmath0 $ ] and a remaining stochastic contribution , from the noise term in ( [ eq : sette ] ) . however , the use of the equilibrium functional @xmath0 $ ] only makes sense with an ensemble averaged density distribution . the delta - function peaks in @xmath33 would give infinite contributions to the first term in ( [ eq : f ] ) , because they correspond to a single microscopic state and not any statistical ensemble average . to make the connection between ( [ eq : sette ] ) and a density functional description one has to implement some kind of averaging over the instantaneous distribution of particles . within the microscopic brownian dynamics the obvious way to proceed is to average over the realizations of the random noise @xmath34 . we denote by brackets , @xmath35 , the results of this averaging and in particular we define the noise - averaged density @xmath36 . in the equilibrium limit , when the system has been allowed to relax for long enough time under the brownian dynamics , this average would correspond precisely to the gibbsian equilibrium average . in the study of the dynamics , out of equilibrium , the density @xmath37 has to be interpreted as an ensemble average , as in a collection of colloidal systems with the colloidal particles at the same initial conditions but with different ( thermalized ) microscopic states for the bath . we may hope , without demonstration , that this brownian ensemble may also give an accurate description of dense systems with newtonian dynamics near the crystallization , for which the ensemble should be interpreted as a collection of systems with the same initial positions but different ( thermalized ) velocities for the particles . when we proceed to take the noise average over the brownian evolution equation ( [ eq : sette ] ) the first effect is to cancel out the noise contribution in the last term . the stochastic equation for time evolution of @xmath38 becomes a deterministic equation for @xmath37 , @xmath39 , \label{eq : otto}\ ] ] in contrast with previous authors @xcite , @xcite , @xcite , who maintain the stochastic character of the dynamic df evolution keeping a random noise term @xcite . systems of non - interacting ideal particles , @xmath40 , provide an exact test of this point . in that case the free energy density functional @xmath41 $ ] reduces to the first two contributions in ( [ eq : f ] ) , since @xmath32=0 $ ] , and ( [ eq : otto ] ) may be written as a closed deterministic equation for the density distribution , @xmath42 = { \bf \nabla } \cdot \left [ \rho({\bf r},t){\bf \nabla } \frac{\delta f_{id } [ \rho({\bf r},t ) ] } { \delta \rho({\bf r},t ) } \right ] , \label{eq : dfid}\ ] ] which is the exact fokker - planck equation for the diffusion and drift of an ideal brownian gas . in the proposals of dean @xcite and kawasaki @xcite this equation would include a stochastic noise term , leading to an overcounting of the fluctuations . in particular the boltzmann equilibrium state predicted by ( [ eq : dfid ] ) in the static limit would be spoiled by the presence of the random noise , which would be equivalent to an overestimation of the temperature . in the case of interacting particles equation ( [ eq : otto ] ) is not a closed relation , since in order to obtain @xmath43 one needs the equal - time two - point correlation @xmath44 . the simplest mean field approximation assumes @xmath45 and gives a closed equation for @xmath43 , but it would give quite pathological results for the molecular core repulsions . following the same procedure used for @xmath43 we may obtain an equation for the time evolution of @xmath46 which in turn depends on the three - point correlation . in fact eq . ( [ eq : otto ] ) is only the first member of an infinite hierarchy of relations known as born - bogolubov - green - kirkwood - yvon ( bbgky ) integro - differential equations connecting n - point functions to ( n+1)-point functions @xcite . as in the equilibrium case one can get approximated results breaking the chain at any level . the so called kirkwood superposition approximation , replaces the three - point correlation by a product of two - point correlations as the next step after the mean field approximation and may already give reasonable results for hard core interactions . here we propose a different strategy , a density functional approach in which the two - point correlation function may be approximated with the help of equilibrium free energy density functionals . the excess free energy density functional @xmath32 $ ] in principle contains all the equilibrium correlation structures in the system and , although the exact functional form is known only for very few systems , there are workable and very accurate approximations for most systems of interest . we may use the information contained in @xmath32 $ ] , about the correlation structure at equilibrium , to approximate @xmath47 in a system out of equilibrium . in this way we get a generic , closed , dynamic density functional relation for @xmath43 equivalent to ( [ eq : dfid ] ) with a clear interpretation of the approximations involved . let us consider an equilibrium state of the system characterized by an arbitrary profile @xmath48 ( the subscript @xmath49 indicates the equilibrium average ) , which we shall eventually take equal to the profile @xmath43 at a given instant @xmath29 . such an equilibrium state certainly exists and represents a minimum of the grand potential functional provided we add an appropriate equilibrating external potential . one can prove , in fact , that for fixed temperature , chemical potential , and pair interactions there always exists a unique external potential @xmath50 which induces the given @xmath48 . in other words , upon adding the external potential @xmath50 , we would pin the system to be at equilibrium in a configuration corresponding to the instantaneous average density @xmath43 ; the potential @xmath50 is a functional of @xmath48 and changes with @xmath29 as @xmath43 varies @xcite . from the general properties of the equilibrium functionals we have that the following two exact equilibrium relations must be satisfied by @xmath48 and @xmath50 . first , the local balance of momentum at any point implies the bbgky relation , @xmath51 = -\beta \frac{1}{\rho_0({\bf r})}\int d r ' \rho^{(2)}_0({\bf r , r ' } ) \nabla v({\bf r - r ' } ) . \label{eq : bbgky}\ ] ] second , the thermodynamic equilibrium implies that the functional derivative of @xmath0 $ ] at any point is equal to a uniform chemical potential @xmath52 . taking the gradient gives the equation , first obtained by lovett et al . @xcite , @xcite , @xmath53= -{\bf \nabla}\frac{\delta}{\delta \rho_0({\bf r})}[\beta \delta f[\rho_0 ] ] = \int d { \bf r ' } c^{(2)}({\bf r},{\bf r ' } ) \nabla \rho_0({\bf r ' } ) , \label{eq : lovett}\ ] ] where @xmath54 } { \delta \rho_0({\bf r } ) \delta \rho_0({\bf r'})}\ ] ] is the direct correlation function , related to the functional inverse of the equilibrium two point density - density correlation @xmath55 . the functional @xmath56 $ ] serves to generate the sequence of inverse linear response or direct correlation functions @xcite upon functional differentiation with respect to @xmath57 . let us emphasize that eqs ( [ eq : bbgky ] ) and ( [ eq : lovett ] ) are exact for the instantaneous potential @xmath50 . comparing these two equations and assuming that the equal time correlation @xmath58 , averaged over the brownian noise , may be approximated by that of the equilibrium system with the same density distribution , we get the last term in eq . ( [ eq : otto ] ) as : @xmath59}{\delta \rho({\bf r},t ) } . \label{eq : df1}\ ] ] in summary , we have used the fact that at any instant we can find a fictitious external potential @xmath50 which equilibrates the system , i.e. constrains its grand potential to be minimal . this minimum is characterized by the imposed density profile @xmath60 and by equilibrium correlations @xmath61 consistent with it . the present approximation replaces the true off - equilibrium pair distribution function @xmath62 by the equilibrium @xmath61 , and then uses the equilibrium density functional @xmath32 $ ] to obtain the relevant information on this function . the assumption that the two routes , eqs . ( [ eq : bbgky ] ) and ( [ eq : lovett ] ) , are equivalent implies that the fluctuation dissipation theorem holds , while in general , out of equilibrium , it is violated @xcite . in fact the relation connecting @xmath61 to @xmath63 ( the o.z equation ) is an exact equilibrium property and is based on the idea that the correlation function is the matrix inverse of the second derivative of the functional @xmath64 with respect to @xmath48 . with ( [ eq : f ] ) and ( [ eq : df1 ] ) we may recast ( [ eq : otto ] ) into the main result of the dynamic density functional approach , based on the use of the equilibrium functional @xmath0 $ ] : @xmath65}{\delta \rho({\bf r},t ) } \right ] , \label{eq : df}\ ] ] which has the form of a continuity equation , @xmath66 , with the current of particles given by : @xmath67}{\delta \rho(r)}\right|_{\rho({\bf r},t)}. \label{eq : current}\ ] ] the main features of this approach are the following : \a ) @xmath0 $ ] is a functional solely of the density field and thus eq . ( [ eq : df ] ) is a closed non linear equation for @xmath68 . \b ) the equation is deterministic , but the variable @xmath68 has to be interpreted as the instantaneous density operator averaged over the realizations of the random noise @xmath69 . the contribution from the ideal gas entropy generates the diffusion term in ( [ eq : dfid ] ) and reflects the presence of the thermal noise . \c ) the only assumptions leading to ( [ eq : df ] ) are that the systems follows a relaxative dynamics , which may be described by the brownian motion of the particles in a thermalized bath , and that the instantaneous two - particle correlations are approximated by those in an equilibrium system with the same density distribution , as given by the ( exact or approximated ) density functional @xmath0 $ ] . before concluding this section we note an interesting feature of the dynamics : using the equation of evolution for @xmath43 it is straightforward to show that for any system with closed or periodic boundary conditions , the dynamics always tends to decrease the free energy functional , i.e. @xmath70}{dt}=-\int d { \bf r } \ \rho({\bf r},t ) \ \left [ \nabla \frac{\delta f}{\delta \rho } \right]^2 \leq 0 . \label{eq : decrease}\ ] ] in the long time limit , the evolution of the system leads to its equilibrium density distribution , which corresponds to a uniform value of @xmath71 , i.e. the usual euler - lagrange equation in the equilibrium df formalism . however , the trajectories that lead to the minima of @xmath64 are not necessarily along the directions corresponding to the maximum slope . the continuity equation , implies that the local conservation of particles is built in and imposes important constrains on the local changes of @xmath68 . for any system with a finite number of particles there is a unique canonical equilibrium density distribution , @xmath48 , which corresponds to the unique local ( and global ) minimum of the exact free energy density functional @xmath0 $ ] . however , the use of approximations for @xmath0 $ ] may lead to the existence of several local minima in which the dynamics of eqs . ( [ eq : df],[eq : decrease ] ) may get trapped ; this deserves further comment . notice first that these local minima of the free energy density functional , and the barriers between them , can not be directly associated to the local minima , and to the barriers , of the potential energy in the langevin description . consider the dynamics of an ideal brownian gas in an external potential with two local minima , separated by a barrier @xmath72 ; the langevin representation requires the gaussian noise @xmath69 to allow the particles to go over the barrier . the probability of such a jump is proportional to @xmath73 and , for large barriers , it sets the scale of time for the equilibration of the system . the same time scale appears in the exact fokker - planck representation ( [ eq : dfid ] ) through a different mechanism : the free energy landscape has a single minimum , at the equilibrium density @xmath74 , there are no free - energy barriers and the deterministic time evolution is set by the particle current ( [ eq : current ] ) along the functional path to equilibrium . when the equilibration requires moving particles across a large potential barrier , the relaxation time is imposed by the density there , @xmath75 , which may produce a very weak current even if the gradient of the local chemical potential is large . the presence of different local minima in approximate free energy density functionals is often found in the use of the density functional formalism for equilibrium properties . the usual interpretation is that the global minimum gives the true equilibrium state , while other local minima are associated with metastable states , and phase transitions are described as the crossover of different minima @xcite . this is a ( generalized ) mean - field level of description , in which the phase transition is described in terms of an order parameter related to the one - particle distribution functions . the exact description should always give convex thermodynamic potentials and it requires the description of the phase transitions in terms of the n - particle correlations , rather than in terms of a one - particle order parameter . in the present context , the use of approximate free energy density functionals with more than one local minima poses a problem of interpretation . when the deterministic time evolution ( [ eq : df ] ) gets trapped at a metastable state it can never reach the true equilibrium state . in previous attempts to use @xmath0 $ ] for dynamics , both dean @xcite and kawasaki @xcite have kept an extra random noise term in ( [ eq : df ] ) , as a remnant of the original random noise in the langevin dynamics . this random noise , allows the system to jump over any metastability barrier in a finite time but it leaves the problem of identifying its origin and intensity . from our analysis it is clear that the use of the free energy density functional makes sense only for the density @xmath76 , averaged over the realizations of @xmath69 in the langevin dynamics , and this averaging gives the deterministic equation ( [ eq : df ] ) . we proceed here to explore the results of the deterministic ddf formalism and come back to this point in the conclusion section . the practical use of ( [ eq : df ] ) to study problems like the growth of a liquid drop from an oversaturated vapour seems to be limited . the classical lifshitz - slyozov - wagner theory @xcite for the late stages of growth , beyond the critical droplet size , is given directly from ( [ eq : df ] ) , even if we use the simplest local density approximation for @xmath0 $ ] . however , any available approximations for @xmath0 $ ] , will be unable to describe the early stages of nucleation , as they do not include the effects of long - range critical - like correlations . however , we now have density functional approximations with a good description of the short range correlation structure in highly packed systems . the df description of the freezing of a liquid @xcite,@xcite or the study of fluids confined to narrow pores @xcite are the most remarkable achievements of the non - local density functionals developed in the last decades . we believe that the ddf formalism may be used to study the dynamics of densely packed fluids taking full advantage of the description of the correlation structure at short range , given by suitable approximations for the equilibrium @xmath0 $ ] . the most interesting ( and difficult ) problem would be the study of a closely - packed system , which has not crystallized . the langevin description ( [ eq : uno ] ) may give extremely slow dynamics since either the random noise has to take the system over large energy barriers ( like in the ideal gas described above ) , or it has to produce rearrangements which are only possible through unlikely correlations of many particles . within the deterministic dynamic density functional approach ( [ eq : df ] ) this situations may be seen in two possible ways : \1 ) the free energy landscape becomes rough and displays many local minima , whose number grows exponentially with the size of the system @xcite , and within our description the system would remain indefinitely trapped in any of these minima , unless it is annealed at a higher temperature . the idea is that when a uniform liquid is forced to have a density larger than the one corresponding to a liquid at coexistence with the solid , the system starts developing inhomogeneous patterns , which are associated with local minima of the free energy . one conjecture relates the origin of the glassy behavior in liquids to the existence of these minima , as an extension of the well founded theory of mean - field spin - glasses @xcite , @xcite . \2 ) alternatively , the free energy landscape @xmath0 $ ] may be smooth , with a single minimum , because the correlations required to relax the system are described well enough by the free energy functional . in this case , the system will never get really frozen , until it reaches the true equilibrium state , but the dynamics may become so slow that it may appear to be frozen in any practical computation . the slow dynamics of ( [ eq : df ] ) may be a result of having very low density along current path ( again , as in the ideal gas above ) or of requiring very unlikely ( but not impossible ) correlations of many particles . this case would correspond to the conjecture that the glassy behaviour in liquids is due to a divergence in the viscosity , and in the relaxation times , rather than to the existence of a metastable freezing. within the dynamical density functional approach having one or the other way would be a result of using a worse of a better approximation for @xmath0 $ ] . the configurations with a very large escape time , when described by a good density functional , may appear like permanent stable states when the dynamics is described with a poorer approximation for @xmath0 $ ] . in the next section , we present explicit results for a simple model in one dimension which displays these features and illustrates how the structure of the nearly - frozen states may be given by simple df approximations , while the actual calculation of the escape time may require a density functional giving good account of the many - particle correlations . since neither the correctness nor the feasibility of the approach presented have been tested so far , we shall begin comparing the two levels of description : the brownian dynamics and df dynamics and in order not to introduce unnecessary sources of discrepancies we consider a one dimensional hard - rod system , whose exact equilibrium df is known @xcite , @xcite and which includes strong correlation effects due to the infinite repulsion between the particles . long ago , percus @xcite was able to determine the exact form of the free energy density functional for an assembly of hard rods of length @xmath77 . the excess functional due to the interactions reads : @xmath78 = \int_{-\infty}^{\infty } dx \ \phi(\eta(x ) ) \ \frac{\rho(x+\sigma/2)+\rho(x-\sigma/2)}{2 } , \label{eq : percus}\ ] ] where @xmath79 and the local packing fraction @xmath80 is defined as : @xmath81 there are other equivalent ways to calculate @xmath64 ( like taking the weighting graining only towards the right , or towards the left ) which yield the same value of @xmath64 for any density distribution . using eq . ( [ eq : df ] ) we obtain the following equation for the time evolution of the density : @xmath82 + \frac{\partial } { \partial x } \left [ \rho(x , t ) \frac{d v_{ext}(x)}{dx } \right ] , \label{eq : rho}\ ] ] where we have chosen the energy units such that @xmath83 . the first term in ( [ eq : rho ] ) represents the diffusion equation for ideal gas case , the second term is the correction due to the hard - rod interaction . it is worthwhile to point out that @xmath84 is just the two - point equilibrium correlation function @xmath85 evaluated at contact , i.e. when @xmath86 , so that this term takes into account the collisions of the rod at @xmath87 with the remaining particles on the right hand side . similarly the other term describes the interactions with the left sector . when the density profile varies very smoothly ( compared with the hard - rod length ) the second term may be written in terms of the local chemical potential and the compressibility and one obtains a diffusion equation with a renormalized constant . our first check has been to compare the results of langevin simulations eq . ( [ eq : uno ] ) with the results obtained by means of the ddf eq . ( [ eq : df ] ) and ( [ eq : rho ] ) , using the exact @xmath0 $ ] , for the free expansion of @xmath2 hard rods in absence of external potential . in figure ( [ fig : dos ] ) we present the density profiles for a system of @xmath88 hard rods of unit length , @xmath89 . the temperature is fixed to @xmath83 and in the initial configuration , at @xmath90 , the rods are set at fixed positions separated by a distance @xmath91 between their centers . in the first stages of the time evolution each rod develops a gaussian density distribution , and the superposition of all the rods gives a total density distribution with strong oscillations , typical of tightly packed hard molecules . with increasing time the packet expands , the oscillations become weaker and then disappear . for very large @xmath29 ( not shown in the figure ) the packet becomes very wide , with @xmath92 everywhere , so that the collision term in ( [ eq : rho ] ) is nearly irrelevant . in that limit the system evolves like an ideal gas with a gaussian distribution of width proportional to @xmath93 . the dots in figure([fig : dos ] ) are the average over 2000 langevin simulations for the same system . the qualitative trend is similar to the ddf results , but the damping of the oscillations is clearly slower in the simulation . in figure([fig : uno ] ) we present the time evolution of @xmath94 for packets with @xmath95 , @xmath96 and @xmath97 , to give a measure of the rate of expansion . the dotted line is the exact result for the ideal gas , @xmath98 , which is independent of @xmath2 . for hard rods there is a clear enhancement of the effective expansion rate , because the rods at the two ends of the packet have a strong bias towards moving away from the the packet . the result of this effect increases with @xmath2 because it acts until the whole packet has expanded . for @xmath99 and any value of @xmath2 the slope of @xmath98 goes to the ideal value , but the enhanced expansion at small @xmath29 produces a shift of the values with respect to the ideal gas . the comparison between the ddf and the average over 2000 langevin simulations again shows the same overall trend and also the same dependence with @xmath2 , but the ddf gives always a slightly larger expansion rate . the case with @xmath95 offers a clear explanation for this discrepancy : with a single rod in the system there should be no collisions and the results should be those of the ideal gas , but the second term in the ddf equation ( [ eq : rho ] ) , obtained with the exact @xmath0 $ ] , still gives a contribution , unless @xmath100 for any @xmath87 . the reason is that in the exact @xmath0 $ ] ( [ eq : percus ] ) , as for any density functional used in the ddf , the density distribution has to be interpreted in the grand - canonical ensemble . it corresponds to a system in contact with a particle reservoir , in which the chemical potential is set to give the average value of @xmath2 ; however , the configurations contributing to the density distribution may have any number of particles . in the system with @xmath101 , there would be contributions from the density distributions with @xmath102 , @xmath103 , ... ( compensated with the contribution with @xmath104 ) , and these contributions include the effects of the collisions . in general , the fluctuation in the number of particles opens a relaxation path which is not present in the langevin simulations , carried with fixed @xmath2 . this extra relaxation path produces the faster damping of the oscillations in the density profiles and the larger diffusion rate of the ddf . the effect is important only in the intermediate stage of the expansion because for @xmath105 the compressibility of the the system is too low to have important fluctuations in @xmath2 and for @xmath99 the system is so diluted that the total effect of collisions is negligible . in figure([fig : tres ] ) we display the density profiles for a system of four hard rods falling to the bottom of a parabolic potential well , @xmath106 , with @xmath107 . at @xmath90 the rods are located at @xmath108 and @xmath109 , well separated from each other . during the first stages of the time evolution each rod follows a steady drift due to the external force and , at the same time , they develop gaussian peaks of increasing width , due to the random diffusion . the collisions between the two rods at each side of the potential well become important for @xmath110 ; later the two packets collide and relax to the equilibrium density distribution , which is reached for @xmath111 , within our numerical precision . the results of the ddf and the average over 2000 langevin simulations are in good agreement , although small discrepancies may be observed both in the early drifting peaks and in the final ( equilibrium ) profiles . the origin of these discrepancies is again the difference between the langevin simulation with fixed @xmath2 and the grand - canonical ddf . the final equilibrium density distribution with the langevin simulation corresponds to the canonical ensemble and for systems with small number of particles it is known to be different from the grand - canonical distribution @xcite . in our case , the difference depends on the value of the parameter @xmath112 , in the external potential : for @xmath113 the rods are very tightly packed , the compressibility is very low and the fluctuations in the number @xmath2 are very small ; for @xmath114 the final equilibrium profile is very broad , without oscillations , and the effect of the collision is too weak to produce observable differences ; but for intermediate values @xmath115 the difference between the canonical and the grand canonical ensemble may be quite important and it is reflected in the different time evolution predicted by the ddf and by the langevin simulations . the use of a canonical @xmath0 $ ] in the ddf equation ( [ eq : df ] ) would , in all probability , give a better agreement between the two methods , but unfortunately we are not aware of any explicit canonical density functional for interacting particles . nevertheless , the use of the equilibrium @xmath0 $ ] to include the role of the molecular correlations is always an approximation , so that even without considering the difference in the thermodynamic ensemble , it was not obvious that the agreement between the ddf and the langevin simulation would be as good as observed in the preceeding figures . the third check of the theory , being considered , is the relaxation in a system which requires strongly correlated motions of all the particles . we place the hard rods in a periodic external potential @xmath116 , with minima at any integer values of @xmath87 . the hard - rods length is taken as @xmath117 , so that two rods can not be at the bottom of nearest - neighbour wells . we take periodic boundary conditions with total length @xmath118 and set @xmath119 rods , which at the initial time are at the bottom of every second well of @xmath120 . the equilibrium density distribution , which may be obtained directly by the minimization of the exact free energy density functional , has the full symmetry of the the external potential , so that the relaxation process has to shift ( on average ) half a particle from the initially occupied wells to those wells which are initially empty . however , the jump of a rod over the barrier , to the next potential well , is not compatible with keeping the next rod at the bottom of its potential well . the system has to pay the extra energy of keeping the two consecutive rods away from the minima or it has to relay on a correlated motion of the @xmath2 rods , to shift from one subset of minima to the other one . in figure ( [ fig : cuatro ] ) we present the time evolution of the density profiles with @xmath121 showing different times , for both the ddf equation ( [ eq : df],[eq : rho ] ) and the average of the langevin simulations over @xmath122 realizations of the noise . in both cases the relaxation is slow ( compared with the previous examples ) and it becomes much slower for increasing values of @xmath123 or @xmath77 . in agreement with our general prediction eq.([eq : decrease ] ) , the system flows to the unique equilibrium state , density peaks grow at the positions of the potential wells which were initially empty until the exact equilibrium density profile is obtained . the comparison of the result clearly shows that the ddf equation , with the exact percus free energy , approaches the equilibrium state faster than the average of the langevin simulation . this difference is related again to the use of different statistical ensembles : the canonical langevin equation keeps constant @xmath2 while the grand - canonical @xmath0 $ ] allows for fluctuations in the number of particles , keeping only the average . the changes in @xmath2 in the ddf open a new relaxation path and gives a faster relaxation , even is the final equilibrium density profiles in the canonical and the grand - canonical ensembles are very similar . to get a quantitative description of the relaxation times we define an order parameter which gives , at any time , the relative difference between the occupation of the odd and the even potential wells , @xmath124 . the initial condition sets @xmath125 and the final equilibrium state corresponds to @xmath126 . both in the case of the ddf and of the langevin dynamics we observe a pure exponential decay , @xmath127 so that whole process may be described by the relaxation time @xmath128 . the importance of the particle correlations in the slowing of the relaxation dynamics is shown in figure ( [ fig : cinco ] ) through the dependence of the relaxation time with on the size of the hard - cores , @xmath129 , keeping the same external potential . both the langevin dynamics and the ddf show a fast decrease of @xmath128 , by nearly two orders of magnitude , when the rod size decreases from @xmath117 to @xmath130 , which makes the occupancy of neighbour potential wells easier . the increase of the external potential amplitude , for fixed @xmath117 , produces similar results , as presented in figure ( [ fig : seis ] ) . the relaxation time grows faster than exponentially with @xmath123 , and the results of the ddf follow from below the general trend of the langevin dynamics , over several order of magnitude for @xmath128 . for comparison we present in the same figure the relaxation times for the ideal gas in the same external potential . the conclusion is that the ddf formalism , with the the exact equilibrium free energy density functional , is able to reproduce qualitatively the low relaxational dynamics of a system produced by the packing constraints of molecular hard cores . in our example the final equilibrium state , with equal occupancy of all the potential wells , represents a superposition of highly correlated states , in which all the particles are in alternate wells , either in the odd or the even positions . the relaxation from the asymmetric initial state requires strong correlations , with all the particles moving together or in cascade , and the unlikelyness of these correlations gives the long relaxation time . this process may be regarded as a ( very simplified ) simile of the relaxation from a very dense , non - crystalline initial state : the approach to equilibrium may require the correlated rearrangement of many molecules , the unlikelyness of these correlations produces relaxation times which are so long that the system may appear as frozen in an ill crystallized state . the one - dimensional hard - rod system provides an interesting test to the ddf , because we may use the exact equilibrium density functional . unfortunately , given the hamiltonian describing a set of @xmath2 interacting particles in a d - dimensional space , the exact free energy functional is in general unknown , since the exactness would be equivalent to calculating the exact partition function of the model in any external potential . consequently one must rely on suitable approximation schemes and , in the case of repulsive hard - core interactions , there exist several approximate methods which yield accurate results @xcite . the requirements imposed are : first , one must recover the thermodynamic properties of the homogeneous fluid ; second , one must reproduce the structure of highly inhomogeneous systems ; finally , the approximation must satisfy a number of exact relations . among the most successful density functionals approximations for the free energy of hard spheres we have those known under the generic name of weighted density approximations ( wda ) . the main assumption is that for each particle , at some point @xmath131 , there is a contribution to the excess free energy , @xmath132 , which is a local function of a weighted density @xmath133 , obtained by averaging the true density profile @xmath134 over a small region centered at @xmath131 : @xmath135 the excess free energy per particle , @xmath136 , is obtained from the bulk equation of state and the weight function is set to get a good ( approximate ) description of the bulk correlation structure . different recipes have been used and in general they require a density dependent weight function @xcite , @xcite . however , even the simplest version in which @xmath137 is taken to be density independent and equal to the normalized mayer function , @xmath138 , is enough to give a qualitative description of the hard - sphere freezing @xcite . in this section we explore the consequences of using the ddf formalism , ( [ eq : df ] ) with approximate forms for @xmath0 $ ] , so that the molecular correlations are represented by an approximation to the true equilibrium correlations . to carry out this analysis we study the same systems described in the previous section but instead of taking the exact percus density functional , ( [ eq : percus ] ) , we use an approximate description of the hard rods similar to those developed for hard spheres . in particular , the wda for hard rods with the exact bulk equation of state and the zeroth - order constant weight function takes @xmath139 = \int_{-\infty}^{\infty } dx \ \phi(\overline \rho(x ) ) \ \rho(x ) , \label{eq : wda1}\ ] ] with the exact @xmath136 given by ( [ eq : nonideal ] ) and @xmath140 the time evolution equation from the ddf ( [ eq : df ] ) with this density functional approximation is similar to ( [ eq : rho ] ) but with the bracket in the second term given by @xmath141}{\delta \rho(x ) } = \frac{1}{2 } \left ( \frac{\rho(x+\sigma , t ) } { 1-\overline \rho(x+\sigma , t ) } + \frac{\rho(x+\sigma , t ) } { 1-\overline \rho(x , t ) } -\frac{\rho(x-\sigma , t)}{1-\overline \rho(x-\sigma , t ) } -\frac{\rho(x-\sigma , t)}{1-\overline \rho(x , t ) } \right ) . \label{eq : rhowda}\ ] ] notice that the weighted density @xmath142 averages the density profile from @xmath143 to @xmath144 , while the variable @xmath145 , in the exact df , takes the average only over half that distance . there is also a different combination of the functions evaluated in ( [ eq : rho ] ) and in ( [ eq : rhowda ] ) , but it is easy to check that in the low density limit , truncating at second order in the density , both equations become identical . this reflects that the wda includes the exact leading term of the molecular correlations in a density expansion , while the higher - order terms are approximated . for the equilibrium properties it is known that this type of zeroth - order wda overestimates the effects of the hard - core packing and it becomes more accurate when the weight function is allowed to depend on the weighted density . the results obtained with this approximate free energy for the problems a and b in the previous section are qualitatively similar to those with the exact @xmath0 $ ] . in the free expansion of a dense state the wda keeps the oscillations in @xmath146 for a longer time than the results with ( [ eq : rho ] ) , reflecting the overestimation of the hard core packing , but the mean square displacements are very similar to those in figure ( [ fig : uno ] ) . the equivalent trends are observed in the problem b ; although there is some difference between the exact and the wda equilibrium states , the dynamics is rather similar . we have also checked , in some cases , that the use of a better wda approximation , with the density dependence of the weight function expanded up to second order , gives results closer to those with the exact @xmath0 $ ] . in this way we have checked that the practical use of the ddf formalism does not depend on having the exact equilibrium @xmath0 $ ] . the use of wda or other density functional approximations opens a wide road for the study of the dynamical properties of many systems , with a level of approximation comparable to the accuracy of the results for the equilibrium properties . the system studied in the last subsection provided the most severe test for the ddf , since it was the case in which the molecular correlations played the most important role , and this system is the one in which the the use of an approximated @xmath0 $ ] may lead to qualitative differences . we have solved ( [ eq : df ] ) with ( [ eq : rhowda ] ) for the same system of @xmath119 hard rods in a periodic external potential as before and look for the relaxation from the asymmetric density distribution . for the system presented in figure ( [ fig : cuatro ] ) , with @xmath117 and @xmath121 , there is a qualitative difference with the results of using the exact @xmath0 $ ] . with the wda free energy the time evolution of @xmath146 goes to a stable very asymmetric profile , with an order parameter @xmath147 , indicating that the particles remain mainly in the potential wells where they are initially located . the freezing of the initial asymmetry is only possible because the wda equilibrium free energy for this system has two different minima , with positive and negative values of @xmath148 , instead of the single minimum with @xmath126 given by the exact @xmath0 $ ] . the approximate inclusion of the hard - core correlations in the wda is not good enough to average over the two types of configurations which appear as separated equilibrium states and , for the same reason , the ddf equation can not include the dynamic path between the two types of configurations . for lower values of the external potential ( or higher temperature ) we find that the final order parameter predicted by the wda decreases , as in a phase transition . indeed , the splitting of the equilibrium state is nothing but a spurious phase transition predicted by an approximate @xmath0 $ ] . below a critical value of @xmath149 the wda recovers a single symmetric equilibrium state with @xmath126 . the relaxation time diverges as we approach the phase transition and it is presented in figure ( [ fig : seis ] ) together with the results of the exact @xmath0 $ ] . this divergence is obviously a spurious result of the approximation used for @xmath0 $ ] . the only reasonable interpretation is that whenever an approximate free energy density functional has different local minima , in which the ddf equation may get trapped , the exact dynamics of the system would require rather long times to relax the density along the the functional path related to the order parameter of the spurious phase transition . the ddf with the wda is good enough to identify the slow density path , along the parameter @xmath148 , and it is good enough to describe the relaxation dynamics along other functional directions , but it can not be used to describe the relaxation of @xmath148 . we have checked that on improving the wda , by including the density dependence of the weight function , the spurious phase transition is shifted towards larger values of @xmath123 . again a better approximation for the equilibrium free energy functional also gives a better account of the molecular correlations in the ddf equation , since it is able to find relaxation paths which were not seen by simpler df approximations . however , when @xmath123 goes beyond a new threshold value ( i.e. when the exact relaxation time becomes too large for the approximate @xmath0 $ ] ) the spurious phase transition reappears . we have also tested the predictions of the ddf with an even simpler approximation for the equilibrium density functional . this is the ramakrishnan - youssouf functional @xcite,@xcite , sometimes referred as the hnc density functional approximation , which considers a functional taylor expansion of the free energy about a global mean density @xmath150 . truncating the series at the second order ( higher order coefficients are in general not known ) one gets : @xmath151 = -\frac{1}{2}\int d { \bf r}\int d { \bf r ' } ( \rho({\bf r})-\rho_b ) c_b({\bf r - r'})(\rho({\bf r'})-\rho_b ) , \label{eq : ry}\ ] ] where @xmath152 is the direct correlation function of a uniform fluid with density @xmath150 . the merit of the ry functional is that it is perhaps the simplest recipe and is easy to implement numerically . however it requires that the system have a well - defined mean density , @xmath150 , and it may it become unreliable when the density varies rapidly over a length scale of the order of molecular size . notice that in our case using the ry approximation is equivalent to saying that during its evolution the system has the same two point correlations that it would have at constant density @xmath150 . in our one - dimensional hard - rod system we may use the exact direct correlation function of the bulk hard - rods system , and the ddf equation is again similar to ( [ eq : rho ] ) but with the second term bracket given by @xmath153 } { \delta \rho(x)}= \left(\frac{\rho(x+\sigma , t)-\rho(x-\sigma , t ) } { 1-\eta_b}+ \rho_b \frac { \eta(x+\sigma/2,t)- \eta(x-\sigma/2,t)}{1-\eta_b^2 } \right ) . \label{eq : rhory}\ ] ] the application to examples a and b is uncertain because they do not have a clearly defined mean density . in the third example we have used ( [ eq : df],[eq : rhory ] ) with the obvious choice @xmath154 to get the relaxation of the density distribution in the periodic external potential . the results , in figure ( [ fig : seis ] ) , show a surprisingly fast relaxation towards the symmetric density profile . this density functional approximation never has spurious minima , so that the dynamics is never trapped in asymmetric distributions , but the relaxation is even faster than in the ideal gas case , in clear contradiction with the langevin simulation . this shows an example opposite to the wda result , the effect of the approximation for the free energy may be to underestimate the packing effects . in this case there is no risk that the system gets trapped out of the true equilibrium , but the approximation fails to describe the long relaxation times imposed by these packing constraints . we have presented a dynamic density functional approach for the relaxation of a classical system in terms of its equilibrium free energy density functional @xmath0 $ ] . the approach is valid only when the velocity correlations ( not included in @xmath0 $ ] ) are irrelevant , which excludes hydrodynamic modes and temperature gradients but may include problems like the brownian motion of colloidal particles and the molecular rearrangements of highly packed systems , when the collision time is much shorter than the relaxation time . in the derivation of the approach we start with the stochastic equations for the langevin dynamics of brownian particles and get a deterministic ddf equation for the time dependence of the density distribution , @xmath43 , which has to be interpreted as the ensemble average over the realizations of the random noise in the langevin dynamics . the equilibrium df formalism is used to go from a formal bbgky hierarchy , for the coupled dynamics of the n - particle distribution functions , to a closed ddf equation for @xmath43 . the assumption leading to this result is that the correlation structure in the system out of equilibrium is replaced by that in the equilibrium system with the same density distribution . moreover , we do not need to calculate the correlation structure explicitly because its effects on the dynamics are given directly in terms of the functional derivative of @xmath0 $ ] . the main advantage of the approach , for practical purposes , is being able to use the good approximations developed for the free energy functional of hard - core molecules to include the packing constraints in the dynamics of dense systems . we have presented several examples of one - dimensional hard rods to compare the ddf with the average of langevin simulations . the results , with the exact @xmath0 $ ] , are always in qualitatively good agreement , even in those cases in which the dynamics becomes extremely slow , when the relaxation requires very unlikely correlations between the particles . the main source of discrepancy is probably due to the fact that the free energy density functional developed for equilibrium always refers to the grand - canonical ensemble , so that the ddf includes the relaxation through changes in the number of particles , while the langevin dynamics keeps @xmath2 fixed . with the use of approximate density functionals , of the same type as developed for realistic models in three dimensions , we get an approximate description of the relaxation dynamics of a quality comparable to that for the equilibrium properties arising from the same @xmath0 $ ] . in some cases , when the role of the correlations is not too important and the relaxation times are not too long , the approximate @xmath0 $ ] is fairly accurate for the dynamic properties . in the third case , in which strong effects of the hard core packing lead to very long relaxation times , the difference between the results of the exact and the approximate @xmath0 $ ] are qualitative both for the equilibrium and for the dynamic properties . the presence of different local minima in the equilibrium free energy density functional produces , in the approximate ddf equation , the permanent freezing of system in any of these states , in contrast with the results of the exact free energy functional . in an interesting series of papers kawasaki @xcite derived a dynamic equation , in terms of a density functional hamiltonian @xmath155 $ ] , by a method different from ours , but his resulting fokker - planck equation for the probability distribution of the density : @xmath156}{\partial t}= -\int d{\bf r}\frac{\delta}{\delta \rho({\bf r})}{\bf \nabla}\cdot\rho({\bf r } ) { \bf \nabla } \bigl[t\frac{\delta}{\delta\rho({\bf r})}+\frac{\delta h[\rho]}{\delta \rho({\bf r})}\bigr ] p[\rho({\bf r}),t ] \label{eq : fpe}\ ] ] is equivalent to our eq . ( [ eq : df ] ) , apart from the presence of a term due to the stochastic noise . if @xmath57 in kawasaki equation ( [ eq : fpe ] ) is interpreted as a density operator , @xmath157 , his approach is equivalent to dean s equation ( [ eq : otto ] ) ; leading to a stochastic equation for this density operator . the main qualitative difference of this approach with our deterministic equation ( [ eq : df ] ) is that , when the approximate density functional for the equilibrium free energy has different local minima , the random noise in the ddf equation would always give a chance for changing from one minimum to another . the long time average of the density would always be a superposition of the density in the different local minima and the long relaxation times would appear as the result of high barriers between the local minima . all these features may appear to be physically correct and to represent a qualitative improvement over the deterministic ddf developed here . however , the use of the equilibrium free energy density functional requires always a density defined as a thermal ensemble average , while ( [ eq : otto ] ) refers to the instantaneous density operator . the hamiltonian @xmath158 in ( [ eq : fpe ] ) should also be a functional of the density operator , @xmath28 , and it is a completely different mathematical object that the excess free energy @xmath32 $ ] , as a functional of the equilibrium density @xmath159 . thus , equations ( [ eq : otto ] ) and ( [ eq : fpe ] ) are correct , for the density operator @xmath28 but impossible to translate in terms of the equilibrium free energy density functional . on the other hand , if these equations are interpreted as equations for the ensemble averaged density , @xmath43 , the random noise term would lead to a double counting of thermal fluctuations in the equation for @xmath43 . in particular , it would lead to a wrong equilibrium distribution , as can be verified in the simple case of non interacting particles for which the fpe ( [ eq : fpe ] ) converges to a probability distribution : @xmath160\sim \exp(-f[\rho]/k_b t)= \exp \left [ -\int d r \ \rho({\bf r } ) \ ( \ln \rho({\bf r})-1)+ \beta v_{ext}({\bf r } ) \right ] , \label{eq : false}\ ] ] where the first term in the exponential is clearly due to the overcounting of thermal fluctuations . moreover , we have shown that if we use the exact functional @xmath0 $ ] the ddf equation does not require random noise to give the correct results : the system always flows towards the true and unique equilibrium state . the ddf relaxation time may become very long when the system has to go through highly correlated states , but this effect corresponds to systems in which the true relaxation time ( in the langevin description ) is also very long . when we use an approximate @xmath0 $ ] these long relaxation times may become infinite , with the system trapped at a local minima of @xmath0 $ ] . an attempt to avoid this effect would require the intensity of random noise not to be proportional to the temperature , but to the error made by the approximation to @xmath0 $ ] , an error which is obviously unknown until we make a better approximation . otherwise , the relaxation time given by the stochastic ddf in these cases would just be a direct result of the uncontrolled level of noise kept in the functional equation . within our deterministic ddf , the existence of frozen states in the local minima of the approximate @xmath0 $ ] should be interpreted as the signature of very long relaxation times , but the only way to calculate how long these times are is to improve the approximation for the equilibrium free energy . nevertheless , knowing the existence and the approximate structure of these states , as given by workable approximations for @xmath0 $ ] , is already an interesting use of the ddf , together with its use to study the relaxation process in those cases in which there are no problems with different local minima . finally , as a plan for future work , we can consider a systematic way to improve the use of the equilibrium @xmath0 $ ] to estimate the correlation structure in the following terms : we tag particle number 1 and follow its position @xmath161 separately , while all the other particles ( @xmath162 ) are included in a density description , with @xmath163 as the noise - averaged conditional probability of finding a particle at position @xmath131 , and time @xmath29 , if the tagged particle is at position @xmath164 . now , we may consider that the @xmath165 particles are moving in an effective external potential , @xmath166 , which also contains the interaction with the tagged particle . the equivalent to the ddf equation ( [ eq : df ] ) may be applied to the conditional density @xmath167 , which is now coupled to the stochastic equation for @xmath161 . the advantage is that the correlation structure becomes partially described at the level of an effective one - particle density in an effective external potential , which is in principle much easier to describe with approximate free energy density functionals . this type of description would be similar to the reaction path description of chemical reactions , in which one , or a few , variables are used to describe the relevant functional directions for the changes in the molecular conformations . however , the formal and the practical use of density functional approximations along this line is still an open problem . it is a pleasure to thank p. tartaglia and f. sciortino for many illuminating discussions , and j.p . hernandez for carefull reading of the manuscript . p.t acknowledges financial support from the direccin general de enseanza superior e investigacin cientfica of spain under grant number pb97 - 1223-c02 - 01 . umbm was partially supported by istituto nazionale di fisica nucleare , project fi11 . bray , adv.phys . * 43 * , 357 ( 1994 ) j.d . gunton , m. san miguel and p.s . sahni , in _ phase transitions and critical phenomena _ edited by c. domb and j.l . lebowitz ( academic press , london 1983 ) , vol . r. evans , adv.phys . , * 28 * , 143 ( 1979 ) and in _ inhomogeneous fluids _ edited by d. henderson ( dekker ) ( 1991 ) . d.s . dean , j.phys . a * 29 * l613 ( 1996 ) k.kawasaki , physica a * 208 * 35 ( 1994 ) . k.kawasaki and s. miyazima , z. phys . b * 103 * 423 ( 1997 ) t.r . kirkpatrick and d. thirumalai , j.phys . a * 22 * l 149 ( 1989 ) . kirkpatrick , phys.rev a * 31 * 939 ( 1985 ) . g. goetze and l.sjogren , rep.progr.phys . * 55 * , 241 ( 1992 ) . r.wild and p. harrowell , phys.rev . e * 56 * 3265 ( 1997 ) . . de schepper and e.g.d . cohen , j.stat.phys . * 27 * , 223 ( 1982 ) . gardiner , _ handbook of stochastic methods _ ( springer , new york 1985 ) . the derivation of the equation for the averages could have been performed by employing the stratonovich convention with the same result . hansen and i.r . macdonald , _ theory of simple liquids _ ( academic , london ) ( 1986 ) r. lovett , c.y . mou and f.p . buff , j.chem.phys . * 65 * 570 ( 1976 ) . cugliandolo and j.kurchan , phys.rev.lett * 71 * , 1 ( 1993 ) . wolynes in _ proceedings of international symposium on frontiers in science _ edited by s.s . chen and p.g . debrunner ( aip , new york ) ( 1988 ) . i.m . lifshitz and l.p . pitaevski , _ physical kinetics _ , vol . 10 ( pergamon , new york , 1981 ) . p.tarazona , phys . rev a * 31 * 2672 ( 1985 ) and phys . rev a * 32 * 3148 ( 1985 ) . a.r . denton and n.w . ashcroft , phys . a * 39 * 4701 ( 1989 ) . r. evans , u. marini bettolo marconi and p.tarazona , mol . phys . * 60 * 573 ( 1987 ) . m. mezard , g. parisi and m.a . spin glass theory and beyond _ , ( world scientific , singapore ) ( 1987 ) .

we expose the relation between the properties of the three - body continuum states and their two - body subsystems . these properties refer to their bound and virtual states and resonances , all defined as poles of the-matrix . for one infinitely heavy core and two non - interacting light particles , the complex energies of the three - body poles are the sum of the two two - body complex pole - energies . these generic relations are modified by center - of - mass effects which alone can produce a borromean system . we show how the three - body states evolve inhe ,li , andbe when the nucleon - nucleon interaction is continuously switched on . the schematic model is able to reproduce the main properties in their spectra . realistic calculations for these nuclei are shown in detail for comparison . the implications of a core with non - zero spin are investigated and illustrated forne (o++ ) . dimensionless units allow predictions for systems of different scales .

nucl-th0512002

the three - body problem has a long history from macroscopic celestial classical mechanics , e.g. sun - earth - moon @xcite to microscopic quantum mechanics , e.g. the helium atom @xcite , three nucleons @xcite or three quarks @xcite . the modern treatment was boosted by the formulation of the faddeev equations @xcite originally aimed at scattering problems , but also successfully applied for bound states @xcite . the more recent interest in bound state halo structures and borromean systems are by now fairly well understood in terms of the basic two - body interactions @xcite . the success is at least indisputable for three - body systems with only one or a few bound states . on the other hand , the corresponding three - body properties for positive energies ( energies above breakup ) are much less established although studied thoroughly for both short and long range interactions @xcite . contributions from both short and long - range interactions make computations numerically difficult . the three - body coulomb problem itself is still considered unsolved @xcite and three - body resonances for strongly interacting particles are still debated @xcite . this is unfortunate , since the continuum structure often form the basis in descriptions of the dynamic behavior of a given system . crucial properties of the continuum are revealed by information about the resonances and virtual states defined as poles of the @xmath0-matrix . together with the discrete set of bound negative - energy states we also have a discrete set of unbound complex - energy states with their corresponding wave functions . for completeness also the continuous non - singular background states are needed , but the singularities of the scattering matrix are very often decisive . important examples using different methods within few - body physics are astrophysical reaction rates @xcite , adiabatic reaction processes arising at low energies or for large impact parameters @xcite , three - body decays @xcite , three - body resonances for faddeev operators @xcite , for nucleons @xcite , for electrons and positrons @xcite and four - body nuclear continuum states @xcite . this list could be extended . continuum structure is in general more difficult than bound state problems although various methods have been designed to overcome the technical problems at least for resonances , see for example @xcite . an understanding of the generic origin of @xmath0-matrix poles would be tremendously helpful especially if recognizable traces of a well - structured origin are left in the realistic spectra . no doubt this would allow easier interpretation of complicated numerical results , allowed design of better methods , and indicate which effects to look for in different contexts . the purpose of this work is to investigate how the two - body interactions determine the three - body continuum structure . we focus on three - body continuum states where none of the two - body subsystems is bound ( borromean systems ) . when two - body bound states are possible , different three - body structures can appear , as for instance , unbound three - body states with negative energy . however , these systems can very often be understood as two - body systems made by a two - body bound state and the remaining third particle in the continuum . with the help of the hyperspheric adiabatic expansion method we determine which three - body bound or virtual states and resonances result from given sets of corresponding two - body properties . taking a simple system as starting point , it is then possible to observe how the three - body states evolve in the complex energy plane when more and more realistic features are incorporated into the calculation . in section ii we give details about the complex scaled adiabatic expansion methods . in section iii the schematic system of an infinitely heavy core and two mutually non - interacting light particles is described analytically , while in section iv the general properties of the system are described after numerical studies of specific systems . in section v we relate the spectra from the schematic model with those obtained in realistic calculations for systems with zero ( @xmath1he , @xmath1li , @xmath1be ) and non - zero core - spin ( @xmath2ne ) . we finish in section vi with some qualitative estimates for other systems and in section vii we give a short summary and the conclusions . to describe a three - body system we use the standard coordinates : @xmath5 where @xmath6 , @xmath7 , and @xmath8 are the masses of the three particles and @xmath9 is an arbitrary normalization mass . from the jacobi coordinates we define the hyperspheric coordinates @xmath10 , where @xmath11 , @xmath12=@xmath13 , @xmath14=@xmath15 , and @xmath16 and @xmath17 give the directions of @xmath18 and @xmath19 , respectively . the three - body wave function is written as a sum of three components @xmath20 , each corresponding to one of the three possible sets of jacobi coordinates @xcite . these three components satisfy the three faddeev equations @xmath21 where @xmath22 is the kinetic energy operator , @xmath23 is the two - body interaction between particles @xmath24 and @xmath25 , and @xmath26 is the three - body energy . here @xmath27 is a cyclic permutation of @xmath28 . we employ the hyperspheric coordinates to expand each component @xmath20 in terms of a complete set of angular functions @xmath29 @xmath30 where the additional index @xmath31 labels different solutions we later on want to consider . usually the expansion ( [ eq3 ] ) converges rather fast , and only a few terms ( typically no more than three ) are needed . by rewriting eq.([eq2 ] ) in terms of the hyperspheric coordinates , and inserting the expansions in eq.([eq3 ] ) we separate the faddeev equations into angular and radial parts : @xmath32 f^{(n_0)}_n(\rho ) = \nonumber \\ \sum_{n ' } \left ( 2 p_{n n ' } \frac{d}{d\rho } + q_{n n ' } \right ) f_{n'}^{(n_0)}(\rho ) \ ; , \label{eq5}\end{aligned}\ ] ] where @xmath33 is a hyperangular operator which together with expressions for the functions @xmath34 and @xmath35 can be found in @xcite . the complete set of angular functions @xmath29 in the expansion ( [ eq3 ] ) is precisely the eigenfunctions of the angular part of the faddeev equations . the index @xmath36 labels the the corresponding eigenvalue @xmath37 , which enters in the coupled set of radial equations ( [ eq5 ] ) as effective potentials . the index @xmath31 is related to the boundary condition for continuum wave functions , which for short - range interactions and no two - body bound states has the asymptotics given by @xcite : @xmath38 , \end{aligned}\ ] ] where @xmath31 then labels the incoming channel , the @xmath0-matrix @xmath39 is a factor depending on the complex momentum @xmath40 related to the complex energy @xmath26 by @xmath40=@xmath41 , and @xmath42 and @xmath43 are hankel functions of first and second kind . the hypermomentum @xmath44 is given by the asymptotic behaviour of the angular eigenvalue @xmath45 that approaches @xmath46 @xcite . for three - body bound states @xmath47 must fall off exponentially at large distances . then only one hankel function is present and the @xmath0-matrix @xmath39 has a pole for the imaginary momentum @xmath40 ( @xmath48 ) and the energy @xmath26 is negative . poles of the @xmath0-matrix in the lower half of the complex momentum plane correspond to three - body resonances , and their asymptotic radial wave function is given only by the @xmath49 part in eq.([eq6 ] ) . asymptotically the hankel function @xmath50 vanishes exponentially like @xmath51 , while @xmath49 grows like @xmath52 . this means that the radial coefficients of the continuum wave functions are dominated by the @xmath49 part . therefore , it is very difficult to distinguish the resonance wave function , where only the @xmath49 term is present in the asymptotics , from an ordinary continuum wave function . this problem is solved by applying the complex scaling method @xcite , where the jacobi coordinates @xmath53 and @xmath54 are rotated into the complex plane by an arbitrary angle @xmath55 ( @xmath56 , @xmath57 ) . this means that only the hyperradius @xmath12 is rotated ( @xmath58 ) , while the five hyperangles @xmath59 remain unchanged . after this rotation the radial wave functions of the resonances behave asymptotically like @xmath60 where @xmath61 is the argument of the complex momentum @xmath40 ( @xmath62 ) . from ( [ eq7 ] ) we observe that when @xmath63 the radial wave function falls off exponentially , exactly as a bound state . continuum wave functions are dominated in the asymptotics by the @xmath64 term , that diverges exponentially when @xmath63 . thus , after complex scaling , resonances can be easily distinguished from ordinary continuum states , and furthermore , the numerical technique used to compute bound states can be used for resonances . in particular , after solving the complex scaled equations ( [ eq4 ] ) and ( [ eq5 ] ) with the boundary condition ( [ eq7 ] ) three - body resonances and bound states are simultaneously obtained . solving eq.([eq5 ] ) with a box boundary condition ( @xmath65=0 , @xmath66 being a large value of the hyperradius ) discretizes the continuum spectrum , and the continuum states are rotated by an angle @xmath55 in the momentum plane and @xmath67 in the energy plane @xcite . the box boundary condition is often enough to obtain accurate bound state and resonance solutions . we analyze in this section the schematic model of an infinitely heavy core and two light particles . these two mutually non - interacting light particles interact with the core via short - range interactions . we shall first assume that these three particles have zero spin and then later generalize to non - zero particle spins . from the jacobi coordinates ( [ eq1 ] ) we can construct the corresponding conjugated momenta : @xmath68 in precise analogy to the hyperradial coordinates , we now introduce the hypermomentum coordinates @xmath69 defined as @xmath70 , @xmath71 , @xmath72 . the three - body momentum @xmath40 defined earlier is independent of the jacobi coordinate system labeled @xmath73 . the directions of @xmath74 and @xmath75 are described by @xmath76 and @xmath77 , respectively . when one of the three spinless particles has infinite mass ( let us take @xmath8=@xmath78 ) , eqs.([eq1 ] ) and ( [ eq8 ] ) simplify , in the three - body center - of - mass system where @xmath79=0 and @xmath80+@xmath81+@xmath82=0 , to the form in fig.[fig1 ] . therefore , in two of the three possible sets of jacobi coordinates , the coordinates are simply proportional to the distances between the infinitely heavy core and each of the two light particles . if the two - body interactions depend only on the distance between particles , the three - body hamiltonian @xmath83 can be written as @xmath84 ( see fig.[fig1 ] ) where @xmath85 and @xmath86 are the two - body hamiltonians describing the corresponding two - body systems . from this separability , the three - body bound state energies are obviously given by the sum of the corresponding two - body bound state energies . for three - body resonances and virtual states the same happens . three - body resonances show up at energies equal to the sum of the two - body resonance energies of the @xmath87 and @xmath88 subsystems . from the separability of the hamiltonian this result may be accepted as obvious . however , the only trivial deduction using the separability is that the three - body energy is given by the sum of the two two - body energies . this does not necessarily imply that a three - body energy equal to the sum of two two - body resonance energies must correspond to a three - body resonance . bound states have discrete energies , and the split of a given three - body bound state energy into two two - body energies can be made only in specific ways , since the two - body energies can have only the specific values corresponding to the discrete two - body bound state spectrum . however , resonances are continuum states , and then a given three - body energy can be obtained by infinitely many pairs of two - body energies . therefore , if a given three - body energy matches with the sum of two two - body resonance energies , there are also infinitely many other pairs of continuum ( non - resonant ) two - body energies whose sum gives the same three - body energy . thus , it is not a trivial conclusion that only the particular three - body state of matching energy corresponds to a three - body resonance defined as a pole in the three - body @xmath0-matrix . let us then investigate this more closely . from the separability of the hamiltonian it follows that the three - body wave function is the product of the wave functions describing the @xmath87 and @xmath88 two - body subsystems . for unbound two - body systems the two - body wave function can be expanded in partial waves , and the three - body wave function takes then the form : @xmath89 where @xmath90 refer to the jacobi system in fig.[fig1 ] and the momenta of the corresponding subsystems are @xmath91 and @xmath92 . the spherical harmonics and the two - body radial functions are denoted @xmath93 and @xmath94 . following @xcite , also the continuum wave function for a three - body system of spinless particles can be written as : @xmath95 where @xmath90 , @xmath96 , @xmath97 and @xmath98 could correspond to any of the three possible sets of jacobi systems . we omitted the index @xmath73 . the hyperradial functions , @xmath99 , are solutions to eq.([eq5 ] ) where we now specified the indices @xmath36 and @xmath31 and explicitly included the factor @xmath40 in the argument . the hyperspheric harmonics @xmath100 can for instance be found in @xcite . using the orthogonality of the hyperspheric harmonics , we immediately have @xmath101 and by inserting eq.([eq10 ] ) into ( [ eq11 ] ) we get @xmath102 where @xmath103 is the normalization constant appearing when expressing the hyperspheric harmonics in terms of the spherical harmonics and the jacobi polynomials @xcite . due to the delta functions , the three - body hyperradial functions in eq.([eq12 ] ) are block - wise diagonal and each block matrix can , for given @xmath104 , be labeled by the indexes @xmath105 . note also that the wave function does not depend on the total orbital angular momentum @xmath106 . all @xmath106-values allowed by coupling are degenerate . in the limit of zero interactions between the core and the light particles the two - body radial wave functions @xmath107 become the spherical bessel function @xmath108 , and eq.([eq12 ] ) becomes : @xmath109 where @xmath110 is a bessel function of the first kind . inserting ( [ eq13 ] ) into ( [ eq9 ] ) we then obtain the three - body wave function @xmath111 and the plane waves are recovered from the hyperradial partial wave expansion . the large - distance asymptotic behaviour of the three - body radial wave functions , eq.([eq12 ] ) , can be obtained from the asymptotics of the two - body wave functions , that for the case in which none of the two - body core - particle subsystems is bound has the form : @xmath112 where @xmath113 are the corresponding two - body hankel functions of first and second kind , and @xmath114 is the two - body @xmath0-matrix . inserting eq.([eq15 ] ) into ( [ eq12 ] ) and using the expressions given in the appendix , we obtain @xmath115 , \;\end{aligned}\ ] ] where we omitted the unimportant block indexes @xmath116 . then eq.([eq16 ] ) is identical to eq.([eq6 ] ) after division by the phase factor @xmath117 in eq.([eq3 ] ) . the three - body @xmath0-matrix takes the form : @xmath118 where @xmath119 is a smooth function of @xmath120 given in terms of the jacobi functions @xmath121 with @xmath122 and @xmath123 . the integrands in eq.([eq17 ] ) contain the two - body @xmath0-matrices depending on corresponding momenta . the integration variable maintains the total energy of the two subsystems while varying the ratio of their momenta . assume that one pole of one of the subsystems is a first order pole which does not coincide with any pole of the other subsystem , then the integrations in eq.([eq17 ] ) across the pole give a smooth function without trace of the pole . however , if two poles from different subsystems exist simultaneously the product of two - body @xmath0-matrices in the numerator of eq.([eq17 ] ) results in a pole term of second order , which after integration across the pole , leaves a first order pole in the three - body @xmath0-matrix . the energy of this three - body pole corresponds then precisely to the sum of the energies of the two two - body poles . if one pole is of higher order , and not coinciding with a pole from the other subsystem , it would survive the integration in precisely in the same way in numerator and denominator . thus no trace is left in the three - body @xmath0-matrix . a three - body pole only arises when two two - body poles belonging to different subsystems exist simultaneously . combining any order of coinciding poles from the two subsystems is then seen to produce a three - body pole of an order equal to the smallest order of the two - body poles involved . this proves that the three - body poles appear if and only if the energy is equal to the sum of energies of two two - body poles from different subsystems . the smallest order of the two coinciding poles reappears in the three - body pole . in case of having a two - body bound state , the corresponding two - body asymptotics is not given by eq.([eq15 ] ) but by a falling off exponential @xmath124 , where @xmath25 is determined by the two - body binding energy . the three - body properties are then directly dictated by the two - body @xmath0-matrix describing the remaining core - particle subsystem . if this two - body @xmath0-matrix has a pole corresponding to a two - body bound state , the three - body system is obviously bound with a binding energy equal to the sum of the two two - body binding energies . if the two - body @xmath0-matrix has a pole corresponding to a two - body resonance ( or virtual state ) , the three - body system has then a resonance ( or virtual state ) that actually corresponds to a resonance ( or virtual state ) of the two - body system made by the bound two - body state and the remaining third particle . at this point it is also important to mention that we are identifying poles of the @xmath0-matrix in the fourth quadrant of the energy plane and resonances . however , strictly speaking , for a pole to be considered a resonance it is required that its width is clearly smaller than the energy separation between resonances . in this way resonances are the poles of the @xmath0-matrix close enough to the real energy axis . it can certainly happen that when two complex energies corresponding to two singularities in the @xmath0-matrix are added , the final energy can be close to the negative imaginary axis , or even in the third quadrant of the energy plane . these singularities should not be considered as resonances in the sense given above . the formalism leading to the expression for the three - body @xmath0-matrix is applicable for a very different system , i.e. three particles of finite mass with only one non - vanishing two - body interaction . if we choose the jacobi coordinate system where @xmath125 is related to the non - zero two - body interaction the wave function is again given by eq.([eq10 ] ) with a bessel function instead of @xmath126 . all derivations remain unchanged and we arrive at the @xmath0-matrix in eq.([eq17 ] ) . the difference is that now only the two - body @xmath0-matrix related to the @xmath125-degree of freedom has poles . therefore the three - body system has no @xmath0-matrix poles . this system effectively also consists of two independent two - body systems , but now reached in a completely different limit with one non - zero interaction and arbitrary masses . the result then illustrates other aspects of the continuum properties for three - body systems . in both cases the conclusion is that one subsystem alone can not produce three - body resonances or virtual states . at least two different subsystems must collaborate and simultaneously contribute with coinciding poles . it then seems inevitable that , if all the three two - body subsystems only have poles , then fully realistic three - body systems must also have only poles . the results in the previous subsection remain valid for particles of non - zero spin if the two - body interactions are spin - independent , because then the spin part of the wave function is decoupled from the coordinate part . in fact , even in the case of identical particles the spin part can always be used to establish the proper ( anti)symmetry of the wave function without changing the coordinate part . the only exception is two identical bosons with zero spin where the spin part of the wave function always is symmetric under exchange of the two bosons . therefore odd values of the relative orbital angular momentum between the bosons are strictly forbidden . this fact leads to important differences compared to non - identical particles @xcite . when the two - body interactions are spin - dependent , the separability of the three - body hamiltonian for the system shown in fig.[fig1 ] can disappear . we denote the spins of particles @xmath73 , @xmath24 , and @xmath25 by @xmath127 , @xmath128 , and @xmath129 , respectively . let us for simplicity assume that the particle - core interactions of the system in fig.[fig1 ] contain a central plus a spin - spin term , i.e. , @xmath130 , where @xmath131 and @xmath132 is the relative orbital angular momentum between the core and particle @xmath73 ( and similarly for the @xmath133 interaction ) . as shown in @xcite , this type of spin - spin operator is especially convenient , since it guarantees conservation of the quantum number @xmath134 in agreement with the intrinsic motion of particles in the core possibly being identical to the particles of the three - body system . the two - body hamiltonian @xmath85 is then diagonal in the basis @xmath135 where @xmath136 is the total two - body angular momentum after coupling of @xmath134 to the spin of the core @xmath129 . we denote the corresponding two - body eigenvalues by @xmath137 . the natural basis to describe the three - body system is then @xmath138 j_{ki } , ( \ell_{kj},s_j)j_j ; j \rangle \right\}$ ] , where @xmath139 is the total three - body angular momentum . the two - body hamiltonian @xmath85 with eigenvalues @xmath137 is diagonal in this three - body basis . this is in contrast to the remaining two - body hamiltonian @xmath86 . we can calculate the matrix elements of the three - body hamiltonian in this basis . with the expression @xmath140 we obtain by recoupling that @xmath141 where @xmath142 therefore the three - body hamiltonian is diagonal in blocks defined by the quantum numbers @xmath143 . as an example , we show the @xmath144 block corresponding to a core and two light particles all three with spin 1/2 . we further assume @xmath145 , and @xmath146 confining @xmath136 and @xmath147 to the values 0 and 1 . we note that eq.([eq20 ] ) depends on the two - body relative orbital angular momenta , @xmath148 and @xmath149 , only through the two - body energies . then the corresponding block is given by : @xmath150 this illustrates that even for the schematic case in fig.[fig1 ] , where particles @xmath73 and @xmath24 do not interact with each other , the three - body hamiltonian is not separable anymore . the three - body eigenvalues are then not given by the sum of the two - body eigenvalues . when particles @xmath73 and @xmath24 are identical and @xmath148=@xmath149=@xmath151 , then @xmath152= @xmath153=@xmath154 and @xmath155= @xmath156=@xmath157 , and the eigenvalues of the matrix ( [ eq21 ] ) are @xmath158+@xmath159 and @xmath160+@xmath161 . only the first of these eigenvalues corresponds to an antisymmetric eigenfunction under exchange of particles @xmath73 and @xmath24 . it is precisely twice the average energy of the two two - body energies . due to the pauli principle , the two light fermions must occupy both the two possible two - body states with angular momentum @xmath162 . then the relevant energies are not the individual energies of the two two - body states , but their average value . when none of the two - body interactions contains the spin - spin term we have that @xmath163 and the hamiltonian ( [ eq21 ] ) is diagonal , with identical diagonal terms given by @xmath164 + @xmath165 , which is the result obtained in the previous section for spin - zero systems . when only one of the two - body interactions has a non - zero spin - spin term ( e.g. , @xmath166 , eq.([eq22a ] ) is invalid and eq.([eq22b ] ) is valid ) the hamiltonian is diagonal in the basis @xmath167 j_{ki } , ( \ell_{kj},s_j)j_j ; j\rangle \right\ } $ ] with the energies @xmath168+@xmath165 and @xmath169+@xmath165 . there are then two possible three - body states , that again are given by the sum of the two - body energies . the results obtained in the previous section for the system in fig.[fig1 ] can be taken as a test for the numerical method used to compute three - body states . taking then the schematic system as starting point we analyze the main properties of the three - body states , and how they evolve in the energy plane when different ingredients are added to the calculations . we maintain the schematic model where @xmath170=0 and @xmath171=@xmath78 . we first assume that both @xmath172 and @xmath173 only act in relative @xmath4-waves producing resonances of complex energies @xmath174 and @xmath175 , respectively . the three - body computation then leads to the results shown in fig . [ fig2 ] for a specific set of parameters . in the small panel , together with the discretized complex rotated three - body continuum states ( plotted for three different scaling angles ) , there are two additional complex rotated branch cuts starting at each of the two - body resonance energies ( crosses in the figure ) . these branch cuts correspond to two of the particles in a two - body resonant state and the third particle in the continuum . to keep the figure cleaner these cuts have not been plotted . taking @xmath176 as an arbitrary length unit , the three - body system has a resonance at the energy @[email protected] , clearly distinguishable from the continuum background and independent of the rotation angle . the computed energy is precisely at @xmath26=@xmath174+@xmath175 , as found in section iii . this numerical result is then proving the efficiency of the numerical method . to illustrate that this indeed is a three - body resonance we also show the lowest radial wave function computed for a given rotation angle @xmath55 . both the real and imaginary parts vanish asymptotically following the corresponding hankel function @xmath49 as required for a pole of the @xmath0-matrix as seen from eq.([eq6 ] ) . when one of the two - body systems is bound and the other has a resonance the three - body state with energy equal to the sum of the two - body energies is simply a two - body resonance of one particle relative to the two - body bound state . this is illustrated in fig . [ fig3 ] , where one two - body resonance remains unchanged while the other two - body state is varied from resonance to bound state . when one two - body system is bound , and thus appearing on the negative real energy axis ( open circles ) , the three - body energies ( squares ) appear at the same distances to the left of the fixed two - body resonance ( black circle ) . if the attractive interaction binding the two - body system decreases , at some point the two - body state enters through the origin into the fourth quadrant of the energy plane and becomes a two - body resonance . in parallel , the three - body energy ( squares ) approaches the two - body resonance energy ( black circle ) and continues through the two - body resonance in southeastern direction all the time following the addition rule . the complex scaling method can be viewed as an analytic continuation into complex coordinates , such that resonances are `` pulled out '' of the continuum . unfortunately the method fails when the pole corresponds to a virtual state , due to the necessary large rotation angle . virtual states remain in the unphysical riemann sheet , and numerical investigation of their effects is more difficult . let now the interactions @xmath172 and @xmath173 correspond to a virtual @xmath179-state and a @xmath4-resonance of energies @xmath174 and @xmath175 , respectively ( @xmath170=0 , @xmath171=@xmath78 ) . then , a 1@xmath180 three - body @xmath0-matrix pole should be present at @xmath26=@xmath174+@xmath175 . to test this numerically we include an attractive @xmath179-wave interaction @xmath170 which , combined with the other interactions , is sufficiently strong to reveal the existence of a 1@xmath180 three - body resonance . we show in fig . [ fig4 ] how the complex energy of this resonance moves as the strength of @xmath170 is varied . the strongest attraction ( the point closest to the origin ) corresponds to a virtual state in the 2 - 3 subsystem at an energy of about @xmath181 in fig.[fig4 ] . a very small additional attraction would bind the three - body system , which would be borromean if the three - body system becomes bound before the 2 - 3 subsystem . when the strength of @xmath170 decreases the three - body resonance moves ( closed circles ) towards the energy @xmath175 of the @xmath4-resonance . if we could track the three - body resonance for even weaker @xmath170 we would for some non - zero value find that it precisely coincides with @xmath175 ( cross ) , reaching then the discontinuity cut of the riemann sheet . for weaker @xmath170 the three - body pole could move through this cut into any other riemann sheet , but it is tempting to suggest that these states could be interpreted as two - body virtual @xmath179-states of particle @xmath182 relative to the resonant state of particles @xmath183 and @xmath184 . thus such a three - body state is not a three - body resonance but a virtual state on the unphysical riemann sheet , although it is not possible to see numerically how this energy ( square ) is reached when @xmath170=0 . this can be better understood if the @xmath179-state is bound as illustrated in fig . when the @xmath179-wave attraction is reduced until the bound state energy is zero the three - body state moves in parallel towards the resonance energy . a continued decrease of the attraction turns the two - body bound state into a virtual @xmath179-state , and the two - body pole moves continuously from the physical to the unphysical riemann sheet . in parallel , the three - body pole moves continuously through the @xmath4-resonance onto another riemann sheet . the connection between two and three - body resonances are very intimate even when all three masses and interactions are non - zero . this is emphasized in fig.[fig4 ] where @xmath173 is varied for fixed @xmath170 . then the 1@xmath180 three - body resonances ( open circles ) follow precisely the motion of the two - body resonances ( open triangles ) . the double arrows connect each two - body resonance with its corresponding three - body state . a variation of the two - body complex energy produces a similar energy change in the three - body state . for completeness we also notice that virtual states in both subsystems lead to virtual three - body states with an energy following the sum rule and sitting in the negative energy axis of the unphysical riemann sheet . in a less schematic model also @xmath171 has to be finite . still often the three - body structure is mostly influenced by two two - body subsystems each dominated by one partial wave . thus we relax the condition @xmath171=@xmath78 but maintain the reduced masses @xmath185 and @xmath186 by adjusting @xmath187 and @xmath188 . the two - body properties then remain unchanged for the same interactions @xmath172 and @xmath173 . the finite masses destroy the separability into two independent subsystems , and the eigenvalues of the three - body hamiltonian are not any more given by the sum of the two - body eigenvalues . also , a given combination of partial waves may couple to several total angular momenta with different energies . when two partial @xmath4-waves both contribute , the total angular momentum and parity must be @xmath189=0@xmath190 , 1@xmath190 , 2@xmath190 , and combining a @xmath4-wave and a @xmath191-wave we get @xmath189=1@xmath180 , 2@xmath180 , 3@xmath180 . the corresponding three - fold degeneracy is broken for finite @xmath171 as illustrated in the upper part of fig . [ fig5 ] . in general , for fixed structure of the system ( fixed reduced masses , interactions , relative distances between particles ... ) , the three - body relative kinetic energy is smaller for finite mass of the core than for infinite mass , as seen by comparing @xmath192 in both cases . thus the three - body resonances often tend to move towards the origin . however , simultaneous conservation of angular momentum and center - of - mass might require a change of structure resulting in less total interaction energy than from the two resonances . thus , the three - body resonance can move in all directions ( fig . [ fig5 ] ) . as in the previous calculations , only the components corresponding to two - body resonances are included . the finite mass effects are relatively small in all cases except for two @xmath4-waves coupled to @xmath189=1@xmath190 . this can be understood by expressing the faddeev component related to the 1 - 3 subsystem in the jacobi coordinates of subsystem 1 - 2 . for @xmath106=1 a @xmath4-wave `` rotates '' fully into a @xmath4-wave . then the @xmath4-wave attraction is fully exploited and the resulting three - body state has the lowest possible energy . when @xmath106 and the two two - body angular momenta are equal there is a similar tendency , decreasing with @xmath106 , to maximize the total attraction . other components obtained by rotation are less important and uneven two - body angular momenta can not exploit the attraction as efficiently , producing a much smaller change with @xmath171 . in practice detailed quantum mechanical computations are needed for each case to determine the size of the effects . it is remarkable that the effects of the coupling and the center - of - mass motion can lower the energy sufficiently to produce a 1@xmath190 three - body bound state ( lower part of fig . [ fig5 ] ) . this is a borromean system obtained by only two two - body interactions . when one resonance is replaced by a virtual state we have not numerically been able to find any three - body resonance produced entirely by finite mass effects and without change of interactions . the origin of the resonances and bound states can be illustrated by different examples connecting the bare schematic model with well known realistic nuclear structures . we first study two nucleons outside the spin - zero @xmath193-particle core , then we present detailed realistic computations for the same @xmath194 systems , and finally we extend to the non - zero core - spin of the borromean nucleus @xmath2ne . let us consider a system made by an infinitely heavy core and two non - interacting spin 1/2 particles . the core is assumed to be charged ( with twice the proton charge ) and the particle - core reduced mass is 0.8@xmath9 , where @xmath9 is the nucleon mass . this system is very similar to an @xmath193-particle and two nucleons . the particle - core reduced mass is the same , and the center - of - mass effects are very small and not relevant for our present purpose @xcite . let us also consider a simple particle - core interaction given by a central short - range @xmath4-wave interaction producing a particle - core ( uncharged ) resonance at an energy of 0.74 mev with a width of 0.60 mev . when the light particle is assumed to have the proton charge , the same short - range interaction produces a core - particle resonance with energy and width of 1.60 mev and 1.26 mev , respectively . these values are consistent with the lowest @xmath4-resonances in @xmath195he and @xmath195li , respectively @xcite . in fig.[fig6 ] , we show the results for one heavy and two light particles resembling the @xmath1he , @xmath1li , and @xmath1be systems . the upper , middle and lower parts correspond to both uncharged , one neutral and one with the proton charge , and both with the proton charge , respectively . when the nucleon - nucleon interaction is equal to zero , the three - body system must have a resonance at an energy equal to the sum of the two two - body resonance energies . this is again confirmed by calculations using the complex rotated adiabatic expansion method . we obtain the three - body resonances indicated by the big open circle , big open square , and big open triangle in the upper , central , and lower parts of fig.[fig6 ] . they match precisely with the sum of the two - body energies each indicated by a cross . the known spectrum for the schematic three - body system , determined completely by the internal two - body states , can be used as starting point to trace the connection with the realistic " three - body system . this must be done by including the effects produced by i ) an additional interaction between the two light particles , ii ) quantum numbers conservation ( like the angular momentum ) , and iii ) center - of - mass effects originating from the finite mass of the core . effects i ) and ii ) are closely connected . once the interaction between the two light particles is included the three - body wave function , initially independent of the total angular momentum ( eq.([eq12 ] ) ) , is now depending on @xmath106 . this can be seen in the upper part of fig.[fig6 ] , where we include the neutron - neutron interaction multiplied by a global factor varying from 0 to 1 . then the 0@xmath190 and 2@xmath190 states evolve as shown by the solid and open circles , respectively . when the full neutron - neutron interaction is included ( last point on each curve ) , the system similar to @xmath1he has a bound borromean 0@xmath190 state ( with a binding energy close to @xmath196 mev ) and a very narrow 2@xmath190 three - body resonance with energy 0.34 mev . for the system similar to @xmath1li shown in the middle part of the figure one of the light particles has the charge of the proton . the neutron - proton interaction is again continuously switched on from zero to full strength . the 0@xmath190 and 2@xmath190 states ( @xmath22=1 ) , analogous to those in the upper part , move as given by the closed and open squares , respectively . the final 0@xmath190 state is still below the three - body threshold , with a binding energy of @xmath197 mev . the final @xmath198 resonance has an energy of 1.1 mev and a width of 0.1 mev . the shift in energy compared to the states in the upper part is produced by the coulomb repulsion between the core and the charged particle . now the allowed @xmath22=0 coupling gives rise to additional three - body states , e.g. the 1@xmath190 state shown by the stars . when the three - body 1@xmath190 state becomes bound the system is still borromean . however , for some threshold strength the two light particles become bound , and the system is not borromean anymore . when the full interaction is included the two light particles then form a deuteron nucleus , and the three - body 1@xmath190 state becomes the ground state of the system , very similar to @xmath1li , with a binding of about @xmath199 mev ( outside the scale of the figure ) . in the lower part of the figure we show results for two light particles each with the proton charge . as in the upper part only the @xmath200 states are allowed , and since all the three two - body subsystems feel the coulomb repulsion , the energies of the @xmath201 states ( closed triangles ) and the @xmath198 states ( open triangles ) are clearly larger than in the previous cases . in fact , when the full proton - proton interaction is included the @xmath201 ground state is unbound , with resonance energy and width of 0.5 mev and 0.2 mev , respectively . for the 2@xmath190 resonance the energy and width are 2.3 mev and 0.5 mev . .computed and experimental energies ( in mev ) for @xmath1he , @xmath1li , and @xmath1be . the energies within brackets are @xmath202 , that give the resonance energy and its width . the computations correspond to the schematic system described in the text ( second column ) , realistic calculations without inclusion of an effective three - body force ( third column ) , and the same calculation when a three - body force is used ( fourth column ) . the experimental values ( last column ) are taken from @xcite . when not specified , the error bars are smaller than the last digit . [ cols="^,^,^,^,^,^",options="header " , ] the effects produced by the finite mass of the core ( center - of - mass effect ) , do not change the previous energies significantly ( typically no more than 100 kev @xcite ) . the spectra obtained for the simplified " @xmath1he , @xmath1li , and @xmath1be nuclei ( only a simple @xmath4-wave particle - core interaction has been used ) , indicate where to find the true states . this is seen in table [ tab1 ] , where the second column gives the computed energies . a comparison with the experimental results ( last column ) reveals that in all the three nuclei the main features of the spectrum are well reproduced . therefore , starting from the schematic case where the three - body energy is given by the sum of the two - body energies and taking into account the main characteristics of the remaining two - body interaction , it is possible to estimate where the three - body states must be placed . as seen in the table , the computed energies are always lower than the experimental ones . this is because the nucleon - core interaction is without any spin - orbit term , and consequently the two possible @xmath4-resonances ( @xmath203 and @xmath204 ) appear at the same energy . this is overbinding the system , since the @xmath203-resonance is known to be a few mev higher @xcite , and a realistic calculation must include a repulsive term pushing up this resonance to the correct value . on top of this , other than @xmath4-waves should be included in the calculation . for @xmath179-waves the effect of the pauli principle must be accounted for , since the @xmath205-shell is fully occupied in the @xmath193-core . this effectively amounts to a highly repulsive @xmath179-wave potential at short distances , also pushing the computed energies towards higher values . furthermore , it is well known that three - body calculations using only realistic two - body interactions typically underbind the system . this can be cured by inclusion of an additional potential taking into account possible three - body effects to reproduce the experimental values . realistic detailed calculations using the ( complex scaled ) hyperspheric adiabatic expansion method concerning @xmath1he can be found in @xcite . for completeness we show in the next section the results obtained for @xmath1li and @xmath1be when similar calculations are performed . in @xcite the use of phase equivalent potentials is suggested as an efficient method to take into account the pauli principle in three - body calculations . this method , together with the hyperspheric adiabatic expansion method , is used to compute the ground state of @xmath206li and @xmath1he . thus , for @xmath1he the alpha - neutron @xmath179-wave interaction is able to bind the neutron into a pauli forbidden state . to exclude this state we use a phase equivalent potential with exactly the same phase shifts for all energies , but with one less bound state . the @xmath4-wave interaction contains a central and a spin - orbit term , such that the @xmath204 and @xmath203 resonances in @xmath195he are placed at the experimental energy values @xcite . the neutron - neutron interaction can also be found in @xcite . the calculation using only the two - body interactions underbinds the three - body nucleus ( third column in table [ tab1 ] ) , and an additional effective three - body force is needed to recover the experimental value ( fourth column ) . in @xcite the same method supplemented by complex scaling is used to compute the 2@xmath190 resonance in @xmath1he . in the present work the calculation is slightly simplified by using a repulsive @xmath179-wave alpha - neutron interaction to take the pauli principle into account . when the same phase equivalent potential as for the ground state is used , the computed 2@xmath190 state in @xmath1he is also given in the third and fourth columns of table [ tab1 ] . as before , the calculation without a three - body force is underbinding the system by about 0.5 mev , and an effective three - body force is again needed . in all the calculations the three - body force is assumed to be gaussian with a range of 3 fm and the strength is adjusted to fit the experimental value . for bound states it is obviously possible to find a strength fitting precisely the experimental value . for resonances , this single parameter is fitting simultaneously the energy and the width of the resonance . for @xmath1li and @xmath1be we perform exactly the same calculations but adding the corresponding coulomb interaction for the two cases . the additional proton - core repulsion in @xmath1li is pushing up the energies of the @xmath201 and 2@xmath190 states compared to the energies obtained for @xmath1he , and even more for @xmath1be , where all the three two - body subsystems feel the coulomb repulsion . the computed energies are again given in the third column of table [ tab1 ] when only the two - body interactions are used . as before , an effective three - body interaction is needed to fit the experimental energies ( fourth column ) . the @xmath201 state in @xmath1li is still below the two - nucleon separation energy threshold , although when the three - body interaction is suppressed the computed state appears to be unbound . again , when the strength of the three - body force is used to match the experimental resonance energies , the widths are also well reproduced , except for the @xmath198 state in @xmath1be , where the width is significantly larger than the experimental value . in addition to the @xmath201 and 2@xmath190 @xmath200 states , in @xmath1li it is also possible to have states with @xmath207 . these three - body states correspond to structures where the neutron and the proton form a deuteron ( @xmath191 ) nucleus , suggesting a description of the nucleus as a @xmath191+@xmath193 two - body system . therefore , we have performed a three - body calculation , identical to the ones described above , but where the nucleon - nucleon interaction has been substituted by the @xmath207 interaction used in @xcite . this potential reproduces the experimental deuteron binding energy , root mean square radius , electric quadrupole moment , and provides a @xmath191-wave content of 5.6% . the correspondig computed energies for the @xmath208 ( ground ) state are also given in table [ tab1 ] . the three - body thomas - ehrman shifts @xcite of these isobaric analog @xmath201 and @xmath198 states are then obtained from table [ tab1 ] , i.e. in mev ( 1.40,1.17,1.05 ) and ( 1.83,1.80,1.65 ) for the realistic cases without and with three - body potential , respectively . the latter results coincide by definition with the experimental values . the decreasing tendency in the computed differences are due to the smaller energy and the larger radii which in turn leads to smaller effects of the coulomb interaction . the underbinding is responsible for this discrepancy with measurements . when the three particles have non - zero spins and spin - dependent two - body interactions , as in any realistic nuclear potential , then the three - body hamiltonian is not even separable in the schematic model with an infinitely heavy core and two mutually non - interacting particles , see section [ sec4 ] . the hamiltonian matrix is then organized in blocks with the matrix elements given by eq.([eq20 ] ) . an example is for @xmath2ne , that is well described as a three - body system made by an @xmath3o core and two protons @xcite . the core of the system has negative parity and spin 1/2 . the two - body subsystem , @xmath209f ( @xmath3o+@xmath4 ) , has four low - lying resonances , two of them ( @xmath210 and @xmath211 ) arise from the coupling of a relative @xmath205-wave and the spin of the core , and other two ( 2@xmath180 and @xmath212 ) result from the coupling between a @xmath213-wave and the spin of the core . the experimental energies and widths ( @xmath214,@xmath215 ) of these two - body resonances are ( 0.535,[email protected] ) mev , ( [email protected],@xmath2170.040 ) mev , ( [email protected] , [email protected] ) mev , and ( [email protected] , @xmath2170.004 ) mev , respectively @xcite . in @xcite several proton - core interactions , all of them reproducing the experimental energies and widths , are given . these interactions contain spin - dependent operators , in particular the spin - splitting operator @xmath218 , where @xmath219 is the spin of the core and @xmath220 results from the coupling between the relative orbital angular momentum and the spin 1/2 of the proton . in the following calculations we shall use the @xmath179 and @xmath191-wave interactions corresponding to a gaussian two - body potential with the parameters given in table 1 of @xcite , but where the spin - orbit strength has been modified to push away the @xmath221-resonances in @xmath209f , but keeping the two @xmath213-resonances at the right energy . in this way our slightly simplified @xmath2ne is characterized exclusively by the @xmath210 , @xmath211 , @xmath222 , and @xmath212 resonances in @xmath209f , and we can perform for @xmath2ne the same kind of analysis as in the previous cases . let us begin by performing calculations for the schematic model without any proton - proton interaction and for an infinitely heavy core . the effect of the finite mass of the core is clearly smaller than for @xmath1he , @xmath1li , and @xmath1be , and the core can safely be assumed to be infinitely heavy . when both protons are in an @xmath179-wave relative to the core , only the @xmath210 and @xmath211 states in @xmath209f are obtained , and only the components with relative core - proton orbital angular momentum equal to zero are included in the calculation . in this case the total three - body angular momentum can only take the value @xmath146 @xcite , and the block to be diagonalized is given in eq.([eq21 ] ) . from the two possible eigenfunctions only one of them is antisymmetric under exchange of the two protons , and corresponds to the eigenvalue @xmath223 , that is twice the average energy of the two @xmath205-resonances . this gives rise to a @xmath224 energy of 1.32 mev , that is recovered when the calculation using the complex scaled hyperspheric adiabatic expansion method is performed . when the proton - proton interaction @xmath225 is progressively introduced , the contribution from the @xmath191-waves starts to be relevant , and the components with relative core - proton orbital angular momentum equal to 2 are also included . the energy of the @xmath224 state then changes as shown by the solid circles in fig.[fig7 ] . around 90% of the proton - proton interaction is already binding the three - body system , becoming then borromean . the full proton - proton interaction gives a binding of about 0.40 mev and a @xmath191-wave content of 37% . this binding is 0.3 mev smaller than the one given in table 8 of @xcite . this is due to the fact that in this calculation , in order to clean the structure of @xmath2ne , we have artificially pushed up the @xmath221-resonances by using a large spin - orbit splitting . this is also reducing the @xmath191-wave content compared to the result in @xcite . in ref.@xcite we show how the 3/2@xmath180 state in @xmath2ne is dominated by the @xmath226-interferences between the components . to investigate this resonances both ( 0 and 2 ) core - proton relative orbital angular momenta are then needed and therefore included in the calculation . for the schematic @xmath2ne model ( @xmath227 ) , one of the protons can be in one of the @xmath179-resonances in @xmath209f ( 0@xmath180 or 1@xmath180 ) , and the other can be in one of the @xmath213-resonances ( @xmath222 or @xmath212 ) . assuming @xmath228 , the block in eq.([eq20 ] ) is already diagonal with only one non - zero energy equal to @xmath229 . this energy of 1.65 mev is recovered in the numerical calculation , which also reveals that the only non - vanishing components correspond to one proton in the @xmath179-wave @xmath211 resonance and the other in the @xmath191-wave @xmath222 resonance . the evolution of the energy when the proton - proton interaction is introduced is shown by the white circles in fig.[fig7 ] , reaching a final value of 0.94 mev . this value is also around 0.3 mev less bound than the one given in table 8 of @xcite . finally , in the figure we also show the evolution of the lowest 5/2@xmath180 resonance in @xmath2ne ( solid squares ) when @xmath225 is introduced . this state is also dominated by @xmath226-interference components . using the schematic case in which @xmath230 as starting point , it is possible to trace the behaviour of the three - body resonances ( figs . [ fig2 ] , [ fig3 ] , and [ fig4 ] ) when the center - of - mass effects and the @xmath170 interaction are included . taking into account the main features of the @xmath170 interaction , it is then possible to make crude estimates concerning different realistic systems . for nuclear systems we take the mass unit @xmath9 equal to the nucleon mass and a length unit equal to a typical range for the nuclear interaction ( @xmath231 fm ) . with this choice , the lowest @xmath4-resonance in the neutron - alpha interaction has dimensionless energy @xmath232 . therefore , in @xmath1he ( @xmath193+@xmath36+@xmath36 ) , before including the neutron - neutron interaction , the three - body system must have a resonance with dimensionless energy of about 0.30 . the neutron - neutron interaction has a low lying virtual state at @xmath233 , which is enough to bind the three - body state as evidenced by the bound state in @xmath1he at @xmath234 . for @xmath206li ( @xmath235li+@xmath36+@xmath36 ) the @xmath179-waves are pronounced and the lowest neutron-@xmath235li virtual @xmath179-state would after spin averaging appear at about 0.07 which , combined with the neutron - neutron interaction , is sufficient to bind the three - body system . substituting @xmath235li by a @xmath236-particle we arrive at an unstable system . with the @xmath236-nucleon @xmath179-wave scattering length of 2 fm obtained from hypertriton computations @xcite the virtual @xmath179-state would be at around 1.7 , too large to bind the @xmath236-neutron - neutron system . we can continue to a more speculative system obtained by combining the recently discovered pentaquark resonance with a third particle like a meson . the pentaquark , @xmath237 , has a mass of 1540 mev and decays mainly into a kaon @xmath238 and a neutron @xcite . the energy above threshold is then around 107 mev , that corresponds to a dimensionless energy of about 20 . thus , the neutron - neutron interaction , that was able to bind @xmath193+@xmath36+@xmath36 and @xmath235li+@xmath36+@xmath36 respectively with 0.15 and 0.07 , but not @xmath236+@xmath36+@xmath36 with 1.7 , can not bind the @xmath239 system . thus a nucleon - kaon - meson cluster system would also exploit the intrinsic quark degrees of freedom and not resemble any three - body structure although additional binding would be picked up . in any case the narrow width of @xmath237 strongly suggests that the resonance can not have nucleon - kaon ( two - particle ) structure , but rather is a genuine five - quark system or perhaps a more exotic structure . the semi - quantitative knowledge obtained with the schematic model can also be used for molecules due to the use of dimensionless parameters . for molecules we choose as typical scale units the mass of the @xmath240he atom and the range @xmath241 . the @xmath242he+@xmath240he dimer has then @xmath243 @xcite . we can then expect that , except for center - of - mass effects , the trimers @xmath242he+@xmath240he+@xmath240he and @xmath240he+@xmath242he+@xmath242he should both have an unbound state with dimensionless energy of about 1.50 when the interaction between the two identical particles is neglected . however , including these interactions of dimensionless energies 0.02 for @xmath240he+@xmath240he and 5.2 for @xmath242he+@xmath242he , we arrive at trimers @xmath242he+@xmath240he+@xmath240he and @xmath240he+@xmath242he+@xmath242he respectively bound and unbound as observed . the two - body interactions are assumed to determine completely the three - body structure including the continuum properties . close to the threshold of binding , slightly above or below , the structure of the three - body system mostly depends on the two - body low - energy scattering properties . these properties are periodically repeated with increasing strengths of these attractive potentials corresponding to one or many bound states in each of the investigated channels . therefore it is sufficient to study unbound or weakly bound two- and three - body systems . the low - energy scattering properties are determined by the phase shifts and reflected in the poles of the @xmath0-matrix , i.e. by the resonances and virtual states . thus , changing the two - body interactions simultaneously change the energies of both two- and three - body resonances . the relative changes of these energy observables are expected to be intuitively easier to understand than by using the connection via the two - body interaction strengths . the strategy is then to vary the two - body resonances and virtual states and study the changes of the corresponding quantities in the three - body system . we employ the efficient and well tested method of complex scaled hyperspherical adiabatic expansion . this method is first briefly sketched to define the notation . then we define a schematic model with an infinitely heavy core and two mutually non - interacting particles . we prove mathematically for spin - zero particles that our formulation provides a pole in the three - body @xmath0-matrix if and only if the complex energy is equal to the sum of two complex energies each corresponding to poles of different two - body @xmath0-matrices for the two particle - core subsystems . the generalization to non - zero spins are formulated and shown to involve diagonalization of simple block - diagonal hamiltonians . the general properties are demonstrated numerically in schematic examples involving both resonances and virtual states . we then investigate the sizes and trends resulting from a finite core mass , non - zero interaction between the light particles and the coupling of different orbital angular momenta . we use dimensionless units to allow easy application on physics systems of different scales . each of these effects can be substantial but the structure of the states can be uniquely traced back to the origin in the structure of the schematic model . borromean systems can arise with only two interactions from center - of - mass effects and a favorable coupling of two angular momenta . for further numerical illustration we use essentially realistic nuclear three - body systems , @xmath1he , @xmath1li , and @xmath1be , consisting of two nucleons and an @xmath193-particle to trace back their measured spectra to our bare schematic model . the effects of isospin symmetry and the mixing short - range and coulomb interactions are then seen . for completeness we also present unpublished and fully realistic calculations of @xmath1li and @xmath1be . the origin of the structure is still apparent , but to get accurate energies and wave functions we must include effects of spin - orbit couplings and the pauli principle . three - body thomas - ehrman shifts can then be studied for these isobaric analog states . finally the effects of non - zero core spin is investigated for the borromean nucleus @xmath2ne consisting of @xmath3o and two protons . in the last section we explain , and illustrate by examples , how to make qualitative estimates of the three - body energies and their structure from the two - body properties of the subsystems . first we test by known examples of borromean halo nuclei , hypertriton , molecular helium clusters and the recently highlighted more speculative pentaquark . in conclusion , we have demonstrated the strong correlation between two- and three - body resonances . the three - body energy and structure can be traced back to the properties of an infinitely heavy core and two non - interacting light particles . substantial changes are often needed to arrive at accurate and realistic properties but the generic origin is apparent and revealing both structures and energies . the three - body @xmath0-matrix for the schematic model in eq.([eq17 ] ) is obtained through the definition in eq.([eq16 ] ) by inserting eq.([eq15 ] ) into eq.([eq12 ] ) . then the following integrals are needed :

the investigation of the neuronal environment allows us to better understand the activity of a cerebral region as a whole . the recent experimental evidences of the presence of transporters for glutamate and gaba in both neuronal and astrocyte compartments raise the question of the functional importance of the astrocytes in the regulation of the neuronal activity . we propose a new computational model at the mesoscopic scale embedding the recent knowledge on the physiology of neuron and astrocyte coupled activities . the neural compartment is a neural mass model with double excitatory feedback , and the glial compartment focus on the dynamics of glutamate and gaba concentrations . using the proposed model , we first study the impact of a deficiency in the reuptake of gaba by astrocytes , which implies an increase in gaba concentration in the extracellular space . a decrease in the frequency of neural activity is observed and explained from the dynamics analysis . second , we investigate the neuronal response to a deficiency in the reuptake of glutamate by the astrocytes . in this case , we identify three behaviors : the neural activity may either be reduced , or enhanced or , alternatively , may experience a transient of high activity before stabilizing around a new activity regime with a frequency close to the nominal one . after translating theoretically the neuronal excitability modulation using the bifurcation structure of the neural mass model , we state the conditions on the glial feedback parameters corresponding to each behavior . + * keywords : * model in neuroscience , qualitative analysis of dynamical systems , bifurcations , neuro - glial interactions , gabaergic and glutamatergic neurotransmissions , excitability modulation , neuronal hyperexcitability + * ams classification : * 34c15 , 34c23 , 34c28 , 34c46 , 34c60 , 34h20 , 92c20 , 92b25 .

1510.01313

for several years , to understand the mechanisms of the cerebral metabolism has become an important issue in neuroscience . the investigation of the neural environment allows us to better understand the activity of a cerebral region as a whole . for instance , the neural activity is composed of an interplay between excitation and inhibition where the cerebral blood flow ( cbf ) dynamics is an essential element as it reflects nutriments supplies such as oxygen and glucose . synaptic transmission @xcite and neural activity @xcite are regulated by neurotransmitters , ions and molecules . broadly speaking , at the microscopic scale , the presynaptic neuron releases neurotransmitters in the synaptic cleft which may bind to postsynaptic neuron receptors . these receptors , when `` activated '' by neurotransmitters , trigger ions exchanges between the extracellular space and the postsynaptic neuron that can induce the activation of the postsynaptic neuron . parallely , neurotransmitter in the extracellular space may bind to astrocytic receptors and transporters inducing the activation of the glial calcic cycle allowing to release glutamate in the extra - synaptic space . the demonstration of the presence of transporters for glutamate and gaba in both neural and astrocyte compartments raises the question of the functional importance of the astrocytes in the regulation of the neural activity @xcite . it has been shown that glial cells and particularly astrocytes have a great impact on both the metabolic regulation @xcite and the cbf dynamics @xcite . indeed , astrocytes modulate the dynamics of neurotransmitter concentrations , and thus the neuronal excitability , the synaptic transmission and the neural activity . consequently , astrocytes dysfunctions are involved in several brain pathologies @xcite . studying the interactions between the neurons and astrocytes cells has therefore become an essential problem in neurophysiology and biophysics . in this context , computational models provide a key tool for interpreting the observed electrophysiological data and for revealing the different ( patho-)physiological mechanisms which may underlie the observational data . several models including the metabolic regulation mechanisms have been proposed in the literature . some authors have built models of tripartite synapse @xcite considering an astrocyte coupled with a presynaptic neuron and a postsynaptic one and the dynamics of neurotransmitters , ions and molecules . other authors considered a single neuron coupled with an astrocyte @xcite and sometimes with a hemodynamic compartment @xcite and the dynamics of neurotransmitters , ions and molecules between and in these elements . recently , a new mesoscopic computational model @xcite focusing on the astrocyte dynamics has been proposed . this model links the neural activity measured by local field potentials ( lfp ) to the cbf dynamics measured by laser doppler ( ld ) recordings through glial activity . the model incorporates the astrocyte cells via their role in neurotransmitters ( glutamate and gaba ) recycling , with physiologically - relevant relationships between these variables . in this article , we propose a model extending the one presented in @xcite for studying the neuro - glial interactions and more particularly the impact that the astrocyte activity may have upon the neuronal one . we use the same organization embedding a neural compartment and a glial one . to reproduce the mesoscopic neural activity , we use the neural mass model ( nmm ) with double excitatory feedback presented in @xcite which generalizes the jansen - rit model used in @xcite . we keep the glutamate and gaba dynamics presented in @xcite for the glial compartment . in this model , the neural activity acts on glutamate and gaba dynamics through the pyramidal and interneuronal firing rates respectively . the essential extension in our model consists in embedding the influence of the glial dynamics upon neural activity through the glutamate and gaba extracellular concentrations . indeed , physiologically , a pyramidal cell ( resp . an interneuron ) releases glutamate ( resp . gaba ) in the synaptic cleft from where it binds to receptors on the postsynaptic neuron and the astrocyte . reuptake processes ( referred to as `` reuptakes '' in this article ) of the neurotransmitters by the local astrocyte and presynaptic neuron regulates their concentration in the extracellular space ( figure [ schema_ng_physio ] ) . in the presynaptic neuron , the reuptake completes the stock whereas the reuptake by the astrocyte triggers a cascade of reactions linked with the modulation of synaptic transmission ( differentially according to the type of neurotransmitters ) and the hemodynamics . hence , we introduce a feedback coupling from the glial compartment upon the neural one in our model . this feedback , referred to as the `` glial feedback '' in this article , allows us to study how different astrocyte deficiencies impact the local neuro - glial activities . if the mechanism of glutamate or gaba reuptake by the astrocytes is deficient , the neurotransmitter accumulates in the synaptic cleft , which leads to an increase in its concentration in the extracellular space . when this concentration reaches a threshold the synaptic transmission to postsynaptic neuron becomes abnormal and the postsynaptic neuron excitability threshold changes . the proposed model provides a unified framework in which knowledge of the physiology of neuron and astrocyte activities , as well as their couplings , can be incorporated , in order ( i ) to simulate the output signals for chosen parameter values , ( ii ) to identify the various qualitative responses of the whole system to physiological disorders , ( iii ) to study theoretically the conditions over the parameters corresponding to each type of response . the neural mass approach offers therefore an optimal compromise between the compactness of the dynamics , the richness of the physio - pathological mechanisms that can be reproduced , and the interpretability of the parameters from the biophysical viewpoint . the range of parameter values can be inspired from physiology - based studies in actual animal models . once the parameter range is defined , one can test the influence of varying a reduced set of parameters ( for example , ratio between excitation and inhibition or glutamate and gaba reuptake ) on the output signals . the paper is organized as follows . in section 2 , we discuss the main pathways of the neuro - glial interactions . we recall the dynamical features that explain the qualitative and quantitative properties of the time series generated by each compartment . in particular , we describe the bifurcation structure of the neural compartment underlying the generation of noise induced spiking ( nis ) outputs on which we focus in this article @xcite . we introduce the bilaterally coupled model and illustrate the main difference with the feedforward model by mimicking the injection of a gaba bolus in the extracellular space . in section 3 , we study theoretically the effect of a deficiency of the gaba reuptake by the astrocyte upon the neural system behavior , we illustrate the result by numerical simulations and link the observed outputs with known biological results . in section 4 , we identify and illustrate numerically the three possible responses of the neural compartment to a deficiency of the glutamate reuptake by the astrocyte : reduced activity , transient and permanent hyperexcitability . we explain this spectrum of responses using the analysis of the dynamical structure of the model and we derive explicit conditions on the parameters involved in the glial feedback corresponding to each type of behaviors . finally we interpret the conditions on the parameters in terms of physio - pathological features and discuss possible applications of this study for investigating experimentally the precise role of astrocyte deficiencies in neuronal hyperexcitability . in this section , we briefly recall the structure of the nmm generalizing the jansen - rit model and studied in @xcite , the way to analyze its properties and the predominant time series pattern that it generates . then we describe the model introduced in @xcite to reproduce the glial dynamics . then we explain our choice of bilateral coupling between these two compartments . finally , we illustrate its main dynamical differences with the feedforward coupling system ( _ i.e. _ unilateral coupling from the neural compartment upon the glial one ) . the nmm represents the dynamical interactions between two neural populations at a mesoscopic scale : a main population of pyramidal cells ( p ) and a population of inhibitory interneurons ( i ) . it also involves the interactions of p with a general population p representing neighboring pyramidal cells and interacting with p through synaptic connections . there are three feedback loops on population p activity : an inhibitory feedback through the interneuron population i , a direct excitatory feedback of p onto itself ( referred to as `` direct feedback '' ) and an indirect excitatory feedback ( referred to as `` indirect feedback '' ) involving the population p ( figure [ nmm](a ) ) . @xmath0 ) : second order process converting action potentials into excitatory ( resp . inhibitory ) post - synaptic potential . box @xmath1 : process converting average membrane potential into average action potential density discharge by neurons of populations p , p and i respectively . @xmath2 for @xmath3 : coupling gain parameters depending on the maximal number @xmath4 of synaptic connections between two populations . @xmath5 : direct feedback coupling gain . @xmath6 : excitatory input . @xmath7 , @xmath8 , @xmath9 : state variables . @xmath10 , @xmath11 , @xmath12 : intermediary variables.,scaledwidth=65.0% ] the conversion process of average pulse density into excitatory and inhibitory postsynaptic potential respectively are based on the following functions introduced by van rotterdam _ et al . _ @xcite : @xmath13 these functions are the basic solutions of the differential operators @xmath14 and @xmath15 respectively : [ transfer_dyn ] @xmath16 in this framework , parameter @xmath17 ( resp . @xmath18 ) tunes the amplitude of excitatory ( resp . inhibitory ) postsynaptic potentials and @xmath19 ( resp . @xmath20 ) represents the time constant of excitatory ( resp . inhibitory ) postsynaptic potentials representative of the kinetics of synaptic connections and delays introduced by circuitry of the dendritic tree @xcite . following freeman s work @xcite , the sigmoidal functions converting the average membrane potential into an average pulse density have the following form : @xmath21 where @xmath22 represents the maximum discharge rate , @xmath23 the excitability threshold and @xmath24 the sigmoid slope at the inflection point . finally the nmm receives an excitatory input @xmath6 standing for the action on population p of neural populations in other areas through long - range synaptic connections . classically one consider @xmath6 a gaussian variable to represent a non - specific input and generate the model outputs . now we can write the dynamics for the intermediary variables @xmath10 , @xmath11 and @xmath12 which represent the outputs of the population p , the population p and the population i respectively ( figure [ nmm](b ) ) : [ x_init ] @xmath25 parameters @xmath2 , @xmath26 , represent the average number of synapses between two populations . following @xcite , each @xmath2 is proportional to the maximal number @xmath4 of synapses between two populations . the excitation of p by its own output , resulting from the intra - population synaptic connections , is weighted by the coupling gain @xmath5 . for sake of comparison , we use a variable change to obtain the same state variables as in the jansen - rit model @xcite : the excitatory output ( @xmath27 ) and the excitatory ( @xmath28 ) and inhibitory ( @xmath29 ) inputs of the main population p. the output @xmath7 acts on the secondary pyramidal cell population p and on the interneuron population i. to analyze the model , we write the dynamics of the state variables @xmath7 , @xmath8 and @xmath9 as a system of first order differential equations : [ y_init ] @xmath30 in this article , we consider the local field potential ( lfp ) as the main model output . following @xcite , it is defined by @xmath31 . it is important to note that , generally , studies of neural mass models , such as jansen - rit model , only considered the case with the same constant excitability thresholds for all populations , _ @xmath32 the behavior of nmms can be deduced from the bifurcation diagram according to the value @xmath33 considered as a parameter , as it has been performed in @xcite on the jansen - rit model . in @xcite , we have classified the types of time series patterns generated by model and the associated bifurcation structures according to the strengths of the different excitative feedbacks applied to population @xmath34 . let us recall the bifurcation diagram underlying the predominant type of generated time series , which we will consider in this article . model has the following useful particularities that have been highlighted in @xcite . first , for a fixed value of parameter @xmath35 , the @xmath7 value of a singular point suffices to have explicit expressions of all the other components . second , for a given @xmath7 value , there exists a unique value of @xmath35 such that @xmath7 corresponds to a singular point . in other terms , the set of singular points obtained for the different values of @xmath35 is a graph over @xmath7 . hence , we can visualize the shape of the singular point locus in the plane @xmath36 : in the case presented here ( see figure [ nis_db ] ) , this curve of singular points is s - shaped . in the following description , for a given bifurcation `` bif '' according to parameter @xmath35 , we note @xmath37 the bifurcation value and , if the bifurcation involves a singular point , we note @xmath38 the corresponding @xmath7 value . ( left ) and associated lfp and @xmath7 time series ( right ) . blue curves : stable singular points . cyan ( resp . green ) curves : singular points with one ( resp . two ) eigenvalues with positive real parts . black curves : @xmath7 extrema along stable limit cycles . black points ( sn@xmath39 and sn@xmath40 ) : saddle - node bifurcations . red point ( h@xmath39 ) : supercritical hopf bifurcation . dashed orange line : saddle - node on invariant circle ( snic ) bifurcation . horizontal gray bar : confidence interval @xmath41 $ ] of the gaussian variable @xmath6 used to generate the time series.,scaledwidth=95.0% ] two saddle - node bifurcations sn@xmath39 and sn@xmath40 split the curve of singular points into three branches . we name `` lower branch '' , `` middle branch '' and `` upper branch '' the sets of singular points satisfying @xmath42 , @xmath43 and @xmath44 respectively . singular points on the lower branch are stable ( blue ) and those on the middle branch are unstable ( cyan ) . singular points on the upper branch are unstable ( green ) for @xmath45 and stable ( blue ) otherwise . at @xmath46 the system undergoes a supercritical hopf bifurcation h@xmath39 giving birth to a stable limit cycle for @xmath45 that persists until @xmath47 where it disappears by a saddle - node on invariant circle ( snic ) bifurcation ( dashed orange line ) . the existence of the snic bifurcation is essential because it implies the appearance of a large amplitude stable limit cycle with large period . thereby , according to the value of @xmath35 , the system alternates between oscillatory phases ( for @xmath48 ) and quiescent phases ( for @xmath49 ) . in other terms , the value @xmath50 plays the role of an activation threshold for the neural compartment , which is a key point of the subsequent analysis . note that the oscillation frequency in the generated oscillatory pattern is driven by the value of @xmath35 : as @xmath35 tends to @xmath50 from above , the limit cycle period tends to infinity . hence , the closest @xmath35 is to @xmath50 , the lowest is the frequency . consequently , when considering a gaussian input for the model , the occurrence of spikes and their frequency depend on the features of the normal distribution generating @xmath6 , which led us to refer to the corresponding pattern as noise - induced spiking ( nis ) in @xcite . such lfp activity , _ i.e. _ sparse large amplitude spikes , corresponds to episodic synchronization of the neuron activities among the populations . physiologically , this pattern of activity arise , as for interictal spiking activity , and is symptomatic of a strong excitability of the neuronal system that can turn into hyperexcitability during pathological crisis . for fixed parameters , the activity is stable in the sense that the oscillation frequency does not change much along time . in the following , we study the variations of the activity when the neural dynamics are altered by the surrounding activity , _ i.e. _ the glial feedback . for reproducing the glial activity , we use the model introduced in @xcite . it focuses on the dynamics of glutamate and gaba concentrations , which are the main neurotransmitters of the central nervous system . in @xcite , the neuro - glial coupling is feedforward : the glial dynamics is driven by the neural activity , generated by the jansen - rit model , but does not impact the neural compartment . the model considers the dynamics of glutamate and gaba concentrations , locally to the main population p of pyramidal cells , at different stages of the recycling mechanism . the local nature of this interaction implies that the firing rate of the secondary population p of pyramidal cells does not impact the glial dynamics associated with the neighboring astrocytes of the main population p. the mechanism is as follows ( figure [ ng_sans_retroaction ] ) : excited pyramidal cells ( resp . interneurons ) release glutamate ( resp . gaba ) in the extracellular space ( synaptic cleft ) . astrocytes and pre - synaptic neurons reuptake the neurotransmitters . astrocytes recycle or consume the neurotransmitters while the presynaptic neurons capture them to complete their stock . : excitatory input on population p. @xmath51_{e}}$ ] and @xmath52_{e}}$ ] : glutamate and gaba extracellular concentration . @xmath51_{a}}$ ] and @xmath52_{a}}$ ] : glutamate and gaba glial concentrations . red arrows : p@xmath53p , p@xmath53i and p@xmath53p couplings . orange arrow : p@xmath53p coupling . green arrow : i@xmath53p coupling . cyan arrows : glutamate and gaba release by populations p and i into extracellular space . purple arrows : glial and neural reuptakes of neurotransmitters.,scaledwidth=60.0% ] following @xcite the glial compartment is built on the firing rate ( @xmath54 ) of the pyramidal cell population and the firing rate ( @xmath55 ) of the interneuron population . the state variables are * @xmath51_{ne}}$ ] and @xmath52_{ie}}$ ] : the fluxes of glutamate and gaba from neurons to extracellular space , * @xmath51_{e}}$ ] and @xmath52_{e}}$ ] : the neurotransmitter concentrations in the extracellular space , * @xmath51_{a}}$ ] and @xmath52_{a}}$ ] : the quantity of neurotransmitters recycled and consumed by the astrocytes . naturally , the dynamics governing @xmath51_{ne}}$ ] and @xmath52_{ie}}$ ] are driven by second - order differential operators similar to the synaptic transfer dynamics introduced in @xcite : @xmath56 as for the synaptic transfer functions , parameter @xmath57 ( resp . @xmath58 ) tunes the peak amplitude of glutamate ( resp . gaba ) concentrations and parameters @xmath59 and @xmath60 ( resp . @xmath61 and @xmath62 ) tune the rise and decay times of glutamate ( resp . gaba ) release transfer function . these dynamics are well - suited for reproducing the qualitative and quantitative properties of rise and decay in neurotransmitter concentrations . the reuptakes of glutamate from the extracellular space by astrocyte ( @xmath51_{ea}}$ ] ) and neurons ( @xmath51_{en}}$ ] ) are triggered when extracellular concentration of glutamate reaches a threshold . moreover the efficiencies of these processes saturate for high concentrations values , which leads to model these dynamics using sigmoidal functions . gaba reuptakes ( @xmath52_{ea}}$ ] and @xmath52_{ei}}$ ] ) are modeled with michaelis - menten dynamics following the experimental literature @xcite . the dynamics of the extracellular concentrations ( @xmath51_{e}}$ ] and @xmath52_{e}}$ ] ) are derived from the input and output fluxes described above . the astrocyte concentration dynamics ( @xmath51_{a}}$ ] and @xmath52_{a}}$ ] ) result from the glial reuptake ones and a linear consumption term . in sigmoidal functions for @xmath51_{e}}$ ] and @xmath52_{e}}$ ] , parameters @xmath63 and @xmath64 are the maximal velocities for glutamate reuptakes by the neurons and the astrocytes respectively , @xmath65 represents the activation threshold and @xmath66 the sigmoidal slope at the inflection point . parameters @xmath67 and @xmath68 ( resp . @xmath69 and @xmath70 ) are respectively the maximal velocity and concentration for glial ( resp . neural ) gaba transporter . finally , @xmath71 and @xmath72 are the glutamate and gaba degradation rates in astrocytes . we refer the reader to @xcite for a detailed explanation of the dynamics . hence , the feedforward model obtained by coupling the nmm defined by and the glial dynamics introduced in @xcite reads [ sys_ng_sr ] @xmath73_{ne } } ' & = { \rm d[glu]_{ne } } \\ { \rm d[glu]_{ne } } ' & = w\,w_1\,{\rm sigm}(y_1-y_2,v_{\rm p})-(w_1+w_2)\,{\rm d[glu]_{ne}}-w_1\,w_2\,{\rm [ glu]_{ne } } \label{glune } \\ { \rm [ glu]_{e } } ' & = { \rm [ glu]_{ne}}-\frac{v_{\rm glu}^{\rm ea}}{1+e^{r_g\,s_g - r_g\,{\rm [ glu]_e}}}-\frac{v_{\rm glu}^{\rm en}}{1+e^{r_g\,s_g - r_g\,{\rm [ glu]_e } } } \\ { \rm [ glu]_a } ' & = \frac{v_{\rm glu}^{\rm ea}}{1+e^{r_g\,s_g - r_g\,{\rm [ glu]_e}}}-v_{\rm cglu}\,{\rm [ glu]_a } \\ { \rm [ gaba]_{ie } } ' & = { \rm d[gaba]_{ie } } \\ { \rm d[gaba]_{ie } } ' & = z\,z_1\,{\rm sigm}(c_3\,y_0,v_{\rm i})-(z_1+z_2)\,{\rm d[gaba]_{ie}}-z_1\,z_2\,{\rm [ gaba]_{ie } } \label{gabaie } \\ { \rm [ gaba]_e } ' & = { \rm [ gaba]_{ie}}-\frac{v_{\rm gba}^{\rm ea}}{k_{\rm gba}^{\rm ea}+{\rm [ gaba]_e}}\,{\rm [ gaba]_e}-\frac{v_{\rm gba}^{\rm en}}{k_{\rm gba}^{\rm en}+{\rm [ gaba]_e}}\,{\rm [ gaba]_e}\\ { \rm [ gaba]_a } ' & = \frac{v_{\rm gba}^{\rm ea}}{k_{\rm gba}^{\rm ea}+{\rm [ gaba]_e}}\,{\rm [ gaba]_e}-v_{\rm cgba}\,{\rm [ gaba]_a}\end{aligned}\ ] ] table [ par_table ] specifies the parameter values used for the simulations in the following sections . the values of parameters associated with the nmm have been chosen to reproduce nis behavior using the analysis in @xcite . .values of the neuro - glial model parameters . [ cols= " < , < , < , < " , ] we note that the fixation mechanisms of glutamate on pyramidal cells and interneurons are the same since the neurotransmitter transporters are independent on the type of neuron . thus , only parameters @xmath74 and @xmath75 representing the maximal coupling gains of the glutamate - related component of the glial feedback discriminate between the coupling functions @xmath76 and @xmath77 , since the synaptic sensitivities may not be the same in pyramidal cells and interneurons . at the beginning of this subsection , we evoked that the glial feedback acts on the excitability thresholds of neurons . more specifically , if there is an excess of neurotransmitter in a synapse from a neuron @xmath78 of population @xmath79 to a neuron @xmath80 of population @xmath81 , the extracellular concentration of neurotransmitter acts on the postsynaptic neuron @xmath80 by changing its excitability threshold . in system the excitability threshold of neurons , that is a parameter at the individual scale , does not appear explicitly . however , when the excitability of the population @xmath81 neurons changes at the individual scale , the number of neurons activated in this population by a given input changes as well and thus the output of this population is also modified . consequently , we choose to change parameter @xmath82 in the equation corresponding to the output of population @xmath81 since this parameter represents a modulation of the threshold of the sigmoidal function @xmath1 . let us now describe how we build the feedbacks on the dynamics of the neural compartment using the sigmoidal functions of the neurotransmitter concentrations introduced in . we need to consider separately each type of synapse in the nmm , and the variables @xmath10 , @xmath11 an @xmath12 for the feedbacks building . the nmm embeds five types of synaptic connections between populations : * @xmath83 from p to p , * @xmath84 from p to p , * @xmath85 from p to i , * @xmath86 from i to p , * @xmath87 from p to itself in the following we detail the modulation of neural intermediary variables for each kind of synapse separately , then we gather these changes to specify the coupling terms reproducing the glial feedback . : excitatory input on population p. @xmath51_{e}}$ ] and @xmath52_{e}}$ ] : glutamate and gaba extracellular concentrations . @xmath51_{a}}$ ] and @xmath52_{a}}$ ] : glutamate and gaba glial concentrations . red arrows : p@xmath53p , p@xmath53i and p@xmath53p couplings . orange arrow : p@xmath53p coupling . green arrow : i@xmath53p coupling . cyan arrows : glutamate and gaba release by populations p and i into extracellular space . purple arrows : glial and neural reuptakes of neurotransmitters . red dashed arrows : glutamate feedbacks on populations p and i. brown dashed arrow : gaba feedback on population p.,scaledwidth=70.0% ] in the framework of the local neuro - glial mass model , the glial feedback does not impact the synaptic connections of type @xmath83 or @xmath84 . as a matter of fact , the glial compartment only takes into account neurotransmitters released locally by neurons of populations p and i , whereas population p is non local to population p. hence , extracellular concentrations of neurotransmitters in the vicinity of p have no impact on the neuronal activity of p and the concentrations in the neighborhood of p and i do not influence postsynaptic neurons of population p. in the discussion , we evoke the fact that network models based on the local model presented in this article may naturally take into account such modulation of mid - range synaptic connections . in the current study focusing on the local model , we consider the case of constant @xmath88 . a synaptic connection of type @xmath85 concerns the variable @xmath12 . in case of glutamate excess in the extracellular space , the postsynaptic neuron is more excitable . consequently , more neurons are activated in the population i. we model this mechanism by introducing a dependency of population @xmath89 excitability threshold @xmath90 on the extracellular glutamate concentration and set in equation : @xmath91_e}).\ ] ] on one hand , a synaptic connection of type @xmath86 is concerned by extracellular concentrations of gaba since it involves gabaergic interneurons . in case of a gaba excess in the extracellular space , the inhibition of the postsynaptic neuron is strengthened , _ i.e. _ less neurons are activated in population p which is translated in the nmm by an increase of the threshold of the sigmoidal term in the @xmath10 dynamics . on the other hand , a synaptic connection of type @xmath87 is impacted by the extracellular concentration of glutamate implying a modulation of variable @xmath10 dynamics as well . in case of an excess of glutamate in this kind of synapse , the postsynaptic neuron is more excitable . hence , more neurons are activated in population p which can be reproduced by a decrease in the threshold parameter appearing in . gathering both modulations impacting the excitability of population p , we set in equation @xmath92_e})-{\rm si_{glup}}({\rm [ glu]_e}).\ ] ] the new neuro - glial mass model embedding the glial feedback is obtained from model by considering the dynamical entries @xmath93 and @xmath94 mentioned above . accordingly , the sigmoidal functions appearing in equations , , , and become : @xmath95_e})-{\rm si_{glup}}({\rm [ glu]_e})-(y_1-y_2 ) ) } } \\ { \rm sigm}(c_1\,y_0,v_{\rm p ' } ) & = \frac{2\,e_0}{1+e^{r\,(v_0-c_1\,y_0 ) } } \\ { \rm sigm}(c_3\,y_0,v_{\rm i } ) & = \frac{2\,e_0}{1+e^{r\,(v_0-{\rm si_{glui}}({\rm [ glu]_e})-c_3\,y_0)}}\end{aligned}\ ] ] we compare time series generated by the neuro - glial model with and without glial feedback to illustrate its impact on the model behavior ( figure [ bolus_gaba ] ) . to this aim , we mimic the same gaba bolus injection ( @xmath96 au ) in the extracellular space at @xmath97 with both models . for obtaining regular patterns and ease the comparison between the outputs , we consider a constant input @xmath33 with a value of @xmath35 close to and greater than @xmath50 ( figures [ bolus_gaba ] , panels ( a1 ) and ( b1 ) ) . hence , from initial time to @xmath97 , both model generate pacemaker nis oscillations at a low frequency . note that , even if gaba concentration is low and the corresponding sigmoidal feedback is consequently very weak , it already impacts the neural activity , which implies a difference in the spike frequency between the two lfp time series . ( top panels ) computed for the model without feedback for all @xmath98 ( a1 ) and for the model with feedback at @xmath99 ( b1 ) and @xmath97 ( b2 ) . time series ( bottom panels ) corresponding to @xmath7 , lfp , extracellular concentrations of glutamate and gaba generated by the models without feedback ( left ) and with feedback ( right ) . the purple lines on the bifurcation diagrams materialize the fixed value of input @xmath35 . the dark red lines on the time series materialize the time of gaba bolus injection ] in the time series generated by the model without feedback ( left panels of figure [ bolus_gaba ] ) , the neural activity and the glutamate concentration dynamics remain unchanged ( figure [ bolus_gaba](a ) ) after the artificial and instantaneous increase in @xmath52_{e}}$ ] that aims to mimic a gaba bolus injection . in contrast , in the model with glial feedback ( right panels of figure [ bolus_gaba ] ) , the strong increase in gaba concentration implies a break in neural activity , and thus a decrease in glutamate concentration . once gaba concentration has become sufficiently low , neural activity starts again . the glutamate and gaba concentrations come back to their respective basal lines and oscillate under the effect of the neural spikes . these phenomenons can be explained using bifurcation - based arguments ( top panels of figure [ bolus_gaba ] ) . in the system without feedback , the neural dynamics is entirely decoupled from the glial one . hence , the bifurcation diagram of the nmm remains unchanged during all the simulation ( figure [ bolus_gaba](a1 ) ) . on the other hand , in the system with feedback , the bifurcation diagram of the neural system is deformed along time , in particular the value of @xmath50 changes with the glial variables . consequently , right after the gaba bolus , @xmath50 becomes greater than the input value @xmath35 ( figure [ bolus_gaba](b2 ) ) . moreover , we recall that @xmath50 plays the role of an activation threshold . thus , as long as gaba concentration remains high , the neural variables are at steady state and the neural compartment remains quiescent . a direct calculation shows that the glial compartment has a single stable singular point : extracellular gaba concentration decreases towards this attractive state implying a slow decrease in @xmath50 . once , gaba concentration is low enough , @xmath50 becomes smaller than @xmath35 , and the system oscillates again . this analysis shows how the model with feedback can take into account changes in glutamate and gaba dynamics to modify all the dynamics of the system and illustrates the interest of embedding the glial feedback in such neuro - glial model . our model allows us to study the effects of variations in glutamate or gaba dynamics on neural activity . in the following , we study the effects of deficiencies in the reuptake of neurotransmitters by the astrocytes both on extracellular concentrations and neural activity . for both types of deficiency ( gaba and glutamate ) , we first describe the biologic context and mechanisms and their outcomes , then we provide a mathematical analysis of the underlying dynamical mechanisms to explain the effects that can be expected in the biological system . we consider an astrocyte presenting a deficiency in its gaba transporters which implies a low capacity to reuptake the extracellular gaba . a glial cell is linked with several neurons , more specifically with several synapses . thus the gaba in synaptic clefts linked with the defective astrocyte increases . consequently , the concerned post - synaptic neurons receive more inhibition from the extracellular gaba and release less neurotransmitters in the following synapses . thus , considering several defective glial cells , all neurons in a local neighborhood are affected . in summary , when glial cells present a deficiency in gaba reuptake , local neurons are more inhibited , and we expect a decrease of their activities in the corresponding simulations . in the model , parameter @xmath67 stands for the maximum velocity of the gaba flux from the extracellular space to the astrocytes , _ i.e. _ in case of gaba saturation in the extracellular space . in that sense , it is related to the efficiency of the main glial transporter of gaba and modulates the glial reuptake dynamics . consequently , to simulate a deficiency in the gaba glial reuptake , we decrease the value of this parameter . at the neuronal level we are interested in @xmath50 value according to the feedback sigmoidal functions . for sake of simplicity in this mathematical analysis , we set : @xmath100_e } ) & \rightarrow v_1 \\ { \rm si_{glup}}({\rm [ glu]_e } ) & \rightarrow \frac{m_{\rm glup}}{m_{\rm glui}}\,v_1 \\ { \rm si_{gaba}}({\rm [ gaba]_e } ) & \rightarrow v_2 \end{aligned}\ ] ] the ranges of @xmath101 and @xmath102 are defined by the limits of @xmath103_e})$ ] and @xmath104_e})$ ] respectively : @xmath105 \ , \text{and } \ , v_2 \in [ 0,m_{\rm gaba}]\ ] ] with these new notations , the dynamical excitability thresholds @xmath94 , @xmath106 and @xmath93 of populations p , p and i become : @xmath107 with these new parameters , an increase or a decrease in gaba ( resp . glutamate ) extracellular concentration is represented by an increase or a decrease in the value of @xmath102 ( resp . @xmath101 ) respectively . the natural effect of a deficiency of gaba glial reuptake on neural activity is an increase in the extracellular gaba concentration . thereby , we characterize the dependency of @xmath50 on the value of @xmath102 . the assumption that @xmath101 can be kept constant is justified in the following remark [ v1constant ] . [ prop1 ] @xmath50 is linear and increasing according to @xmath102 the set of the system singular points obtained for the different values of parameter @xmath35 can be explicitly expressed according to @xmath7 , @xmath101 and @xmath102 all other parameters being fixed . the @xmath7 components of the singular points for given values of @xmath35 , @xmath101 and @xmath102 are characterized as solutions of @xmath108 where @xmath109 all the other components of a given singular point result by direct calculation from its @xmath7 component . we rewrite equation as follows @xmath110 obviously the two saddle - node bifurcation values @xmath111 and @xmath112 are local extrema of function @xmath113 . in particular @xmath114 is the local maximum of @xmath113 and is defined as the solution of [ ysnic ] @xmath115 since @xmath116 is independent on @xmath102 , so is @xmath117 and it can be considered as a parameter in equation . from and we obtain the following expression for @xmath50 @xmath118 let us consider the model generating an oscillatory output with a fixed value of @xmath35 ( ( @xmath119 ) . if the extracellular concentration of gaba increases ( e.g. by an injection of a gaba bolus as in figure [ bolus_gaba ] ) , the value of @xmath102 increases and proposition 1 asserts that the value of @xmath50 also increases . as already explained , the closest @xmath50 is to @xmath35 , with @xmath119 , the largest is the limit cycle period , thus the oscillation frequency of the outputs decreases . if @xmath50 increases enough such that @xmath120 , the stable limit cycle of the system disappears , and the neural compartment becomes quiescent . in the case of a deficiency of gaba glial reuptake , the extracellular concentration of gaba increases , and we can use proposition 1 to explain the subsequent effects . for that , we use the following _ in silico _ protocol : we initialize the neuro - glial model in an oscillatory phase with a low oscillation frequency and consider @xmath6 a gaussian input . at @xmath121 , we turn off the gaba glial reuptake by setting @xmath122 ( figure [ v2_psnic ] ) . the result is an increase in gaba extracellular concentration implying an increase in @xmath50 . as @xmath50 increases , the probability for @xmath6 to overcome @xmath50 along the associated brownian motion decreases , and also does the oscillation frequency ( figure [ v2_psnic ] ) . consequently , we observe a decrease in the oscillation frequency after @xmath123 . in the time series , the oscillation frequency decreases gradually during a transient ( @xmath124 ) until reaching its minimum . this can be explained by the slow increase of gaba extracellular concentration that reaches its new baseline at @xmath125 . [ v1constant ] a deficiency in the gaba reuptake by the astrocyte implies a decrease in the neural activity . hence , the glutamate extracellular concentration remains close to the baseline . consequently , the impact of the changes in @xmath101 value can be neglected and , under this approximation , proposition [ prop1 ] characterizes the global effect of such deficiency on the neural compartment excitability . value according to @xmath102 ( left ) . time series corresponding to @xmath126 , @xmath52_e}$ ] , @xmath51_e}$ ] and @xmath127_e})$ ] ( right from top to bottom ) for @xmath6 a gaussian variable . at @xmath128 s , the gaba glial reuptake is artificially altered by setting @xmath122 . ] in this section , we investigate the impact of a deficiency of glutamate reuptake by the astrocytes upon the neuronal activity . such deficiency provoke an increase in the extracellular concentration of glutamate and , consequently , neurons in this neighborhood are more excitable . yet , it is important to note that interneurons release more gaba implying an increase in the gaba extracellular concentration as well , and an enhancement of the inhibition of the pyramidal activity . hence , the possible balance between glutamate - induced over - excitation and subsequent gaba - induced over - inhibition may lead to different types of response of the neuronal compartment . as previously , for studying theoretically the underlying mechanisms , we consider the nmm with two parameters @xmath101 and @xmath102 representing glutamate- and gaba - related feedbacks respectively . in the last section , we have characterized the linearity of @xmath50 according to @xmath102 for any fixed value of @xmath101 . now , let us fix the value of @xmath102 and study the variations of @xmath50 according to @xmath101 . we recall that @xmath50 is a key value of the system structure since it represents the excitability threshold of the neural compartment . it is important to note that in a specific case , the snic bifurcation disappears without disappearance of this excitability threshold . in this case , the supercritical hopf bifurcation occurring for a large value of @xmath35 and the saddle - node bifurcation ( sn@xmath39 ) previously linked to the snic bifurcation , are preserved and a subcritical hopf bifurcation appears , close to sn@xmath39 , giving birth to an unstable limit cycle . this limit cycle persists for a very small interval of @xmath35 values before it disappears through a fold bifurcation of limit cycles . we refer the reader to @xcite for more details about this bifurcation structure . for certain parameter values of the whole model , the bifurcation structure may therefore be lost when @xmath101 varies . computing the region of the parameter space ( of high dimension ) for which it remains unchanged for any @xmath101 is difficult . yet , the previous analysis of the nmm @xcite ensures us that this region is large . thus , in the following , we assume that @xmath50 exists and the associated saddle - node bifurcation is not degenerated for all @xmath129 $ ] , _ i.e. _ the maximal interval of values taken by @xmath130_e})$ ] , which is the case , in particular , for the parameter values given in table [ par_table ] and [ par_feedback ] , that have been used for the simulations . we recall that @xmath50 can be written as follows : @xmath131 where @xmath132 is given by . since we consider @xmath102 fixed we introduce the function @xmath133 as explained above , for each @xmath101 , there exists a unique bifurcation value @xmath50 occurring at the non - hyperbolic ( saddle - node ) singular point characterized by @xmath117 which is defined by @xmath134 this value satisfies @xmath135 . we can not find the explicit expressions of @xmath136 and @xmath137 . thus , for characterizing the variations of @xmath50 with @xmath101 , we take advantage of the implicit definitions above and focus on localizing the extrema of @xmath137 . assume that for all @xmath138 $ ] , @xmath50 exists and the associated saddle - node bifurcation is not degenerate . then 1 . if @xmath139 , @xmath137 has no local extremum , 2 . if @xmath1400 , \frac{b\,e_0\,r\,c_4}{2\,b}\right[$ ] , @xmath137 may admit two local extrema : a minimum at @xmath141 and a maximum at @xmath142 . if both exist , then @xmath143 . let us search for local extrema of function @xmath137 which is implicitly defined by . hence , we are interested in solving the following problem of optimization under constraint : @xmath144 we introduce the associated lagrangian function @xmath145 the necessary condition for the existence of an extremum of @xmath146 under the constraint @xmath147 is @xmath148 that is [ cn_lagrangien ] @xmath149 by assumption , the saddle - node bifurcation associated with the snic bifurcation is not degenerate , _ i.e. _ every solution of for @xmath129 $ ] satisfies @xmath150 . thus , system reads [ splag ] @xmath151 consequently , if the problem under constraint admits an extremum , this extremum satisfies @xmath152 . following the assumption that a snic bifurcation occurs for any value of @xmath129 $ ] , equation admits a solution for any @xmath101 . hence , if the problem under constraint admits an extremum , it corresponds to a snic bifurcation occurring at @xmath153 such that @xmath154 from , we obtain @xmath155 using the facts that , for any fixed values of @xmath7 , function @xmath156 is bell - shaped and its maximal value does not depend on @xmath7 ( see figure [ diff_sigm_v1 ] ) , one obtains that function @xmath157 vanishes in @xmath101 if @xmath1580 , \frac{b\,e_0\,r\,c_4}{2\,b}\right[.\ ] ] if @xmath139 , function @xmath157 admits no zero , which proves the first item of the proposition 2 . now , we assume that condition is fulfilled and we search the values of @xmath101 satisfying @xmath159 , _ i.e. _ @xmath160 which reads @xmath161 setting @xmath162 we obtain the two solutions @xmath163 of @xmath159 : @xmath164 note that @xmath141 ( resp . @xmath142 ) corresponds to the extremum when the saddle - node sn@xmath39 ( resp . sn@xmath40 ) crosses the fold of the surface @xmath165 . we consider @xmath166 and we note @xmath167 the value of @xmath7 corresponding to the snic connection for this value of @xmath101 , _ i.e. _ the solution of @xmath168 to prove that @xmath50 reaches a local minimum at @xmath166 , we introduce the bordered hessian matrix @xmath169 associated with the lagrangian function at its singular point @xmath170 ( solution of system ): @xmath171 the determinant of @xmath172 is given by @xmath173\ ] ] on one hand , the saddle - node associated with the snic bifurcation is not degenerate and is a local maximum of @xmath174 , thus @xmath175 . on the other hand , for any @xmath7 , @xmath156 is increasing at @xmath176 ( see figure [ diff_sigm_v1 ] ) , thus @xmath177 . finally @xmath178 and @xmath179 corresponds to a local minimum of @xmath50 . a similar argument proves that @xmath180 corresponds to a local maximum of @xmath50 ( where @xmath181 is the @xmath7 value corresponding to @xmath182 bifurcation for @xmath183 ) . the above proposition can be interpreted as a necessary condition for having a change in the sense of variations of @xmath50 when @xmath101 varies in @xmath184 $ ] . the following result gives a sufficient condition for @xmath141 actually lying in @xmath184 $ ] . we have @xmath185 $ ] if and only if @xmath186 $ ] where @xmath187 since @xmath141 satisfies @xmath188 , one obtains , from equation , @xmath189 for any @xmath7 , function @xmath190 is strictly increasing over @xmath191 $ ] and @xmath192 $ ] if and only if @xmath193 $ ] defined by and ( see figure [ diff_sigm_v1 ] ) . defined by and interval @xmath194 $ ] for which @xmath195$].,scaledwidth=45.0% ] in conclusion , for a fixed value of @xmath102 , @xmath50 reaches a local minimum at a value @xmath192 $ ] if and only if @xmath1960 , \frac{b\,e_0\,r\,c_4}{2\,b}\right [ \ , \cap \ , \left[i_1,i_2\right].\ ] ] moreover , in section [ gaba_deficiency ] about the gaba glial deficiency , we proved that , for a fixed value of @xmath101 , @xmath50 is linear and increasing with @xmath102 . both results allow us to predict that there exist three shapes of @xmath197 according to the value of @xmath198 . * if @xmath199 then @xmath200 and @xmath50 strictly increases with @xmath101 and @xmath102 . * if @xmath201 , then @xmath202 and @xmath50 strictly decreases when @xmath101 increases ( for @xmath102 fixed ) and strictly increases with @xmath102 ( for @xmath101 fixed ) . * if @xmath186 $ ] , then @xmath192 $ ] and @xmath50 decreases when @xmath101 increases in @xmath203 $ ] and increases with @xmath204 ( for @xmath102 fixed ) . in the following we illustrate the three qualitative types of neural activity resulting from an astrocyte deficiency to capture glutamate using the following values : @xmath205\end{aligned}\ ] ] for each case , we provide simulations representing the value of @xmath50 in @xmath206 space and time series generated by the model when the glutamate glial reuptake is altered ( figures [ mglup2 ] , [ mglup3 ] and [ mglup25 ] ) . in @xmath207 \times [ 0,0.1]$ ] space ( left panel ) , and time series corresponding to @xmath126 , @xmath52_e}$ ] , @xmath51_e}$ ] and @xmath208_e})$ ] ( right panels ) obtained with @xmath209 . black curve on the colormap : trace of ( @xmath103_e})$ ] , @xmath104_e})$ ] ) along the orbits . at @xmath210 , we alter the glutamate glial reuptake by setting @xmath211.,scaledwidth=93.0% ] in @xmath207 \times [ 0,0.4]$ ] space ( left panel ) , and time series corresponding to @xmath126 , @xmath52_e}$ ] , @xmath51_e}$ ] and @xmath208_e})$ ] ( right panels ) obtained with @xmath212 . black curve on the colormap : trace of ( @xmath103_e})$ ] , @xmath104_e})$ ] ) along the orbits . at @xmath210 , we alter the glutamate glial reuptake by setting @xmath211.,scaledwidth=95.0% ] in @xmath207 \times [ 0,0.2]$ ] space ( left panel ) , and time series corresponding to @xmath126 , @xmath52_e}$ ] , @xmath51_e}$ ] and @xmath208_e})$ ] ( right panels ) obtained with @xmath213 . black curve on the colormap : trace of ( @xmath103_e})$ ] , @xmath104_e})$ ] ) along the orbits . at @xmath210 , we alter the glutamate glial reuptake by setting @xmath211.,scaledwidth=95.0% ] for @xmath214 ( case @xmath215 , figure [ mglup2 ] ) , @xmath141 is negative and thus @xmath50 increases with @xmath101 . thereby , when we reproduce a glutamate glial deficiency triggering an increase in @xmath101 value , we observe a decrease in the oscillation frequencies in the neural activity . moreover , reduction of the glutamate glial reuptake together with the strong decrease of neural activity , triggers an increase in the baseline of glutamate extracellular concentration . indeed , the lack of reuptake involves the accumulation of extracellular glutamate . since neurons are less activated , the glutamate release is lowered and the neural reuptake can stabilize the glutamate extracellular concentration . figure [ mglup3 ] illustrates case @xmath216 ( @xmath217 ) . since @xmath218 , @xmath50 decreases as @xmath101 increases . thereby , a glutamate glial deficiency triggers an increase of @xmath101 value , and we observe an increase in the oscillation frequencies in the neural time - series . since the glial reuptake is reduced , glutamate accumulates in the extracellular space and the corresponding concentration baseline increases drastically . figure [ mglup25 ] illustrates intermediary case @xmath219 ( @xmath220 $ ] ) . since @xmath2210,1[$ ] , @xmath50 decreases for @xmath222 and increases otherwise . thereby , when we simulate a glutamate glial deficiency triggering an increase in @xmath101 value , we observe an increase in the oscillation frequencies in lfp time series , followed by a decrease . indeed , after the alteration of the glutamate glial reuptake , the glutamate extracellular concentration increases and excites more pyramidal cells and interneurons . subsequently , interneurons release more gaba , which implies an increase of the inhibition of pyramidal cells and a decrease in the neural activity . this sequence of events explains the delay in the regulation of the oscillation frequencies . this kind of behavior is physiologically relevant . indeed , it is conceivable that an excess of glutamate extracellular concentration is regulated after a delay , triggering a decrease of neural activity after the initial increase . moreover , the frequency after the regulation delay can be greater or lower than the initial one , depending on the value of the ratio @xmath198 . note that this value can be tuned to obtain @xmath141 small enough and @xmath50 large enough so that the frequency after regulation is equal or lower than the one before reuptake deficiency . this property offers the possibility of fitting the model outputs to experimental data and allows us to propose hypotheses about physiological and pathological mechanisms . in this article , we have introduced a new neuro - glial mass model built on a bilateral coupling of the nmm studied in @xcite and the glial model proposed in @xcite focusing on gaba and glutamate concentration dynamics . the model is based on recent biological knowledge resulting from experimental data @xcite to ground the interaction between the neural and extracellular / glial compartments . note that , as explained in the beginning of section 2.2 . , only basic properties of the dynamical coupling are needed to prove the qualitative dependencies studied in sections 3 and 4 . by lack of experimental data in the literature , we chose sigmoidal functions for relaying the glial feedbacks in the simulations because they represent a paragon of bounded increasing functions involving a significant threshold effect . then , using the interpretation of the aggregated yet biophysical parameters involved in this model , we have reproduced _ in silico _ various types of deficiencies in the reuptake of gaba or glutamate by the astrocytes and studied their impact upon the neural activity . we took advantage of the bifurcation analysis performed in @xcite to characterize theoretically the dynamical mechanism leading to local neuronal hyperexcitability through a modulation of the snic bifurcation standing for the excitability threshold of the neural compartment . the first result concerns the impact of a deficiency in the reuptake of gaba by astrocytes which implies an increase in the gaba concentration in the extracellular space . we have shown ( proposition 1 ) that the excitability threshold @xmath50 increases linearly with the feedback term value depending on the gaba concentration in the extracellular space . hence , such an astrocyte deficiency simulated in the model results in a decrease in the frequency of neural activity , which is consistent with the biological knowledge @xcite . the second result concerns the neuronal response to a deficiency in the reuptake of glutamate by the astrocytes . in this case , the neural activity may either be reduced or enhanced or , alternatively , may experience a transient of high activity before stabilizing around a new activity state with a frequency close to the nominal one ( _ i.e. _ before induction of astrocyte deficiency ) . we have characterized ( proposition 2 and corollary 1 ) the relationships between parameters of the model for predicting the neuronal response . it is worth noticing that it is possible to calculate explicitly the snic bifurcation for the uncoupled nmm ( see @xcite ) , but not for the bilaterally coupled model embedding the glial compartment . hence , we have expressed the question of characterizing the variations in the neuronal excitability as an optimization problem under an equality constraint resulting from the implicit characterization of the saddle - node bifurcation . the two model parameters @xmath74 and @xmath75 involved in proposition 2 and corollary 1 represent the maximal strengths of the glutamate concentration impacts on the excitability of the pyramidal cell and the interneuron populations respectively . our model - based study shows that the neural compartment may `` resist '' to the impact of glutamate excess in the extracellular space only if the value of @xmath223 lies in a positive interval . for small values ( @xmath224 ) , the neural activity frequency is lowered , as shown in figure [ mglup2 ] , while , for high values ( @xmath225 ) , the activity frequency increases drastically and permanently , as shown in figure [ mglup3 ] . for intermediary values , the neural activity recovers after a high frequency transient to a comparable mode despite the high values of extracellular glutamate and gaba concentrations . note that @xmath226 and @xmath227 explicitly depend on the coupling strength modulating the inhibition of the pyramidal cells activity by the interneurons ( _ i.e. _ parameter @xmath228 ) . on the other hand , only @xmath227 depends on @xmath75 . hence , one might interpret this later dependency as a balance between the glutamate - related glial feedbacks upon pyramidal cells and interneuron that the system should fulfill for being able to recover from a dysfunction in the astrocyte activity and avoid hyper - excitable behaviors . an interesting fact should be noticed for future experimental investigation based on this study : for @xmath229 , the astrocyte deficiency in capturing glutamate induces , in addition to the increase in the extracellular glutamate concentration , an increase in the gaba concentration , even greater than the one resulting from a default in gaba capture . dynamically , this increase is necessary so that the system could reach an alternative state and recover a low frequency activity , but may not be sufficient if the glial feedback on the pyramidal cells and the interneurons is unbalanced , _ i.e. _ for to large values of @xmath223 . the gaba increase induced by an excess of extracellular glutamate and the corresponding neuronal response form an experimental benchmark to identify the mechanism of the transition through hyperexcitability and potential neuronal recovery . future theoretical works will extend the analysis of the local neuro - glial dynamical interactions by studying the model response when the neuronal compartment undergoes a drastic change in its bifurcation structure induced by a deficiency in the astrocyte compartment dynamics . as a matter of fact , for other parameter values corresponding to different inter- and intra - population connectivity strengths , the neural compartment may generate other types of time series than noise - induced spiking ( nis ) among those identified in @xcite . therefore , the present study can be continued to study the dynamical mechanisms underlying successive transitions through various activity regimes . another perspective will consist to consider two coupled neuro - glial models , the input @xmath6 in each one being replaced by the output of the other one . the extension of the results presented in this article to such a system may provide useful methodological tools to tackle the hectic question of neuro - glial network modeling . this work was performed within the labex smart ( anr-11-labx-65 ) supported by french state funds managed by the anr within the investissements davenir program under reference anr-11-idex-0004 - 02 . we thank h. berry for helpful comments on this work . we also thank s. blanchard and f. wendling for allowing us to use the glial model they submitted . a. aubert and r. costalat . a model of the coupling between brain electrical activity , metabolism , and hemodynamics : application to the interpretation of functional neuroimaging . , 17(3):11621181 , 2002 . s. blanchard , s. saillet , a. ivanov , p. benquet , c. benar , h. benali , and f. wendling . the neuro - glio - vascular coupling through a new computational model : astrocytes activation can explain the cerebral blood flow nonlinear response to interictal events . . y. h. huang , s. r. sinha , k. tanaka , j. d. rothstein , and d. e. bergles . astrocyte glutamate transporters regulate metabotropic glutamate receptor - mediated excitation of hippocampal interneurons . , 24(19):45514559 , may 2004 . b. molaee - ardekani , j. mrquez - ruiz , i. merlet , r. leal - campanario , a. gruart , r. snchez - campusano , g. birot , g. ruffini , j .- delgado - garca , and f. wendling . effects of transcranial direct current stimulation ( tdcs ) on cortical activity : a computational modeling study . , 6(1):2539 , jan . 2013 . zhang , h .- k . wang , c .- q . ye , w. ge , y. chen , z .- jiang , c .- p . poo , and s. duan . released by astrocytes mediates glutamatergic activity - dependent heterosynaptic suppression . , 40(5):971982 , dec .

the statistical approach for derivation of the clump mass function ( clmf ) developed by donkov , veltchev & klessen is put to observational test through comparison with mass distributions of clumps from molecular emission and dust continuum maps of galactic cloud complexes , obtained by various authors . the results indicate gravitational boundedness of the dominant clump population , with or without taking into account the contribution of their thermal and magnetic energy . the clmf can be presented by combination of two power - law functions separated by a characteristic mass from about ten to hundreds solar masses . the slope of the intermediate - mass clmf is shallow and nearly constant ( ) while the high - mass part is fitted by models that imply gravitationally unstable clumps and exhibit slopes in a broader range ( ) , centered at the value of the stellar initial mass function ( ) . [ firstpage ] ism : clouds - ism : structure - turbulence - methods : statistical

1304.5902

dense clumps in molecular clouds ( mcs ) are typical sites of star formation as they are often associated with young stellar objects . their origin can be sought in the early epoch of cloud evolution when supersonic turbulence creates sets of condensations in the cold , mainly molecular gas . recent numerical simulations indicate their mean densities , sizes and masses vary in ranges @xmath3 @xmath4 , @xmath5 pc and @xmath6 @xmath7 , respectively @xcite , in consistency with extensive observational data about mc clumps ( e.g. , * ? ? ? clump morphology is also diverse : from filamentary to compact , quasi - spherical shapes @xcite . gravitational stability analysis shows that some clumps are subject to further contraction and collapse and eventually give birth to single stars or stellar clusters . numerous individual clumps were initially identified on maps of molecular line emissions which trace different density regimes in mcs : @xmath8co ( @xmath9 @xmath4 ) , @xmath10co and c@xmath11o ( @xmath12 @xmath4 ) , cs ( @xmath13 ) and h@xmath10co@xmath14 ( @xmath13 ) ( @xmath15 @xmath4 ) . such surveys were performed in nearby ( at distances @xmath16 pc ) galactic cloud complexes like orion a @xcite , orion b @xcite , taurus @xcite , ophiuchus @xcite , lupus @xcite and many others . some of these results on the physical parameters of clumps were included in the statistical study of @xcite who considered a sample of 9 galactic star - forming regions . in the last decade , further intensive research of mcs by use of some high - density tracers like h@xmath10co@xmath14 ( @xmath13 ) @xcite as well of dust continuum @xcite and dust extinction ( e.g. * ? ? ? * ) observations allowed for more precise mapping of cloud structure . some clumps originally found on emission maps were further decomposed and more compact ( typical sizes @xmath17 pc ) , very dense ( @xmath18 @xmath4 ) and probably collapsing clumps were delineated . some authors call such objects ` dense cores ' ; hereafter , we label them simply _ cores_. it is suggested that cores eventually form stars @xcite and thus the study of their mass function ( cmf ) will enable a better understanding of the physical origin of the stellar initial mass function ( imf ) and its possible variations . indeed , numerous dust continuum and dust extinction observations demonstrate that the cmf resembles the imf in its shape when fitted by a single power - law @xcite , a combination of two power - law @xcite or a lognormal function @xcite . it therefore has been proposed that the imf is a direct product of the cmf and a uniform star formation efficiency @xcite . yet it is still unclear how the cmf originates from the mass distribution of the initially formed mc clumps . the clump mass function ( clmf ) , as derived from molecular line emission surveys of galactic star - forming regions using low- and moderate - density tracers , varies in shape and slope and often differs substantially from the imf . most of the earlier studies of nearby complexes from co mapping resulted in a single power - law clmf of a rather shallow slope @xmath19 without any characteristic mass @xmath20 @xcite in contrast to the salpeter slope @xmath21 @xcite of the high - mass imf over @xmath22 . moreover , the clmf slope @xmath23 was found to remain nearly constant for a large range of clump masses @xmath24 , in different mcs @xcite . later works allowed for a more detailed mapping of mcs . emission from co molecules was found to trace lower density cloud regions . tracers like c@xmath11o revealed structures with @xmath25 @xmath4 but essentially larger ( @xmath26 pc ) than prestellar cores @xcite . such compact clumps encompass about 10% of the cloud mass . it was demonstrated that the derived clmfs could be represented by a combination of 2 power - law functions as @xmath27 @xmath7 separates intermediate - mass from high - mass clmf , with slopes @xmath28 and @xmath29 , respectively . this general picture was confirmed from submillimeter continuum studies of mc complexes at kpc - scale distances . using @xmath30 and @xmath31 scuba maps , reid & wilson derived 2 power - law clmfs for two massive star - forming regions . for ngc7538 , they obtained @xmath32 and @xmath33 while the variations of the high - mass slope seem to be more sensitive to the wavelength and close to the salpeter value : @xmath34 @xcite . for m17 , the characteristic mass turned out to be an order of magnitude less ( @xmath35 ) , the intermediate mass slopes are positive ( @xmath36 ) and the high - mass ones are about the value for co clumps @xmath37 @xcite . extending further their study to 11 low- and high - mass star - forming regions , those authors conclude that a single power - law clmf is clearly ruled out and argue for a two power - law function with @xmath38 and @xmath39 about the salpeter slope , regardless of the diversity of characteristic masses @xcite . the predictions of our statistical model ( * ? ? ? * hereafter , paper i ) are generally consistent with a two power - law clmf with shallow intermediate - mass slope @xmath40 and steeper high - mass slope @xmath39 , allowing for a variety of @xmath20 within two orders of magnitude . in this work they are compared with results on the observational clmf : i ) from molecular - line studies of three nearby star - forming regions : orion a @xcite , orion b @xcite , taurus @xcite , and of a sample of 9 regions @xcite ; ii ) from dust continuum surveys _ and _ co mappings of 2 kpc - away regions : m17 @xcite and rosette @xcite . we define _ clumps _ as condensations formed through a turbulent cascade during the early mc evolution . the ensemble of clumps generated at given spatial scale @xmath41 obeys a power - law relationship between clump masses and densities @xmath42 where the exponent @xmath43 is calculated considering equipartition relations between various forms of energy : gravitational , turbulent ( kinetic ) , thermal ( internal ) and magnetic ( * ? ? ? * hereafter , dvk11 ) . the clmf is derived as a superposition of clump mass distributions over a range of scales , taking into account the fractal structure of the cloud ( paper i ) . the physical basis of the model , its free parameters and the construction of the clmf are recalled in section [ clmf_model ] . the clump mass distribution derived from modeled structure of each considered individual cloud as well as different distributions that fit the composite observational clmf are presented and commented in section [ results ] . section [ discussion ] contains a discussion on the applicability and the restrictions of our approach to predict the observational clmf and envisions its possible extensions . our conclusions are summarized in section [ summary ] . our statistical model for clump description is presented in details in dvk11 ( sect . 2 and 3 ) and paper i ( sect . its main assumptions can be summarized as follows : 1 . _ scaling laws within the turbulent cloud : _ we consider fully developed supersonic turbulence that implies homogeneous and isotropic stochastic medium with a fractal structure and well - defined scaling laws of turbulent velocity , mean density and mean magnetic field . turbulent flows create density structures at any scale in the inertial range @xmath44 through a cascade possibly driven by the very process of cloud formation @xcite . we estimate the upper limit @xmath45 pc adopting a typical size of giant mc @xmath46 pc as an injection scale and taking in view that the largest scale of the turbulence inertial range is about a factor of 3 less @xcite . the lower limit @xmath47 is imposed above the actual end of the inertial range from the construction of our model : the sizes @xmath48 of all clumps generated at given scale @xmath41 must be within the inertial range and are typically an order of magnitude less than @xmath41 . various observations show that the inertial range spans at least 3 orders of magnitude , i.e. the size of the smallest generated object @xmath49 must be about @xmath50 pc . we consider mainly molecular and isothermal gas with temperatures @xmath51 k. requiring supersonic medium at all fractal scales and by use of a typical velocity scaling ( see equation [ eq_larson_1 ] below ) , one obtains @xmath52 pc which yields a scale of its generation @xmath53 pc . + turbulent velocity dispersion @xmath54 and mean mass density @xmath55 are assumed to scale according to `` larson s first and second laws '' @xcite whereas the scaling relation of the mean magnetic field @xmath56 is obtained from its observationally verified relation to the mean mass density ( @xmath57 : * ? ? ? * ) : @xmath58~,\ ] ] @xmath59~,\ ] ] @xmath60~,\ ] ] a fixed velocity scaling index @xmath61 is chosen . while observational and numerical studies indicate a broad range @xmath62 , the results in this paper narrow it to @xmath63 ( ` soft ' velocity scaling ) . the density scaling index @xmath64 is derived self - similarly from the assumption of mass - density relationship for clumps generated at a given scale @xmath41 ( see below ) . 2 . _ lognormal clump density distribution : _ such volumetric distribution of mass density @xmath65 is testified from numerous numerical simulations of supersonic turbulence and is described through a standard lognormal probability density function ( pdf ) : @xmath66}\,ds~,\ ] ] where @xmath67 is the log density and @xmath68 are the distribution peak and the standard deviation , respectively . in our model , this density statistics is used as the basis of clump statistics as follows . let @xmath69 is the total number of pixels in a cloud map and @xmath70 is the number of those with density @xmath65 . then @xmath71}~.\ ] ] + since turbulence is assumed to be homogeneous and isotropic and gravity is a central force , one can consider the generated clumps as homogeneous spheres , characterized solely by their size ( diameter ) @xmath48 . of course , the real fragments in a turbulent mc with high mach numbers ( @xmath72 ) delineated by isodensity contours @xmath73 $ ] can be with complex shapes and even not necessarily connected regions . the essence of our statistical approach is to describe them by an ensemble of @xmath74 spherical homogeneous clumps with density @xmath65 . thus the ` average clump ensemble ' generated at given spatial scale @xmath41 follows a lognormal density distribution like the pdf at that scale . + the parameters of the density distribution : @xmath75 depend on the spatial scale through the sonic mach number @xmath76 where @xmath77 is the sound speed . the turbulence forcing parameter @xmath78 spans values between @xmath79 , for purely solenoidal forcing , and @xmath80 , for purely compressive forcing ( see @xcite ) . the modeled clmfs presented in this paper favor mainly solenoidal forcing , mixed in a natural way with compressive modes : @xmath81 . a comment on that is included in the discussion . 3 . _ clump mass - density - size relationship : _ masses , densities and sizes of clumps in an ` average ensemble ' are assumed to obey the statistical relationships : @xmath82 @xmath83 with a choice of normalization units : @xmath84 the dimensionless parameter @xmath85 accounts for the precision of the clump size ` measurement ' and can be interpreted as mapping resolution of the scale of clump generation . it is appropriate to set it as a small constant of order of several percent since @xmath48 is typically an order of magnitude less than @xmath41 . + according to our turbulent scenario of clump formation , the clump density - size relationship ( equation [ eq_n - l ] ) is taken as a self - similar extension of the density scaling law ( equation [ eq_larson_2 ] ) and hence : @xmath86 4 . _ equipartitions between clump energies : _ a general type of equipartition relation between gravitational @xmath87 and kinetic @xmath88 energy per unit volume with some coefficient @xmath89 of proportionality @xmath90 is expected to hold for structures shaped by turbulence in which gravity gradually takes over ; e.g. in regions where turbulence decays locally or where it accumulates material reaching a state of local gravitational instability . a fiducial range @xmath91 is testified from simulations of the early stage of the clump evolution , before stars have been formed @xcite . the analysis of mhd simulations of cloud formation at galactic scale ( length unit @xmath921 kpc ; * ? ? ? * ) showed that the considered equipartition can include contributions of thermal ( internal ) @xmath93 and magnetic @xmath94 energy as well @xcite . + in our model , we assume such equipartition relations to hold for the ` average clump ensemble ' and use them to derive the mass - density scaling index @xmath95 at each scale : * ` wkin2 ` or ` wkin4 ` , equipartition of the gravitational vs. kinetic energy : @xmath96 * ` wkin2mag ` , equipartition of the gravitational vs. kinetic and magnetic energy : @xmath97 * ` wkin2th2 ` , equipartition of the gravitational vs. kinetic and thermal energy : @xmath98 the corresponding equations for the typical member of the ` average clump ensemble ' which are used to derive @xmath95 are listed in paper i ( appendix a and b ) . in our model , a mc is considered as a hierarchical set of spatial scales and its structure is described through the solutions @xmath95 obtained for a chosen equipartition relation . they depend on four free parameters : velocity - scaling index @xmath61 , turbulent forcing parameter @xmath78 , mapping resolution @xmath85 ( equation [ eq_norm_l ] ) and temperature @xmath99 . the parameters ranges of variation that yield plausible solutions can be restricted by use of observational studies of mc structure . the work of ( * ? ? ? * hereafter , lal10 ) , based on dust extinction maps of galactic cloud complexes , corresponds best to our approach . these authors define an effective radius @xmath100 of a subregion or a set of subregions with total area @xmath101 which we interpret as an observational counterpart of our notion of ` spatial scale ' , i.e. @xmath102 . then the mc structure is described by the relationship between the effective radius and the mass within the total area @xmath101 : @xmath103 ( see fig . 2 in lal10 ) . on the other hand , the total mass contained within a scale @xmath41 depends , in our approach , on the index @xmath95 due to the self - similarity assumption ( cf . equations [ eq_n - l ] and [ eq_dens_scaling ] ) : @xmath104 . hence , the mass - scale power - law index @xmath105 or : @xmath106 comparison between @xmath107 obtained in that way from the work of lal10 and our model predictions @xmath95 for sets ( @xmath61 , @xmath78 , @xmath85 , @xmath99 ) is described in paper i ( see fig . 1 there ) . the best fits @xmath95 for the cloud complexes taurus , orion a and orion b studied in this paper are plotted in the left panels of fig . [ fig_taurus ] , [ fig_oria ] and [ fig_orib ] , respectively . we comment them in section [ results ] . a parameter set ( @xmath61 , @xmath78 , @xmath85 , @xmath99 ) corresponding to the obtained best fit @xmath95 of the lal10 data for a chosen mc yields : i ) a lognormal clump mass distribution @xmath108 ; and ii ) a measure of the total number of clumps @xmath109 at each scale ( dvk11 , eq . 17 ) : @xmath110 the clmf of an individual cloud is derived as a superposition of the lognormal clump mass distributions . each of them is discretized through sets of weights @xmath111 where the mass range is centred at the peak of the distribution and the mass limits are determined from the requirement : @xmath112 also , to construct the clmf correctly , one must require mass conservation throughout the mc fractal structure @xcite . the whole cloud with size equal to the upper limit of the inertial range @xmath45 pc is to contain @xmath113 substructures of mass @xmath114 at given scale @xmath41 . as shown in paper i ( sect . 3.2 ) : @xmath115 eventually , one obtains for the clmf value in a selected mass bin @xmath116 : @xmath117 if only gravitationally unstable clumps are considered , one should take also into account their contraction in timescales given by the free - fall time @xmath118 . their _ time - weighted _ mass function can be derived by permitting constant replenishment of the clump population and introducing a weighting factor of each scale of clump generation : @xmath119 ( cf . equations [ eq_larson_2 ] and [ eq_dens_scaling ] ) . then the time - weighted clmf is obtained as the number of substructures @xmath113 in equation [ eq_clmf ] is modified by factor @xmath120 . we put to test the clmf , derived in paper i , using molecular line observations of nearby galactic mc complexes and dust continuum studies of further 3 regions at kpc - scale distances , mapped also in co lines ( table [ table_mc ] ) . subsamples of the original clump data are considered taking into account the lower mass limit of confidence in our models and/or the observational completeness limit ; the corresponding size and mass ranges are specified in columns 3 and 4 . lcccccccccc sf region & d & ref . & sizes & masses & @xmath121 & @xmath20 & & cf method & note + & [ kpc ] & & [ pc ] & [ @xmath7 ] & & [ @xmath7 ] & i m & hm & & + + taurus & 0.14 & 1 & @xmath123 & @xmath124 & @xmath125 & @xmath126 & @xmath127 & & phm & mostly gb + orion a & 0.45 & 2 & @xmath128 & @xmath129 & @xmath130 & @xmath131 & @xmath132 & @xmath133 & ppv & @xmath9290% starless , gb ? + orion b & 0.45 & 3 & @xmath134 & @xmath135 & @xmath125 & @xmath136 & @xmath137 & @xmath138 & gaussclumps & ub + gb + mc sample & @xmath139 & 4 & @xmath140 & @xmath141 & @xmath142 & @xmath143 & @xmath144 & @xmath145 & various & mainly starless , gb ? + & @xmath139 & 4 & @xmath140 & @xmath146 & @xmath142 & @xmath136 & @xmath147 & @xmath148 & various & mainly starless , gb ? + * m17 * & 1.6 & 5 & @xmath149 & @xmath150 & @xmath151 & @xmath152 & @xmath153 & @xmath154 & gaussclumps & ub + gb ? + * rosette * & 1.6 & 6 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath160 & clumpfind & starless , mostly ub ? + + * m17 * & 1.6 & 7 & @xmath161 & @xmath162 & @xmath163 & @xmath164 ? & @xmath165 & & clumpfind & mainly starless + & 7 , 8 & @xmath166 & @xmath167 & @xmath168 & @xmath164 & @xmath144 & @xmath133 & clumpfind & ub + gb ? + * rosette * & 1.6 & 9 & @xmath169 & @xmath170 & @xmath171 & @xmath172 & @xmath173 & @xmath174 & getsources & starless + [ 1 ] @xcite ; [ 2 ] @xcite ; [ 3 ] @xcite ; [ 4 ] @xcite ; [ 5 ] @xcite ; [ 6 ] @xcite ; [ 7 ] @xcite ; [ 8 ] @xcite ; [ 9 ] @xcite + three individual nearby star - forming regions of comparable size are selected : i ) with clumps delineated through similar density tracers : @xmath10co ( orion b ; * ? ? ? * ) , c@xmath11o ( taurus ; * ? ? ? * ) and cs ( @xmath13 ) ( orion a ; * ? ? ? * ) , consistent with the typical clump densities in our approach @xmath175 @xmath4 ; ii ) with known general structure from dust extinction maps of lal10 . the spatial extent of the studied regions in the chosen complexes falls within the adopted inertial range of turbulence ( sect . [ physical_framework ] , ( i ) ) . dust extinction data reveal some diversity of structure in terms of mass - size relation of regions delineated by isodensity contours : taurus exhibits monotonically ` steep ' structure in terms of @xmath95 while orion a and orion b have ` shallow ' internal regions and ` steep ' external , less dense regions ( fig . [ fig_taurus]-[fig_orib ] , left panels ; cf . fig . 2 in lal10 ) . taurus is a typical low - mass star - forming region @xcite , while numerous high - mass stars have been formed in orion a and b in the recent past . as an additional test of the predictive power of the models for a variety of environments , we perform comparisons with observational clmf for a sample of 9 mc complexes @xcite wherein star - forming , cluster - forming and starless clumps have been detected . an equipartition relation and a parameter set were sought that yield the best fit @xmath95 of the cloud structure as traced by the lal10 data ( fig . [ fig_taurus]-[fig_orib ] , left panels ) . the modeled clmf derived from them is compared with the observational one from molecular line maps ( fig . [ fig_taurus]-[fig_orib ] , right panels ) . the range of spatial scales considered in the fitting procedure ( shaded areas , left panels ) corresponds to the mass range of the referred observational clmf ( shaded areas , right panels ) , additionally restricted by the lower clump mass limit of confidence in the model . the scales are treated as those where the observed clumps were generated and hence their range is obtained through the clump mass - scale diagram of the tested model ( paper i , sect . 3.1 ) . [ fig_taurus ] ( left ) demonstrates that the equipartition choice ` wkin4 ` yields a very good fit of the inner parts of the complex . that seems physically consistent to us since this case describes structures which have been shaped under significantly influence of self - gravity . the typical dense clumps generated in them have mass distributions in a relatively good agreement with the observational clmf ( fig . [ fig_taurus ] , right ) , obtained by @xcite . the best - fit parameters point to a mainly solenoidal turbulent forcing ( see * ? ? ? * ) and a ` soft ' velocity scaling , typical for incompressible turbulence . on the other hand , the equipartition relation ` wkin4 ` is obviously not applicable to the less dense extensive parts of the cloud . we stress , however , that in our model scales @xmath176 pc do not produce clumps with masses within the considered observational range . it should be pointed out that the clmf obtained by @xcite differs significantly from clmfs in other clouds as derived from molecular line emission data . using essentially the lowest mass bin ( @xmath177 @xmath7 ) , these authors argue for a flat mass function in taurus , although they remain open for an alternative interpretation based on an `` unbiased survey ... in a wide mass range '' . we believe that indeed their data cover the mass range around the turn - over of the clmf while its high - mass part ( if existing ) falls beyond their scope note that characteristic masses @xmath20 of hundreds @xmath7 are derived in various mc complexes ( table 1 , column 7 ) . the power - law slope of the modeled clmf over the lower mass limit of confidence in our model is slightly shallower than the typical one of the mass distribution of co clumps @xcite . it would steepen ( @xmath178 ) if the observational data contained clumps with masses up to the expected @xmath179 @xmath7 cf . the modeled clmfs from ` wkin4 ` ( fig . 3 in paper i ) . we obtained good fits of the orion a structure in the case ` wkin2mag ` at intermediate to large spatial scales ( fig . [ fig_oria ] , left ) . the corresponding clmfs generally agree with the result of @xcite . enhancement of the model temperature , compatible with the derived @xmath180 k in the complex @xcite , leads to larger , more plausible values of @xmath78 and @xmath61 and yields a wider range of scales of clump generation as the cloud structure is better fitted through the curve @xmath95 ( fig . [ fig_oria ] , bottom ) . depending on the model temperature , the case ` wkin2mag ` in our approach generates mostly / only gravitationally unstable clumps ( paper i , fig . 3 ) . on the other hand , @xcite claim that all clumps in their sample are close to virial equilibrium . therefore a time - weighted clmf ( cf . [ derivation_clmf ] ) is more appropriate for comparison with the referred observational study . time - weighting steepens the high - mass slope of the model from @xmath181 , typical for fractal clouds , to @xmath182 , i.e. steeper than that of the stellar imf ( @xmath183 ; * ? ? ? * ) ( fig . [ fig_oria ] , bottom right ) . this leads to agreement ( within the data uncertainties ) with the observational clmf for clump masses @xmath184 @xmath7 while for lower masses the deviation is drastic . in contrast , the non - weighted modeled clmf is generally consistent with the observational clmf below @xmath20 and for @xmath185 k. we note the high completeness limit ( @xmath186 @xmath7 ) of the data which might artificially constrain the intermediate - mass range . the referred observational clmf in this region is part of the work of @xcite who analyzed co data sets for 8 galactic mcs . they argued for a single power - law clmfs in all cases , with a universal mean slope @xmath178 , over a wide range of clump masses . however , a closer look at their result for orion b reveals that this mc may be an exception . the conclusion of @xcite about a single power - law with shallow slope is based mostly on the single mass bin @xmath152 @xmath7 ( fig . [ fig_orib ] , right ) . inspection of the statistically rich data in the range @xmath187 @xmath7 indicate rather a combination of two power laws : a shallow intermediate - mass slope below @xmath188 @xmath7 and a significantly steeper high - mass slope . adopting the latter value as @xmath20 and the completeness data limit of the authors @xmath189 @xmath7 , one obtains @xmath190 similar to that in orion a and a salpeter - like @xmath191 . and adopting characteristic mass @xmath143 @xmath7 ( top ) and @xmath136 @xmath7 ( bottom ) . see text for the considered mass ranges . other designations are the same like in figs . [ fig_taurus]-[fig_orib].,width=317 ] modeling of the orion b structure as traced by the lal10 data yields good fits in the cases ` wkin2th2 ` and ` wkin2 ` ( fig . [ fig_orib ] , left ) . for such choices of equipartition and temperature @xmath192 k , the critical mass over which most of the generated clumps are gravitational unstable is about the adopted characteristic mass from the used observational data . therefore we apply time - weighting for the high - mass regime and obtain @xmath193 close to the salpeter value . the modeled clmfs fit remarkably well the high - mass observational ones , excluding the sole bin at the upper limit of the distribution ( fig . [ fig_orib ] , right ) . non - weighted clmfs obviously fail to reproduce the observational one although the high - mass slope @xmath194 is close to the estimate of @xcite . the model is not applicable to fit the intermediate - mass clmf because of the high lower mass limit of confidence . the presented best - fit descriptions @xmath95 of mc structure in taurus , orion a and orion b and their corresponding clmfs with different characteristic masses @xmath20 and hm slopes ( cf . table [ table_mc ] ) are derived by use of different equipartition relations that may reflect a variety of physical conditions . what model case would describe appropriately a statistical clmf obtained from a large set of clump data in different mcs ? the sample in the study of @xcite could give a clue to the answer it encompasses nearby cloud complexes of diverse star - forming type as the vast majority of clumps are associated with no or a small number young stellar objects and have masses below several dozens @xmath7 . clumps associated with young stellar clusters were found in two out of nine sampled clouds and their fraction is only 5 % of the total number of clumps . all objects with masses @xmath195 @xmath7 are from this group . one may take the latter value as an upper mass limit when fitting the clmf , in order to avoid extreme environments and to secure sufficient statistics . the two power - law shape of the clmf is evident whereas the characteristic mass is ambiguous . we consider two choices which both allow for very good fits : i ) @xmath196 @xmath7 , adopting the estimate of @xcite and decreasing the upper mass limit further by one bin ; and ii ) @xmath197 @xmath7(fig . [ fig_mcs ] ) . the lower mass limit in both cases is the lower limit of confidence of the tested models . the choice i ) yields excellent agreement of the model case ` wkin2th2 ` with the observational clmf : time - weighted model for the hm part and non - time - weighted for the i m part ( fig . [ fig_mcs ] , top ) . the i m slope is very similar to the ones found in the considered individual mc complexes while the hm slope is about the salpeter value , like in orion b ( sect . [ sect_orion_b ] ) . the discrepancy with the data for @xmath198 @xmath7 could be attributed to incompleteness of the sample in this mass range . a salpeter - like slope is to be expected for mass distributions of ( nearly ) virialized clumps assuming their formation at a constant rate . the effect will be similar if the clmf is derived from a statistically significant sample containing clumps in complexes at different evolutionary stages of clump formation . note also that the variation of the velocity scaling index @xmath61 in the fitting models is consistent with most observational and numerical works ( e.g. * ? ? ? * ; * ? ? ? the values of the @xmath78 parameter indicate preliminary solenoidal turbulent forcing with small contributions of the compressive mode . considering the choice ii ) , an upper mass limit @xmath199 @xmath7 is adopted as mentioned above . the model case ` wkin2 ` provides an excellent fit of the intermediate - mass clmf and a problematic one for the high - mass clmf . again , the completeness of the clump mass bin @xmath200 @xmath7 is an open issue which is crucial for the correct slope estimation . on the other hand , the obtained low velocity scaling index @xmath61 is consistent with results for taurus , orion a and orion b. modeling of the observational clmf from the equipartition cases ` wkin2 ` and ` wkin2th2 ` suggests that most of the clumps generated in mc complexes of diverse star - forming activity and with mean density about the typical local density ( cf . eq . 7 in paper i ) are in a state close to virial equilibrium . an additional test for adequacy of the proposed statistical modeling of the clmf is to construct clump mass - size diagrams . the comparison between the used observational data and the best - fit models of the observational clmfs is displayed in fig . [ fig_ml_mr ] . there are two difficulties with such analysis . first , the ` average clump ensembles ' in our model occupy narrow strips on @xmath201 diagrams to be compared with the radii of observed clumps . ] whereas the dispersion of the observational samples is huge . second , mass and ( especially ) size determinations from molecular - line maps depend essentially on the clump - finding technique which is different in each considered reference work ( table [ table_mc ] , column 10 ) . a meaningful criterion is to compare the slope @xmath202 on the observational @xmath201 diagram with the predicted slopes @xmath203 as the latter vary with the chosen equipartition case and the velocity scaling index @xmath61 ( cf . paper i , fig . the predicted slopes from the best - fit clmf model are consistent with the ones from the referred molecular - line studies within @xmath204 limit for taurus and orion a and within about @xmath205 limit for orion b and for the sample of @xcite . in two cases , the ` average clump ensemble ' strips cross the region of the observational data ( fig . [ fig_ml_mr ] , bottom ) . taking into account the good agreement between observational and modeled clmfs ( figs . [ fig_taurus]-[fig_mcs ] ) , the shift between the two sets on the @xmath201 diagrams with a factor of 2 toward larger ( orion a ) or smaller sizes , is to be interpreted with the variety of the clump _ size _ definition . the issue will be commented further in the next section wherein we compare our model with clmfs from dust continuum studies and in the discussion . sensitive submillimeter continuum maps are appropriate for probing the mc structure on small scales due to the optically thin dust emission . in the last decade , the achieved angular resolutions of @xmath206 from ground - based observations ( e.g. scuba , bolocam ) and space missions like herschel allowed for study of clumps with sizes @xmath207 pc in complexes at distances over @xmath208 kpc . therefore it is instructive to compare results from such studies with our model . in view of the complexes remoteness , the considered regions were not included in the work of lal10 and we do nt have at our disposal their dust extinction mapping as a tool to trace and fit the general cloud structure ( cf . figs . [ fig_taurus]-[fig_orib ] , left ) . to compensate this lack , we chose star - forming regions which clump population has been probed also by use of molecular - line maps . the observational clmf was fitted directly and the clump mass - size diagram was used as an additional test . large part of this massive star - forming region was mapped by @xcite in two scuba bands . we use here the clump data from their @xmath209 m map because of the richer statistics . the fitting of the observational clmf and the corresponding clump - mass size diagram are plotted in fig . [ fig_m17 ] ( top ) . a very good fit of the clmf is achieved from the model case ` wkin4 ` that points to strongly gravitationally bound clumps . the predicted and the observational clump mass - size diagrams are also in agreement ( fig . [ fig_m17 ] , right top ) . note that we assumed a higher gas temperature @xmath185 k which is about the average clump temperature derived by the authors . although @xcite suggest a two power - law clmf with @xmath210 @xmath7 , it seems that their result hints rather at an order - of - magnitude higher characteristic mass . this is supported by the slope in the mass range @xmath211 @xmath7 which is similar to the i m slopes found from the referred molecular - line studies ( cf . table [ table_mc ] ) . however , only two clumps with masses @xmath212 @xmath7 in the observational sample are far from sufficient for a plausible estimate of the characteristic mass . to clarify the issue , we apply our method to a c@xmath11o study of the south - western sector of m17 @xcite . as demonstrated in fig . [ fig_m17 ] , bottom , the results for an intermediate - mass clmf with @xmath213 @xmath7 are virtually the same , fitted from the same model case and free parameter values and slightly higher temperature . moreover , the high - mass clmf is also well fitted by the time - weighted mass distribution of gravitationally unstable clumps . neglecting the mass bin @xmath214msol , which contains only 2 clumps , the slope @xmath39 is shallower than that of the stellar imf . for independent confirmation of this result , we composed a twice larger sample , including clump data from a similar study of ngc7538 @xcite . as shown in fig . [ fig_m17_ngc7538 ] , the clmf is fitted again from the model case ` wkin4 ` and a similar parameter set @xmath215 for a large range of clump masses @xmath216 @xmath7 , applying time - weighting for the more massive clumps . in comparison to the mass distribution from the m17 sample , the i m slope shallows slightly and approaches the values from the molecular - line studies of nearby mc complexes ( cf . table [ table_mc ] ) while the hm slope is much steeper than that from stuzki & gsten s study and is identical with the one in orion a ( fig . [ fig_oria ] ) although with a smaller @xmath20 . the agreement of the model with the clump data on the @xmath201 diagram is again excellent ( fig . [ fig_m17_ngc7538 ] , right ) . the abundant observational studies of this complex in the outer galaxy indicated active star formation in the past and nowadays . its clump population has been studied both on molecular - line and dust continuum maps ; we chose the works of @xcite and @xcite , respectively . the high - quality herschel data used by the latter authors allow for derivation of the clmf in a mass range spanning three orders of magnitude ( fig . [ fig_rosette ] , top left ) . some incompleteness is sensible in the mass bins @xmath217 @xmath7 and has in our view , two possible explanations . first , for the sake of our study we selected only starless clumps from the original sample . second , @xcite applied the getsources algorithm @xcite for clump decomposition which favors identification of more compact ( and less massive ) objects . because of the lower mass - limit of confidence , our statistical method is not able to fit the clmf below @xmath218 @xmath7 . the observed mass distribution for larger masses could be interpreted as single power - law clmf with a slope about @xmath219 ( not shown ) . in view of the possible incompleteness of the high - mass clump data , we suggest rather a two power - law mass function , fitted from the model case ` wkin2 ` ( fig . [ fig_rosette ] , top left ) . the characteristic mass for ` wkin2 ` and for the obtained best - fit parameters is @xmath220 @xmath7 ( see fig . 3 in paper i ) . adopting this value for @xmath20 , one gets a bit steeper intermediate - mass clmf than found for other mc complexes in this paper and a high - mass clmf , with a slope typical for fractal clouds @xcite . on the clump mass - size diagram , a shift of the model from the observed clumps is evident like in some other studied regions : by a factor of 2 to 4 ( fig . [ fig_rosette ] , top right ; cf . [ fig_ml_mr ] ) . nicola schneider ( private communication ) provided for us estimates of the cloud mass @xmath221 as a function of spatial scale @xmath41 , calculated from the original herschel column density map of rosette . that enabled a check of the derived best - fit model with the observed cloud structure , in terms of the lal10 work ( figs . [ fig_taurus]-[fig_orib ] , left ) . the shape of the modeled curve @xmath95 is consistent with the dust - continuum data but with a shift toward lower @xmath222 . decreasing the mapping resolution parameter down to @xmath223 , we found excellent agreement with the data ( fig . [ fig_rosette ] , right , embedded diagram ) . that is illustrative how the observed general cloud structure affects the @xmath85-parameter space of the clmf fitting . @xcite derived the observational clmf from co mapping of rosette mc . as seen in fig . [ fig_rosette ] ( bottom ) , their clump size range practically does not overlap with the one of @xcite while the mass range spans from @xmath224 to @xmath225 @xmath7 . the clmf is well fitted from the model case ` wkin2mag ` and a very ` soft ' velocity scaling , @xmath226 . such values of the scaling index were found from magneto - hydrodynamic simulations of @xcite . the comparison with the considered molecular - line and dust - continuum studies of galactic mc complexes demonstrates the applicability of a statistical approach to derive the clmf presented in paper i. the results on fitting the observational clmf through our model are summarized in table [ table_fit ] . they are obtained by use of two different primary fitting criteria : fitting of the general cloud structure @xmath227 or direct fitting of the clmf . regardless of the type of observational data , the applied fitting criterion and of the variety of model cases , the best - fit parameters span relatively narrow ranges : * ` soft ' velocity scaling : @xmath228 * mainly solenoidal forcing : @xmath229 * mapping resolution from few percent to one tenth of the spatial scale : @xmath230 * typical temperatures for molecular gas phase : @xmath231 k. a small velocity scaling index @xmath61 , close or equal to the kolmogorov value for incompressible turbulence ( @xmath79 ) , may seem unrealistic in view of the high compressibility of interstellar turbulence which implies @xmath232 . in fact , numerical simulations of magnetized clouds show that the velocity power spectrum can be even shallower than in the kolmogorov theory . @xcite measured @xmath233 for thermal - to - magnetic pressure ratio in range @xmath234 . ( in our modeling , the latter value could be as low as @xmath235 in case ` wkin2mag ` ; see eq . [ eq_mag_scaling ] . ) apparently , the velocity scaling index for incompressible turbulence @xmath236 should not be necessarily treated as the lowest possible value from theoretical considerations . the best - fit range of the turbulent forcing parameter @xmath78 is narrower , with typical values @xmath237 , in most considered complexes . since @xmath78 can vary within a star - forming region , this result might be explained as a statistical effect . indeed , @xmath238 corresponds to a natural mixture between solenoidal and compressive modes as the latter represent longitudinal waves , occupying one of the three spatial dimensions ( see * ? ? ? * and fig . 8 there ) . the typical best - fit values of the mapping resolution parameter @xmath85 are about several percent . these are the expected values , appropriate to distinguish substructures which are significantly smaller than the spatial scale @xmath41 and significantly larger than the scale of dissipation . generally , our results lend support to a two power - law shape of the clmf : intermediate - mass and high - mass part , with two distinct values of the characteristic mass @xmath20 : @xmath143 and @xmath239 @xmath7 . ( the molecular - line study of rosette mc with its high @xmath20 is the only exception . ) the first value is consistent with model cases of a `` virial - like '' equipartition between gravitational and turbulent energy , possibly with contribution of the thermal energy at small scales and negligible contribution of the magnetic energy ( ` wkin2 ` , ` wkin2th2 ` ) . these equipartition relations are not sufficient to argue that the modeled clumps are in virial equilibrium @xcite but rather indicate their gravitational boundedness or contraction . on the other hand , if the considered model cases hold for a whole cloud , they are indicative for its global collapse @xcite . large characteristic masses of hundreds @xmath7 are obtained for strongly gravitating clumps , with possible magnetic support ( ` wkin4 ` , ` wkin2mag ` ) . such are evidently the cases in taurus , orion a , m17 and ngc7538 . the slope of the modeled intermediate - mass clmf is shallow and does not vary significantly from complex to complex ( @xmath240 ) while the variety is larger for the high - mass one : from typical slopes for fractal clouds to slopes , a bit steeper than the one of the stellar imf ( @xmath241 ) . it seems that the single power - law observational clmfs of slope @xmath242 derived from molecular - line studies in 1990s @xcite have been products of a combination between intermediate- and a few bins from high - mass clmf . on the other hand , the variety of high - mass clmf slopes probably reflects a real variety of physical conditions in individual complexes and the gravitational balance and evolution of the dense fragments in them . its model slope @xmath243 similar to that of the stellar imf when the clmf is time - weighted . when derived from molecular - line mapping of orion a @xcite or the dust - continuum studies of m17 and ngc7538 by @xcite , @xmath244 exceeds noticeably the salpeter value but lies still within the observed variability range @xcite . the dust extinction mapping of taurus , orion a and orion b by lal10 enabled us to make a link between the predicted general cloud structure and the clmf . we point out also the excellent agreement applying this criterion to the dust - continuum study of rosette , making use of a column density map of rosette ( fig . [ fig_rosette ] , top right , embedded ) obtained from herschel data ( see * ? ? ? * for details ) . ( note that the extinction map derived from near - ir extinction using 2mass @xcite delivers similar values for @xmath245 . ) generally , good agreement is found for the chosen scales of clump generation @xmath41 between @xmath246 and @xmath247 pc while in taurus the upper limit is about half of that . those are the conservative limits of the inertial range of turbulence , i ) ] that is a corner stone in our framework of clump description . the discrepancies for larger @xmath41 are to be expected and can be interpreted as reflecting changes in the physical conditions such scales may lie outside the inertial range and/or the assumption of isothermality does not hold for them ( e.g. * ? ? ? * ) . on the other hand , at @xmath248 pc the density pdf deviates from the lognormal shape and develops a power law tail in the high - density part @xcite . the predicted typical clump size in such density regimes approaches @xmath249 pc which is about the sonic scale for temperature range @xmath250 k , i.e. the assumption for supersonic turbulence as a generator of clumps breaks down . @l@ c@ c@ c@ c@ c@ c@ c@ c@ c@ c@ c@ c sf region & fc & mc & & & + & & & @xmath61 & @xmath78 & @xmath85 & @xmath99 [ k ] & @xmath251 & @xmath252 & @xmath244 & @xmath243 & @xmath253 & @xmath203 + + taurus & @xmath254 & ` wkin4 ` & 0.30 - 0.36 & 0.38 - 0.42 & 0.07 - 0.09 & 10 - 20 & @xmath255 & @xmath256 & & & @xmath257 & @xmath258 + orion a & @xmath254 & ` wkin2mag ` & 0.38 - 0.42 & 0.38 - 0.42 & 0.07 - 0.09 & 18 - 22 & @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath263 & @xmath264 + orion b & @xmath254 & ` wkin2 ` & 0.30 - 0.36 & 0.30 - 0.36 & 0.055 - 0.065 & 10 - 15 & @xmath265 & & @xmath266 & @xmath267 & @xmath268 & @xmath269 + & @xmath254 & ` wkin2th2 ` & 0.30 - 0.36 & 0.30 - 0.36 & 0.050 - 0.065 & 10 - 15 & @xmath265 & & @xmath266 & @xmath270 & @xmath268 & @xmath271 + mc sample & clmf & ` wkin2 ` & 0.30 - 0.36 & 0.42 - 0.46 & 0.02 - 0.10 & 10 - 12 & @xmath272 & @xmath273 & @xmath274 & @xmath275 & @xmath276 & @xmath271 + & clmf & ` wkin2th2 ` & 0.42 - 0.46 & 0.36 - 0.40 & 0.02 - 0.10 & 8 - 10 & @xmath277 & @xmath278 & @xmath279 & @xmath280 & @xmath276 & @xmath281 + * m17 * & clmf & ` wkin4 ` & 0.43 - 0.45 & 0.37 - 0.40 & 0.02 - 0.10 & 28 - 35 & @xmath282 & @xmath283 & @xmath284 & @xmath285 & @xmath286 & @xmath287 + * rosette * & clmf & ` wkin2mag ` & 0.18 - 0.23 & 0.30 - 0.36 & 0.02 - 0.10 & 8 - 15 & @xmath288 & @xmath289 & @xmath290 & @xmath291 & @xmath292 & @xmath293 + + * m17 * & clmf & ` wkin4 ` & 0.40 - 0.44 & 0.37 - 0.40 & 0.02 - 0.10 & 28 - 35 & @xmath294 & @xmath295 & & & @xmath296 & @xmath297 + * m17*+ & clmf & ` wkin4 ` & 0.43 - 0.46 & 0.38 - 0.42 & 0.02 - 0.10 & 28 - 35 & @xmath298 & @xmath299 & @xmath300 & @xmath301 & @xmath302 & @xmath303 + * rosette * & clmf & ` wkin2 ` & 0.44 - 0.46 & 0.41 - 0.45 & 0.02 - 0.03 & 18 - 25 & @xmath304 & @xmath305 & @xmath306 & @xmath307 & @xmath308 & @xmath309 + & ( @xmath254 ) & & & & & & & & & & & + probed parameter spaces : @xmath310 ; @xmath311 ; @xmath312 ; @xmath313 k + inclusion of data from dust - continuum studies in the analysis of the clump mass - size ( @xmath201 ) diagrams confirms the results from molecular - line mappings ( sect . [ mr_diagrams ] ) : consistency between the slopes within 2@xmath314 and , occasionally , a shift by a factor of up to 3 on the size axis . generally , the agreement between the locations of the ` average clump ensemble ' and the observed clumps is better ; especially , for the combined clump sample from m17 and ngc7538 . the discrepancies on the @xmath201 diagrams could be attributed to the various clump - finding techniques used in the referred observational studies ( cf . table [ table_mc ] , column 10 ) and hence to different clump size definitions . the latter would not lead to substantially different mass estimates if a typical clump structure is a dense central region and diffuse outer shell . a careful comparative analysis of the basic clump - finding algorithms , applied to identical observational or numerical datasets , is necessary for a more comparison between our statistical clump description and the observed clumps in mc complexes . the role of the magnetic fields in shaping the physical characteristics of mcs and of their substructures can be accounted for more thoroughly in a further extension of this work . @xcite proposed an analytical model of the relation between the width of the pdf @xmath314 in magnetized turbulent medium . it includes a modification of equation [ eq_sigma_pdf ] , consistent with a scaling law of magnetic field @xmath57 ( equation [ eq_mag_scaling ] ) and can be easily incorporated in our modeling . finally we caution the reader that the predictions of our model hold for less evolved clumps of sizes @xmath315 pc and must not be compared with observational mass distributions of dense ( prestellar ) cores . the latter ones will be subject of another study which considers the high - density power - law tail of the density pdf ( cf . * ; * ? ? ? * ) in active star forming regions as a basis for cores description . the essence of this approach is presented in @xcite while the model is in process of development . such study may answer the question whether the difference between the clump mass function and the core mass function is physical or statistical . our statistical approach for physical description of condensations ( clumps ) formed through a turbulent cascade during the early mc evolution predicts : i ) the cloud structure in terms of effective size - mass scaling relations , through the mass - density exponent @xmath222 as a function of the spatial scale @xmath41 ( dvk11 ) ; and ii ) the composite clump mass function ( clmf ; paper i ) . different models within this framework are generated by choosing an appropriate energy equipartition relation and a set of 4 free parameters : velocity scaling index @xmath61 , turbulent forcing parameter @xmath78 , mapping resolution parameter @xmath85 and temperature @xmath99 . in this paper we compared clmfs from molecular - line and dust - continuum studies of galactic cloud complexes with ones derived from our model applying alternative fitting criteria : fitting the structure @xmath95 of individual complexes ( when additional dust - extinction data are available ) or direct fitting of the observational clmf . both fitting criteria lead to modeled clump mass distributions , in good or excellent agreement with the considered observational data . the equipartition relations which yield these fits indicate gravitational boundedness of the dominant clump population in the considered clouds , possibly including the contribution of magnetic or thermal energy : model cases ` wkin2 ` , ` wkin2th2 ` or ` wkin2mag ` . on the other hand , the results for taurus , m17 and ngc7538 rather hint at strongly gravitating or contracting inner parts of those complexes ( model case ` wkin4 ` ) . the derived best - fit values of the parameters @xmath61 , @xmath78 and @xmath99 for all studied individual clouds span relatively narrow ranges . in most mc complexes the typical velocity scaling index is found to be similar to the original `` larson s first law '' ( eq . [ eq_larson_1 ] , @xmath316 ) or to @xmath317 testified from recent observations @xcite . the best - fit values of the turbulent forcing parameter concentrate around @xmath238 which corresponds to a natural mixture between compressive and solenoidal modes @xcite . the modeled clump mass distributions support a clmf which might be represented as a combination of two - power law functions . the latter are separated by a characteristic mass @xmath20 that varies typically within one order of magnitude or more : from about ten to hundreds solar masses . the slope of the intermediate - mass clmf is shallow and nearly constant : @xmath0 . the high - mass part of the clmf corresponds to gravitationally unstable clumps in all considered model cases and hence a more appropriate description should take into account the dynamical clump evolution . such description is achieved through time - weighting of the clump mass distribution . ( the effect will be similar if the clmf is derived from a statistically significant sample containing clumps in mc complexes at different evolutionary stages of clump formation . ) we obtained slopes within a broader range @xmath318 that includes the typical value for fractal clouds @xmath219 @xcite as well that of the stellar initial mass function @xcite . comparison between the observational and the modeled clump mass - size diagrams reveals agreement of the slopes within the @xmath205 limit and , in most cases , a systematic shift toward smaller or larger clump sizes . the latter is to be attributed to variety of size definitions used in the different clump - finding techniques . _ acknowledgement : _ we are grateful to the anonymous referee for his / her insightful comments and recommendations which helped us to improve substantially this paper . we thank j. di francesco for the list of clump properties in rosette mc derived from herschel data , and n. schneider for providing us the mass / spatial scale values shown in fig . 8 from the same herschel column density map . we appreciate the help of v. ossenkopf and r. simon who provided for us data on the individual clump characteristics in orion b from the original study of @xcite . alves , j. , lombardi , m. , lada , c. , 2007 , a&a , 462 , l17 ballesteros - paredes , j. , & vzquez - semadeni , e. , 1995 , revmexaa , ser . conf . , 3 , 105 ballesteros - paredes , j. , 2006 , mnras , 372 , 443 banerjee , r. , vzquez - semadeni , e. , hennebelle , p. , klessen , r. s. , 2009 , mnras , 398 , 1082 bergin , e. , tafalla , m. , 2007 , ara&a , 45 , 339 blitz , l. , 1993 , in : levy , e. , lunine , j. , eds . , protostars and planets iii . the university of arizona press , tuscon & london collins , d. , kritsuk , a. , padoan , p. , li , h. , xu , h. , ustyugov , s. , norman , m. , 2012 , apj , 750 , 13 crutcher , r. , 1999 , apj , 520 , 706 di francesco , j. , sadavoy , s. , motte , f. , schneider , n. , hennemann , m. , csengeri , t. , bontemps , s. , balog , z. , et al . , 2010 , a&a , 518 , l91 donkov , s. , stanchev , o. , veltchev , t. , 2012 , proc . of the viii serbian - bulgarian astron . , leskovac , serbia , may 8 - 12 , 2012 , eds . m. k. tsvetkov , m. s. dimitrijevic , k. tsvetkova , o. kounchev , z. mijajlovic ( arxiv 1206.1444 ) donkov , s. , veltchev , t. , klessen , r. s. , 2011 , mnras , 418 , 916 ( dvk11 ) donkov , s. , veltchev , t. , klessen , r. s. , 2012 , mnras , 423 , 889 ( paper i ) elmegreen , b. g. , 1997 , apj , 486 , 944 enoch , m. , evans , n. , sargent , a. , glenn , j. , rosolowsky , e. , myers , p. , 2008 , apj , 684 , 1240 federrath , c. , klessen , r. , schmidt , w. , 2008 , apj , 688 , l79 federrath , c. , roman - duval , j. , klessen , r. , schmidt , w. , mac low , m .- , 2010 , a&a , 512 , 81 hara , a. , tachihara , k. , mizuno , a. , onishi , t. , kawamura , a. , obayashi , a. , fukui , y. , 1999 , pasj , 51 , 895 heithausen , a. , bensch , f. , stutzki , j. , falgarone , e. , panis , j. , 1998 , a&a , 331l , 65 hennebelle , p. , banerjee , r. , vzquez - semadeni , e. , klessen , r. s. , audit , e. , 2008 , a&a , 486 , l43 ikeda , n. , sunada , k. , kitamura , y. , 2007 , apj , 665 , 1194 ikeda , n. , kitamura , y. , sunada , k. , 2009 , apj , 691 , 1560 johnstone , d. , fich , m. , mitchell , g. , moriarty - schieven , g. , 2001 , apj , 559 , 307 johnstone , d. , matthews , h. , mitchell , g. , 2006 , apj , 639 , 259 kainulainen , j. , beuther , h. , henning , t. , plume , r. , 2009 , a&a , 508 , l35 kenyon , s. , gmez , m. , whitney , b. , 2008 , in handbook of star forming regions , vol . b. reipurth ( san francisco , ca : asp ) , 405 kerton , c. , martin , p. , johnstone , d. , ballantyne , d. , 2001 , apj , 552 , 601 klessen , r. s. , 2000 , apj , 535 , 869 klessen , r. s. , & hennebelle , p. , 2010 , a&a , 520 , a17 knyves , v. , andr , p. , menshchikov , a. , schneider , n. , arzoumanian , d. , bontemps , s. , attard , m. , motte , f. , et al . , 2010 , a&a , 518 , l106 kramer , c. , stutzki , j. , rohrig , r. , corneliussen , u. , 1998 , a&a , 329 , 249 kritsuk , a. , norman , m. , padoan , p. , & wagner , r. , 2007 , apj , 665 , 416 kritsuk , a. , norman , m. , & wagner , r. , 2011 , apj , 727 , l20 kroupa , p. , 2001 , mnras , 322 , 231 larson , r. , 1981 , mnras , 194 , 809 lombardi , m. , alves , j. , lada , c. , 2010 , a&a , 519 , 7 ( lal10 ) menshchikov , a. , andr , p. , didelon , p. , motte , f. , hennemann , m. , schneider , n. , 2012 , a&a , 542 , 81 molina , f. , glover , s. , federrath , c. klessen , r. , 2012 , mnras , 423 , 2680 motte , f. , andr , p. , neri , r. , 1998 , a&a , 336 , 150 motte , f. , andr , p. , 2001 , asp conf . series , vol . 243 , eds . t. montmerle & p. andr , 301 nutter , d. , ward - thompson , d. , 2007 , mnras , 374 , 1413 onishi , t. , mizuno , a. , kawamura , a. , ogawa , h. , fukui , y. , 1996 , apj , 465 , 815 onishi , t. , mizuno , a. , kawamura , a. , tachihara , k. , fukui , y. , 2002 , apj , 575 , 950 padoan , p. , juvela , m. , kritsuk , a. , norman , m. , 2006 , apj , 653 , l125 padoan , p. , juvela , m. , kritsuk , a. , norman , m. , 2009 , apj , 707 , l153 passot , t. , vzquez - semadeni , e. , & pouquet , a. , 1995 , apj , 455 , 536 reid , m. , wilson , c. , 2005 , apj , 625 , 891 reid , m. , wilson , c. , 2006 , apj , 644 , 990 reid , m. , wilson , c. , 2006 , apj , 650 , 970 salpeter , e. , 1955 , apj , 121 , 161 schneider , n. , bontemps , s. , simon , r. , ossenkopf , v. , federrath , c. , klessen , r. s. , motte , f. , andr , ph . , 2011 , a&a , 529 , 1 schneider , n. , csengeri , t. , hennemann , m. , motte , f. , didelon , p. , federrath , c. , bontemps , s. , di francesco , j. , et al . , 2012 , a&a , 540 , 11 shetty , r. , collins , d. , kauffmann , j. , goodman , a. , rosolowsky , e. , & norman , m. , 2010 , apj , 712 , 1049 stanke , t. , smith , m. , gredel , r. , khanzadyan , t. , 2006 , a&a , 447 , 609 stutzki , j. , & gsten , r. , 1990 , apj , 356 , 513 tachihara , k. , mizuno , a. , fukui , y. , 2000 , apj , 528 , 817 tachihara , k. , onishi , t. , mizuno , a. , fukui , y. , 2002 , a&a , 385 , 909 tatematsu , k. , umemoto , t. , kameya , o. , et al . , 1993 , apj , 404 , 643 testi , l. , sargent , a. , 1998 , apj , 508 , l91 vzquez - semadeni , e. , gmez , g. , jappsen , a. , ballesteros - paredes , j. , gonzlez , r. , klessen , r. s. , 2007 , apj , 657 , 870 williams , j. , blitz , l. , stark , a. , 1995 , apj , 451 , 252

the standard post - newtonian approximation to gravitational waveforms , called t - approximants , from non - spinning black hole binaries are known not to be sufficiently accurate close to the last stable orbit of the system . a new approximation , called p - approximants , is believed to improve the accuracy of the waveforms rendering them applicable up to the last stable orbit . in this study we apply p - approximants to the case of a test - particle in equatorial orbit around a kerr black hole parameterized by a spin parameter that takes values between and in order to assess the performance of the two approximants we measure their _ effectualness _ ( i.e. larger overlaps with the exact signal ) , and _ faithfulness _ ( i.e. smaller biases while measuring the parameters of the signal ) with the exact ( numerical ) waveforms . we find that in the case of prograde orbits , that is orbits whose angular momentum is in the same sense as the spin angular momentum of the black hole , t - approximant templates obtain an effectualness of for spins for the effectualness drops to about 0.82 . the p - approximants achieve effectualness of for all spins up to the bias in the estimation of parameters is much lower in the case of p - approximants than t - approximants . we find that p - approximants are both effectual and faithful and should be more effective than t - approximants as a detection template family when for both t- and p - approximants perform equally well so that either of them could be used as a detection template family .

gr-qc0406038

stellar mass compact binaries consisting of double neutron stars ( ns ) , double black holes ( bh ) or a mixed binary consisting of a neutron star and a black hole , are the primary targets for a direct first detection of gravitational waves ( gw ) by interferometric detectors , ligo @xcite , virgo @xcite , geo600 @xcite , and tama @xcite . under radiation reaction the orbit of a binary slowly decays , emitting a signal whose amplitude and frequency increases with time and is termed a `` chirp '' signal . while it is believed that there is a greater population of ns - ns binaries @xcite , it is the bh - bh binaries that are the strongest candidates for detection since they can be seen from a greater volume , about two orders - of - magnitude greater than ns - ns binaries @xcite . in order to detect such sources one employs the method of matched filtering @xcite . briefly , the method works as follows : firstly , one creates a set of waveforms , or templates as they are called , that depend on a number of parameters of the source and its location and orientation relative to the detector . these templates are then cross - correlated with the detector output weighted by the inverse of the noise spectral density . if a signal , whose parameters are close to one of the template waveforms , is actually present in the detector output then the cross - correlation builds up , with the dominant contribution coming from frequencies where the noise spectral density is low . thus , in the presence of a sufficiently strong signal the correlation will be much larger than the rms correlation in the absence of any signal . how large should it be before we can be confident about the presence of a signal depends on the combination of the rate of inspiral events and the false alarm probability ( see e.g. ref . @xcite for a simple estimation ) . the effectiveness of matched filtering depends on how well the phase evolution of the waveform is known . even tiny instantaneous differences , as low as one part in @xmath10 in the phase of the true signal that might be present in the detector output and the template that is used to dig it out could lead to a cumulative difference of several radians since one integrates over several hundreds to several thousands of cycles . in view of improving the signal - to - noise ratio for inspiral events there has been a world - wide effort in accurately computing the dynamics of a compact binary and the waveform it emits or to use phenomenologically defined detection template families @xcite . there have been parallel efforts on using two different approximation schemes : on the one hand the post - newtonian ( pn ) expansion of einstein s equations has been used to treat the dynamics of two bodies of comparable masses with and without spin , in orbit around each other . this approximation is applicable when the velocities involved in the system are small but there is no restriction on the ratio of the masses @xcite . on the other hand , black hole perturbation theory has been used to compute the dynamics of a test particle in orbit around a spin - less or spinning black hole . black hole perturbation theory does not make any assumptions on the velocity of the components , but is valid only in the limit when the mass of one of the bodies is much less than the other @xcite . the post - newtonian approximation is a perturbative method which expands the equations of motion , binding energy and gw flux as a power series in @xmath11 , where @xmath12 is a typical velocity in the system and @xmath13 is the speed of light . in the early stages of an inspiral , the radiation reaction time - scale @xmath14 where @xmath15 is the angular velocity and @xmath16 its time - derivative , is much greater than the orbital time - scale @xmath17 . it is during this adiabatic regime that the post - newtonian approximation works best . at present , the pn expansion for the case of comparable - masses is known to order @xmath18 @xcite and @xmath19 @xcite , for the energy and flux functions , respectively . however , at this order an arbitrary parameter exists in the expression for the flux . in order to see how well pn theory performs , we can compare two different systems . if we assume a ns - ns binary of masses ( 1.4,1.4 ) @xmath20 and a lower frequency cutoff of the detector at 40 hz , then the orbital velocity " of the binary is small , @xmath21 , ] when it enters the detector bandwidth and the two stars are still largely separated , @xmath22 . the ratio of time - scales in the most sensitive regime of the detector is in the range @xmath23 . if on the other hand we take a bh - bh binary of masses ( 10,10)@xmath20 , the orbital velocity is quite large , @xmath24 and the separation is quite small , @xmath25 upon entering the detector bandwidth . this is very close to the regime , @xmath26 , @xmath27 , where the background curvature becomes strong and the motion relativistic . once again , comparing time - scales , we obtain @xmath28 , where the final value is taken at the last stable orbit at @xmath29 hz . it is known that pn theory becomes inaccurate at an orbital separation of @xmath30 @xcite . therefore , post - newtonian approximation becomes less valid for higher mass systems in the ligo band but well describes the early stages of the inspiral of a ns - ns system visible in ligo . as previously stated , black hole perturbation theory makes no assumptions about the orbital velocity of the components , but does restrict their masses . one assumes that a test particle of mass @xmath31 is in orbit about a central bh of mass @xmath32 such that @xmath33 . assuming this restriction is satisfied we have an analytical expression for the energy . however , no analytical expression has been worked out for the gravitational wave flux emitted by such a system . using black hole perturbation theory , a series approximation was initially calculated to @xmath34 by poisson for a test particle in circular orbit around a schwarzschild black hole @xcite . the series was extended numerically to @xmath35 by cutler et al . @xcite , and then to @xmath36 by tagoshi and nakamura @xcite and confirmed analytically by tagoshi and sasaki @xcite . the most recent progress is an extension of the series to @xmath37 by tagoshi , tanaka and sasaki @xcite . for a test particle in circular orbit about a kerr black hole , the initial progress was again made by poisson @xcite . the series approximation was improved from @xmath38 to @xmath35 , and subsequently to @xmath39 by tagoshi , tanaka , shibata and sasaki @xcite . several authors @xcite have shown that the convergence of both post - newtonian approximation and black hole perturbation theory is too slow to be useful in constructing accurate templates . more recently , damour , iyer and sathyaprakash ( hereafter dis ) showed for the case of a test - mass in orbit about a schwarzschild bh , that by using properly defined energy and flux functions that have better analytical properties , combined with pad techniques , it was possible to take the existing series expansion and improve its convergence properties @xcite . the new approximation in which pad approximants of new energy and flux functions are used to derive improved templates is called p - approximant . using a fiducially defined `` exact '' waveform , it was shown that the p - approximant templates were both more _ effectual _ ( i.e. larger overlaps with the exact waveform ) and _ faithful _ ( i.e. larger overlaps with the exact waveform and lower biases in the estimation of parameters ) than the corresponding post - newtonian ( hereafter t - approximant ) templates . while in general , more templates are needed for p - approximant templates to cover the same volume of parameter space @xcite , the extra computational cost is preferred for the increased performance in p - approximants . the failure of the pn expansion to converge sufficiently quickly in the case of a test particle orbiting a schwarzschild bh @xcite does not bode well for the modelling of even a test particle inspiralling into a kerr bh . another motivation for this work is that at present there is effort to use _ effective one body _ ( eob ) waveforms @xcite to detect the inspiral and merger signals from two comparable - mass kerr bhs in gw data . as the eob waveforms are based partially on p - approximants , any development and concretization of the benefits of p - approximant templates will boost our confidence in using these improved waveforms as detection templates . in this paper we will extend the p - approximant technique to the case of a test particle orbiting a kerr black hole . the reason for focusing on test - mass systems is that we can use the exact numerical fluxes @xcite from black hole perturbation theory with which to compare our results and thereby reliably demonstrate the usefulness of the technique . we begin in sec . [ sec : waveform ] with a summary of the matched filtering and the signal - to - noise ratio achieved by the first generation of geo , ligo and virgo interferometers for spinning black hole binaries . we then go on to discuss in sec . [ sec : energyandflux ] the current state - of - the - art in our understanding of the evolution of a test particle in orbit around a kerr black hole . in particular , we shall discuss the time - evolution of the orbital energy and gravitational wave flux as a function of the spin of the central black hole at various post - newtonian orders , and the locations of the last stable and unstable circular orbits . we shall see that the post - newtonian expansion of the flux does not show a regular behaviour as we move from low to high orders in the post - newtonian expansion , becoming worse for more rapidly spinning black holes . in order to improve the convergence properties of the flux function , in sec . [ sec : fluxmodel ] we shall introduce a modified form of the flux function and its pad approximant . we shall demonstrate in sec . [ sec : results ] the improved behaviour of the pad approximant , at first graphically and then by showing that the overlaps , of the inspiral waveform based on it with the exact waveform , are close to unity . we shall use a number of different test systems in our comparison : these will range from systems with dissimilar masses , such as a ns - bh binary , to comparable - mass systems , such as a bh - bh binary . to deal with the comparable mass systems , we shall introduce in the energy and flux functions a term dependent on the symmetric mass ratio @xmath40 while not being entirely consistent , because of the fact that no finite mass correction terms are included , it allows us to examine the performance of various templates as we move from the test - mass systems to comparable - mass systems . the emphasis of the current paper is also on the estimation of parameters . we have carried out a detailed study of how good t- and p - approximants are in measuring the parameters of the binary . we shall show in sec . [ sec : results ] that as a result of not being _ effectual _ representations of the exact signal , t - approximants also turn out not to be _ faithful _ representations either . in other words , the systematic error in the estimation of the parameters caused by the wrong phasing of the signal is much larger in the case of t - approximants than in the case of p - approximants . in summary , p - approximants are not only _ effectual _ but they are _ faithful _ too as in the case of non - spinning black hole binaries . in this section we will discuss the nature of the post - newtonian waveform from an inpiralling compact binary and the response of an antenna to such a signal . we will then use the fourier transform of the waveform to compute the signal - to - noise ratio ( snr ) that can be achieved for these signals when they are detected using matched filtering . in the transverse - traceless gauge gravitational waves are represented by two polarisation amplitudes @xmath41 and @xmath42 the response @xmath43 of an antenna to an incoming signal is expressed as a combination of the two polarisation states and the beam pattern functions @xmath44 and @xmath45 of the antenna as @xcite : @xmath46 for a wave from a binary of masses @xmath47 and @xmath48 ( total mass @xmath49 and mass ratio @xmath50 ) that is inclined with respect to the plane of the sky at an angle @xmath51 , propagating in the direction @xmath52 ( see ref . @xcite for exact definitions ) , frequency @xmath53 and polarization angle @xmath54 with respect to the antenna , the polarisation amplitudes , in the so - called _ restricted _ post - newtonian approximation @xcite , are @xmath55 with @xmath56 a velocity parameter , and the beam - pattern functions are @xmath57 using the above expressions for the beam - pattern functions and gravitational wave amplitudes the response takes the form @xmath58,\ ] ] where the amplitude coefficient @xmath59 and phase @xmath60 can be assumed to be constant for signals lasting for a short duration ( say , less than about 30 mins ) : @xmath61 . \label{kt8}\end{aligned}\ ] ] thus , gravitational wave antennas are not able to extract the two polarisations separately and the data analysis problem boils down to matching the time - varying phase @xmath62 and to a lesser extent the amplitude @xmath63 of the antenna response function . post - newtonian theory and the quadrupole formula applied to a binary give the relativistic binding energy @xmath64 per unit mass , and the flux of the waves @xmath65 as series expansions in the parameter @xmath66 once we have the binding energy and the flux we can use the energy balance argument , namely that the flux of gravitational waves is completely balanced by the negative rate of change of the binding energy [ @xmath67 , in order to arrive at a parametrized equation for the evolution of the phase @xmath68 of gravitational waves . integrating the energy balance equation supplemented by @xmath69 one obtains : @xmath70 where @xmath71 @xmath72 is a reference time and @xmath73 is a reference phase of the signal , when the velocity parameter is @xmath74 the numerical integration of the above equations is more economical when the following differential equations are solved instead , @xmath75 the parametric representation , eqs . ( [ eq : timevsv ] ) and ( [ eq : phivsv ] ) , of the phasing formula @xmath76 holds under the assumption of ` adiabatic inspiral ' , _ i.e. _ , that gravitational radiation damping can be treated as an adiabatic perturbation of a circular motion . however , the effective one - body approach @xcite has allowed a treatment of the radiation damping to proceed beyond the adiabatic approximation . in order to extract an inspiral signal that may be buried in noisy data by the method of matched filtering , we need to employ post - newtonian accurate representations for the two functions @xmath77 and @xmath65 that appear in the above phasing formulas . given an approximant @xmath78 , @xmath79 , one can define by replacing @xmath80 , @xmath81 in eqs . ( [ eq : timevsv ] ) and ( [ eq : phivsv ] ) some approximate parametric representation , @xmath82 , @xmath83 , and therefore a corresponding approximate template @xmath84 obtained by replacing @xmath85 , in the following @xmath85-parametric representation of the waveform @xmath86 by the function of time @xmath87 obtained by inverting @xmath82 . there are several ways of performing this inversion which leads to the different t - approximants , _ taylort1 _ [ keep the rational polynomial @xmath88 in eqs . ( [ eq : timevsv ] ) and ( [ eq : phivsv ] ) as is and solve the integrals numerically ] , _ taylort2 _ [ re - expand the rational polynomial as a post - newtonian series , truncate terms to the appropriate order and solve the integrals analytically to get series expansions , @xmath89 and @xmath90 but solve numerically for @xmath68 from the parametric equations ] and _ taylort3 _ [ do as in taylort2 but also invert the series expansions to obtain the phasing as an explicit function of time : @xmath91 , as discussed in ref . @xcite . it is often convenient to deal with the fourier representation of the waveform . in the stationary phase approximation the fourier transform , defined as @xmath92 , for positive frequencies reads @xcite @xmath93 } , \label{d4.6a}\ ] ] and , since @xmath43 is real , @xmath94 here @xmath95 is the stationary point of the phase @xmath96 ( i.e. , @xmath95 is defined by @xmath97 ) , @xmath98 @xmath99 is the time - derivative of the instantaneous gravitational wave frequency evaluated at the stationary point given by , @xmath100 and @xmath101 is the phase of the fourier transform in the stationary phase approximation given by @xmath102 instead of solving the integrals in the above equation it is numerically efficient to solve the following equivalent differential equations for the phasing formula in the fourier domain : @xmath103 and in this study we have used the above differential equations in computing the waveforms . for the energy we have used the exact analytical formula ( see next section ) ; for the flux we have used the exact numerical flux to define the exact waveform and the perturbative expansions , or their re - summed improved versions , to define the approximate waveforms . in order to estimate the signal - to - noise ratio ( snr ) we shall assume that the signal is detected using matched filtering and that the bank of templates used in matched filtering has a waveform that matches the unknown signal very closely . to compute the optimum snr we need to known the noise spectral density of the detectors . for initial ligo , the one - sided noise power spectral density ( psd ) from the design study @xcite is given by @xcite @xmath104\ { \rm hz}^{-1},\ ] ] where @xmath105 and @xmath106hz is the `` knee - frequency '' of the detector . we take the psd to be infinite below the lower frequency cutoff of @xmath107hz . for the virgo detector the psd is given by @xmath108\ { \rm hz}^{-1},\ ] ] where @xmath109hz and @xmath110hz . finally , for the geo detector the expected noise psd is @xmath111\ { \rm hz}^{-1}\ ] ] where @xmath112 hz and the noise is assumed to be infinite below a lower frequency cutoff of @xmath107 hz . next , let us define the scalar product of two waveforms @xmath113 and @xmath114 by @xmath115 , \label{eq : scalarprod}\ ] ] where the * denotes complex conjugate and @xmath116 are the fourier transforms of @xmath117 . the ( square ) of the signal - to - noise ratio obtained by matched filtering a signal @xmath113 with a template @xmath114 is given by @xmath118 if the template perfectly matches the signal then the snr is simply @xmath119 for the inspiral signal given in eq . ( [ d4.6a ] ) , and using the _ newtonian values _ for the energy and flux functions , namely @xmath120 @xmath121 and @xmath122 the snr reduces to : @xmath123^{1/2},\ \\ ] ] here @xmath124 is the frequency where the data analyst chooses to terminate the signal . the cutoff frequency is usually chosen to be the frequency of the waves at the last stable orbit of the binary ( cf . sec . [ subsec : energy ] ) which depends on the total mass of the two stars and their spin magnitude . note that the snrs depend only on the combination @xmath125 the chirp mass and not on individual masses of the system . for a given total mass the snr at a fixed distance ( alternatively the span of the antenna at a fixed snr ) is the highest for binaries of equal masses ( when @xmath126 ) and reduces by the fraction @xmath127 for binaries of unequal masses . the snr decreases inversely as the distance to the source and depends not only on the intrinsic parameters @xmath128 @xmath129 and @xmath130 but also on the relative orientations of the source and the antenna specified by the angles @xmath131 @xmath132 @xmath133 @xmath134 we can compute the snr for a source of rms orientation by averaging the amplitude factor @xmath59 over all the angles . performing the average over the angles leads to : @xmath135 for an ideally oriented source , that is for a source that produces the highest snr , @xmath136 thus , the rms and ideal snrs , obtained by using @xmath137 and @xmath138 respectively , are given by because here we use one - sided psd as opposed to two - sided psd used in ref . @xmath139^{1/2},\ \ \rho_{\rm ideal } = \frac{5}{2 } \rho_{\rm rms}. \label{final snr}\ ] ] in fig . [ fig : snr ] we have plotted the _ rms _ snr for the ligo , virgo and geo detectors , as a function of the spin magnitude and the total mass of the system . clearly , systems with large spins in which the test mass is in prograde orbit produce the highest snrs since their last stable orbit frequency is higher than those of retrograde orbits and they last longer . to capture such signals , however , one must know the phasing of the waves very well since the bodies spend a substantial fraction of their time in the detector band in the strongly non - linear regime of the evolution . we shall show later in this paper that there could be a significant drop in the snr , and a corresponding drop in the number of candidate events that can be detected , while p - approximants lose little snr and therefore capture almost all events . we can see from eqs . ( [ eq : timevsv ] ) and ( [ eq : phivsv ] ) that the phase of the gravitational wave depends both on the energy and flux functions of the binary system . in the test - mass case , that is when one of the masses is far smaller than the other ( @xmath140 ) so that the dynamics is governed entirely by the kerr geometry of the massive body , we have an exact expression for the energy @xmath141 , but only a series representation for the flux @xmath142 . in the comparable - mass case , wherein one can not neglect the perturbation caused by the companion , both functions are represented only by post - newtonian expansion . the standard approximants for @xmath143 and @xmath144 are simply their successive taylor approximants @xmath145 and @xmath146 respectively . the gravitational wave signal constructed using these taylor approximants are called t - approximant signals or waveforms are simply t - approximants . it was shown in ref @xcite that the convergence properties of both the energy and flux functions can be improved by using pad approximation of well - defined energy and flux functions . damour , iyer and sathyaprakash followed a two - pronged approach to construct improved energy and flux functions @xcite : starting from the more basic energy - type and flux - type functions , @xmath147 and @xmath148 ref . @xcite constructs pad - type approximants , say @xmath149 , @xmath150 , of the `` basic '' functions @xmath147 , @xmath151 one can then compute the required energy and flux functions entering the phasing formula . the successive approximants @xmath152 $ ] and @xmath153 $ ] have better convergence properties than their taylor counterparts @xmath154 $ ] and @xmath155 $ ] . in this study we will restrict ourselves to the test mass approximation and , therefore , employ the exact energy function . however , in the same approximation the flux function has been computed analytically only as a taylor series although numerically the flux has been computed exactly . since our aim is to draw conclusions on how effective p - approximants are in the comparable mass case , wherein one has only a taylor expansion of the flux , we construct p - approximants of flux and compare it with the numerical results . the reasons for constructing a new flux function are the following : firstly , it is well known , in the schwarzschild case , that a simple pole exists at @xmath156 @xcite in the expression for the flux . we might , therefore , expect that a similar pole exists in the kerr case . a taylor approximant of the flux function will never produce a pole while the equivalent pad approximant , which is essentially a rational polynomial , will have a pole . although it is well known that pad approximants have the ability to model functions which are subject to singularities @xcite it is not guaranteed that the rational polynomial will reproduce the exact pole . secondly , it is often possible to recover from the _ divergence _ of a taylor series by using pad approximation . again , one can not be certain that the resulting pad approximant , even when it converges , will be closer to being exact than any of the taylor approximants , but experience with the test mass approximation in the schwarzschild case render some optimism to this expectation . we create a pad approximation of a truncated power series of an analytic function containing @xmath157 terms by equating it to a rational function such that the rational function when expanded as a power series and truncated to order @xmath157 coincides with the original power series . more precisely , pad approximation can be thought of as an operator @xmath158 that acts on a polynomial @xmath159 to define a rational function : @xmath160{\frac{\displaystyle # 1}{\displaystyle # 2 } } p^{n}_{m } \left ( \sum_{k=0}^{n } a_{k}v^{k } \right ) = \dfrac{\sum_{k=0}^{n } a_{k}\,v^{k}}{\\ 1+\sum_{k=1}^{m } b_{k}\,v^{k } } , \label{eq : pade}\ ] ] such that the number @xmath161 of coefficients in the rational polynomial on the right hand side is the same as the number @xmath162 of taylor coefficients on the left hand side . by setting @xmath163 with @xmath164 we can define two types of pad approximants : these are the super - diagonal , @xmath165 , and sub - diagonal , @xmath166 , approximants . normally , the sub - diagonal approximants are preferred over super - diagonal approximants . this is because when @xmath167 the right hand side of eq . ( [ eq : pade ] ) can be re - expanded as a continued fraction which have the property that as we go to each new order of the power series only one new coefficient needs to be calculated . conversely , with the super - diagonal approximants , we would have to re - calculate all the @xmath168 s and @xmath169 s in the above equation as we go to higher orders in the taylor expansion . this means that the sub - diagonal pad approximants are more _ stable _ and if we see a trend of convergence in the coefficients the addition of a term is not likely to spoil this convergence . we refer the reader to appendix [ sec : padeapp ] for a more summary of the properties of pad approximations and how to find the diagonal pad coefficients using continued fraction expansion . in the case of both schwarzschild and kerr black holes we have an exact expression for the orbital energy of a test particle in a circular orbit around the parent black hole . for a black hole of mass @xmath32 the energy @xmath170 in terms of the dimensionless magnitude of velocity @xmath171 @xmath172 being is the radial coordinate in the boyer - lindquist coordinates , takes the form @xcite @xmath173 where @xmath0 is a dimensionless spin parameter given in terms of the spin angular momentum @xmath174 of the black hole by @xmath175 with @xmath176 spin angular momentum per unit mass in the kerr metric . it is actually the derivative of the orbital energy that appears in the phasing formula given by : @xmath177 where a prime denotes a derivative with respect to the velocity parameter @xmath12 . for black holes with spin , the positions of the last stable circular orbit is a function of the spin parameter @xmath0 . in order to find the position of the last stable circular orbit we take @xmath178 which gives @xcite : @xmath179\,\left [ 3 + z_{1}(q ) + 2\,z_{2}(q ) \right]}\right],\ ] ] where @xmath180\\ \nonumber \\ z_{2}(q ) & = & \sqrt{3\,q^2 + z_{1}^{2}}.\end{aligned}\ ] ] the @xmath181 @xmath182 sign on @xmath183 corresponds to prograde ( retrograde ) orbits . now the position of the last unstable circular orbit or `` photon ring '' occurs where the energy function exhibits a pole . thus , the photon orbit is found by solving a cubic equation . of the three poles , the physically relevant pole is given by @xcite @xmath184\right],\ ] ] where once again the different signs define a prograde ( retrograde ) orbit . in the limit of @xmath185 , both @xmath183 and @xmath186 move towards the horizon of the black hole , @xmath187 in the maximally rotating case of @xmath188 , we can not distinguish between the last stable orbit and the light ring . this means , the greater the spin of the bh , the closer the particle can approach to it before beginning the plunge phase . on the other hand , in the limit @xmath189 , the position of @xmath183 moves outwards from the bh . thus the particle begins its plunge much earlier than in the prograde case . finally , for a particle in an orbit about a schwarzschild black hole the above equations give the familiar position of @xmath190 for the lso and @xmath191 for the photon ring . as discussed in the introduction it has not been possible to derive an exact analytical expression for the flux of gravitational waves emitted by a binary system although analytic approximate expressions , and exact numerical results , have been computed . in the interesting case of two comparable masses in orbit around each other post - newtonian methods have been used to derive an expression for the flux to quite a high order in the expansion parameter . however , since post - newtonian theory is known to be poorly convergent , especially when the expansion parameter approaches unity , it has been suggested to employ in its place an equivalent rational polynomial , or pad approximation , to the ( modified ) flux function . since the purpose of this paper is to test the effectiveness of such an approximation we need a firm ground on which we can conduct our test . in the case of a test mass in an equatorial orbit around a kerr black hole numerical methods have been used to compute the flux to all post - newtonian orders . this is , of course , valid only in the limit of vanishingly small mass of the test body as compared to the central object . however , all the relativistic corrections , including hereditary effects , such as the back scattering of gravitational waves off the curved background geometry , would be present in this computation . the deformation of the geometry due to the presence of the second body , or the back reaction of the waves on the motion of the test body , which would in turn affect the emission process , will not have been included in such a calculation . the deformation of the geometry , parametrized by the symmetric mass ratio of the system @xmath192 is important only a few orbits before the two objects merge . this should be expected from the fact that both in newtonian and einsteinian gravity it has been possible to construct an effective one - body formalism to describe the dynamics of a binary and the dynamics derived within this formalism is expected to be valid close to the point when the two bodies plunge towards each other . therefore , we expect that the deformation of the system will not bring about a major change in the analytic behaviour of the flux and lessons learnt in the test mass approximation will be applicable , albeit qualitatively , in the comparable mass case . thus , our strategy is to employ the test mass exact flux computed numerically together with the analytic expression for the energy function discussed in the previous section . these two exact functions , together with the energy balance equation , can be used to construct an exact phasing formula for the inspiral signal . we shall also define an approximate phasing formula using the corresponding post - newtonian expansions of the two functions . our approximations at each post - newtonian order will be one of two types : ( 1 ) the standard post - newtonian or ( 2 ) the improved pad approximation . in the next section we will discuss how these approximations can be further improved and in the last section we will measure the quantitative performance of the two approximations . for a test - particle in a circular equatorial orbit the post - newtonian expansion of the flux function has been calculated up to @xmath193 ( @xmath194 ) in the case of a schwarzschild bh @xcite , and to @xmath193(@xmath195 ) in the case of a kerr bh @xcite . the general form of the flux function in both these cases is given by the expression @xmath196 . \label{eq : flux}\ ] ] where the expansion is known to order @xmath197 and 11 , in the kerr and schwarzschild cases , respectively , @xmath0 is the dimensionless spin parameter introduced earlier and @xmath198 is the dominant _ newtonian _ flux function given by @xmath199 here , @xmath200 is the magnitude of the invariant velocity parameter observed at infinity which is related to the angular frequency @xmath201 by @xmath202 . the relation between the parameter @xmath200 and the local linear speed @xmath12 in boyer - lindquist coordinates is given by @xmath203^{1/3},\ ] ] which reduces , in the schwarzschild limit , to @xmath204 note that this local linear velocity is related to the boyer - lindquist radial coordinate @xmath172 by @xmath205 the spin - dependent coefficients , @xmath206 , and the log - term coefficients , @xmath207 , are given by has the wrong sign in some later references but ref . @xcite has the correct sign . ] , @xmath208 where @xmath209 is euler s number . for graphical purposes , it is more appropriate to deal with the newton - normalized fluxes defined by @xmath210 we will now outline the method of calculation the p - approximant of the gw flux as proposed by dis @xcite . we notice from the form of the series expansion for the flux , eq . ( [ eq : flux ] ) , that we begin to encounter logarithmic terms at order @xmath211 and above . in general , series approximations of this form have slow convergence properties . in order to prepare the series representation of the flux for creating the pad approximation , it is convenient if we factor out the logarithmic terms . we can then write eq . ( [ eq : flux ] ) as @xmath212 \left[\sum_{k=0}^{n}\,c_{_{k}}x^{k}\right],\ ] ] where the new coefficients @xmath213 and @xmath214 are functions of the old coefficients @xmath215 and @xmath216 as in ref . @xcite the log - terms have been `` normalized '' using the value of the velocity parameter at the lso ; this helps in reducing the importance of the log - terms . factoring out the logarithmic terms aids in constructing the rational polynomial , or the pad approximation , of the reminder . moreover , since we expect the flux to have a pole at the location of the light ring it is best to factor out the expected pole so that the reminder has good analytical properties . to this end we create the factored flux function , @xmath217 by the operation @xmath218 factoring out the pole also helps to alleviate the problem arising from the absence of the linear term in the pn expansion of the flux . ( note that @xmath219 in both the schwarzschild and kerr cases . ) we can see from appendix [ sec : padeapp ] that in the absence of such a term the continued fraction form of the pad approximation , the so - called diagonal pad approximant , would lead either to zero or infinite pad coefficients . the above operation rectifies this problem by introducing a linear term into the taylor series for the flux . if we write the expression in full we obtain @xmath220 \left[\sum_{k=0}^{n}\,f_{_{k}}x^{k}\right ] , \label{eq : fluxf}\ ] ] where @xmath221 and @xmath222 @xmath223 . we can now construct a new flux function by using the pad approximant of the factored flux given above . indeed , we can construct two variants of the new flux : the first one is what we call the _ direct _ or d - pad approximant , which is obtained by directly starting from the flux function @xmath224 in eq . ( [ eq : fluxf ] ) and constructing the equivalent rational polynomial . this is the approach followed in dis . an alternative approach to this is motivated by the fact that in the gravitational wave phasing formula the flux appears in the denominator . thus , instead of constructing the pad approximant of the flux function one could first construct the polynomial expansion of the inverse of the flux function and construct the pad approximant of the resulting polynomial . we call the approximant constructed this way as _ inverse- _ or i - pad approximant because it is obtained from the taylor expansion of the _ inverse _ of the flux function @xmath224 in eq . ( [ eq : fluxf ] ) . thus , our two improved versions of the flux are defined as follows : the direct pad approximant is defined by @xmath225p^ { m}_{m+\epsilon}\left[\sum_{k=0}^{n}\,f_{_{k}}x^{k}\right],\ ] ] where @xmath226 is the diagonal or sub - diagonal pad approximant , @xmath227 with @xmath228 or @xmath229 depending on whether @xmath157 is even or odd . the inverse pad approximant of flux is defined by @xmath230p^ { m}_{m+\epsilon}\left[\sum_{k=0}^{n}\,d_{_{k}}x^{k}\right],\ ] ] where the coefficients @xmath231 in the taylor expansion are defined by @xmath232 one finds @xmath231 by first expanding the rhs in a binomial series and then identifying the coefficients of the various terms with those on the lhs . having constructed the rational polynomials equivalent to a given truncated taylor series of the modified flux function we can return to the original flux function that appears in the phasing formula . we shall call the flux function so constructed as _ p - approximant , _ and not just pad approximant to remind ourselves that the new flux has been obtained by improving the convergence properties in two steps ( i.e. , definition of a new flux function and the construction of its rational polynomial ) and not just a direct application of pad approximation . thus , we define the _ direct p - approximant _ of flux as @xmath233 and _ inverse p - approximant _ as , @xmath234^{-1}.\ ] ] detailed investigation shows that inverse p - approximants of flux , namely @xmath235 have better convergence properties than direct p - approximants . we shall therefore use only @xmath236 in all our investigations in the rest of this paper . in this section we will look at the convergence properties of the post - newtonian series representations and the improved p - approximants of the flux function discussed in the last section . to this end we shall employ all the post - newtonian orders up to which the taylor expansions have been worked out in the kerr case . we shall investigate the convergence of the approximants in the case of a test particle ( i.e. , @xmath237 limit ) either co - rotating ( i.e. , @xmath238 _ prograde orbits _ ) or counter - rotating ( i.e. , @xmath239 _ retrograde orbits _ ) , with respect to a central black hole whose spin parameter @xmath0 takes nine values over the full range from @xmath240 in this section we shall make a graphical comparison of the analytical and numerical fluxes and assess the performance of the flux with respect to variation in the spin - parameter values at increasingly higher post - newtonian orders . we will find that neither the post - newtonian series nor their improved versions fit the numerical flux exactly when the parameters are chosen to be the same for analytical fluxes as the numerical flux . however , the p - approximant fluxes agree with the numerical fluxes quite closely when their spin values are slightly mis - matched with the true values of the numerical fluxes . the post - newtonian approximants do show this improved behaviour , albeit not to the extent of p - approximants . thus , we can expect the p - approximants to define better signal models than post - newtonian approximants with regard to both faithfulness and effectualness . to @xmath241 for a test particle in prograde orbit around a kerr black hole . the solid lines represent the exact numerical fluxes with thicker lines corresponding to larger spin magnitudes , @xmath242 respectively . the correspondingly thickened dashed lines represent the taylor- and p - approximants.,title="fig:",width=288 ] to @xmath241 for a test particle in prograde orbit around a kerr black hole . the solid lines represent the exact numerical fluxes with thicker lines corresponding to larger spin magnitudes , @xmath242 respectively . the correspondingly thickened dashed lines represent the taylor- and p - approximants.,title="fig:",width=288 ] the numerical fluxes have been computed exactly @xcite , but in the test mass approximation , at a set of test values of the spin parameter @xmath244 for prograde orbits the flux has been computed at @xmath245 . for the sake of completeness , we shall also include in our comparison the numerical flux for a particle in orbit about a schwarzschild bh , i.e. @xmath246 @xcite . in fig . [ fig : ktp ] we have plotted the newton - normalized numerical fluxes alongside analytical fluxes for both the truncated taylor approximants ( four panels on the left ) and ( inverse ) p - approximants ( four panels on the right ) . for each approximant , we consider four post - newtonian orders at \{@xmath247 } and five spin values . the solid lines with increasing thickness correspond to the exact numerical fluxes at progressively larger values of the spin parameter starting at @xmath248 ( thinnest line ) and ending at @xmath249 ( thickest line ) . the corresponding dashed lines represent the analytical approximant to the gw flux . let us first concentrate on the post - newtonian fluxes : we notice immediately that at all values of @xmath0 , and large values of @xmath250 the 2.5 post - newtonian approximant @xmath251 is highly divergent . indeed , this order has been known to exhibit improper behaviour in the schwarzschild case and continues to be so for prograde orbits in kerr . the situation improves at @xmath252 but begins to worsen again at @xmath253 becoming highly divergent at @xmath254 . the plots serve to illustrate the principal problem with the pn expansion : going to higher orders of the approximation does not necessarily correspond to a better accuracy . the sequence of panels on the right - hand - side of fig . [ fig : ktp ] demonstrates that the p - approximant flux has improved convergence properties over the post - newtonian flux for the same range of post - newtonian orders and spins . the extreme divergence which the post - newtonian flux exhibited at @xmath251 and @xmath255 has been cured . let us note , however , that sometimes the rational polynomial approximation of a taylor series can produce spurious poles in the region of interest . an example of this can be seen at @xmath256 which shows a spurious pole at every value of @xmath244 the main factor that stands out in the right panels of fig . [ fig : ktp ] is the following : the p - approximant fluxes display the same characteristic shape as the numerical fluxes although they do nt agree with the exact flux at higher values of the parameter @xmath257 to @xmath241 for a test particle in retrograde orbit around a kerr black hole . the solid lines represent the exact numerical fluxes with thicker lines corresponding to larger spin magnitudes , @xmath258 respectively . the correspondingly thickened dashed lines represent the taylor- and p - approximants.,title="fig:",width=288 ] to @xmath241 for a test particle in retrograde orbit around a kerr black hole . the solid lines represent the exact numerical fluxes with thicker lines corresponding to larger spin magnitudes , @xmath258 respectively . the correspondingly thickened dashed lines represent the taylor- and p - approximants.,title="fig:",width=288 ] in the case of prograde orbits the p - approximants graphically display a closer convergence to the numerical fluxes : while being better than the taylor approximation , they are still not as good as we would like . however , we must remember that what is more important is how the phasing of the waveforms defined by the two approximants match with that defined by the exact flux , rather than a simple graphical agreement of the flux . the graphical display is useful as it gives us a rough idea on any improvement in performance of a particular model . but as the noise of the interferometer weights the inner - product between templates , it may emphasize a particular range of frequency of the waveform more than the other . it may turn out that the divergent region of a particular model is outside the detector bandwidth for most masses . in such cases the approximant may work well in spite of the divergences or poles . we shall return to this question in sec . [ sec : results ] when we consider the matching of waveforms . we now focus on the gw flux from a test - mass counter - rotating with respect to the black hole . we will again use a similar set of spin values , i.e. @xmath260 as well as the schwarzschild flux for comparative purposes . in fig . [ fig : ktr ] , left panels , we plot the taylor approximant flux for retrograde orbits . the solid lines in the figure correspond to the exact numerical fluxes with thicker lines corresponding to smaller spin values @xmath261 and @xmath262 respectively . the correspondingly thickened dashed lines represent the taylor series approximation to the gw flux with the same spin magnitude as the numerical flux . the right panels are the same except that in place of taylor approximant fluxes we plot the inverse p - approximant fluxes . the first thing we notice is that just as in the case of prograde motion the @xmath251 and @xmath255 approximants are again highly divergent although at @xmath255 , the convergence of the pn approximation improves as we go to larger negative spins . this is not entirely surprising : as we go to larger counter - rotating spins the higher order taylor expansion coefficients become smaller and the non - linear general relativistic corrections are not as important as in the case of prograde orbits . moreover , the position of the lso moves outwards from the bh as @xmath0 becomes more negative . at @xmath263 , the lso is at @xmath264 . this corresponds to @xmath265 at @xmath266 the lso has moved outwards to @xmath267 , corresponding to @xmath268 . if we compare this to the equivalent spins in the prograde case we find that for @xmath269 , @xmath270 and @xmath271 . at @xmath249 , @xmath272 and @xmath273 . in other words , retrograde orbits are weakly bounded orbits in comparison to prograde orbits and experience smaller relativistic corrections . recalling that gravitational wave luminosity @xmath274 retrograde flux for @xmath266 at lso is smaller by a factor @xmath275 compared to prograde flux for @xmath249 at lso . consequently , the pn approximation seems to work well in the retrograde case . in the case of retrograde orbits p - approximants provide , at order @xmath276 only a marginal improvement over t - approximants . this is in sharp contrast to the prograde case where the improvement was excellent . at orders @xmath277 and @xmath255 are again quite good , with no poles at @xmath278 for most spin values . while there seems to be a less obvious benefit in using the p - approximant flux in the retrograde case , it is again important to note that the p - approximants show the same general behaviour as the exact flux in the retrograde case too and can therefore be expected to match the true general relativistic signal more closely . in the last two sections we noted how the p - approximant sequence seems to have a character similar to the sequence given by the numerical computations although the two do not agree perfectly . this suggests that if the value of the spin parameter used in the t- or p - approximant fluxes is varied relative to the true value of the spin parameter in the exact flux , we should be able to obtain a better fit to the numerical fluxes . while we will initially do this graphically , we remind the reader that this exercise is purely demonstrative . as the fluxes are newton - normalized , it is difficult to say in what region of the velocity we should attempt to make the best fit . added to this is the fact that when matching templates we should take the weighting of the psd into account . however , if the approximant fluxes agree graphically well with the numerical fluxes then it would mean that we should be able to match the t- or p - approximant flux with the numerical result without introducing any new parameters . to examine how best mis - matching spins might work we will focus on one of the post - newtonian orders , say @xmath252 the order at which the t - approximants work well and p - approximants have no spurious poles . in fig . [ fig : fitt ] , left panel , we have plotted the best - fit for the t - approximant flux for various values of the spin magnitude from @xmath279 ( top most curve ) to @xmath280 ( bottom most curve ) . for each spin magnitude we have fitted the t - approximant with the numerical result as best as possible by varying the spin - magnitude of the approximant . in best fitting the numerical flux we incurred errors ( defined as @xmath281 ) in the spin magnitude of the approximant relative to the exact case . these are listed in table [ table : bestfit ] . although the errors decrease as we approach the extreme prograde flux , we can see from the figure that the fit is not very good for the taylor approximant for @xmath282 schwarzschild case . .the `` best - fit '' value of the spin parameter , @xmath0 , for t and p - approximant fluxes for @xmath279 to @xmath280 . for the schwarzschild case we have an absolute error of 0.11 in the case of t - approximant and @xmath283 in the case of p - approximant . [ cols="^,^,^,^,^,^,^,^,^,^",options="header " , ] [ table : schwarzoverlaps ] we have applied p - approximant templates to the case of a ns orbiting kerr bhs of varying mass and spin . when graphically matching the numerical fluxes we found that the p - approximant fluxes gave a better fit as compared to the pn fluxes , especially when we varied the value of the spin parameter and best fitted the exact flux . however , the true test of how well a template performs is reflected by the fitting factors achieved . using a signal waveform constructed from the exact expression for the orbital energy and numerical fluxes from black hole perturbation theory , we were able to compare the performance of t and p - approximant templates . in the case of retrograde , schwarzschild and prograde orbits , not only did the p - approximants gave better and more reliable fitting factors , they also gave smaller biases in the estimation of parameters . we also saw the true power of the p - approximant templates in that we were able to generate templates right up to the lso . this is something that was not possible with the t - approximant templates due to the approximation for the flux function becoming negative before the lso is reached . this restricted just how much of the numerical waveform we could model . while not being completely correct due to the fact that we omitted the finite - mass correction terms , we also saw that the p - approximant templates gave the best performance when in the equal - mass case . the bias in the estimation of parameters is particularly poor in the case of t - approximants especially for high - mass systems . since high - mass systems can be seen to a greater distance , and therefore likely to be the first sources to be detected , it is important to keep in mind that the bias in the estimation of parameters could be as large as the random errors associated with the measurement . such biases could seriously affect the test of general relativity @xcite that are envisaged to be carried out with the aid of high - mass systems . it is clear that for the type of systems examined in this paper , namely equatorial test - mass circular orbits in kerr , p - approximant templates are to be preferred over their pn counterparts in both detection and measurement . it remains to be seen if p - approximants ( or their improved versions ) prove to be good enough when considering generic orbits in kerr and for comparable mass systems with spinning black holes on generic orbits . we would like to thank prof . m. shibata for his permission to use the numerical fluxes in this work . we would also like to thank t. damour , b. iyer and the members of the cardiff relativity group for the stimulating conversations and helpful suggestions . ekp would like to thank the school of physics and astronomy , cardiff , for its hospitality in the final stages of this work . this research was supported in parts by the particle physics and astronomy research council and the leverhulme trust , both in the uk . 99 a. abromovici et al . , _ science _ * 256 , * 325 ( 1992 ) c. bradaschia et al , _ nucl . instrum . methods phys . _ sec . a , 518 ( 1990 ) j. hough and k. danzman , _ proposal for a @xmath284 laser interferometric gravitational wave antenna _ , ( 1994 ) k. kuroda et al in _ gravitational waves : sources and detectors _ , ed . by i. cinfolini and f. fidecaro , 100 ( world scientific , 1997 ) l. p. grishchuk , v. m. lipunov , k. a. postnov , m. e. prokhorov and b. s. sathyaprakash , _ phys . _ * 44 * , 1 ( 2001 ) e. phinney , _ astrophys . j. _ * 380 * , l17 ( 1991 ) r. narayan , t. piran and a. shemi , _ astrophys . j. _ * 379 * , l17 ( 1991 ) i. h. stairs et al . , _ astrophys . j. _ * 505 * , 352 ( 1998 ) k. belczynski , v. kalogera and t. bulik , astrophys . j. , * 572 * 407 ( 2001 ) . astro - ph/0111452 . k. a. postnov and m. e. prokhorov , _ binary black hole formation and mergings , _ in proceedings of the xxxiv rencontres de moriond `` gravitational waves and experimental gravity '' , january 23 - 30 , 1999 , astro - ph/9903193 . c. w. helstrom , _ statistical theory of signal detection _ ( pergamon press , london 1968 ) s. v. dhurandhar and b. s. sathyaprakash , _ phys . rev . _ d * 49 * , 1707 ( 1994 ) a. buonanno , y. chen , and m. vallisneri , phys . d * 67 , * 024016 ( 2003 ) . a. buonanno , y. chen , and m. vallisneri , phys . d * 67 , * 104025 ( 2003 ) . y. pan , a. buonanno , y. chen , and m. vallisneri , _ a physical template family for gravitational waves from precessing binaries of spinning compact objects : application to single - spin binaries , _ gr - qc/0310034 l. blanchet , t. damour , b. iyer , c. m. will and a. g. wiseman , _ phys . rev . lett . _ * 74 * , 3515 ( 1995 ) l. blanchet , t. damour and b. iyer , _ phys _ d * 51 * , 5360 ( 1995 ) c. m. will and a. g. wiseman , _ phys . _ d * 54 * , 4813 ( 1996 ) l. blanchet , b. iyer , c. m. will and a. g. wiseman , _ class . quantum grav . _ * 13 * , 575 ( 1996 ) l. blanchet , _ phys _ d * 54 * , 1417 ( 1996 ) t. damour , p. jaranowski and g. schfer , _ phys . _ d * 63 * , 044021 ( 2001 ) ; v. c. de andrade , l. blanchet and g. faye , _ class . quantum grav . _ * 18 * , 753 , ( 2001 ) . l. blanchet , g. faye , b. r. iyer and b. joquet , _ phys . _ d * 65 * , 061501 ( 2002 ) ; l. blanchet , b. r. iyer and b. joquet , _ phys . _ d * 65 * , 064005 ( 2002 ) e. poisson , _ phys . _ d * 47 * , 1497 ( 1993 ) c. cutler , l. s. finn , e. poisson and g. j. sussman , _ phys . _ d * 47 * , 1511 ( 1993 ) h. tagoshi and t. nakamura , _ phys . _ d * 49 * , 4016 ( 1994 ) m. sasaki , _ prog . _ * 92 * , 17 ( 1994 ) h. tagoshi and m. sasaki , _ prog . theor . phys . _ * 92 * , 745 ( 1994 ) t. tanaka , h. tagoshi and m. sasaki , _ prog . _ 1087 ( 1996 ) p. r. brady , j. d. e. creighton and k. s. thorne , _ phys . rev . _ d * 58 * , ( 1998 ) e. poisson , _ phys . _ d * 49 * , 1860 ( 1993 ) m. shibata , m. sasaki , h. tagoshi and t. tanaka , _ phys . _ d * 51 * , 1646 ( 1995 ) h. tagoshi , m. shibata , t. tanaka and m. sasaki , _ phys . _ d * 54 * , 1439 ( 1996 ) c. cutler et al . , _ phys . _ * 70 * , 2984 ( 1993 ) e. poisson , _ phys _ d * 52 * , 5719 ( 1995 ) e. poisson , _ phys . _ d * 55 * , 7980 ( 1997 ) t. damour , b. r. iyer and b. s. sathyaprakash , _ phys . _ d * 57 * , 885 ( 1998 ) e. k. porter , _ class . quantum grav . _ * 19 * , 4343 , ( 2002 ) y. mino , m. sasaki , m. shibata , h. tagoshi and t. tanaka prog . theor . suppl . * 128 * 1 ( 1997 ) . a. buonanno and t. damour , _ phys . _ d * 59 * , 084006 ( 1999 ) t. damour , _ phys . rev . _ d * 64 * , 124013 ( 2001 ) t. damour , b. r. iyer and b. s. sathyaprakash , _ phys . _ d * 63 * , 044023 ( 2001 ) t. damour , b. r. iyer , p. jaranowski and b. s. sathyaprakash , _ phys . rev . _ d * 67 * , 064028 ( 2003 ) k. s. thorne , _ gravitational radiation _ , in s. w. hawking and w. israel ( eds ) , _ 300 years of gravitation _ ( cambridge uni . press , cambridge ) b. s. sathyaprakash and s. v. dhurandhar , _ phys . _ d * 44 * , 3819 ( 1991 ) s. v. dhurandhar , j. watkins and b. f. schutz , _ fourier transform of a coalescing binary signal _ ( unpublished preprint , 1989 ) t. damour , b. r. iyer and b. s. sathyaprakash , _ phys . _ d * 62 * , 084036 ( 2000 ) b. s. sathyaprakash , _ phys . _ d * 50 * , r7111 ( 1994 ) w. h. press , b. p. flannery , s. a. teukolsky and w. t. vetterling , _ numerical recipes in c : the art of scientific computing _ ( cambridge uni . press , cambridge 1992 ) 2nd edition . c. m. bender and s. a. orszag , _ advanced mathematical methods for scientists and engineers _ , mcgraw - hill international book company , singapore , 1978 . j. m. bardeen , w. h. press and s. a. teukolsky , _ astrophys . j. _ * 178 * , 374 ( 1972 ) s. chandrasekhar , _ the mathematical theory of black holes _ ( oxford uni . press , new york , 1983 ) m. shibata ( private communication ) c. cutler and e. flanagan , _ phys . _ * d 49 * 2658 ( 1994 ) gr - qc/9402014 . r. balasubramanian , b. s. sathyaprakash , and s. v. dhurandhar , _ phys . _ d * 53 * , 3033 ( 1996 ) ; b. j. owen , phys . rev . d * 53 * , 6749 ( 1996 ) ; b. j. owen and b. s. sathyaprakash , _ phys . _ d * 60 * , 022002 ( 1999 ) . b. f. schutz in _ the detection of gravitational radiation _ , ed . by d. blair ( cambridge uni . press , cambridge 1989 ) l. blanchet and b. s. sathyaprakash , _ phys . _ * 74 * , 1067 ( 1995 ) . let @xmath285 be an analytic function whose taylor expansion in @xmath250 denoted @xmath286 is @xmath287 we construct the pad approximation @xmath288 of @xmath289 by equating the polynomial expansion with a rational function to the same order according to @xmath290 where @xmath160{\frac{\displaystyle # 1}{\displaystyle # 2 } } \frac{b_{n}(x)}{d_{m}(x ) } = \dfrac{\sum_{n=0}^{n } b_{n}\,x^{n}}{\\ \sum_{n=0}^{m } d_{n}\,x^{n } } , \label{eq : ratfunc2}\ ] ] such that @xmath291 . the benefit of using pad approximants over taylor approximants is that the error in the approximation after @xmath292 terms reduces by a factor of @xmath293 @xcite . if we let @xmath294 , where @xmath228 or @xmath295 , we construct the `` diagonal '' and `` sub - diagonal '' pad approximants respectively . the advantage of this is that we can cast the rational function in the form of a continued fraction , i.e. @xmath296 if we use the rational function interpretation , eq . [ eq : ratfunc2 ] , each time we go to a higher order @xmath157 of approximation , we have to recalculate each of the coefficients , @xmath297 and @xmath298 . however , by using the continued fraction formalism , we only have to calculate the newest coefficient , @xmath299 . as an example , consider the taylor expansion @xmath300 by matching this to a pad approximation of the same order we get @xmath301 solving for the pad coefficients , @xmath299 , gives us @xmath302 if we now go to the next order , i.e. @xmath303 and solve for the coefficients @xmath299 once more , we obtain @xmath304 the example shows that as we go to higher and higher orders in @xmath12 , we only have to calculate one new coefficient as all the lower order coefficients remain constant . consequently , there is a sense of stability in the behaviour of the continued fraction coefficients and convergence is easier to test . there are , unfortunately , problems with pad approximation that one should bear in mind : firstly , there are regions where pad approximants diverge while taylor approximants converge . the second , and perhaps most important feature is that when pad approximation does indeed converge , we can not be sure that it converges to the exact function that we had in mind @xcite . it should therefore always be kept in mind that there are inherent difficulties in using pad approximants . finally , pad approximants sometimes exhibit spurious poles in the region of interest . however , as one will be able to test _ a priori _ the existence of poles one can avoid the use of such approximants .

a systematic fieldtheoretic description of critical adsorption on curved objects such as spherical or rodlike colloidal particles immersed in a fluid near criticality is presented . the temperature dependence of the corresponding order parameter profiles and of the excess adsorption are calculated explicitly . critical adsorption on elongated rods is substantially more pronounced than on spherical particles . it turns out that , within the context of critical phenomena in confined geometries , critical adsorption on a microscopically thin ` needle ' represents a distinct universality class of its own . under favorable conditions the results are relevant for the flocculation of colloidal particles . = 23.0 true cm = 16.0 true cm = 0 true cm -0.25 cm 0 cm 1.5 cm 1.5em 0.5 mm

cond-mat9809351

in colloidal suspensions the interaction between the mesoscopic dissolved particles and the solvent is of basic importance @xcite . for example , the solvent generates effective interactions between the colloidal particles which can even lead to flocculation . the richness of the physical properties of these systems is mainly based on the possibility to tune these effective interactions over wide ranges of strength and form of the interaction potential . traditionally this tuning is accomplished by changing the chemical composition of the solvent , e.g. , by adding salt , polymers , or other components @xcite . compared with such modifications , changes of the temperature or pressure typically result only in minor changes of the effective interactions . this , however , is only true as long as the solvent is not thermodynamically close to a phase transition of its own . for example , if the solvent consists of a binary liquid mixture close to a _ first - order _ demixing transition into a a - rich and a b - rich liquid phase , even slight changes of the temperature or of the partial pressures of the two species a and b can lead to massive changes of the effective interactions between dissolved colloid particles induced by the occurrence of wetting transitions . they lead to wetting films of the preferred phase coating the colloidal particles @xcite . these wetting films can snap into bridges if the particles come close to each other leading to flocculation @xcite . for charged colloidal particles such as silica spheres immersed in the binary liquid mixture of water and 2,6-lutidine @xcite flocculation can also be influenced by screening effects generated by the adsorbed layers @xcite . similarly drastic effects can occur if the solvent is brought close to a _ critical point_. the inevitable preference of the surfaces of the colloidal particles for one of the two solvent species of a binary liquid mixture near its critical demixing point or for the liquid phase of a one- or two - component solvent fluid near its liquid - vapor critical point results into the presence of effective surface fields leading to pronounced adsorption profiles of the preferred component . this so - called ` critical adsorption ' becomes particularly long - ranged due to the correlation effects induced by the critical fluctuations of the order parameter of the solvent . in the case of a planar wall critical adsorption has been studied in much detail @xcite . asymptotically close to the critical point @xmath0 it is characterized by a surface field which is infinitely large so that the order parameter profile actually diverges upon approaching the wall up to atomic distances @xmath1 . as compared to a planar surface critical adsorption on a spherical particle is expected to exhibit important differences in behavior because the confining surface has a positive curvature and because a sphere represents only a quasi - zero - dimensional defect floating in the critical fluid . the interference of critical adsorption on neighboring spheres gives rise to the so - called critical casimir forces @xcite which have been argued to contribute to the occurrence of flocculation near @xmath0 @xcite . a quantitative understanding of these phenomena requires the knowledge of the critical adsorption profiles near the colloidal particles and the resulting effective free energy of interaction in the whole vicinity of the critical point , i.e. , as function of both the reduced temperature @xmath2 and the field @xmath3 conjugate to the order parameter . this ambitious goal has not yet been accomplished . instead , the introduction of a surface curvature has limited the knowledge of the corresponding critical adsorption so far to the case of spheres for the particular thermodynamic state @xmath4 of the solvent @xcite . only recently at least the temperature dependence of the critical casimir force between a sphere and a planar container wall has been addressed @xcite . thus the present study of the temperature dependence of critical adsorption on a single sphere contributes one step towards reaching the aforementioned general goal . apart from spherical particles , also rodlike particles play an important role @xcite . rodlike objects are provided , e.g. , by fibers or colloidal rods @xcite , semiflexible polymers with a large persistence length such as actin @xcite , microtubuli @xcite , and carbon nanotubes @xcite . moreover the knowledge of the general curvature dependence of critical adsorption is also relevant , e.g. , for curved membranes @xcite dissolved in a fluid near criticality or for the liquid - vapor interface between a binary liquid mixture near its critical demixing point and its noncritical vapor , which exhibits rippled configurations due to the occurrence of capillary waves @xcite . near criticality the relevant length scales of the solvent structures are dominated by the diverging bulk correlation length @xmath5 , where @xmath6 is the standard universal bulk critical exponent ; @xmath7 and @xmath8 are nonuniversal amplitudes in the one- ( @xmath9 ) and two - phase region ( @xmath10 ) , respectively , with values typically in the order of a few . in practice the correlation length can span the range between @xmath11 - @xmath12 depending on @xmath13 . in the present context this length scale is played off against the length scale @xmath14 of the radius of the dissolved particles . we note that the available systems can realize both the limit @xmath15 as well as the opposite limit @xmath16 . in the case of ludox silica particles @xmath17 @xcite so that the limit @xmath16 can be easily achieved even with the upper limits for @xmath18 set by finite experimental resolutions . the ratio of the length @xmath19 and the radius @xmath14 of rodlike particles can be quite large , in conjunction with a small radius such as @xmath20 in the case of colloidal boehmite rods @xcite . in this work we consider _ long _ rods , i.e. , @xmath21 , and neglect effects which may arise due to their finite length @xmath19 . in the present contribution we investigate systematically the temperature dependence of the critical adsorption on a single spherical or rodlike particle , i.e. , the case @xmath22 . ( the generalization to the case @xmath23 is straightforward but tedious . ) in order to be able to treat spheres and cylinders in a unified way within a field - theoretical approach and for general spatial dimensions @xmath24 it is helpful to consider the particle shape of a _ generalized cylinder _ @xmath25 @xcite with an infinitely extended ` axis ' of dimension @xmath26 . the ` axis ' can be the axis of an ordinary infinitely elongated cylinder ( @xmath27 ) , or the midplane of a slab ( @xmath28 ) , or the center of a sphere ( @xmath29 ) . for general integer @xmath24 and @xmath26 the explicit form of @xmath25 is @xmath30 with @xmath31 and @xmath32 perpendicular and parallel to the axis , respectively ( see fig.[fig_cyl ] ) . note that @xmath31 is a @xmath33-dimensional vector with @xmath34 the radius @xmath14 of the generalized cylinder @xmath25 is the radius in the cases of an ordinary cylinder or a sphere and it is half of the thickness in the case of a slab . for the slab the geometry reduces to the much studied case of ( two decoupled ) half spaces . the generalization of @xmath24 to values different from three is introduced for technical reasons because @xmath35 marks the upper critical dimension for the relevance of fluctuations of the order parameter leading to a behavior different from that obtained from mean - field theory valid for @xmath36 . in sec.[sec_op ] we discuss the general scaling properties of the local order parameter profiles for critical adsorption on spheres and cylinders , in particular the behavior close to the particle surfaces and for small particle radii , respectively . in sec.[sec_excess ] we consider the corresponding properties of the excess adsorption . in sec.[sec_mf ] we present explicit results both for the order parameter profiles and for the excess adsorption in mean - field approximation . section [ summary ] contains our conclusions . in appendix [ app_ex ] we discuss the two - point correlation function near a microscopically thin ` needle ' at criticality . in appendix [ app_neu ] we determine a universal amplitude and a universal scaling function as needed in secs.[sec_op ] and [ sec_excess ] , respectively . in appendix [ appb ] , finally , we consider the general curvature dependence of the excess adsorption . ( 16,10 ) ( -0.7,-1.3 ) ( 8.0,-1.3 ) asymptotically , i.e. , in case of an infinitely large surface field , critical adsorption on the surface of a sphere or cylinder with radius @xmath14 is characterized by an order parameter profile @xmath37 which close to @xmath0 and for @xmath38 takes the scaling form @xmath39 for radial distances @xmath40 from the surface larger than a typical microscopic length @xmath1 . here @xmath41 denotes the thermal average in the presence of a sphere or a cylinder and @xmath42 is the standard universal bulk critical exponent . the scaling functions @xmath43 depend on two scaling variables : @xmath44 the scaling functions @xmath43 are _ universal _ while the nonuniversal _ bulk _ amplitudes @xmath45 and @xmath46 are determined by the value @xmath47 for @xmath48 , @xmath49 of the order parameter in the unbounded bulk and by the amplitudes of the correlation length , respectively . ( in this work we are always concerned with the ising universality class . ) we note that the form of @xmath43 depends on the definition used for the correlation length @xmath50 . for definiteness we adopt @xmath50 as being the _ true _ correlation length fixed by the exponential decay of the two - point correlation function in real space @xcite . bearing this in mind one finds that @xmath51 , @xmath52 , and @xmath53 where the prefactor in front of the exponential may by an algebraic function of @xmath54 ( see , c.f . , sec.[sec_mf ] ) . on the other hand , in the limit @xmath55 with @xmath54 fixed the scaling functions @xmath56 reduce to @xmath57 corresponding to the half - space bounded by a planar surface . in recent years the scaling functions @xmath58 for the half - space have been intensively studied theoretically @xcite and compared with experiments @xcite . for later reference we quote some of their properties . first , we note the asymptotic behavior @xcite @xmath59 & & \ , \ , \times \ , \big [ 1 \ , + \ , \bar{a}_{\pm } \ , x_{\pm}^{\,1/\nu } \ , + \ , \bar{a}_{\pm}^{\ , ' } \ , x_{\pm}^{\,2/\nu } \ , + \ , \bar{b}_{\pm } \ , x_{\pm\,}^{\,d } \ , + \ , \ldots \ , \big ] \nonumber\end{aligned}\ ] ] with universal amplitudes @xmath60 , @xmath61 , @xmath62 , and @xmath63@xcite which depend , however , on the definition of the correlation length @xcite . the ellipses stand for contributions which vanish more rapidly than @xmath64 . + ( 16,20 ) ( -0.5,12.3 ) ( -0.5,0.1 ) [ fig_hs ] the exponent @xmath65 is the bulk ` scaling dimension ' @xcite of the order parameter @xmath66 . for convenience of the reader we quote their numerical values @xcite : [ a45 ] @xmath67 @xmath68 so that @xmath69 the leading power law in eq.([a40 ] ) is a consequence of the fact that due to the presence of the symmetry breaking surface field the limit @xmath70 must lead to a nonvanishing order parameter profile at @xmath71 so that @xmath72 ( the subscript ` hs ' stands for half - space ) . thus the amplitude of the power law @xmath73 in eq.([a50 ] ) consists of the combination @xmath74 of nonuniversal bulk amplitudes and of the universal surface amplitude @xmath60 . because the lhs of eq.([a50 ] ) does not depend on how the limit @xmath71 is approached , the universal amplitudes @xmath75 and @xmath76 are related via the universal bulk ratio @xmath77 : @xmath78 the terms proportional to @xmath61 and @xmath62 in eq.([a40 ] ) correspond to _ regular _ contributions for @xmath70 of the order parameter since @xmath79 \ , \ , , \quad t \to 0 \ , \ , , \ ] ] where @xmath80 and @xmath81 are _ independent _ of the sign of @xmath13 @xcite . the term proportional to @xmath63 in eq.([a40 ] ) corresponds to the first singular contribution @xcite for @xmath70 of the order parameter leading to the term @xmath82 in eq.([a60 ] ) with the bulk exponent @xmath83 and @xmath84 . figure [ fig_hs ] summarizes theoretical results for @xmath58 for the spatial dimensions @xmath36 , @xmath85 , and @xmath86 . in @xmath36 the functions @xmath43 are given by the following mean - field expressions : @xmath87 the results for @xmath88 represent recent monte carlo ( mc ) simulations @xcite and field - theoretical renormalization group ( rg ) calculations @xcite . for @xmath89 exact results are available from the semi - infinite two - dimensional ising model @xcite . note that @xmath58 for any fixed value of @xmath54 _ decreases _ with decreasing @xmath24 @xcite . this reflects the general trend that critical fluctuations , which reduce the mean value of the order parameter , are more pronounced in lower spatial dimensions . the curves in the inset of fig.[fig_hs ] are scaled so that they demonstrate the leading behavior of @xmath90 . ( a similar representation will be used in sec.[sec_mf ] . ) in table [ tab_ca ] we quote the corresponding numerical values of the surface amplitudes @xmath60 and @xmath61 according to eq.([a40 ] ) . in order to achieve a presentation in the inset of fig.[fig_hs](b ) which reflects both the leading behavior of @xmath91 and the exponential decay @xmath92 we introduce the function @xmath93 the curves in the inset of fig.[fig_hs](b ) comply with the above condition since @xmath94 and @xmath95 so that in both limits the leading asymptotic behaviors of @xmath96 are not changed . + .numerical values of the universal amplitudes @xmath60 and @xmath61 ( see eq.([a40 ] ) ) . [ cols="<,^,^,^,^,^,^",options="header " , ] next , we consider the cylinder , i.e. , @xmath97 and @xmath98 , which represents a relevant perturbation ( see sec.[subsec_sre ] ) . in this case eq.([a280 ] ) holds for the behavior of @xmath99 . thus @xmath100 becomes independent of @xmath101 in the limit @xmath102 and tends to the constant given by the integral in eq.([e130 ] ) with @xmath101 replaced by zero . this leads to @xmath103 with the universal numbers @xmath104 \ , \ , .\ ] ] numerical values of @xmath105 are only known for @xmath36 at present ( see table [ tab_neu ] and sec.[subsec_exmf ] ) . note that the numbers @xmath105 are not defined for @xmath89 ( compare sec.[subsec_sre ] ) . the exponent in the power law in eq.([e180 ] ) differs from that in eq.([e150 ] ) @xmath10 apart from the difference generated on purely geometric grounds @xmath10 by @xmath65 due to the different small @xmath101 behavior of the critical adsorption profiles for spheres as compared with cylinders . equation ( [ e180 ] ) in conjunction with eq.([e70 ] ) leads to @xmath106 for @xmath70 the universal function @xmath107 in eq.([e90 ] ) tends to the universal number @xmath108 with @xmath105 given by eq.([e190 ] ) . again , the limit @xmath109 is independent of the definition for the correlation length ( compare eq.([e170 ] ) ) . the critical fluctuations of the fluid are described by the standard hamiltonian @xcite @xmath110 for a scalar order parameter field @xmath111 supplemented by the boundary condition @xmath112 at the surface of the sphere or the cylinder corresponding to the critical adsorption fixed point @xcite . the position vector @xmath113 covers the volume @xmath114 accessible to the critical fluid . the parameter @xmath115 is proportional to @xmath2 and @xmath116 is the @xmath117 coupling constant . the thermal average @xmath118 corresponding to the hamiltonian in eq.([m10 ] ) with the boundary condition for critical adsorption at the surface of @xmath25 can be systematically expanded in terms of increasing powers of @xmath116 , i.e. , @xmath119 \ , \ , , \quad s = r_{\perp } - r \ , \ , .\ ] ] the leading contribution to this expansion corresponds to the mean - field result for the order parameter profile which becomes exact in the limit @xmath120 . the profile @xmath121 is determined by minimizing @xmath122 which leads to the ginzburg - landau type equation of motion @xmath123 which is supplemented by the boundary conditions @xcite [ m70 ] @xmath124 @xmath125 the parameter @xmath115 is related to @xmath2 and to the correlation length @xmath50 by @xmath126 and ( compare the text following eq.([a20 ] ) ) @xmath127 equations ( [ m40 ] ) and ( [ m70 ] ) uniquely determine the profile @xmath128 , which is related to the universal scaling functions @xmath56 in eq.([a10 ] ) by @xmath129 within the present mean - field approach this scaling form holds because the lhs of eq.([m80 ] ) is dimensionless so that it depends only on the dimensionless variables @xmath130 and @xmath131 , where @xmath132 is related to @xmath50 by eq.([m50 ] ) . similarly , the universal scaling function @xmath133 in eq.([a62 ] ) is given by @xmath134 equations ( [ m80 ] ) and ( [ m90 ] ) lead to exact results for @xmath120 and generalized cylinders @xmath25 with arbitrary @xmath33 in the interval @xmath135 ( compare fig.[diagram ] ) . equations ( [ m40 ] ) and ( [ m70 ] ) can be solved numerically , e.g. , by means of a shooting method @xcite . in some cases analytical solutions are available . in the remaining part of this section we present the corresponding explicit results for the problems discussed in secs.[sec_op ] and [ sec_excess ] . we start with the universal scaling functions @xmath136 according to eqs.([m80 ] ) and ( [ m40 ] ) . figure[kugel ] shows their behavior as function of @xmath137 for various values of the parameter @xmath138 in the case of a sphere ( i.e. , @xmath139 with @xmath36 , compare sec.[subsec_sre ] and fig.[diagram ] ) . for our presentation we choose a scaled form which reflects the behavior of @xmath140 ( compare the inset of fig.[fig_hs ] ; here @xmath141 ) . accordingly , the curves start at the mean - field values @xmath142 and @xmath143 ( see eq.([a150 ] ) and table[tab_ca ] ) , respectively , and decay exponentially for @xmath144 . + ( 16,20 ) ( -0.5,12.8 ) ( -0.5,0.8 ) [ kugel ] ( 16,20 ) ( -0.5,12.8 ) ( -0.5,0.8 ) in order to deal with the rapid decay of @xmath145 for small values of @xmath146 in the case @xmath147 we replace the function @xmath148 in eq.([a61 ] ) by @xmath149 so that @xmath150 . we emphasize again that the function @xmath151 is introduced only in order to facilitate an appropriate representation of @xmath152 which reflects the behavior for both small and large values of @xmath153 for all values of @xmath146 ( compare the discussion related to fig.[fig_hs ] ) . the overall dependence of @xmath56 on the parameter @xmath101 is in line with eq.([a240 ] ) and the related discussion . starting from the half - space behavior for @xmath154 the profiles @xmath155 decrease with decreasing @xmath101 for any fixed value of @xmath54 and vanish in the limit @xmath102 . this latter behavior for a sphere differs from the corresponding one for a cylinder ( i.e. , @xmath97 and @xmath36 , compare sec.[subsec_sre ] and fig.[diagram ] ) . figure[cyl ] shows the corresponding behavior of @xmath56 for various values of @xmath101 . for the presentation of @xmath152 in the case of the cylinder the function @xmath156 is suitable for all values of @xmath146 . the overall dependence on @xmath101 is characterized by the fact that @xmath56 does not vanish in the limit @xmath102 but rather tends to the finite limit function @xmath157 corresponding to the critical adsorption profile on a thin needle . this is in line with eq.([a280 ] ) and the related discussion . according to eq.([a290 ] ) the curves for @xmath158 start at the mean - field values of the universal amplitudes @xmath159 which are given by @xmath160 next , we consider the universal scaling function @xmath133 in eq.([m90 ] ) corresponding to @xmath161 . figure[amp ] shows its behavior for the present case @xmath36 and several values of @xmath33 . the function @xmath133 starts at @xmath162 ( see the text following eq.([a65 ] ) ) and for @xmath163 it vanishes in the case of the sphere ( i.e. , @xmath139 with @xmath36 ) while it tends to the finite ( 16,12 ) ( -0.5,4 ) number @xmath164 from eq.([m110 ] ) in the case of the cylinder ( i.e. , @xmath165 ) . in the marginal case @xmath166 it vanishes only logarithmically ( compare sec.[subsec_sre ] and fig.[diagram ] ) . for @xmath161 analytical solutions for eq.([m40 ] ) are available in some special cases even for @xmath167 . for example , for @xmath161 and @xmath168 the differential equation can be solved by relating it to the generalized emden - fowler equation @xcite , or by realizing that it represents a so - called ` equidimensional ' equation ( see , e.g. , section 1.4 in ref.@xcite ) . the result is @xmath169 which agrees with eqs.([m90 ] ) and ( [ a85 ] ) . it is worthwhile to note that an analytical solution also exists for the particular case @xmath161 and @xmath170 for which @xmath171 from which the corresponding result for @xmath133 can be inferred from eq.([m90 ] ) . finally , for @xmath161 , @xmath172 , and arbitrary values of @xmath173 one finds @xmath174 so that @xmath175 for @xmath173 . this is in line with eq.([m114 ] ) as well as with fig.[diagram ] and the related discussion , according to which in @xmath36 a generalized cylinder @xmath25 with @xmath173 represents a relevant perturbation . in particular , eq.([m116 ] ) leads to @xmath176 representing the mean - field profile near a thin needle ( for which @xmath97 , compare sec.[subsec_sre ] ) at criticality . in appendix[app_ex ] we calculate the corresponding two - point correlation function in the presence of this profile . the asymptotic behavior of @xmath56 for @xmath36 in various limiting cases can be derived directly from the defining eqs.([m80 ] ) and ( [ m40 ] ) even for those cases in which no full analytical solution is available @xcite . \(i ) @xmath177 with @xmath138 fixed . because in this limit @xmath178 is exponentially small , in eq.([m40 ] ) one can neglect powers of it larger than one . this leads to a linear differential equation for @xmath178 which can be solved in terms of modified bessel functions @xmath179 and @xmath180 @xcite with index @xmath181 . here only the decreasing function @xmath182 must be considered . using the asymptotic behavior of @xmath182 for large arguments @xcite and eq.([m80 ] ) one finds @xmath183 where the amplitude - function @xmath184 remains undetermined by the present method and must be evaluated numerically in general . an important feature of eq.([m120 ] ) is that for increasing values of @xmath54 the exponential decay is enhanced by the algebraic prefactor . thus the decay is weaker if @xmath185 and it is weaker for a cylinder ( @xmath97 ) than for a sphere ( @xmath139 with @xmath36 ) . this is expected because the perturbation of the bulk fluid is strongest in case of the half - space ( corresponding to @xmath185 ) , less for a cylinder , and even less for a sphere . \(ii ) @xmath186 with @xmath14 and @xmath50 fixed , i.e. , @xmath187 ( compare sec.[subsec_short ] ) . the leading behavior of @xmath188 can be obtained by inserting the ansatz @xmath189 into eq.([m40 ] ) and fixing the coefficients @xmath190 and @xmath191 such that the prefactors of the most singular powers of @xmath192 cancel . the resulting expression can be arranged so that it takes the form implied by eqs.([a150 ] ) and ( [ m80 ] ) , i.e. , @xmath193 & & \ , + \ , \big [ \frac{5}{9 } \ , \frac{(d-1)^2}{4 } \ , - \ , \frac{1}{3 } \ , \frac{(d-1)(d-2)}{2 } \big ] \frac{s^2}{r^2 } \ , \ , - \ , \frac{1}{6 } \tau s^2 \ , + \ , \ldots \big\ } \ , \ , . \nonumber\end{aligned}\ ] ] equation([m150 ] ) confirms that the leading dependence on @xmath115 is analytic with the same prefactor as for the half - space . in addition , one can read off the curvature parameters according to mean - field theory , i.e. , @xmath194 we note that the derivation of eq.([m150 ] ) implicitly contains a consistency check for the validity of the general curvature expansion ( [ a150 ] ) . for example , the parameters @xmath195 and @xmath196 are fixed by considering only two different types of generalized cylinders @xmath25 with nonvanishing curvature , e.g. , those for which @xmath97 and @xmath166 . equation ( [ m160 ] ) , however , holds for any value of @xmath33 in the interval @xmath135 with @xmath36 which encompasses , in particular , the three integer values @xmath97 , @xmath85 , and @xmath197 ( compare fig.[diagram ] ) . thus the curvature parameters are overdetermined . in the present case , however , this consistency may be regarded as a simple consequence of the differential equation ( [ m40 ] ) in which the perturbation generated by the surface curvature is proportional to @xmath198 . in the next subsection we consider the small curvature expansion ( [ e120 ] ) for the excess adsorption , for which the corresponding consistency check in @xmath36 provides a more stringent test . \(iii ) @xmath102 with @xmath192 and @xmath50 fixed , i.e. , @xmath199 ( compare sec.[subsec_sre ] ) . we consider the case of a sphere , i.e. , @xmath139 with @xmath36 . by inserting the ansatz @xmath200 into eq.([m40 ] ) and keeping only terms linear in @xmath14 one obtains a linear differential equation for @xmath201 , similar as in case ( i ) above . the solution of this differential equation in terms of the modified bessel function @xmath202 @xcite in conjunction with eq.([m80 ] ) leads to @xmath203 here the constant prefactor @xmath204 is fixed because the limit @xmath186 of the rhs of eq.([m170 ] ) must reproduce eq.([a65 ] ) , in which @xmath205 for @xmath206 according to eq.([a85 ] ) . equation ( [ m170 ] ) is in line with eq.([a240 ] ) because for @xmath36 one has @xmath207 and the universal scaling functions @xmath208 are given by @xmath209 for @xmath36 eq.([e70 ] ) reduces to @xmath210 with @xmath211 and @xmath212 . the universal scaling function @xmath213 in eq.([q10 ] ) is given by eq.([e60 ] ) with @xmath56 and @xmath58 from eqs.([m80 ] ) and ( [ a30 ] ) . in the following we discuss the behavior of @xmath213 in the limit of large and small values of @xmath138 and give results for the whole range of @xmath101 representing the crossover between these two limits . \(i ) @xmath213 for @xmath55 ( compare sec.[subsec_big ] ) . by using the differential equation ( [ m40 ] ) in conjunction with eq.([m80 ] ) one can determine the coefficients @xmath214 and @xmath215 in eq.([e110 ] ) for generalized cylinders @xmath25 with @xmath36 and arbitrary @xmath33 in the interval @xmath216 . this calculation is presented in appendix [ appb ] . as a result one finds that the dependences of @xmath217 and @xmath218 on @xmath33 are precisely of the form given by eq.([e125 ] ) . in the present case @xmath36 one has @xmath219 and the curvature parameters are given by @xcite [ q20 ] @xmath220 @xmath221 @xmath222 the consistency with eq.([e125 ] ) for @xmath36 can be traced back to nontrivial properties of the differential equation ( [ m40 ] ) ( see appendix [ appb ] ) and thus provides a valuable and important check for the validity of the general curvature expansion ( [ e120 ] ) . we expect that the curvature expansion ( [ e120 ] ) is also valid in @xmath88 and that the corresponding numerical prefactors multiplying @xmath18 or @xmath223 in the curvature parameters for @xmath88 differ only quantitatively from those in eq.([q20 ] ) valid for @xmath36 . as an illustration , consider a curved _ membrane _ with both sides exposed to a fluid near criticality . in particular we consider the _ total _ excess adsorption , i.e. , the sum of the excess adsorptions on each side of the membrane , per unit area . in this case the contributions to @xmath224 in the expansion ( [ e120 ] ) from each side cancel and the signs in eqs.([q20b ] ) and ( [ q20c ] ) in conjunction with eq.([a140 ] ) imply that the total excess adsorption is _ larger _ near spherical regions ( @xmath139 ) of the membrane as compared to flat regions , whereas near cylindrical regions ( @xmath97 ) it is _ smaller _ as compared to flat regions . \(ii ) @xmath213 for @xmath102 ( compare sec.[subsec_low ] ) . in this case the behaviors for spheres and cylinders are qualitatively different due to the different behaviors of @xmath56 . for a sphere ( i.e. , @xmath139 with @xmath36 ) the power law ( [ e150 ] ) is valid where the exponent @xmath225 equals @xmath226 and according to eqs.([e160 ] ) and ( [ m180 ] ) the universal amplitudes @xmath227 are given by @xmath228 for a cylinder ( i.e. , @xmath97 and @xmath36 ) one finds the behavior ( [ e180 ] ) where the universal numbers @xmath105 in eq.([e190 ] ) can be evaluated numerically with the result @xmath229 \(iii ) the full scaling functions @xmath213 describe the crossover between their analytic behaviors for @xmath230 and the power laws for @xmath102 which have been discussed in ( i ) and ( ii ) above . figures [ fig7 ] and [ fig8 ] show numerical results for @xmath213 corresponding to a sphere and a cylinder , respectively , in @xmath36 . the results confirm the small curvature expansion as implied by eqs.([e110 ] ) - ( [ e125 ] ) , and ( [ q20 ] ) and the properties ( [ e150 ] ) and ( [ e180 ] ) , and they provide the range of validity of the asymptotic behavior . note that in the case of the sphere @xmath231 as function of @xmath232 diverges for @xmath233 with the corresponding power law whereas in the case of the cylinder with increasing @xmath234 it interpolates between one finite value related to @xmath235 to the other finite value @xmath105 . + ( 16,20 ) ( -0.5,12.8 ) ( -0.5,0.8 ) [ fig7 ] ( 16,17.5 ) ( -0.5,7.5 ) we have studied critical adsorption phenomena on spherical and cylindrical particles of radius @xmath14 which are immersed in a fluid near criticality , @xmath236 , for the case that the fluid is at the critical composition . the corresponding adsorption profiles at the radial distance @xmath192 from the surface are characterized by universal scaling functions @xmath237 for @xmath238 , involving the bulk correlation lengths @xmath50 for @xmath239 , and @xmath240 for @xmath241 ( see eqs.([a10 ] ) and ( [ a62 ] ) , respectively ) . in the following we summarize our main results starting with _ local properties _ of @xmath56 and @xmath133 in various limiting cases as indicated by fig.[fig_land ] . 1 . for @xmath241 we have introduced the short distance expansion of the order parameter profile near a weakly curved @xmath242-dimensional surface of _ general _ shape ( see eq.([a110 ] ) ) . this expansion involves the local curvature invariants @xmath243 , @xmath244 , and @xmath245 ( see eq.([a100 ] ) ) . the corresponding expansion parameters @xmath246 , @xmath195 , and @xmath196 appear also in the short distance expansion ( [ a150 ] ) of @xmath56 valid for @xmath239 . the parameters @xmath246 , @xmath195 , and @xmath196 are universal and depend only on the space dimension @xmath24 ( see eqs.([a120 ] ) , ( [ a130 ] ) , and ( [ m160 ] ) ) . 2 . for @xmath247 the order parameter profile near a cylinder becomes independent of @xmath14 in the limit @xmath199 , i.e. , if the cylinder radius is microscopically small ( see eqs.([a260 ] ) - ( [ a290 ] ) ) . in contrast , near a sphere the ` universal part ' of the order parameter profile , i.e. , the one described by eqs.([a230 ] ) and ( [ a240 ] ) , vanishes for @xmath199 . this _ qualitative _ difference in behavior can be explained by means of a small radius operator expansion ( see eqs.([a170 ] ) and ( [ a180 ] ) ) . as indicated by fig.[diagram ] a sphere is an irrelevant perturbation for the fluid near criticality whereas a cylinder is a relevant perturbation . the explicit forms of the universal scaling functions @xmath56 and @xmath133 for a sphere and a cylinder within mean - field approximation ( see figs.[kugel ] - [ amp ] ) corroborate the results described in the above paragraphs ( 1 ) and ( 2 ) . we now turn to the _ excess adsorption _ @xmath248 describing the total enrichment of the preferred component of the fluid near criticality in proximity of a sphere or cylinder ( see eq.([e10 ] ) ) . the curvature dependence of @xmath248 is characterized by universal scaling functions @xmath249 obtained from @xmath250 by integrating over the first variable @xmath251 ( see eqs.([e60 ] ) and ( [ e70 ] ) ) . 1 . for @xmath252 we have introduced the expansion of @xmath248 near a weakly curved @xmath242-dimensional surface of _ general _ shape in terms of the local curvature invariants @xmath243 , @xmath244 , and @xmath245 ( see eq.([e120 ] ) ) . the corresponding expansion parameters @xmath235 , @xmath253 , and @xmath254 depend only on the space dimension @xmath24 and can be expressed in terms of the universal coefficients @xmath214 and @xmath215 appearing in the expansion of @xmath255 ( see eqs.([e110 ] ) - ( [ e125 ] ) ) . the explicit calculation of @xmath235 , @xmath253 , and @xmath254 within mean - field approximation ( see eq.([q20 ] ) and appendix [ appb ] ) provides an important check for the validity of the general curvature expansion of the excess adsorption . 2 . for @xmath256 the behaviors of @xmath248 and @xmath213 are characterized by power laws ( see eqs.([e150 ] ) - ( [ e165 ] ) for a sphere and ( [ e180 ] ) - ( [ e200 ] ) for a cylinder ) . the exponents in these power laws are different for spheres and cylinders beyond the purely geometrical effects . this can be traced back to the corresponding difference in behavior of @xmath56 addressed in paragraph ( 2 ) . the explicit forms of @xmath213 for a sphere and a cylinder in mean - field approximation ( see figs.[fig7 ] and [ fig8 ] ) confirm the general results presented in paragraphs ( 4 ) and ( 5 ) . we conclude by summarizing some of the _ fieldtheoretic developments _ and _ perspectives _ for future theoretical work , simulations , and experiments . 1 . for a systematic field - theoretical analysis of the critical adsorption phenomena on spheres and cylinders it is useful to introduce the particle shape of a ` generalized cylinder ' ( see eq.([i10 ] ) and fig.[fig_cyl ] ) which is characterized by the space dimension @xmath24 and an internal dimension @xmath33 encompassing a sphere , a cylinder , and a planar wall as special cases . 2 . for a single sphere , certain asymptotic behaviors of the order parameter _ profile _ are characterized by universal quantities ( see eqs.([a150 ] ) and ( [ a240 ] ) ) . these quantities are accurately known in @xmath88 ( see eq.([a120 ] ) , ( [ a130 ] ) , and table [ tab_ca ] ; as far as the scaling functions @xmath208 are concerned , only the function @xmath257 appropriate for @xmath258 is known accurately for @xmath88 , see appendix [ app_neu ] ) . the corresponding predictions can be tested using small angle scattering of light , x - rays or neutrons which can probe the enlarged effective size of the colloidal particles due to the adsorption . 3 . for a single sphere , in @xmath88 the behavior of the _ excess adsorption _ in the limit @xmath256 is available as well ( see eqs.([e70 ] ) , ( [ e150 ] ) , and table [ tab_neu ] ) and can be tested experimentally by , e.g. , volumetric measurements . for a disc in @xmath89 the numerical values of the universal amplitudes @xmath227 in eq.([e150 ] ) are also known ( compare appendix [ app_neu ] ) . this can be relevant for protein inclusions in fluid membranes @xcite . 4 . apart from the curvature dependence of the excess adsorption ( see eq.([e120 ] ) ) also the corresponding energetic contributions , i.e. , the change of the surface tension generated by the surface curvature , can be relevant for applications . for example , in case of a membrane immersed in a fluid near criticality such contributions and their temperature dependence are expected to influence the intrinsic bending rigidities of the membrane . these modifications of the bending rigidities can , in turn , induce shape changes of vesicles formed by closed membranes in a controllable way ( see , e.g. , ref.@xcite ) . 5 . a one - dimensional extended perturbation in an ising - like system which breaks the symmetry of the order parameter represents a relevant perturbation ( see sec.[subsec_sre ] ) of the bulk system . this gives rise to new universal quantities such as the critical exponents @xmath259 and @xmath260 characterizing the decay of the structure factor ( see appendix [ app_ex ] and in particular eqs.([g90 ] ) and ( [ g110 ] ) ) and the amplitudes @xmath159 and @xmath105 ( see eqs.([a290 ] ) and ( [ e190 ] ) , respectively ; the mean - field values of @xmath159 and @xmath105 are quoted in eqs.([m110 ] ) and ( [ q40 ] ) ) . apart from a rodlike particle immersed in a fluid , such a one - dimensional perturbation could also be realized in a _ solid _ , e.g. , by a lattice dislocation in a binary alloy if the dislocation locally breaks the symmetry of the order parameter . this would be a natural starting point for testing the numerous fieldtheoretic predictions derived here by monte carlo simulation . as stated in the introduction , the results addressed in paragraph ( 2 ) can be relevant for the flocculation of colloidal particles dissolved in a fluid near criticality . specificly , if one considers the case @xmath241 and compares the behaviors of the order parameter profiles near a sphere and near an infinitely elongated cylinder , respectively , the order parameter of the fluid at a distance @xmath261 from the particle with radius @xmath14 is larger by a factor @xmath262 for the cylinder than for the sphere ( see eqs.([a62 ] ) , ( [ a230 ] ) and ( [ a260 ] ) ) . for rods with a large but finite length @xmath19 and distances @xmath263 this enhancement is of the order of @xmath264 . since the critical casimir forces between particles are partially induced by an overlap of the order parameter profiles generated by the individual particles in the space between them , the aforementioned results suggest that not only spheres but , _ a fortiori _ , also long rods in a fluid near criticality can aggregate due to critical casimir forces . we thank prof . e. eisenriegler for helpful discussions and profs . a. j. bray and m. e. fisher for helpful correspondence . this work has been supported by the german science foundation through sonderforschungsbereich 237 _ unordnung und groe fluktuationen_. the two - point correlation function @xmath265 for gaussian fluctuations around the profile @xmath176 ( see the text following eq.([m116 ] ) ) _ at _ criticality satisfies the differential equation @xmath266 \ , g({\bf r } , { \bf r}\ , ' ) \ , = \ , \delta^{(d)}({\bf r } - { \bf r}\,')\ ] ] where @xmath267 is the laplacian operator in @xmath24-dimensional space . ( although the present mean - field calculation implies @xmath36 we leave the symbol @xmath24 for clarity . ) in coordinates adapted to the geometry of a generalized cylinder @xmath25 with @xmath97 and @xmath172 the laplacian operator @xmath267 has the form @xmath268 with @xmath269 as the laplacian operator in the two - dimensional subspace perpendicular to @xmath25 . here @xmath270 denotes the angle between @xmath31 and a fixed direction in the radial subspace . in order to ( 16,13 ) ( 2,-3 ) solve eq.([g10 ] ) we carry out the fourier transform in the subspace parallel to @xmath25 and a partial wave decomposition in the radial subspace , i.e. , @xmath271 & & = \ , \sum_{n=0}^{\infty } \ , w_{n}(\vartheta ) \ , \int\frac{d^{d-2\,}p}{(2 \pi)^{d-2 } } \ , \exp[i \ , { \bf p } \ , ( { \bf r}_{\parallel } - { \bf r}_{\parallel}^{\,\ , ' } \ , ) ] \ , \ , \widetilde{g}_{n}(r_{\perp},r_{\perp}^{\,\ , ' } , p ) \ , \ , , \nonumber\end{aligned}\ ] ] where @xmath270 is now the angle between @xmath31 and @xmath272 ( see fig.[fig_needle ] ) . the functions @xmath273 are given by @xmath274 with @xmath275 for @xmath276 and zero otherwise . they obey @xmath277 where the support of the @xmath26-function on the rhs is located entirely in the interval @xmath278 so that @xmath279 for any @xmath280 . the propagator @xmath281 in eq.([g40 ] ) satisfies the radial equation @xmath282 \ , \widetilde{g}_{n}(r_{\perp},r_{\perp}^{\,\ , ' } , p ) \ , = \ , \frac{\delta(r_{\perp } - r_{\perp}^{\,\,'})}{r_{\perp } } \ , \ , .\ ] ] for @xmath283 the solutions of this equation are modified bessel functions @xmath284 and @xmath285 @xcite with index @xmath286 the physically acceptable solution @xmath281 of eq.([g60 ] ) is uniquely determined by the requirements that ( i ) it is symmetric in @xmath287 and @xmath288 , ( ii ) it decays for @xmath289 , and ( iii ) it does not diverge as the singularity at @xmath290 in eq.([g60 ] ) is approached . the result is @xmath291 with @xmath292 , @xmath293 , and @xmath294 given by eq.([g70 ] ) . equations ( [ g40 ] ) and ( [ g80 ] ) represent the two - point correlation function near a thin needle in the presence of the adsorption profile @xmath295 at criticality . in the following we infer the asymptotic behavior of @xmath265 in various limits . \(a ) @xmath287 , @xmath288 fixed and @xmath296 . the rhs of eq.([g80 ] ) can be expanded for @xmath297 , e.g. , by relating @xmath298 to @xmath299 and @xmath300 via the corresponding formula in section 9.6.2 of ref.[45(a ) ] and by using the formula in section 9.6.10 of the same reference . apart from terms proportional to integer powers of @xmath301 , which are analytic in @xmath302 yielding only exponentially decaying contributions on the lhs of eq.([g40 ] ) , one finds that the _ leading singular _ contribution behaves as @xmath303 . regarding the leading behavior for @xmath296 in eq.([g40 ] ) only the term with @xmath276 is important so that [ g90 ] @xmath304 with @xmath305 \(b ) @xmath306 , @xmath288 fixed and @xmath289 . for @xmath307 using eq.([g40 ] ) and rescaling @xmath308 leads to @xmath309 here @xmath310 can be expanded for @xmath311 so that in this limit again only the term with @xmath276 is important and [ g110 ] @xmath312 with @xmath313 we note that within the present mean - field theory the relation @xmath314 is fulfilled because @xmath315 . the exponents in eqs.([g90 ] ) and ( [ g110 ] ) for a thin needle are different from those for the half - space for which @xmath316 and @xmath317 in mean - field approximation . thus the correlations in the critical fluid are more suppressed in proximity of the needle than in the bulk fluid , and are even more suppressed near the surface bounding the half - space . this can be understood by noting that a needle represents a weaker , but still relevant , perturbation for the critical fluid than a planar wall ( compare fig.[diagram ] ) . for a sphere one finds that the two - point correlation function at criticality decays as @xmath318 with the bulk exponent @xmath319 if the distance vector @xmath320 is increased in any direction . this reflects the fact that a sphere represents an irrelevant perturbation . in this appendix we determine the universal amplitudes @xmath227 ( see eqs.([e150 ] ) and ( [ e160 ] ) ) and the universal scaling functions @xmath208 ( see eqs.([a240 ] ) and ( [ a250 ] ) ) . the corresponding mean - field results ( see eqs.([q30 ] ) and ( [ m180 ] ) , respectively ) hold in @xmath36 . for @xmath88 and @xmath86 results are available @xcite for the fourier transform of the bulk two - point correlation function introduced in eq.([a250 ] ) , i.e. , @xmath322 which can be written in the scaling form @xmath323 the universal scaling functions @xmath324 are fixed by the normalization condition @xmath325 and by the choice of @xmath50 as the true correlation length for @xmath239 . the amplitudes @xmath326 are nonuniversal and @xmath327 is a universal bulk critical exponent . numerical values of the universal bulk critical exponent @xmath319 are given by @xmath328 and @xmath329 @xcite ( compare eq.([a45 ] ) ) . inserting eq.([a250 ] ) into eq.([n10 ] ) and using eq.([e160 ] ) yields @xmath330 with @xmath331 . the universal numbers @xmath332 are given by @xmath333 n \ , r_{+}^{\,2-\eta } \ , r_{\xi}^{\,\eta - 2 } \ , \displaystyle{\frac{c^{+}}{c^{- } } } \ , q_3 \ , \ , \ , , & t < t_c \ , \ , , \end{array } \right.\ ] ] with the universal amplitude ratio @xmath334 introduced by tarko and fisher @xcite . the amplitudes @xmath335 correspond to correlation length defined via the second moment @xcite of the correlation function , i.e. , @xmath336 . the amplitude @xmath337 is defined by @xmath338 which implies that @xmath339 in eq.([a200 ] ) is related to @xmath337 via @xmath340 with the numbers @xmath341 \ , = \ , 0.654308 \ , \ , \ , , & d = 2 \ , \ , , \\[4 mm ] \gamma(\eta ) \sin(\frac{\eta \ , \pi}{2 } ) / ( 2 \pi^2 ) \ , = \ , 0.078196 \ , \ , \ , , & d = 3 \ , \ , . \end{array } \right.\ ] ] numerical values of @xmath342 are given by @xmath343 and @xmath344 @xcite . in eq.([p20 ] ) @xmath345 , @xmath346 , and @xmath347 are known universal amplitude ratios with @xmath348 ( see eq.(3.7 ) in ref.@xcite ) , @xmath349 @xcite ; @xmath350 , @xmath351 @xcite ; @xmath352 @xcite , @xmath353 @xcite . using eqs.([p10 ] ) and ( [ p20 ] ) in conjunction with the values of @xmath354 from table [ tab_ca ] and with @xmath355 , and @xmath356 yields the values of @xmath227 quoted in table [ tab_neu ] . we note that the accuracy of the quoted value of @xmath357 is unknown because the accuracy of the aforementioned value of @xmath358 is not given reliably @xcite . next we outline how the universal scaling functions @xmath208 introduced in eq.([a250 ] ) can be inferred from presently available results for @xmath324 ( see eqs.([n10 ] ) and ( [ n20 ] ) ) . \(a ) @xmath88 : in this case the scaling functions @xmath208 are given by @xmath359 with the amplitudes @xmath360 fixed by the condition @xmath361 which allows one to express the nonuniversal amplitudes @xmath326 in terms of @xmath339 and @xmath46 . for the case @xmath258 the approximation for @xmath362 proposed by bray @xcite can be regarded to be reliable . accordingly , for values @xmath363 the function @xmath362 can be inferred from the first column in table v of ref.@xcite whereas for @xmath364 the asymptotic expansion by fisher and langer @xcite is applicable , i.e. , @xmath365 \ , \ , , \quad d = 3 \ , \ , , \ ] ] with coefficients @xmath366 and the bulk critical exponents @xmath319 , @xmath367 , and @xmath6 . using in eqs.([n30 ] ) and ( [ n40 ] ) the values @xmath368 , @xmath369 , and @xmath370 as quoted in ref.@xcite yields @xmath371 , @xmath372 , @xmath373 @xcite , and @xmath374 . unfortunately , for @xmath147 and @xmath88 we are not aware of an accurate estimate of @xmath375 . \(b ) @xmath89 : in this case the scaling functions @xmath208 are given by @xmath376 where @xmath377 is a modified bessel function @xcite . in @xmath89 exact results for @xmath324 can be deduced from ref.@xcite . however , some care is necessary regarding the definition of the scaling variable @xmath378 used in ref.@xcite : whereas @xmath379 for @xmath258 , one has @xmath380 for @xmath147 . the numerical prefactor in front of @xmath381 equals @xmath382 with @xmath383 from table iv in ref.@xcite ( see eq.(3.7 ) in ref.@xcite ) and with the universal amplitude ratio @xmath384 in @xmath89 @xcite . for the amplitudes @xmath385 in eq.([n50 ] ) one obtains @xmath386 and @xmath387 . in this appendix we outline how the coefficients @xmath214 and @xmath215 introduced in eq.([e110 ] ) can be calculated within mean - field theory , i.e. , for @xmath36 . since the approaches for @xmath258 ( @xmath9 ) and for @xmath147 ( @xmath10 ) are quite similar we restrict our presentation to the case @xmath258 . according to eqs.([m40 ] ) and ( [ m80 ] ) for @xmath36 the universal scaling function @xmath388 satisfies the nonlinear differential equation ( in the following we drop the subscript ` @xmath9 ' in order to simplify the notation ) [ b10 ] @xmath389 where the derivatives are taken with respect to the variable @xmath390 . the parameter @xmath33 with @xmath216 can be chosen arbitrarily . equation ( [ b10a ] ) is supplemented by the boundary conditions @xmath391 the scaling function @xmath392 from eq.([e60 ] ) then reads with @xmath394 defined by eq.([b10 ] ) and the half - space profile @xmath395 given by the first part of eq.([a30 ] ) . note that the singularity at @xmath396 of the first term in curly brackets in eq.([b20 ] ) is cancelled by the second term . according to eq.([e110 ] ) the function @xmath397 can be expanded as @xmath398 with the coefficients @xmath399 and @xmath400 which we want to determine . to this end we expand @xmath394 as @xmath401 by inserting eq.([b40 ] ) into eq.([b10a ] ) and equating terms with the same power in @xmath378 one derives the familiar nonlinear differential equation for the half - space profile @xmath402 , i.e. , in the integrand on the rhs of eq.([b20 ] ) in conjunction with eq.([b40 ] ) we find for the coefficients in eq.([b30 ] ) the expressions @xmath412 and @xmath413 where @xmath402 , @xmath409 , and @xmath410 are the solutions of eqs.([b50 ] ) - ( [ b70 ] ) . while @xmath402 is given by the first part of eq.([a30 ] ) we now turn to the calculation of @xmath409 and @xmath410 . it is important to retain the dependence of the latter functions on @xmath33 in analytical form in order to be able to carry out the consistency check of eq.([e125 ] ) for the present case @xmath36 . \(i ) function @xmath409 . both the rhs of eq.([b60a ] ) and the boundary condition in the first part of eq.([b60b ] ) depend on @xmath33 via the term @xmath414 . we note that even the full function @xmath409 exhibits this simple dependence on @xmath33 . the reason is that eq.([b60a ] ) is linear with respect to @xmath415 so that eq.([b60 ] ) is solved by the ansatz @xmath416 where @xmath417 is independent of @xmath33 and satisfies the inhomogeneous linear differential equation \(ii ) function @xmath410 . by using eq.([b110 ] ) the differential equation ( [ b70a ] ) turns into @xmath421 with @xmath417 from eq.([b130 ] ) . in this case both on the rhs of eq.([b140 ] ) and in the boundary condition in the first part of eq.([b70b ] ) different dependences on @xmath33 arise via the terms @xmath422 and @xmath414 . again it is crucial to observe that eq.([b140 ] ) is linear with respect to @xmath423 so that its dependence on @xmath33 takes the simple form @xmath424 where @xmath425 and @xmath426 are independent of @xmath33 and satisfy separately the inhomogeneous linear differential equations [ b170 ] @xmath429 @xmath430 equations ( [ b160 ] ) and ( [ b170 ] ) can be solved numerically , which is facilitated by the fact that both @xmath417 and @xmath402 on the rhs of eqs.([b160a ] ) and ( [ b170a ] ) are known in analytical form . by inserting the resulting function @xmath410 from eq.([b150 ] ) and the function @xmath409 from eq.([b110 ] ) into eqs.([b90 ] ) and ( [ b100 ] ) one obtains the dependences of @xmath431 and @xmath432 on @xmath33 as given by eq.([e125 ] ) with the curvature parameters given in the first parts of eq.([q20 ] ) . the procedure in the case @xmath147 is completely analogous to that for @xmath258 as outlined above and yields the curvature parameters in the second parts of eq.([q20 ] ) . for the case of parallel plates see m. krech and s. dietrich , phys . a * 46 * , 1886 ( 1992 ) ; * 46 * , 1922 ( 1992 ) ; m. krech , _ the casimir effect in critical systems _ ( world scientific , singapore , 1994 ) ; m. krech , phys . e * 56 * , 1642 ( 1997 ) . such a generalized cylinder has already been introduced in the different context of the interaction between colloidal particles and flexible polymer chains in a dilute polymer solution , see e. eisenriegler , a. hanke , and s. dietrich , phys . e * 54 * , 1134 ( 1996 ) . any definition @xmath433 different from that of @xmath50 as the true correlation length is related to @xmath50 by the universal ratio @xmath434 . this leads to a corresponding change of the scaling functions @xmath43 and of quantities obtained from them so that experimentally observable quantities remain unchanged ( for more details see appendix [ app_neu ] and ref.@xcite ) . the bars on @xmath61 , @xmath62 , @xmath63 serve to distinguish these coefficients from the corresponding ones @xmath435 , @xmath436 , @xmath437 introduced in ref.@xcite which are defined differently from eq.([a40 ] ) . the first singular contribution proportional to @xmath63 in eq.([a40 ] ) can be attributed @xcite to the leading nontrivial ` surface operator ' @xmath438 with the stress tensor operator @xmath439 whose ` scaling dimension ' is equal to the spatial dimension @xmath24 . we briefly outline the argument : since the curvature contributions must depend on how the surface is embedded in the space @xmath440 they must be derivable from the local extrinsic curvature tensor @xmath441 where @xmath442 is the local unit vector normal to the surface ( we only consider orientable surfaces ) . the principal local radii of curvature @xmath443 are the inverse of the @xmath444 eigenvalues of the matrix @xmath445 . therefore @xmath243 , @xmath244 , and @xmath245 as defined in eq.([a100 ] ) are the only independent scalar quantities to first and second order in @xmath446 which can be deduced from @xmath447 and which are invariant under permutations of the indices @xmath448 . the leading regular term captured by the ellipses in eq.([a150 ] ) which is analytic in @xmath449 is proportional to @xmath450 . since all terms in curly brackets depend only on the dimensionless variables @xmath54 and @xmath101 this regular term must scale with @xmath451 where the exponent is larger than two if @xmath452 . it is instructive to interpret eq.([m40 ] ) as an equation of motion of a particle at ` position ' @xmath456 and ` time ' @xmath192 subjected to the ` potential energy ' @xmath457 @xcite . in general the differential equation can not be solved analytically due to the time - dependent friction term . one seeks solutions @xmath458 which come at rest on the top @xmath459 of the potential hill at time @xmath460 . this problem can be solved numerically , e.g. , by means of a ` shooting method ' as a trial and error approach . \(a ) m. abramowitz and i. a. stegun , _ handbook of mathematical functions _ ( dover , new york , 1972 ) ; ( b ) i. s. gradshteyn and i. m. ryzhik , _ table of integrals , series , and products _ ( academic , london , 1965 ) . the curvature parameters for a convex surface as considered here should , of course , be the same as for a concave surface . in the present mean - field case this does indeed hold . this can be confirmed by using the methods outlined in appendix [ appb ] .

we have carried out a detailed abundance analysis for 21 elements in a sample of four rgb stars in the outer halo globular cluster ngc 7492 ( 25 kpc ) ; we find [ fe / h ] = dex inferred from fe i lines ( from fe ii ) using high dispersion ( r==35,000 ) spectra obtained with hires at the keck observatory . most elements show no sign of star - to - star variation within our limited sample . we have , however , detected an anti - correlation between o and na abundances similar to that seen in our previous analyses of inner halo gcs as well as in studies of relatively nearby gcs by others . we compare the abundance ratios in ngc 7492 with those we previously determined for the much closer old halo gcs m3 and m13 . after making corrections for trends of abundance ratio with metallicity characteristic of halo stars , we find that for for these three gcs for each of the elements in common we deduce identical abundance ratios with respect to fe to within the probable measurement uncertainties . thus , the chemical history of the outer halo as exemplified by the metal - poor outer halo globular cluster ngc 7492is indistinguishable from that of the inner halo , exemplified by m3 and m13 , at least through the epoch of formation of these old globular clusters . this applies to the neutron capture processes as well .

astro-ph0411621

abundance determinations of stars in galactic globular clusters can provide valuable information about important astrophysical processes such as stellar evolution , stellar structure , galactic chemical evolution and the formation of the milky way . with the advent of efficient high resolution spectrographs on 8 - 10 m telescopes , it is now possible to reach at least the luminous rgb stars in even the most distant galactic globular clusters ( gcs ) . in our previous work in this area we have explored the abundances for large samples of stars in the canonical relatively nearby gcs m71 @xcite , m5 @xcite m3 , and m13 @xcite , as well as in pal 12 @xcite , a cluster associated with the sgr dwarf galaxy @xcite . for the case of pal 12 , in addition to the difference in age of several gyr , it being younger than the bulk of the galactic gcs @xcite , we found evidence for substantive differences between the chemical history of pal 12 and that of the `` normal '' halo gcs of similar [ fe / h ] is given by @xmath4 on a scale where @xmath5 h atoms . then [ x / h ] = log[n(x)/n(h ) ] @xmath6 log[n(x)/n(h ) ] , and similarly for [ x / fe ] . ] . in the present work , we study the outer halo gc ngc 7492 , a cluster which is not suspected ( yet ) of being part of any known stream or otherwise abnormal . we compare the abundance ratios deduced from high resolution , high signal - to - noise ratio for four giants in this distant outer halo gc with those from our recent analysis of a large sample of stars in the relatively nearby and well studied gcs m3 and m13 , which have metallicities close to that of ngc 7492 . we look for evidence in the deduced abundance ratios of some difference in the formation mechanisms or chemical history of gcs in the `` normal '' outer halo . there has been no previous high dispersion study of ngc 7492 . @xcite obtained [ fe / h ] = @xmath7 dex using their narrow band q39 photometric system , @xcite derived @xmath8 dex from the @xmath9 method applied to two rr lyrae variables in the cluster , while moderate dispersion spectroscopy in the region of the infrared ca triplet by @xcite gave [ fe / h ] = @xmath10 dex . given the large distance of ngc 7492 , we select the brightest possible stars on the upper rgb for observation , making no attempt to reach fainter luminosities . these stars were picked from the photometric study of this cluster by @xcite . since this is a rather sparse cluster , these stars , which are the four brightest suspected members , lie distributed along the upper rgb , not concentrated at the rgb tip . the positions of these stars in a @xmath12 cmd are illustrated in figure [ figure_cmd ] superposed on the predicted cluster isochrone from the @xcite evolutionary tracks ; see figure 10 of @xcite for a @xmath13 cmd of ngc 7492 . although the cluster is distant , there are so few luminous giants that they must be observed as single stars ; two of them can not be fit into a single 7 arcsec long slit . spectra were obtained with hires @xcite at the keck observatory 20 - 22 august 2003 . the instrument configuration covered the range 4650 to 7010 , with small gaps between the orders at the red end . this is the `` yellow '' configuration described in @xcite . these data were reduced using a combination of figaro @xcite scripts and the software package makee . the desired minimum snr was 90 over a 4 pixel resolution element for a wavelength near the center of the hires detector . this is calculated strictly from the counts in the object spectrum , and excludes noise from cosmic ray hits , sky subtraction , flattening problems , etc . since the nights were relatively dark , sky subtraction is not an issue except at the specific wavelengths corresponding to strong night sky emission lines , such as the na d doublet . the seeing was extremely good during these nights , making the exposures shorter than normal , and enabling us to reach this goal for the stars reported here . table [ table_sample ] gives details of the hires exposures for each star , with the total exposure time for each object . all long integrations were broken up into separate exposures , each 1200 sec long , to optimize cosmic ray removal . the last column of the table gives the heliocentric radial velocity for each star , measured from the hires spectra ; see @xcite for the details of the procedure used to determine @xmath14 . the radial velocity of ngc 7492is large and negative , and the cluster abundance is low . it was easy to tell after one integration whether or not a star is a member of the cluster . approximate measurements of @xmath14 were made on line ; all stars attempted turned out to be radial velocity members of ngc 7492 . the four rgb stars have a mean @xmath14 of @xmath15 km s@xmath16 , agreeing well within the errors with the early measurement of @xcite of a single bright cluster member , but somewhat larger than the value of @xmath17 km s@xmath16 found from moderate resolution spectra by @xcite . our more accurate @xmath14 for ngc 7492 should be used in future computations of the mass of the galaxy which rely on the orbits of its outlying satellites . the velocity dispersion from the four members , with no correction for an instrumental contribution , is @xmath18 = 1.2@xmath191 km s@xmath16 , reflecting the low mass of this sparse cluster ; the observed @xmath18 is only slightly larger than the expected instrumental uncertainties . we follow the procedures developed in our earlier work on globular cluster stars and described in @xcite to determine the stellar parameters for the four rgb stars in ngc 7492 . @xmath11 is derived by comparing reddening - corrected broad band colors with the predictions of grids of model atmospheres . we utilize here the grid of predicted broad band colors and bolometric corrections of @xcite based on the marcs stellar atmosphere code of @xcite . we normally utilize @xmath20 and @xmath21 colors to determine @xmath11 . the infrared colors were taken from 2mass @xcite . @xcite provided @xmath22 photographic photometry over a large field including the entire cluster of ngc 7492 . ccd photometry in @xmath13 was presented by @xcite , but their field is smaller than the cluster and the calibration of their photometry is not secure . @xcite , in the course of a study of blue stragglers in this cluster , obtained ccd @xmath13 photometry for a large sample of stars , but the data tables were never published and have subsequently been lost ( ct , private communication , 2004 ) . the two published photometric studies of ngc 7492 give @xmath23 mags which are inconsistent by several tenths of a mag ; the differences are not just a simple offset . we therefore measured @xmath12 for our sample stars from andicam images taken for this purpose on aug . 1 , 2004 with the 1.3 m telescope at ctio operated by the smarts consortium ; these values are given in table [ table_sample ] . andicam is a dual channel cammera constructed by the ohio state university instrument group , but only the optical channel was used . our andicam program requires photometric conditions , and additional standard star fields , charged to our andicam allocation through noao , are always taken for us . we adopt a distance for ngc 7492 of 26.2 kpc @xcite ( as compared to 7.5 kpc for m13 and 10.4 kpc for m3 ) , with a reddening of e(b v ) = 0.036 mag from @xcite . the relative extinction in various passbands is taken from @xcite ( see also * ? ? ? the adopted stellar parameters are given in table [ table_teff ] . to the maximum extent possible , the atomic data and the analysis procedures used here are identical to those we used in our recently completed analysis of a large sample of stars in m3 and m13 @xcite . in particular , see 3 of that paper for a description of the measurement of the equivalent widths ( listed for the four stars in ngc 7492 in table [ table_eqw ] ) , 4 for a discussion of the atomic parameters , 4.2 for our adopted solar abundances ( tabulated in table 2 of our earlier paper , and repeated as table [ table_sun ] here ) , and 6 for a description of our abundance analysis procedures . as in our earlier work , the abundance analysis is carried out using a current version of the lte spectral synthesis program moog @xcite . we employ the grid of stellar atmospheres from @xcite without convective overshoot , when available . the template file of suitable unblended lines with their adopted atomic parameters is identical to that we used in our recently completed analysis of m3 and of m13 @xcite . three of the four stars gave @xmath24 km s@xmath16 based on deriving a uniform fe abundance as a function of @xmath25 for the large set of fe i lines . the fourth star ( star r ) gave 1.8 km s@xmath16 ; this value was also set to 2.0 km s@xmath16 . lines with @xmath25@xmath26 175 mwere ignored , except for the ba ii lines in the coolest star in our sample in ngc 7492 . the resulting abundance ratios for the four rgb stars in ngc 7492 are given in tables [ table_abund_a ] to [ table_abund_e ] . species with only one detected line are assigned an uncertainty of 0.10 dex . table 7 of @xcite indicates the changes in derived abundance ratios for small changes in the adopted stellar parameters , the [ fe / h ] for the adopted model atmosphere , or the set of @xmath25for the lines of each species , and is appropriate for use here as well . the mean abundance and 1@xmath18 variance for the species observed in ngc 7492 are listed in table [ table_abundsig ] . the oxygen abundance is derived from the forbidden lines at 6300 and 6363 . the subtraction of the night sky emission lines for the forbidden lines was reasonably straightforward given that the radial velocity of ngc 7492 is sufficiently different from 0 km s@xmath16 that their @xmath25 can be reliably measured . the c / o ratio was assumed to be solar . cn and ni i contamination is negligable ( see * ? ? ? [ o / fe ] is given with respect to fe ii ; abundance ratios for all other elements are given with respect to fe i. the deduced mean [ fe / h ] of ngc 7492 value is in good agreement with that inferred by @xcite from moderate dispersion spectra in the region of the infrared ca triplet . the na abundance was obtained from the 5680 doublet for all four stars . we have , as in our previous papers , not used any non - lte corrections for na . calculations by @xcite suggest values of between 0.1 and 0.2 dex are appropriate for our sample in ngc 7492 , with the coolest star having the largest value . calculations by @xcite suggest somewhat smaller values . the ba abundance of star 950 in ngc 7492 appears to be @xmath270.2 dex larger than that of the other cluster members studied here . however , the detected lines of ba ii are all within the range where substantial hfs corrections occur . table 7 of @xcite shows the very high sensitivity of the deduced ba abundances to small uncertainties in the equivalent width and microturbulence of the ba ii lines we use . on the other hand , the deduced [ y / fe ] is also somewhat high for this star , while the deduced fe is the lowest of all the stars in our sample . perhaps a slight adjustment of @xmath11 for this star is required . at this point , we assume this is the result of observational and modelling uncertainties and does not indicate a real spread in [ ba / fe ] within ngc 7492 , but further verification of this is desirable . the abundances of the elements with respect to fe , [ x / fe ] , as a function of @xmath11 are shown in figure [ figure_o_si ] , covering o , na , mg and si , figure [ figure_ca_v ] , which includes ca , sc , ti and v , figure [ figure_cr_ni ] , which includes cr , mn , co and ni , figure [ figure_cu_zr ] , which includes cu , zn , y and zr , and figure [ figure_ba_dy ] , for ba , la , nd , eu and dy . note the apparent star - to - star variation in [ o / fe ] and in [ na / fe ] , which becomes undetectably small , if it exists at all , for the elements heavier than na . the scatter for [ ca / fe ] and for [ ni / fe ] , with 8 to 13 detected absorption lines in each star , is remarkably small , @xmath28 dex over the four star sample in ngc 7492 . to check for the presence of star - to - star variations in abundance ratios within the small sample of rgb stars in ngc 7492 , we use a parameter we call the `` spread ratio '' ( @xmath29 ) . the numerator of @xmath29 is the 1@xmath18 rms variance for the sample of four stars in ngc 7492 about the mean abundance for each atomic species ( @xmath30 ) with detected absorption lines , denoted @xmath18 ; the relevant values are given in the first three columns of table [ table_abundsig ] . the denominator of @xmath29 is the total expected uncertainty , @xmath31 , which is the sum in quadrature of the known contributing terms . included are a term corresponding to an uncertainty of 50 k in @xmath11 , the same for an uncertainty of 0.2 dex in log(@xmath32 ) , and for an uncertainty of 0.2 km s@xmath16 in @xmath33 , and the observed uncertainty [ @xmath34 , the parameter @xmath35 , which is calculated from data given in tables [ table_abund_a ] to [ table_abund_e ] , is taken as the variance about the mean abundance for a given species in a given star , i.e. the 1@xmath18 rms value about the mean abundance of species @xmath30 in a given star/@xmath36 , where @xmath37 is the number of observed lines of species @xmath30 . it includes contributions from errors in the measured @xmath25 , random errors ( i.e. between lines of a given species ) in the adopted @xmath38 values , etc . some species , an example being fe i with its very large value of @xmath37 , have unrealistically small values of @xmath35 ; we adopt a minimum of 0.05 dex for this parameter . the ratio @xmath39 is an indication of whether there is any intrinsic star - to - star variation in [ x / fe ] . a high value of this `` spread ratio '' , tabulated in the fifth column of this table , suggests a high probability of intrinsic scatter for the abundance of the species @xmath30 . ideally the mean @xmath29 for those elements with no star - to - star variation should be unity ; for many species the measured @xmath29 is close to that value , certainly closer here than for the sample in m3 and in m13 we studied earlier @xcite . inspection of table [ table_abundsig ] shows that for all but two species @xmath40 for the sample of four stars in ngc 7492 , indicating little sign of an intrinsic star - to - star range in abundance . o i and na i , however , have the two largest values of @xmath29 , 1.3 and 3.0 respectively . note that @xmath29 for mg i is 0.5 , suggesting no real star - to - star abundance variations for this element . we therefore assume that the range of abundances seen in our sample of rgb stars in ngc 7492for na i and o i represent real star - to - star abundance variations ; while no other element shows definite evidence for such variations from this simple analysis . c , n , o , na , mg , and al are known to show correlated abundance variations from star - to - star among the most luminous stars in globular clusters ; see , e.g. the review of @xcite . our simple spread ratio analysis ( see [ section_spread ] ) shows definite star - to - star variations in abundance of both o and na in our small sample of rgb stars in ngc 7492 . variations in mg , if present are smaller and subtle . it is well established that o and na are anti - correlated among luminous giants in globular clusters , see , e.g. @xcite . furthermore , @xcite compiled the data from the literature , combined it with their own , and showed that the same linear relation can be used to fit the o and na data for all globular clusters studied in detail thus far . the latest addition to the clusters studied in detail , ngc 2808 , by @xcite , does so as well . figure [ figure_ona ] shows the relationship between na and o abundances ( both with respect to fe ) for our sample in ngc 7492 . also superposed is the line representing the fit for this anti - correlation determined by @xcite for the luminous giants in m3 , shifted by @xmath60.07 dex and + 0.1 dex in the vertical and horizontal axis as compared to the relation we found for m13 @xcite . the first and last quartiles of the o na anti - correlation seen by @xcite in their sample of luminous giants in m3 are indicated . there is a reasonably clear anti - correlation which corresponds well with that seen for luminous giants in other well studied galactic gcs ( see , e.g. the compilation of * ? ? ? a similar correlation is detected in the outer halo gc ngc 7006 by @xcite . there is marginally statistically significant evidence for a correlation between [ na / fe ] and [ mg / fe ] for our small sample in ngc 7492 , similar to that shown in other gcs ( see , for example , figures 12 and 13 of * ? ? ? the larger uncertainty in our deduced [ mg / fe ] ratios makes this result quite uncertain . we now turn to what we can learn about the chemical history of the galaxy by comparing the abundance ratios in ngc 7492 , at a galactocentric distance of 25 kpc , with those from our recent analysis of the inner halo gcs , m3 and m13 @xcite in m3 and m13 , with @xmath0 of 12 and 9 kpc respectively . we note that the atomic parameters , the analysis procedures , and the software packages used are identical in both of these studies . hence we should be able to detect small differences in the relative values of the abundance ratios of these three gcs . table [ table_abundcomp ] gives the parameter @xmath41[x / fe ] , which is , for each species with detected lines , the mean abundance ratio [ x / fe ] for ngc 7492 with the average of the same parameter for m3 and for m13 subtracted . because of the large star - to - star differences in o / fe seen in m3 and especially in m13 , we subtract the mean [ o / fe ] for stars in m13 of luminosities comparable to those we observe in ngc 7492 . similarly , @xcite found a luminosity dependence of [ mg / fe ] in m13 , most probably due to the luminosity dependence of non - lte corrections , which were ignored . again in this case we subtract the mean [ mg / fe ] of stars of comparable luminosity in m13 to those observed in ngc 7492 . if we allow @xmath42 dex as a tolerable range given the potential internal and systematic errors in the analyses of these three gcs , we find that 80% of the 20 elements in common have a difference of 0.0@xmath42 dex , with only si , cr and co outside that range . figure [ figure_abundcomp_m3m13 ] shows the resulting differences in [ x / fe ] as a function of atomic number . the abundance ratios of [ x / fe ] for si and co are each larger in ngc 7492 than they are in m3 and m13 , while that of cr is smaller . the largest magnitude of the set of @xmath41[x / fe ] occurs for si , and is + 0.23 dex . galactic chemical evolution produces trends in abundance ratios as a function of metallicity . the case of [ si / fe ] is illustrated in figure [ figure_sife ] , where metallicity is parameterized by [ fe / h ] ; examples for the elements ca , ti and ba are given in figures 21 , 22 and 23 of @xcite . we use the same set of of high precision analyses of gcs as in @xcite , specifically ngc 6528 @xcite , ngc 6553 @xcite , 47 tuc @xcite , m71 @xcite , m5 @xcite , ngc 288 @xcite , ngc 362 @xcite , ngc 6752 @xcite , m3 @xcite , m13 @xcite , ngc 6397 @xcite and m15 @xcite , adding in ngc 7492 as well . to characterize the behavior of the metal - poor halo field stars , we adopt abundance ratios from recent large surveys of @xcite , @xcite , @xcite , and by @xcite . no effort has been made to homogenize these analyses , but since the field star surveys were carried out over the course of more than a decade , we have corrected for the difference in the solar fe abundance adopted by each . given that ngc 7492 has [ fe / h ] 0.35 dex smaller than the mean value for m3 and m13 , we have attempted to evaluate the correction to the difference caused by these global trends . we can only do this for about half of the elements studied here . this correction is given in the last column of table [ table_abundcomp ] , and is to be subtracted from the value of @xmath41[x / fe ] to form @xmath43[x / fe ] . these corrections , which do not exceed 0.1 dex in magnitude , bring @xmath43[si / fe ] , @xmath43[cr / fe ] and @xmath43[co / fe ] within the range consistent with no difference ( 0.0@xmath42 dex ) , while not causing any additional elements to exceed the allowed range for equality . the corrections for si , cr and co are shown on figure [ figure_abundcomp_m3m13 ] as well . thus after implementing the corrections for global chemical evolution , all the elements in common show identical abundance ratios [ x / fe ] for ngc 7492 as for m3 and m13 , allowing for a @xmath42 dex tolerance ; 75% of the 20 values of @xmath43[x / fe ] lie within the range @xmath44 to @xmath45 dex . this suggests that the galactic chemical evolution of the outer halo at @xmath0 25 kpc has been identical to that of the well studied inner halo gcs , at least up to the time of the formation of the old globular clusters ngc 7492 , m3 and m13 . in particular , our limited evidence , based on ba , la and eu abundance ratios , suggests the neutron capture processes , both @xmath46 and @xmath47 , appear to have had similar histories throughout the spatial extent of the halo for old gc stars . we have carried a detailed abundance analysis for 21 elements in a sample of four rgb stars in the metal poor distant outer halo globular cluster ngc 7492([fe / h ] @xmath48 dex ) . the analyzed spectra , obtained with hires at the keck observatory , are of high dispersion ( r=@xmath3=35,000 ) . most elements show no sign of star - to - star variation within our limited sample . we have , however , detected an anti - correlation between o and na abundances similar to that seen in our previous analyses of inner halo gcs as well as in studies of relatively nearby gcs by other . a correlation between mg and na abundance may also be present . we have compared the abundance ratios in ngc 7492 with those we previously determined for the much closer old halo gcs m3 and m13 @xcite , hoping that since all these analyses were carried out by the same two people within a timespan of only a few months in a completely consistent manner , with the same line lists , the same atomic parameters , the same analysis codes and procedures , etc . that small differences in the abundance ratios might be detectable . we evaluate the trends of abundance ratio with metallicity for old halo stars from our data combined with published large surveys of halo field star abundances . we then apply corrections to the abundances we derived for m3 and m13 for each species , when feasible , to extrapolate them to the 0.35 dex smaller [ fe / h ] of ngc 7492 . after making such corrections , all the elements in common show identical abundance ratios for ngc 7492 and for m3 and m13 , allowing for a @xmath42 dex tolerance , and 75% of them are then within the tolerance @xmath190.10 dex . this suggests that the galactic chemical evolution of the outer halo at @xmath0 25 kpc has been identical to that of the well studied much closer inner halo gcs , at least up to the time of the formation of the old globular clusters ngc 7492 , m3 and m13 . in particular , our limited evidence , based on ba , la and eu abundance ratios , suggests the neutron capture processes , both @xmath46 and @xmath47 , appear to have had similar histories throughout the spatial extent of the halo for old gc stars as well . the presence of the o / na anti - correlation in ngc 7492 , with @xmath0 of 25 kpc , and the similarity of its chemical history to that of the well studied nearby gcs , provide new constraints on any model of gc formation in the galactic halo . the entire keck / hires user communities owes a huge debt to jerry nelson , gerry smith , steve vogt , and many other people who have worked to make the keck telescope and hires a reality and to operate and maintain the keck observatory . we are grateful to the w. m. keck foundation for the vision to fund the construction of the w. m. keck observatory . the authors wish to extend special thanks to those of hawaiian ancestry on whose sacred mountain we are privileged to be guests . without their generous hospitality , none of the observations presented herein would have been possible . this publication makes use of data from the two micron all - sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center , funded by the national aeronautics and space administration and the national science foundation . we are grateful to the national science foundation for partial support under grant ast-0205951 to jgc . barnes , s. a. , 1968 , , 73 , 579 carretta , e. , cohen , j. g. , gratton , r. g. & behr , b. b. , 2001 , , 122 , 1469 carretta , e. , gratton , r. , bragaglia , a. , bonifacio , p. & pasquini , l. , 2004 , , 416 , 925 carretta , e. , bragaglia , a. & cacciari , c. , 2004 , , in press cohen , j. g. , frogel , j. a. , persson , s. e. & elias , j. h. , 1981 , , 249 , 481 , 1981 fulbright , j. p. , 2000 , , 120 , 1841 gratton , r. g. , carretta , e. , eriksson , k. & gustafsson , b. , 1999 , , 350 , 955 gratton , r. g. & sneden , c. , 1991 , , 241 , 501 gratton . r. g. , _ et al . _ , 2001 , , 369 , 87 gustafsson , b. , bell , r.a . , eriksson , k. & nordlund , , 1975 , a&a , 42 , 407 skrutskie , m. f. , schneider , s.e . , stiening , r. , strom , s.e . , weinberg , m.d . , beichman , c. , chester , t. _ et al . _ , 1997 , in _ the impact of large scale near - ir sky surveys _ , ed . et al . _ ( dordrecht : kluwer ) , p. 187 sneden , c. , kraft , r. p. , shetrone , m. d. , smith , g. h. , langer , g. e. & prosser , c. f. , 1997 , , 114 , 1964 sneden , c. , kraft , r. p. , guhathakurta , p. , peterson , r. c. & fulbright , j. p. , 2004 , , 127 , 2162 takeda , y. , zhao , g. , takada - hidai , m. , chen , y. , saito , y. & zhang , h. w. , 2003 , chin.j.astron.astrophys . , 3 , 316 thevenin , f. , carbonnel , c. , de freitas pacheco , j. a. , idiart , t. p. , jasniewsicz , g. , de lavery , p. & plez , b. , 2001 , , 373 , 905 vogt , s. e. _ et al . _ 1994 , spie , 2198 , 362 lcrrrrrr h , 231 & 23 08 22.32 @xmath49 37 43 & 14.71 & 13.36 & 20/08/2003 & 800 & 88 & @xmath50 + t , 458 & 23 08 25.75 @xmath49 37 10 & 15.50 & 14.41 & 21/08/2003 & 3600 & 92 & @xmath51 + r & 23 08 29.46 @xmath49 36 32 & 15.51 & 14.40 & 20,21/08/2003 & 3000 & 92 & @xmath52 + k , 950 & 23 08 20.83 @xmath49 36 20 & 15.77 & 14.72 & 22/08/2003 & 3600 & 105 & @xmath15 + llrrrrrrrrrrrrrrr oi & 6300.30 & 0.00 & @xmath69.78 & 30.0 & 23.5 & 13.5 & 27.3 + oi & 6363.78 & 0.02 & @xmath610.30 & 9.0 & & 8.0 & 9.5 + nai & 5682.63 & 2.10 & @xmath60.70 & 46.0 & 32.1 & 8.0 & 16.0 + nai & 5688.19 & 2.10 & @xmath60.42 & 61.0 & 49.6 & 15.0 & 28.5 + nai & 6160.75 & 2.00 & @xmath61.23 & 18.5 & & & + mgi & 4703.00 & 4.34 & @xmath60.67 & 157.2 & 126.7 & 115.8 & 136.0 + mgi & 5528.40 & 4.34 & @xmath60.48 & 170.2 & 148.0 & 133.8 & 145.8 + mgi & 5711.09 & 4.34 & @xmath61.67 & 82.7 & 64.0 & 30.0 & 46.0 + sii & 5690.43 & 4.93 & @xmath61.87 & 27.0 & 20.0 & & 23.5 + sii & 5948.54 & 5.08 & @xmath61.23 & 42.0 & 36.0 & 33.0 & 27.0 + sii & 6155.13 & 5.62 & @xmath60.76 & 24.0 & 20.0 & 17.0 & 31.0 + sii & 6237.32 & 5.62 & @xmath61.01 & 15.0 & & & + cai & 5512.99 & 2.93 & @xmath60.27 & 56.5 & 40.0 & 23.0 & 33.7 + cai & 5581.96 & 2.52 & @xmath60.47 & 82.7 & 58.3 & 44.8 & 48.1 + cai & 5588.75 & 2.52 & 0.44 & 133.4 & 116.7 & 95.3 & 103.0 + cai & 5590.11 & 2.52 & @xmath60.71 & 80.7 & 61.1 & 46.7 & 46.9 + cai & 5601.28 & 2.52 & @xmath60.44 & 77.3 & 68.3 & 49.0 & 41.5 + cai & 6161.30 & 2.52 & @xmath61.03 & 49.2 & 24.0 & & 22.0 + cai & 6162.17 & 1.90 & @xmath60.09 & & 145.4 & 132.0 & 148.0 + cai & 6166.44 & 2.52 & @xmath61.05 & 51.3 & 31.0 & 11.0 & 25.5 + cai & 6169.04 & 2.52 & @xmath60.54 & 72.5 & 54.5 & 35.9 & 43.0 + cai & 6169.56 & 2.52 & @xmath60.27 & 93.8 & 65.6 & 46.7 & 61.2 + cai & 6471.66 & 2.52 & @xmath60.59 & 83.6 & 61.9 & 38.2 & 50.8 + cai & 6493.78 & 2.52 & 0.14 & 123.8 & 90.9 & 77.8 & 89.6 + scii & 5526.79 & 1.77 & 0.13 & 94.7 & 88.6 & 78.2 & 82.6 + scii & 5657.90 & 1.51 & @xmath60.50 & 93.4 & 76.1 & 66.6 & 72.7 + scii & 5667.15 & 1.50 & @xmath61.24 & 54.0 & 41.3 & 31.0 & 32.0 + scii & 5669.04 & 1.50 & @xmath61.12 & 60.8 & 45.0 & 34.8 & 44.8 + scii & 5684.20 & 1.51 & @xmath61.08 & 77.0 & 41.0 & 36.0 & 51.0 + scii & 6245.64 & 1.51 & @xmath61.13 & 56.9 & 35.0 & 32.2 & 34.4 + scii & 6604.60 & 1.36 & @xmath61.48 & 57.2 & 34.0 & 19.0 & 45.0 + tii & 4981.74 & 0.85 & 0.50 & 167.7 & 111.3 & 97.5 & 120.7 + tii & 4999.51 & 0.83 & 0.25 & 157.4 & 113.0 & 86.3 & 106.4 + tii & 5022.87 & 0.83 & @xmath60.43 & 107.2 & 71.0 & 41.8 & 62.3 + tii & 5039.96 & 0.02 & @xmath61.13 & 148.0 & 94.4 & 73.6 & 75.7 + tii & 5426.26 & 0.02 & @xmath63.01 & 33.7 & & & + tii & 5471.20 & 1.44 & @xmath61.39 & 15.0 & & & + tii & 5490.15 & 1.46 & @xmath60.93 & 34.7 & & & + tii & 5644.14 & 2.27 & 0.05 & 41.0 & & & + tii & 5662.16 & 2.32 & @xmath60.11 & 24.5 & & & + tii & 5937.81 & 1.07 & @xmath61.89 & 16.4 & & & + tii & 5941.75 & 1.05 & @xmath61.52 & 43.6 & 17.3 & & 14.0 + tii & 5953.16 & 1.89 & @xmath60.33 & 41.1 & & & + tii & 5978.54 & 1.87 & @xmath60.50 & 29.0 & & & + tii & 6258.10 & 1.44 & @xmath60.35 & 75.0 & 37.3 & 25.4 & 35.7 + tii & 6261.10 & 1.43 & @xmath60.48 & 72.2 & 36.5 & 18.0 & 28.7 + tii & 6303.76 & 1.44 & @xmath61.57 & 9.0 & & & + tii & 6312.22 & 1.46 & @xmath61.55 & 9.5 & & & + tii & 6743.12 & 0.90 & @xmath61.63 & 45.8 & & & + tiii & 4657.20 & 1.24 & @xmath62.32 & 84.0 & 97.0 & 59.3 & 83.9 + tiii & 4708.67 & 1.24 & @xmath62.37 & 96.8 & 81.8 & 71.1 & 87.7 + tiii & 4865.62 & 1.12 & @xmath62.81 & 75.0 & 84.0 & 52.7 & 64.1 + tiii & 5185.91 & 1.89 & @xmath61.46 & 98.0 & 85.9 & 74.8 & 81.9 + tiii & 5336.79 & 1.58 & @xmath61.63 & 121.0 & 105.0 & 93.2 & 105.4 + vi & 5670.85 & 1.08 & @xmath60.43 & 35.0 & & & + vi & 5703.57 & 1.05 & @xmath60.21 & 34.3 & & & + vi & 6081.44 & 1.05 & @xmath60.58 & 29.8 & & & + vi & 6090.22 & 1.08 & @xmath60.06 & 41.7 & & & + vi & 6199.20 & 0.29 & @xmath61.28 & 32.0 & & & + vi & 6251.82 & 0.29 & @xmath61.34 & 35.4 & & & + vi & 6274.64 & 0.27 & @xmath61.67 & 21.8 & & & + vi & 6285.14 & 0.28 & @xmath61.51 & 46.2 & & & + cri & 5345.81 & 1.00 & @xmath60.97 & 134.1 & 104.3 & 75.4 & 85.6 + cri & 5348.33 & 1.00 & @xmath61.29 & 115.8 & 74.5 & 54.7 & 63.9 + cri & 5409.80 & 1.03 & @xmath60.71 & 169.5 & 126.0 & 88.8 & 106.9 + cri & 5787.96 & 3.32 & @xmath60.08 & 12.0 & & & + mni & 4754.04 & 2.28 & @xmath60.09 & 107.2 & 74.2 & 49.8 & 62.2 + mni & 4783.42 & 2.30 & 0.04 & 121.2 & 95.0 & 69.2 & 72.9 + mni & 4823.51 & 2.32 & 0.14 & 116.5 & 75.0 & 67.5 & 78.0 + mni & 5537.74 & 2.19 & @xmath62.02 & 28.0 & & & + fei & 4788.77 & 3.24 & @xmath61.81 & 49.5 & & & + fei & 4891.50 & 2.85 & @xmath60.11 & 170.7 & 144.8 & 137.7 & 142.3 + fei & 4919.00 & 2.86 & @xmath60.34 & 168.5 & 141.0 & 117.4 & 141.4 + fei & 5083.34 & 0.96 & @xmath62.96 & 165.2 & 131.2 & 112.3 & 126.2 + fei & 5166.28 & 0.00 & @xmath64.20 & & 141.4 & 118.9 & 139.7 + fei & 5194.95 & 1.56 & @xmath62.09 & 174.6 & 130.3 & 121.0 & 129.1 + fei & 5232.95 & 2.94 & @xmath60.10 & 172.0 & 139.0 & 129.6 & 139.5 + fei & 5324.19 & 3.21 & @xmath60.10 & 160.4 & 135.8 & 117.7 & 125.8 + fei & 5393.18 & 3.24 & @xmath60.72 & 114.4 & 93.0 & 81.8 & 86.1 + fei & 5410.92 & 4.47 & 0.40 & 87.1 & 73.4 & 56.0 & 68.3 + fei & 5415.21 & 4.39 & 0.64 & 105.0 & 91.0 & 66.4 & 86.9 + fei & 5424.08 & 4.32 & 0.51 & 114.7 & 87.8 & 83.1 & 99.2 + fei & 5445.05 & 4.39 & @xmath60.03 & 79.3 & 70.0 & 47.0 & 67.2 + fei & 5473.90 & 4.15 & @xmath60.69 & 42.2 & 43.9 & 20.0 & 24.5 + fei & 5493.50 & 4.10 & @xmath61.68 & 15.9 & & & + fei & 5497.52 & 1.01 & @xmath62.83 & & 147.7 & 134.3 & 143.9 + fei & 5501.46 & 0.96 & @xmath63.05 & & 138.6 & 118.5 & 136.5 + fei & 5506.79 & 0.99 & @xmath62.79 & & 146.2 & 130.0 & 142.4 + fei & 5525.55 & 4.23 & @xmath61.08 & 21.0 & & & 13.5 + fei & 5554.88 & 4.55 & @xmath60.35 & 33.9 & 35.0 & 22.0 & 25.6 + fei & 5567.39 & 2.61 & @xmath62.67 & 64.8 & 39.0 & & 31.2 + fei & 5569.62 & 3.42 & @xmath60.49 & 123.0 & 92.0 & 81.1 & 91.2 + fei & 5572.84 & 3.40 & @xmath60.28 & 146.1 & 108.6 & 100.1 & 104.3 + fei & 5576.09 & 3.43 & @xmath60.92 & 109.1 & 83.5 & 63.7 & 72.3 + fei & 5586.76 & 3.37 & @xmath60.14 & 144.8 & 117.7 & 105.1 & 114.1 + fei & 5641.44 & 4.26 & @xmath61.08 & 30.0 & & & + fei & 5662.52 & 4.18 & @xmath60.57 & 65.0 & 48.5 & 37.2 & 39.8 + fei & 5679.02 & 4.65 & @xmath60.82 & 25.0 & & & + fei & 5701.54 & 2.56 & @xmath62.14 & 106.0 & 76.3 & 46.9 & 66.7 + fei & 5705.98 & 4.61 & @xmath60.49 & 25.5 & & & 29.0 + fei & 5752.04 & 4.55 & @xmath60.94 & 19.2 & & & + fei & 5753.12 & 4.26 & @xmath60.69 & 52.5 & 38.9 & 30.0 & 30.4 + fei & 5762.99 & 4.21 & @xmath60.41 & 70.2 & 44.9 & 36.5 & 48.0 + fei & 5775.06 & 4.22 & @xmath61.30 & 22.6 & & & + fei & 5778.46 & 2.59 & @xmath63.43 & 19.6 & & & + fei & 5806.72 & 4.61 & @xmath60.95 & 11.0 & & & + fei & 5859.60 & 4.55 & @xmath60.55 & 34.0 & 22.0 & 15.6 & 17.7 + fei & 5862.35 & 4.55 & @xmath60.33 & 52.2 & 30.0 & 21.0 & 24.2 + fei & 5883.81 & 3.96 & @xmath61.26 & 27.0 & 17.5 & & 19.0 + fei & 5930.17 & 4.65 & @xmath60.14 & 47.4 & 36.7 & 21.3 & 18.0 + fei & 5934.65 & 3.93 & @xmath61.07 & 42.0 & 42.0 & 29.5 & 21.8 + fei & 5952.72 & 3.98 & @xmath61.34 & 53.3 & 35.5 & 20.0 & 24.3 + fei & 5956.69 & 0.86 & @xmath64.50 & 110.7 & 60.7 & 34.5 & 51.2 + fei & 5976.79 & 3.94 & @xmath61.33 & 41.7 & 18.0 & 17.5 & 19.3 + fei & 5983.69 & 4.55 & @xmath60.66 & 37.5 & & 16.8 & 20.0 + fei & 6024.05 & 4.55 & 0.03 & 67.4 & 40.5 & 38.9 & 48.9 + fei & 6027.05 & 4.07 & @xmath61.09 & 38.3 & & & + fei & 6055.99 & 4.73 & @xmath60.37 & 29.2 & 28.0 & & 15.0 + fei & 6065.48 & 2.61 & @xmath61.41 & 143.9 & 111.4 & 90.6 & 97.2 + fei & 6078.50 & 4.79 & @xmath60.33 & 29.2 & 25.5 & 10.5 & 12.5 + fei & 6137.69 & 2.59 & @xmath61.35 & 144.3 & 110.6 & 99.2 & 109.8 + fei & 6151.62 & 2.18 & @xmath63.37 & 66.0 & 37.8 & 17.6 & 30.1 + fei & 6157.73 & 4.07 & @xmath61.16 & 42.5 & 26.2 & 8.0 & 20.9 + fei & 6165.36 & 4.14 & @xmath61.47 & 24.9 & & & + fei & 6173.34 & 2.22 & @xmath62.88 & 91.0 & 62.6 & 36.0 & 51.6 + fei & 6180.20 & 2.73 & @xmath62.65 & 55.0 & 29.8 & 14.0 & 27.7 + fei & 6187.99 & 3.94 & @xmath61.62 & 14.0 & & & + fei & 6191.56 & 2.43 & @xmath61.42 & 174.9 & 129.4 & 99.3 & 128.8 + fei & 6200.31 & 2.61 & @xmath62.37 & 88.5 & 59.7 & 31.5 & 47.1 + fei & 6240.65 & 2.22 & @xmath63.17 & 59.3 & 35.2 & 16.0 & 23.0 + fei & 6246.32 & 3.60 & @xmath60.88 & 91.9 & 74.3 & 48.9 & 57.7 + fei & 6252.55 & 2.40 & @xmath61.77 & 146.8 & 106.8 & 100.0 & 112.0 + fei & 6254.26 & 2.28 & @xmath62.43 & 122.1 & 95.9 & 73.0 & 79.3 + fei & 6265.13 & 2.18 & @xmath62.54 & 118.9 & 89.8 & 59.1 & 77.7 + fei & 6297.79 & 2.22 & @xmath62.64 & 108.1 & 79.9 & 44.1 & 59.4 + fei & 6301.51 & 3.65 & @xmath60.72 & 93.7 & 71.0 & 50.0 & 65.1 + fei & 6315.31 & 4.14 & @xmath61.23 & 25.0 & & & + fei & 6355.03 & 2.84 & @xmath62.29 & 69.3 & 45.1 & 19.0 & 33.0 + fei & 6380.75 & 4.19 & @xmath61.38 & 22.3 & & & + fei & 6393.60 & 2.43 & @xmath61.58 & 156.6 & 118.3 & 100.8 & 115.0 + fei & 6408.03 & 3.69 & @xmath61.02 & 86.7 & 59.3 & 40.2 & 49.6 + fei & 6411.65 & 3.65 & @xmath60.72 & 109.2 & 75.1 & 58.2 & 75.7 + fei & 6421.35 & 2.28 & @xmath62.01 & 145.9 & 101.0 & 85.7 & 101.2 + fei & 6475.63 & 2.56 & @xmath62.94 & 58.8 & 33.2 & 14.6 & 19.5 + fei & 6481.87 & 2.28 & @xmath63.01 & 86.3 & 46.5 & 30.6 & 43.2 + fei & 6498.94 & 0.96 & @xmath64.69 & 98.1 & 51.2 & 31.9 & 43.5 + fei & 6546.24 & 2.76 & @xmath61.54 & 137.2 & 99.1 & 74.1 & 92.7 + fei & 6581.21 & 1.48 & @xmath64.68 & 42.7 & & & + fei & 6592.91 & 2.73 & @xmath61.47 & 130.9 & 97.4 & 81.7 & 90.8 + fei & 6593.87 & 2.43 & @xmath62.37 & 99.2 & 67.4 & 48.0 & 64.2 + fei & 6608.02 & 2.28 & @xmath63.93 & 16.0 & & & + fei & 6609.11 & 2.56 & @xmath62.66 & 79.1 & 45.5 & 26.5 & 37.5 + fei & 6633.75 & 4.79 & @xmath60.80 & 27.0 & & & 12.0 + fei & 6648.12 & 1.01 & @xmath65.92 & 13.0 & & & + fei & 6739.52 & 1.56 & @xmath64.79 & 27.0 & & & + fei & 6750.15 & 2.42 & @xmath62.58 & 101.8 & 68.0 & 46.6 & 60.2 + fei & 6839.83 & 2.56 & @xmath63.35 & 27.9 & & & + fei & 6843.65 & 4.55 & @xmath60.83 & 25.6 & & & + fei & 6855.18 & 4.56 & @xmath60.74 & 34.6 & & & + fei & 6861.95 & 2.42 & @xmath63.85 & 20.0 & & & + fei & 6978.85 & 2.48 & @xmath62.45 & 104.8 & 72.6 & 44.6 & 66.5 + fei & 6988.52 & 2.40 & @xmath63.56 & 39.7 & & & + fei & 7022.95 & 4.19 & @xmath61.15 & 38.0 & & & + fei & 7038.22 & 4.22 & @xmath61.20 & 30.1 & & & + feii & 4923.93 & 2.88 & @xmath61.32 & 138.4 & & & + feii & 5197.58 & 3.23 & @xmath62.23 & 74.0 & 64.0 & 70.9 & 76.8 + feii & 5234.63 & 3.22 & @xmath62.22 & 75.1 & 67.6 & 74.3 & 83.6 + feii & 5414.08 & 3.22 & @xmath63.62 & 22.0 & 23.5 & 10.0 & 16.5 + feii & 5425.26 & 3.00 & @xmath63.24 & 29.0 & 27.0 & 27.0 & 37.5 + feii & 5534.85 & 3.25 & @xmath62.64 & 64.3 & 56.8 & 53.2 & 52.9 + feii & 5991.38 & 3.15 & @xmath63.57 & 30.0 & & 12.0 & 27.4 + feii & 6149.26 & 3.89 & @xmath62.69 & 18.0 & 25.2 & 16.4 & 26.9 + feii & 6247.56 & 3.89 & @xmath62.36 & 40.0 & 31.9 & 31.0 & 31.6 + feii & 6369.46 & 2.89 & @xmath64.20 & 10.0 & & & + feii & 6416.92 & 3.89 & @xmath62.69 & 19.0 & 15.9 & & 15.8 + feii & 6516.08 & 2.89 & @xmath63.45 & 64.0 & 58.3 & 47.5 & 51.8 + coi & 5530.79 & 1.71 & @xmath62.06 & 30.0 & & 9.0 & 10.0 + coi & 5647.23 & 2.28 & @xmath61.56 & 21.5 & & & + coi & 6189.00 & 1.71 & @xmath62.45 & 18.0 & & & + nii & 5578.72 & 1.68 & @xmath62.64 & 86.1 & 44.1 & 32.0 & 33.0 + nii & 5587.86 & 1.93 & @xmath62.14 & 81.4 & 36.9 & 20.5 & 39.0 + nii & 5682.20 & 4.10 & @xmath60.47 & 18.0 & & & + nii & 5748.35 & 1.68 & @xmath63.26 & 41.0 & & 7.0 & 14.0 + nii & 5846.99 & 1.68 & @xmath63.21 & 34.1 & & & + nii & 6128.97 & 1.68 & @xmath63.33 & 30.0 & 23.5 & 10.4 & 10.5 + nii & 6175.37 & 4.09 & @xmath60.54 & 22.0 & & & + nii & 6176.81 & 4.09 & @xmath60.53 & 21.8 & & & + nii & 6177.24 & 1.83 & @xmath63.51 & 15.0 & & & + nii & 6482.80 & 1.93 & @xmath62.63 & 47.8 & 21.0 & 13.3 & 19.6 + nii & 6586.31 & 1.95 & @xmath62.81 & 45.5 & 22.3 & 8.0 & 13.0 + nii & 6643.63 & 1.68 & @xmath62.30 & 126.3 & 83.0 & 56.5 & 73.9 + nii & 6767.77 & 1.83 & @xmath62.17 & 106.4 & 74.8 & 51.7 & 64.1 + cui & 5105.54 & 1.39 & @xmath61.50 & 82.0 & 46.0 & 24.0 & 28.0 + cui & 5782.12 & 1.64 & @xmath61.78 & 41.5 & & & + zni & 4722.16 & 4.03 & @xmath60.39 & 45.0 & 36.6 & 41.0 & 44.0 + zni & 4810.54 & 4.08 & @xmath60.17 & 46.0 & 54.6 & 52.4 & 48.5 + yii & 4883.69 & 1.08 & 0.07 & 85.0 & 71.3 & 74.5 & 71.8 + yii & 5087.43 & 1.08 & @xmath60.17 & 71.0 & 55.0 & 56.2 & 60.1 + yii & 5200.42 & 0.99 & @xmath60.57 & 56.0 & 42.1 & 32.7 & 44.0 + zri & 6127.44 & 0.15 & @xmath61.06 & 17.0 & & & + zri & 6134.55 & 0.00 & @xmath61.28 & 13.0 & & & + zri & 6143.20 & 0.07 & @xmath61.10 & 10.0 & & & + baii & 5853.70 & 0.60 & @xmath61.01 & 117.3 & 98.1 & 99.9 & 108.2 + baii & 6141.70 & 0.70 & @xmath60.07 & 172.8 & 134.7 & 152.3 & 155.9 + baii & 6496.90 & 0.60 & @xmath60.38 & 186.6 & 146.8 & 154.9 & 163.0 + laii & 5122.99 & 0.32 & @xmath60.85 & 27.8 & & & + laii & 6390.48 & 0.32 & @xmath61.41 & 21.5 & & & + ndii & 4947.02 & 0.56 & @xmath61.13 & 18.0 & & & + ndii & 4959.12 & 0.06 & @xmath60.80 & 66.6 & 34.0 & 23.5 & 32.0 + ndii & 5092.79 & 0.38 & @xmath60.61 & 44.0 & 28.5 & 17.4 & 16.9 + ndii & 5212.35 & 0.20 & @xmath60.96 & 48.0 & & & + ndii & 5249.58 & 0.98 & 0.20 & 49.9 & 30.5 & 33.5 & 31.5 + ndii & 5319.81 & 0.55 & @xmath60.14 & 61.9 & 33.2 & 33.8 & 27.8 + euii & 6645.11 & 1.38 & 0.12 & 25.5 & & & + lclc o & 8.85 & fe & 7.45 + na & 6.33 & ni & 6.25 + mg & 7.54 & cu & 4.21 + al & 6.47 & zn & 4.60 + si & 7.55 & ba & 2.13 + ca & 6.36 & y & 2.24 + sc & 3.10 & zr & 2.60 + ti & 4.99 & la & 1.14 + v & 4.00 & nd & 1.45 + cr & 5.67 & eu & 0.51 + mn & 5.39 & dy & 1.10 + lcrlrrcrcrcccc 231 & @xmath61.76 @xmath19 0.05 * & 90 & @xmath61.72 @xmath19 0.06 & 12 & 0.26 @xmath19 0.05 * & 2 & 0.10 @xmath19 0.05 & 3 & 0.67 @xmath19 0.16 & 3 + r & @xmath61.77@xmath19 0.05 * & 66 & @xmath61.80 @xmath19 0.08 & 9 & 0.49 @xmath19 0.10 & 1 & 0.19 @xmath19 0.05 * & 2 & 0.52 @xmath19 0.06 & 3 + 458 & @xmath61.79 @xmath19 0.05 * & 70 & @xmath61.77 @xmath19 0.05 & 10 & 0.55 @xmath19 0.05 * & 2 & @xmath60.10 @xmath19 0.05 * & 2 & 0.55 @xmath19 0.20 & 3 + 950 & @xmath61.94 @xmath19 0.05 * & 64 & @xmath61.88 @xmath19 0.05 & 9 & 0.52 @xmath19 0.15 & 2 & @xmath60.28 @xmath19 0.05 * & 2 & 0.45 @xmath19 0.20 & 3 + llrrrrrlrrrlr 231 & 0.33 @xmath19 0.09 & 4 & 0.09 @xmath19 0.05 * & 11 & 0.20 @xmath19 0.07 & 7 & 0.16 @xmath19 0.05 * & 18 & @xmath60.07 @xmath19 0.06 & 8 + r & 0.36 @xmath19 0.10 & 3 & 0.11 @xmath19 0.05 * & 12 & 0.09 @xmath19 0.05 * & 7 & 0.12 @xmath19 0.08 & 7 & & 0 + 458 & 0.45 @xmath19 0.12 & 3 & 0.05 @xmath19 0.05 * & 12 & 0.10 @xmath19 0.06 & 7 & 0.18 @xmath19 0.07 & 7 & & 0 + 950 & 0.42 @xmath19 0.11 & 2 & 0.07 @xmath19 0.05 & 11 & 0.15 @xmath19 0.05 * & 7 & 0.07 @xmath19 0.05 * & 6 & & 0 + llrlllllrrrr 231 & @xmath60.24 @xmath19 0.09 & 4 & @xmath60.24 @xmath19 0.09 & 4 & 0.11 @xmath19 0.05 * & 3 & @xmath60.06 @xmath19 0.05 * & 13 & @xmath60.63 @xmath19 0.06 & 2 + r & @xmath60.16 @xmath19 0.11 & 3 & @xmath60.48 @xmath19 0.13 & 3 & & 0 & @xmath60.07 @xmath19 0.08 & 7 & @xmath60.57 @xmath19 0.10 & 1 + 458 & @xmath60.29 @xmath19 0.06 & 3 & @xmath60.43 @xmath19 0.05 * & 3 & 0.18 @xmath19 0.10 & 1 & @xmath60.09 @xmath19 0.05 * & 8 & @xmath60.72 @xmath19 0.10 & 1 + 950 & @xmath60.29 @xmath19 0.05 * & 3 & @xmath60.39 @xmath19 0.08 & 3 & 0.32 @xmath19 0.10 & 1 & @xmath60.09 @xmath19 0.06 & 8 & @xmath60.62 @xmath19 0.10 & 1 + lrlllrrrrrrrrrr 231 & @xmath60.09 @xmath19 0.08 & 2 & @xmath60.29 @xmath19 0.05 * & 4 & 0.36 @xmath19 0.10 & 3 & 0.24 @xmath19 0.10 & 3 & 0.12 @xmath19 0.16 & 2 + r & @xmath60.09 @xmath19 0.08 & 2 & @xmath60.35 @xmath19 0.05 * & 4 & & 0 & 0.05 @xmath19 0.09 & 3 & & 0 + 458 & @xmath60.16 @xmath19 0.05 & 2 & @xmath60.31 @xmath19 0.06 & 4 & & 0 & 0.34 @xmath19 0.07 & 3 & & 0 + 950 & 0.13 @xmath19 0.05 * & 2 & @xmath60.11 @xmath19 0.08 & 4 & & 0 & 0.50 @xmath19 0.08 & 3 & & 0 + lrrcrrrrrr oi & 0.46 & 0.13 & 0.10 & 1.30 & 4 + nai & @xmath60.02 & 0.21 & 0.07 & 3.00 & 4 + mgi & 0.55 & 0.09 & 0.20 & 0.45 & 4 + sii & 0.39 & 0.06 & 0.10 & 0.60 & 4 + cai & 0.08 & 0.03 & 0.12 & 0.25 & 4 + scii & 0.14 & 0.05 & 0.21 & 0.24 & 4 + tii & 0.13 & 0.05 & 0.12 & 0.42 & 4 + tiii & 0.34 & 0.06 & 0.15 & 0.40 & 4 + vi & @xmath60.07 & & 0.23 & & 1 + cri & @xmath60.25 & 0.06 & 0.17 & 0.35 & 4 + mni & @xmath60.39 & 0.10 & 0.18 & 0.55 & 4 + fei & @xmath61.82 & 0.08 & 0.10 & 0.80 & 4 + feii & @xmath61.79 & 0.07 & 0.13 & 0.54 & 4 + coi & 0.20 & 0.11 & 0.12 & 0.92 & 3 + nii & @xmath60.08 & 0.02 & 0.10 & 0.20 & 4 + cui & @xmath60.64 & 0.06 & 0.16 & 0.38 & 4 + zni & @xmath60.03 & 0.11 & 0.13 & 0.85 & 4 + yii & @xmath60.27 & 0.11 & 0.15 & 0.73 & 4 + zri & 0.36 & & 0.16 & & 1 + baii & 0.28 & 0.19 & 0.22 & 0.86 & 4 + laii & 0.12 & & 0.18 & & 1 + ndii & 0.32 & 0.11 & 0.11 & 1.00 & 4 + euii & 0.61 & & & & 1 + lrrrr o & 0.46 & 0.15 & 0.02 & + na & @xmath60.02 & 0.21 & @xmath60.10 & 0.00 + mg & 0.55 & 0.09 & 0.12 & 0.05 + si & 0.39 & 0.06 & 0.23 & 0.08 + ca & 0.08 & 0.03 & @xmath60.09 & 0.07 + sc & 0.14 & 0.05 & 0.14 & 0.07 + ti & 0.24 & 0.06 & 0.08 & 0.07 + v & @xmath60.07 & & 0.01 & + cr & @xmath60.25 & 0.06 & @xmath60.22 & @xmath60.10 + mn & @xmath60.39 & 0.10 & @xmath60.04 & @xmath60.05 + co & 0.20 & 0.11 & 0.21 & 0.07 + ni & @xmath60.08 & 0.01 & @xmath60.01 & 0.00 + cu & @xmath60.64 & 0.06 & @xmath60.02 & + zn & @xmath53 & 0.11 & 0.02 & + y & @xmath60.27 & 0.11 & @xmath60.04 & + zr & 0.36 & & 0.40 & + ba & 0.28 & 0.019 & 0.05 & @xmath60.04 + la & 0.12 & & 0.03 & + nd & 0.32 & 0.11 & 0.10 & + eu & 0.61 & & 0.07 & 0.00 +

we investigate halo mass selection properties of red - sequence cluster finders using galaxy populations of the millennium simulation ( ms ) . a clear red sequence exists for ms galaxies in massive halos at redshifts , and we use this knowledge to inform a cluster - finding algorithm applied to projections of the simulated volume . at low redshift ( ) , we find that of the clusters found have galaxy membership dominated by a single , real - space halo , and that are blended systems for which no single halo contributes a majority of a cluster s membership . at , the fraction of blends increases to , as weaker redshift evolution in observed color extends the comoving length probed by a fixed range of color . other factors contributing to the increased blending at high- include broadening of the red sequence and confusion from a larger number of intermediate mass halos hosting bright red galaxies of magnitude similar to those in higher mass halos . our method produces catalogs of cluster candidates whose halo mass selection function , , is characterized by a bimodal log - normal model with a dominant component that reproduces well the real - space distribution , and a redshift - dependent tail that is broader and displaced by a factor lower in mass . we discuss implications for x - ray properties of optically selected clusters and offer ideas for improving both mock catalogs and cluster - finding in future surveys . cosmology : clusters of galaxies , large scale structure

0706.0211

the abundance and distribution of massive dark matter halos provide a sensitive probe of cosmology and theories of structure formation . the galaxies within these halos also have their evolution strongly affected by their hosts . clusters of galaxies are the observational realization of such halos which has inspired multi - wavelength campaigns to find and characterize them . with the advent of large format ccd cameras on large telescopes , which can identify galaxies to high redshifts over wide fields , there has been renewed interest in optical searches for clusters using multicolor imaging @xcite , see @xcite for a review of optical cluster finding methods . in particular , methods which identify the cluster red sequence @xcite have attained significant success in identifying cluster candidates over wide fields to @xmath10 and above . because red sequence galaxies dominate the cluster population , including the reddest galaxies at a given redshift and becoming redder with increasing redshift , the restriction to red sequence colors approximately isolates a redshift slice . this redshift filtering increases the signal - to - noise of cluster detection by largely eliminating projection effects from unassociated structures along the line of sight . however , contamination is still expected from blue galaxies at even higher redshift than the cluster and from galaxies near enough to the cluster to lie within the narrow , red - sequence color region . this residual contamination is the focus of this work . we are motivated by current red sequence based cluster searches , such as the sdss @xcite , in particular by those using two filters only such as the rcs and the rcs-2 @xcite and sparcs @xcite.gladders / rcs/ and http://spider.ipac.caltech.edu/staff/gillian/sparcs . ] we investigate the nature of the cluster population selected by a two filter method applied to mock galaxy samples of the millennium simulation ( ms ) @xcite . throughout this paper , we use `` clusters '' to refer to objects found by the algorithm and `` halos '' to refer to the dark matter halos identified in the simulation using the full 3d dark matter distribution . we use joint halo cluster membership identifying the ms halos to which each cluster s galaxies belong to categorize the purity and completeness of the cluster population . ( joint halo cluster membership is defined by taking a cluster , found using the red sequence method below , and then identifying the ms halos to which its galaxies belong . ) our cluster finder is patterned after the scheme used in three dimensions to identify halos . we apply a circular overdensity algorithm , centered on bright @xmath7-band galaxies , to spatial projections of the galaxy populations at the discrete redshifts @xmath11 , @xmath12 and @xmath13 . an advantage of the millennium simulation is that it provides mock clusters situated in their correct cosmological context as part of the evolving cosmic web . including the cosmic web is significant because projections of superclusters , structures that tend to align along filaments meeting at the cluster of interest , provide a major source of confusion for cluster identification that is difficult to otherwise model . by having available the full 3d galaxy and dark matter distribution in the simulation we are able to monitor and isolate different physical effects which can influence red sequence cluster finding . the outline of the paper is as follows . we describe our methods in [ sec : methods ] and give our findings for the ms in [ sec : results ] . we consider some implications and properties of the blends in [ sec : implications ] and discuss properties causing and correlating with the blending which might extend beyond our particular search algorithm and simulation in [ sec : discussion ] . we conclude in [ sec : conclusions ] . the appendix compares different purity and completeness definitions in use . @xmath14 & intercept&slope & max rs dist . @xmath15 & @xmath16 low / high & @xmath17 low / high + 0.41 & 0.31 & 0.51 & 0.52 & 0.028 & 0.078 & -0.72/ 1.30 & -0.56/0.52 + 0.69 & 0.57 & 0.81 & 0.72 & 0.052 & 0.14 & -1.10/0.36 & -0.80/0.64 + 0.99 & 0.85 & 1.14 & 0.75 & 0.060 & 0.18 & -0.40/0.72 & -1.08/1.40 + + [ tab : zchange ] change in redshift across the millennium box at different redshifts , red sequence intercept and slope , maximum distance from red sequence in color - magnitude space ( @xmath15 ) , the @xmath18 color change across the box ( to front , and then to back , per @xmath19 gpc ) , and the @xmath7 magnitude change across the box . color and magnitude changes are taken from the bruzual - charlot ( 2003 ) model as described in the text , see also fig . the context for our study is the model of the spatial distribution of massive halos and the galaxies that inhabit them provided by the millennium simulation @xcite . this is a collisionless dark matter simulation performed in a periodic cube @xmath20mpc ( comoving ) on a side , using @xmath21 particles for a cosmology with parameters @xmath22 . mock galaxies , with luminosities and colors , are generated by post - processing the dark matter halo merger trees with a semi - analytic prescription for the gas dynamics and feedback . for details , see @xcite . in particular , our version is that described in detail in @xcite , however with the updated dust prescription of @xcite which better models dust extinction at higher redshifts . we focus our cluster finding investigation on local confusion , projections on spatial scales @xmath23 of a target halo that will , at these redshifts , be barely resolved by photometric redshifts of the next - generation surveys ( des , cfht - ls , pan - starrs1 , kids , snap , lsst ) . ( although the scales these surveys might resolve are comparable to the box size considered here , these surveys are not necessarily using only the rs method described in this paper . ) we use the simulated galaxy and halo catalogues at three fixed epochs given by redshifts @xmath24 , @xmath12 and @xmath13 . these values span much of the expected redshift range of interest for a survey such as the rcs . halos in the simulation are found by using a friends - of - friends algorithm @xcite and galaxy membership is determined based on this . the friends - of - friends linking length ( 0.2 times the mean interparticle spacing ) can link objects into one large halo which by eye look to be smaller components , we note below where our results show signs of this effect . halo masses are given in terms of @xmath25 ( denoted as @xmath26 henceforth ) , the mass enclosed within a radius interior to which the mean density is 200 times the critical density at that epoch . at our redshifts there were 1268 , 805 and 426 halos with @xmath27 and 113 , 47 and 19 halos with @xmath28 . for the red sequence search , the sdss @xmath29 and @xmath7 filters , which bracket the 4000 angstrom break for approximately @xmath30 , are employed . at the highest redshift , we also considered @xmath31 band , our results for this case are described at the end of 4 ; results below will be given for @xmath18 unless stated otherwise . [ fig : redseq ] shows that a red sequence in @xmath18 vs. @xmath7 exists in rich ms halos over the range of redshifts probed . we use galaxies above @xmath32 , corresponding to @xmath7-magnitudes of @xmath33 , 21 and 22 at redshifts @xmath34 , @xmath12 and @xmath13 , and yielding samples of 942313 , 1005469 and 1054711 galaxies , respectively . the top panels show contours of the full , magnitude - limited population while lower panels show the color - magnitude behavior of galaxies in halos with 8 or more members . taking galaxies within the inner @xmath35mpc of the centers of the latter sample , we fit a linear relation in the @xmath18 vs. @xmath7 plane . following @xcite , we throw out @xmath36 outliers and iterate the fit to find the slope and intercept of the red sequence . the width of the red sequence is set to enclose 90% of the full massive halo galaxy sample . the distance , @xmath15 , is taken perpendicular to the red sequence line in the color - magnitude space . of these galaxies , the red sequence widens and for high redshift slightly increases the contamination from projection under study here . ] table [ tab : zchange ] lists the slopes , intercepts , and widths of the red sequence for all three redshifts . the red sequence color - magnitude relation is a weak function of halo mass or richness , so the parameters are not particularly sensitive to the choice of halos with 8 or more members . defining the red sequence using the ms galaxy population itself means that our color cuts are optimally tuned to the content of the ms massive halos . with observations , one derives color cuts using the color - magnitude data of a target subset of galaxies , such as the approaches used by @xcite and @xcite . comparing the simulation results to observations , it appears that the mock red sequence has the wrong tilt and is slightly wider than observed . we experimented with `` tightening '' the red sequence by moving the galaxy colors closer to the best - fit line , but such a procedure did not have a large effect on our conclusions so we present our results using colors as provided . we wish to use projections of each proper time output to create finite redshift segments of a full sky survey . starting with the coeval ms galaxy samples , we introduce passive color and magnitude evolution into spatial projections to mimic the behavior of a light - cone population . the color evolution with redshift is based on an instantaneous bruzual - charlot ( bc ) burst at @xmath37 and shown for @xmath18 in fig . [ fig : colorchange]a . for comparison , we show the average ( slightly bluer ) color of the ms red sequence galaxies for our three redshifts . the ms red sequence galaxies are expected to be bluer than the bc model , since their stars were not formed in a single burst at high redshift . the ms galaxies are also bluer than bcg s in the sdss @xcite . we use this simple bc model to define piecewise constant color gradients , @xmath38 , along the line of sight , shown by the solid line segments in fig.[fig : colorchange]a . we define a @xmath7 magnitude gradient analogously . foreground and background color - magnitude evolution are modeled separately , with parameters given in table [ tab : zchange ] . fainter galaxies may evolve into the @xmath7 magnitude cut because of the change in observed magnitude with redshift . to catch these potential interlopers , we employ galaxy catalogues half a magnitude fainter in @xmath7-band than required by the unevolved red sequence cuts . note that the applied color gradient becomes progressively shallower at higher redshift . the assumed degree of color and magnitude evolution is key since it controls the redshift filtering power of the red sequence . to foreshadow one of our main results , fig . [ fig : colorchange]a illustrates how the color evolution determines the line - of - sight path length probed by the red sequence color range . the dashed regions in fig . [ fig : colorchange]a are centered at the average color of the red sequence galaxies at each redshift and are bounded vertically by the approximate range of color of the red sequence . they are bounded horizontally by the redshift extents of the comoving @xmath39 sightline available within the ms volume . at @xmath24 , the evolutionary color gradients are strong enough that projected red sequence galaxies will shift out of the target color range before the @xmath39 ms boundary is reached , but this is not quite the case at @xmath40 and @xmath13 . fig.[fig : colorchange]b further illustrates how the imposed color evolution acts as a redshift filter . taking the color and magnitude of each galaxy and our line of sight gradients , fig.[fig : colorchange]b shows the fraction of these galaxies remaining on the red sequence as a function of line - of - sight distance . such galaxies will still be potential members of a cluster centered at the origin . a more narrowly peaked distribution indicates a smaller fraction of galaxies available for inclusion via projection during cluster finding . as can be seen , the fraction of galaxies remaining within the red sequence cut at large distances from the origin increases with redshift ; the red sequence selects a longer path along line of sight at higher redshift . the other source of contamination is galaxies that are shifted into the red sequence by the change in observed color . the number density of these galaxies , normalized by the number of red sequence galaxies at the central redshift , is shown by the light lines in fig.[fig : colorchange]b . except for the most distant part of the box at @xmath24 , this number is relatively small . our use of a uniform color change with redshift for all galaxies is not strictly correct for all galaxy types . however , blue star forming galaxies change in observed color much more slowly with redshift than in this model , so to be shifted erroneously into our red sequence color cut , these galaxies are required to be at significantly higher redshift than the cluster . since they would then lie outside of our 500 @xmath41 mpc box , they are not included in our analysis . the strongest contribution to interloper candidates is from galaxies which have colors within our red sequence color cut even though they are far from the central galaxy along the line of sight . our algorithm defines clusters as circular regions , centered on a bright galaxy , with red - sequence sky surface density equal to a multiple @xmath42 of the mean value at the redshift of interest . this approach is analogous to the spherical overdensity method used to define the halo masses . for target centers , we work in descending order through a list of red - sequence galaxies ranked ( brightest to dimmest ) by apparent @xmath7-band magnitude . this ranking is motivated by a desire to find the rare , high mass halos first , then work down the mass function to more common objects . around a potential cluster center , a radially - sorted list of red sequence neighbors is used to define a mean galaxy number density profile as a function of transverse separation . we use the periodic boundaries of the ms to recenter the simulated volume on each candidate center . the volume extends @xmath43mpc in front and behind , and galaxy colors are adjusted , linearly with distance in the projected direction , as described above . starting with the 8 nearest neighbors , ( to avoid shot noise problems in tracing the cluster profiles at small radii ) , we work outward in radius @xmath44 until the the number of galaxies @xmath45 fails to satisfy the overdensity criterion @xmath46 here @xmath47 is the mean sky surface density of red sequence galaxies in the ms , including the effects of the applied observed color evolution along the projected dimension . if the overdensity criterion is not satisfied for 8 galaxies , the object is discarded , if @xmath45 meets or exceeds a minimum of 8 galaxies , then this cluster is added to the output list . all members are then recorded and removed from the remaining list of potential cluster centers . times the average red sequence ( background ) density , @xmath48 in our case . the approximate change of radius with richness can be read off from eq.[eqn : codefn ] . note too that our cluster - finding algorithm traces galaxy overdensities to radii which can potentially reach greater than 1 @xmath41 mpc . this algorithm increases the survey sensitivity to truly extended structures , but may also increase the cross - section for interlopers relative to algorithms which search for clusters only on a limited , smaller scale ; however , a fixed aperture richness based cluster finder performed significantly more poorly . ] note that area overlap of clusters is allowed , so that a single galaxy can belong to more than one cluster ( 6 - 7% of the galaxies end up in more than one cluster at the lowest 2 redshifts , dropping to @xmath49 at higher redshifts ; in contrast , galaxies only belong to one ms halo ) . to boost statistics , we make three projections of the simulated volume along its principal axes . the choice of @xmath42 is discussed below . the sensitivity of survey purity and completeness to the choice of @xmath42 is further explored in the appendix . the clusters found by the search algorithm can be linked back to the dark matter halos in the simulation using their respective lists of galaxy members . a perfect algorithm would be complete with respect to halos and have no false positives , no clusters that appear rich on the sky but are actually several less rich systems aligned along the line - of - sight . in addition , the halo and cluster richnesses would agree . a perfect algorithm would therefore recover the intrinsic distribution of halo mass @xmath26 as a function of red sequence galaxy richness @xmath45 . this distribution is shown in the top panels of fig . [ fig : ngalm200 ] . at all redshifts , a mean , red sequence richness of @xmath50 above our @xmath7-magnitude limit corresponds to a @xmath51 halo . when fit ( throwing out 3-@xmath52 outliers several times ) above a minimum of 8 members , we find that mass scales with red sequence richness as @xmath53 , with @xmath54 , @xmath55 and @xmath55 at @xmath56 , @xmath12 and @xmath13 respectively . the mass intercepts are @xmath57 , @xmath58 and @xmath59 and there are @xmath60 4100 , 2900 , and 1300 @xmath61 halos at these redshifts , respectively . note that red sequence richness is a fairly noisy tracer of mass ; the rms level of scatter is @xmath62 or higher above the richness cut of @xmath63 ( a detailed discussion of scatter in richness vs. mass can be found in @xcite ) . the richness we use in finding the clusters may not be the best richness to use for getting the cluster mass ( e.g. galaxy counts within some aperture might be useful after the clusters are found , for finding the clusters themselves we found a fixed aperture performed significantly worse ) . some observational surveys for galaxy overdensities account for projections of foreground / background galaxies via a statistical subtraction of the expected number of projected galaxies , calculated from random non - cluster pointings . our cluster richness estimator , @xmath64 , does not include such a correction ; our overdensity requirement means that approximately @xmath65 of the galaxies are from the background . for each cluster identified in projection , we list all halos contributing one or more of its member galaxies . the quality of the cluster detection is measured by the top - ranked matched fraction , @xmath66 , defined as the fraction of cluster members coming from the halo that contributes the plurality of the cluster s red sequence galaxies . we define two classes , _ clean _ and _ blended _ , based on whether the plurality is or is not the majority of the cluster s membership , @xmath67 we assign to each cluster the mass of its top - ranked halo found through member - matching . if two ( or more ) halos contribute the same number of galaxies , and are both top - ranked , we take the most massive . an ideal cluster catalog would be _ pure _ , _ complete _ and _ unbiased _ with respect to halos . a perfectly _ pure _ sample would have no accidental projections ; all the galaxies of any chosen cluster would be common members of a single , dark matter halo . a perfectly _ complete _ sample would be one for which each halo in the survey volume appears once , and only once , in the list of clusters . finally , an _ unbiased _ cluster catalog would contain clusters that reproduce the mean mass - richness relation defined by halos . in this section , we consider these issues , both in the context of setting our circular overdensity threshold and in the results obtained . we will see that high levels of purity and completeness are achieved , and that the cluster samples are nearly unbiased . ( many definitions of purity and completeness exist in the literature , we describe and compare several of them in the appendix , and detail our definitions as we use them below . ) the cluster catalogs produced by the search algorithm depend on the value of the number density threshold @xmath42 . choosing too high a value will pick out only the cores of the richest halos , resulting in a catalog that is pure and complete at very high masses , but is otherwise incomplete . picking too low a value will extend the search into the periphery of halos , leading to a catalog that , although complete across a wide range of masses , suffers from impurities due to blending multiple halos into a single cluster . our choice of @xmath68 and @xmath69 for clusters provides samples that are highly complete for @xmath70 halos . [ fig : ffound ] shows a measure of completeness , the fraction of halos assigned as top - ranged matches to clusters with @xmath61 . the completeness is very high for halos with intrinsic @xmath70 , but it drops considerably for lower - richness halos . more halos are missed at higher redshift , and these tend to have extended , filamentary shapes suggestive of recent ( or imminent ) merging . at higher redshift , the major merger rate increases , leading to a higher fraction of disturbed halos . keeping the cluster richness fixed at @xmath69 in order to define whether a halo is found or not ( completeness ) , samples derived with higher values of @xmath42 will be more pure ( have fewer blends ) but less complete , and vice - versa for samples constructed with lower @xmath42 . further quantitative discussion on purity and completeness can be found in the appendix . fig.[fig : nofn ] shows that , at each redshift , the value @xmath68 produces a cluster catalog with a richness function , @xmath71 , that matches well that of the underlying halo population . averaging the three projections , there are @xmath72 and @xmath73 clusters with @xmath74 at @xmath24 , 0.69 and 0.99 , respectively . these values compare well to the ms halo counts of 4098 , 2926 , 1290 for @xmath74 . the scatter from the average of cluster numbers between different lines of sight is less than a percent at @xmath24 and less than four percent at @xmath75 . the good match in number counts does not imply that the algorithm is perfect . in fact , the typical number of halos contributing to an @xmath76 cluster is @xmath77 . the second and fifth richest clusters found at @xmath24 illustrate the range of behavior in clean and blended clusters . figure [ fig : z041c12 ] shows projected positions and color - magnitude information for sky patches centered on the two clusters . the second richest cluster has 212 members contributed by 21 different halos . members of one of the most massive halos at that epoch , @xmath78 , comprise @xmath79 of the cluster members . the remaining members come from 20 other halos , including some lying in the foreground . a small number of members are contributed by halos in the background . the fifth richest cluster , with 175 members , presents a very different case . its most massive contributing halo has a mass @xmath80 , which contributes almost all of its own galaxies but only 35% of the cluster s members ( @xmath81 ) . a total of 53 other halos also contribute , many lying close ( within @xmath82 ) in the foreground or background . although much richer than most of the halos considered , these two examples illustrate the essential projection problem that is causing the blends ; both sets of galaxies appear to be reasonable clusters in the x - y plane . in the next two sections the statistics of the clean and blended clusters , and their features , will be discussed in more detail . the mass selection function is an important ingredient for cosmological tests with optical cluster surveys ( @xcite , @xcite ) . [ fig : ngalm200 ] ( bottom ) shows the relationship between the observed richness of a cluster and the mass of its top - ranked halo ( see [ sec : match ] ) . circles show clean clusters while small crosses show blends . at each redshift , the clean cluster population displays a power law mean relation remarkably similar to that of the underyling halo population . the slopes of the relations agree at the few percent level ; the values for halos ( clean clusters ) for @xmath74 are 1.07 ( 1.04 ) , 1.10 ( 1.06 ) , 1.10 ( 1.15 ) from low to high redshift , respectively . the intercepts at @xmath83 also agree at the few percent level , and could be further fine - tuned by introducing small changes to the search threshold @xmath42 at each redshift . at all redshifts , the circular overdensity algorithm is effective at identifying the mean richness - mass behavior of the underlying halo population . the dispersion in the observed cluster sample is larger than for halos , due to failure modes of the search algorithm . at fixed observed richness , blending creates a tail to low masses while fragmentation of large halos into multiple clusters introduces a high mass tail . [ fig : mlike ] shows estimates of the conditional halo mass distribution , @xmath8 , derived from cross - sections of the joint likelihood data in fig . [ fig : ngalm200 ] in richness ranges @xmath84 at redshifts @xmath85 , respectively . this choice gives a constant average halo mass , @xmath86 , at all three redshifts . the cluster likelihoods ( shaded in the figure ) are compared with the halo distributions for the same richness ranges , shown by solid lines . the top row shows all clusters , while the middle and bottom rows separate the samples into clean and blended systems , respectively . raw counts rather than normalized likelihoods are shown to give the number of objects . at @xmath56 , more than @xmath3 of clusters in the chosen richness range have their dominant underlying halo contributing at least half of the galaxies . the mass distribution of the found clusters matches well the underlying halo mass likelihood . at higher redshift , the correspondence between halos and clusters weakens somewhat ; the number of blends more than doubles , from @xmath87 at @xmath56 to @xmath6 at @xmath88 . the blended systems contribute a low mass tail to the halo mass likelihood . for the distributions , the central mass of the clean clusters remains at @xmath89 at all 3 redshifts , while the central mass of the blends drops , from @xmath90 at @xmath24 to @xmath91 . thus the ratio of central masses between the clean and blended clusters also increases with redshift . our classification of clean _ versus _ blended clusters is based on a somewhat arbitrary cutoff of @xmath92 in member fraction . figure [ fig : f1hcum ] provides a more complete picture by plotting the cumulative fraction of clusters that have top - ranked halo member fraction @xmath93 . here the same observed cluster richness limits as in fig . [ fig : mlike ] are used . cutting at @xmath94 , the vertical line , gives the clean fractions quoted in fig . [ fig : mlike ] . analogues for other definitions of `` clean fraction '' in terms of @xmath66 can be read off as well . there is a clear trend with redshift , with clusters at @xmath88 being less well - matched to halos than those at @xmath56 . the median value of @xmath66 tells a similar story , decreasing from @xmath95 at @xmath56 to @xmath96 at @xmath88 . blending is clearly increasing at larger redshift . going to a higher central mass gives similar trends , e.g. centering on a richness corresponding to a average @xmath97 halo mass at all redshifts gives a clean fraction of 90% at redshift 0.41 which decreases to 76% at redshift 0.99 for the same @xmath42 as above ( @xmath42 can be increased for higher richness to improve both numbers but the increase of blends at high redshift remains ) . there are several effects which cause an increasing incidence of blends at higher redshift . firstly , the change of observed color with distance is weaker , and secondly , the red sequence is wider , so the color - magnitude cut selects galaxies from a thicker slice along the line of sight . these seem to be the strongest causes and were illustrated in fig.[fig : colorchange ] . another way of seeing the effect of color / magnitude evolution is to remove it entirely at @xmath24 ; the background level then increases and the contrast between the clusters and the background declines . lowering @xmath42 to obtain the same number of clean clusters at the fixed mass range of fig . [ fig : mlike ] , we find that the level of blends increases to @xmath98 , very close to what is seen at @xmath99 . similarly , to increase the clean fraction , one can impose the @xmath24 color evolution on the @xmath75 population . in this case , however , the number of non - red sequence galaxies brought into the red sequence through our evolution increases strongly , limiting the degree to which blends can be reduced . a third contributing factor is that , at earlier times , the mass function is steeper , causing the number of possible interloper halos per target halo ( of mass @xmath100 , for example ) to grow at high redshift . the increase in intermediate - mass halos is also enhanced because the central galaxy magnitude is less well correlated with host halo mass at @xmath88 than at low redshift . over time , central galaxies in massive halos grow and brighten via mergers , leading to a stronger correlation between @xmath7magnitude and halo mass . our cluster finding algorithm works in descending order of luminosity . at low redshift , the luminosity sorting corresponds well to a sorting in halo mass but , at high redshift , more low mass systems are mixed into the range of central galaxy magnitude occupied by high mass halos . as these factors are fairly generic , as expected , the trend toward more blends at @xmath88 appeared in all the cases we considered : changing definition and tightness of the red sequence , changing @xmath64 cuts and changing the spherical overdensity requirement . for a wide range of density cuts and modeling choices the blends have roughly half the mass of the clean matches at @xmath24 , and this mass scale declines at higher redshift . since blended clusters are associated with lower mass halos , they will be evident in follow - up studies as such . their mean lensing signal , x ray luminosity and temperature , and thermal sz decrement should be low relative to clean systems . spectroscopic signatures of substructure , in the form of multiple peaks or other departures from gaussianity , would also be likely in these systems . the imprecise centering of the multiple components along the line - of - sight would tend to flatten the radial number density profile . .expected cluster x - ray properties . [ cols="^,^,^,^,^",options="header " , ] [ tab : xray ] + @xmath101 numbers in parenthesis give the log - normal scatter , @xmath102 . table [ tab : xray ] provides estimates of the soft band x - ray luminosity from our ms blended and clean clusters with richness 18 @xmath103 ( now fixed across redshifts ) , compared to values for halos of the same richness . we assume a power - law relation of the form @xmath104 @xcite , and quote values normalized , arbitrarily , to the luminosity of a @xmath105 halo at each epoch . we also assume scatter in the mass luminosity relation , @xmath106 , and combine this with the dispersion in mass for the chosen richness range ( fig.[fig : mlike ] ) to give the dispersion in luminosity , @xmath102 . lower values have been suggested for @xmath107 @xcite , but the scatter in mass at fixed @xmath45 dominates the intrinsic l - m scatter anyway . the clean clusters have mean x - ray luminosities that tend to be slightly higher than the corresponding values for halos of the same richness . the blended systems are substantially dimmer , by a factor two in the mean at @xmath24 , growing to a factor three at @xmath75 . blends should be a generic outcome of red sequence - based cluster finding methods , and there are indications of this from initial x - ray and dynamical observations of the rcs clusters . in chandra observations of 13 clusters at @xmath108 , @xcite confirm 12 as x - ray sources at 3@xmath109 significance , suggesting that @xmath110 of the cluster candidates are massive structures with deep gravitational potential wells ( see also @xcite ) . however , their x - ray luminosities were systematically lower at a given cluster richness than seen for lower - redshift x - ray selected clusters . most of the clusters lay on a sequence only slightly offset from the expected @xmath111-richness relation , but several clusters were significantly offset . optical spectroscopy of one of these clusters ( at @xmath112 ) showed that it consisted of several structures which are dynamically discrete but whose red sequences were overlapping in the survey data @xcite precisely the sort of blended system expected by the study here ( see also @xcite ) . evidence for large scatter between x - ray luminosity and optical richness has been seen in e.g. @xcite . instead of using only the top - ranked halo mass to determine the x - ray signal , we can instead sum the luminosity of all contributing halos . in this case , all the cluster luminosities go up , with the clean subset increasing by roughly @xmath113 and the blended subset increasing by a larger amount . then the ratio of clean to blended mean luminosities changes to @xmath114 at low redshift and to @xmath115 at high redshift . the luminosity measured by x ray observation will depend on details of the projected spatial arrangement , the noise characteristics and other details that lie beyond the scope of this investigation . it seems reasonable to consider the values quoted for the single halo case as a lower bound , and the values from summing all halos as an upper bound , on what would be observed . another difference between clean and blended systems is in their radial cluster profiles . stacked profiles of the clean and blended clusters are used to produce the density profiles , @xmath116 , shown in fig . [ fig : radial ] . the clean clusters have a significantly steeper mean density profile than the blends . this result suggests that a matched angular filter approach @xcite could offer improvements , particularly one that includes radial distance information from photometric redshifts or colors @xcite . observations of colors with distance to cluster center ( e.g. @xcite ) and other properties ( e.g. @xcite ) are already in place at high redshifts . going further down the luminosity function would provide more galaxies to trace out the profile , but at the risk of including more faint background galaxies redshifted into the color region . the interlopers in both clean and blended clusters , as expected by @xcite , lie ( slightly ) more frequently in the background than the foreground . there does nt seem to be a strong trend in the moment of inertia for clean versus blended clusters ; often the blends are projections , rather than objects which are merely unrelaxed . we also considered using @xmath117 color at high redshift , rather than @xmath18 , and found a similar blend fraction , even though the red sequence turns out to be narrower . this is because the evolution of red sequence galaxies ( now defined with respect to @xmath117 ) remains very slow with redshift , thus , as with @xmath18 color at high redshift , many of the galaxies do not evolve out of the red sequence even when far from the cluster center . similarly , the number of non - red sequence galaxies evolving into the selection window remains small across the @xmath118 projected length . as mentioned earlier , blends can be immediately reduced by increasing the spherical overdensity criterion @xmath42 , but only at the cost of losing true halos as well . an increase in @xmath42 also shifts the mass - richness relation to lower values of @xmath45 compared to the intrinsic case , and decreases the number of clusters found at fixed @xmath45 . these trends reflect the usual tradeoff between purity and completeness for cluster samples ; for more discussion see , e.g. , the appendix of @xcite and the appendix of this paper . in the above analysis , we have found properties and trends for blends as a function of redshift . some of these results depend on particular details of the millennium simulation and our method , and some are likely to be general . most of the increase in blends at @xmath119 comes from the slower change of color with increasing redshift . this color change was not obtained directly from the millennium simulation but from a simple stellar population synthesis model that reproduces observations . we expect this result to be general . our implementation of the color change with redshift is crude but the candidate high redshift interlopers are mostly red sequence galaxies , where our approximation is best expected to hold . as a result , we do not expect more detailed color implementations , such as mock light cones ( e.g. @xcite for the ms ) , to produce substantially different local ( @xmath120 ) projected contamination . the increased width of the red sequence at high redshift is derived from the millennium simulation . however , at @xmath88 , the weak color evolution combined with the deep `` green valley '' separating the red and blue populations in the ms means that our results are reasonably insensitive to the precise width . most of the interloper galaxies are themselves members of the red sequence in their respective projected halos . the @xmath18 color shift for @xmath39 projection at @xmath75 is @xmath121 and @xmath122 , so only by compressing the red sequence to a width well below these values would one have an appreciable effect on the blended fraction . the relative numbers of interloper halos at different redshifts is a property of the underlying dark matter power spectrum and linear growth rate . for a fixed target mass , more interloper halos at higher redshift are expected generically . physically , if we look at the line - of - sight distribution of the contaminating material we find that the contaminating mass at large distances ( @xmath123mpc ) more than doubles between redshift @xmath34 and @xmath13 . this enhanced contamination from large distances is also true on a cluster - by - cluster basis : the fraction of clusters with more than e.g. 30% of their material coming from @xmath124 mpc grows significantly with redshift . this material is far outside the cluster virial radius and not just material which got `` caught '' by the cluster finder before it fell in to truly be part of the cluster . note that superclusters of very large size have been seen out at these redshifts , for example see recent studies of superclusters by @xcite . there are other possible interlopers as well . for instance , adding galaxies that lie outside the ms volume will only increase the amount of blended contamination . also , at faint magnitudes , the increasing numbers of background blue galaxies available to redshift into the red sequence are a potential cause for concern ; increasing numbers of blue galaxies at high redshift are observed ( e.g. @xcite ) . ( it should be noted that there are observational techniques to take many of them out which we do not include here ) . we saw only a small fraction of candidate interlopers from galaxies outside the red sequence , except at low redshift , where the green valley is highly compressed ( see fig.[fig : colorchange ] ) . this is good , as the observed color and magnitude evolution of these galaxies was approximated to be the same as for red sequence galaxies ; we expect the interloper numbers due to these objects is minimized at high redshift because of their small contribution in our approximation . in fact , taking out the observed color and magnitude evolution of the non - red - sequence galaxies entirely ( and adjusting @xmath42 to get the same value of @xmath125 ) gives similar clean fractions as in our fiducial model . to extend our analysis of the ms ( tuning the cluster finder to cluster color profiles in more detail , for example ) requires further developments . the ms utilizes sophisticated physical models for properties such as star formation histories , initial mass function and stellar population synthesis models , dust production and radiative transfer models , and the sensitivity of all of these to local conditions . the resulting ms catalogues match observations of large numbers of properties ( e.g. @xcite ) at many redshifts . a detailed , multi - color comparison to observed galaxy number counts is given in @xcite . some departures from observations are noted there , in particular the over - prediction of the abundance of moderately massive galaxies at high redshifts , notably @xmath126 . for our cluster finding , the only properties used are the galaxy locations and their fluxes in two filters . tuning the cluster finder to more specific properties of these fluxes ( for example , their radial trends within clusters ) in the ms will require higher fidelity galaxy formation models . work is in progress to improve the model s match to observations . for example , known issues in the ms under study include , for @xmath127 clusters , a faint red satellite excess and an excess tail of bright blue objects ( but with overall blue satellite fractions too low ) , and no `` green valley''@xcite . we find a red sequence with the wrong sign for the color - magitude slope ( the brightest galaxies tend to be slightly bluer than the fainter , see fig . [ fig : redseq ] ) , and similarly the blue fraction increases towards brighter luminosity and has the wrong radial evolution within clusters for our three redshifts . in addition the simulation was run with the earlier wmap parameters rather than the wmap3 @xcite current best fit cosmology . future improvements in optical cluster finding will require simulated catalogues that are in better quantitative agreement with the growing body of deep galaxy surveys . to refine and use more sophisticated color finders does not necessarily require all the physics employed in the ms , which aims to explain a multitude of observational properties rather than simply reproduce them . purely statistical mock catalogues can be built on halo model fits tuned empirically to data in the redshift range of interest . the catalogues can focus narrowly on observational properties relevant to the search algorithm . in particular , critical to a quantitative prediction of the amount of contamination in color - selected surveys are accurate colors for galaxies in groups and filaments in the outskirts of clusters , as red - sequence galaxies in these regions are the most likely source of interlopers . such survey - specific catalogues are crucial to understand selection functions , in part because they allowing search algorithms to be tuned to the cluster `` color footprint '' and spatial profile . this approach has already been profitably used at low redshifts ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . at high redshift , data sets large enough to tune such catalogues are just coming into being ; combined with modeling improvements in recent years the construction of such catalogues is now a feasible task . however , without such a catalogue in hand , our primary effect is still simple to illustrate . this effect is that the spatial cut provided by the observed color and magnitude cut widens as redshift increases . conversely a narrow spatial cut reduces the blends strongly . for example , taking an exact spatial cut for the ms , boxes @xmath128mpc wide at all three redshifts , the clean fraction becomes almost 100% at low redshift and 95% at high redshift . ( presumably the remaining blends are due to the other contributing factors mentioned above . ) a slice this thick would corresponds to a fine redshift selection , @xmath129 at redshift 0.99 ( 0.41 ) . this level of accuracy is potentially attainable with next - generation photometric redshifts . with the advent of wide field imagers , optical searches have become a powerful way to compile large samples of high redshift clusters . key to these techniques is the use of multi - color information to reduce the line - of - sight contamination that plagued earlier , single filter , observations @xcite . two - filter information provides only limited redshift filtering , and this paper begins to explore the questions of what types of objects are selected by such techniques , and how this selection evolves with redshift . we use a simple circular overdensity search algorithm on local sky projections of the galaxy population of the millennium simulation , tuned using knowledge of the red sequence present in simulated halos with eight or more galaxies brighter than @xmath130 in the @xmath7-band . the free parameter , the density contrast @xmath42 , is tuned to maximize both purity and completeness , and the choice @xmath68 produces a number of clusters as a function of galaxy richness that is close to the underlying richness function of halos . we find that essentially all clusters have some degree of projected contamination ; a cluster of optical richness @xmath45 typically has red sequence members from @xmath131 halos along the line - of - sight . in the large majority of cases , the contamination is not dominant , and most of a cluster s members are associated with a single , massive halo . a minority are highly blended cases in which projected contamination is dominant , and no single halo contributes a majority of the cluster s members . we find an increased fraction of blends with redshift . although several factors contribute , the most important factor appears to be weaker evolution in the observed color of red sequence galaxies with increasing redshift . this effectively increases the path length searched by the red sequence color cut , leading to a larger cross section for accidental , line - of - sight projections . in addition , at higher redshift , the number of @xmath132 halos relative to a @xmath105 halo is larger , and the central galaxy red magnitudes at these mass scales are more similar . the blends add a low - mass tail to the halo mass selection function for clusters of fixed optical richness . for our found clusters with optical richness targeting @xmath133 halos , we expect that @xmath134 of these systems would be underluminous in x rays by a factor of two at @xmath24 , growing to @xmath98 underluminous by a factor closer to three at @xmath75 . the scatter in individual x ray luminosities for the complete set of clusters is expected to be large , @xmath135 at high redshift , and there is considerable overlap in the distributions of @xmath136 expected for clean and blended clusters . it should be noted that , observationally , high redshift low - luminosity systems are also likely have lower signal to noise . the galaxy number density profiles are slightly shallower for blends than for clean clusters , and a matched spatial filter approach may help identify and eliminate the former . since some fraction of halos , those undergoing mergers especially , will also be spatially extended , careful study of the effect of spatial filtering on halo completeness is needed . alternatively , instead of decreasing the number of blends in searches , our findings here suggest modeling the mass likelihood @xmath8 as a bimodal log - normal distribution , with the fraction of blends , and the location and width of that component , included as nuisance parameters . this expected bimodal distribution can be incorporated into error estimates for cluster number counts as a function of redshift , for instance , along with other expected errors ( such as the 5 - 10% scatter associated with red sequence associated redshifts @xcite ) . understanding the detailed color / magnitude trends within galaxy clusters is key to refining red sequence cluster finding and improving its success rate . fortunately , data sets in hand or on the way , combined with rapidly improving modeling methods , will lead to improvements in our understanding of high redshift colors and their evolution . this work will be driven largely by survey - specific mocks current examples are the 2mass @xcite , the deep2 survey@xcite , the 2dfgrs @xcite and the sdss @xcite and such efforts will be necessary for mining the rich science provided by existing and future high redshift cluster surveys . we thank the anonymous referee for many helpful comments and suggestions . jdc thanks a. albrecht , m. brodwin , c. fassnacht , r. gal , j. hennawi , a. von der linden , l. lubin , g. de lucia , s. majumdar , t. mckay , n. padmanabhan , e. rozo , r. stanek and d. weinberg for helpful discussions and/or questions and the galileo galilei institute , the santa fe cosmology summer workshop , and the aspen center for physics for hospitality during the course of this work and the opportunity to present these results , and lbl for support . aee thanks j. annis and t. mckay for conversations and acknowledges support from nasa grant nag5 - 13378 , from nsf itr aci-0121671 , and especially from the miller institute for basic research in science at uc , berkeley . mw thanks charles lawrence for conversations . djc wishes to thank both the aspen center for physics and the department of physics at the university of michigan for hospitality , and acknowledge support from nsf grant ast507428 . mw acknowledges support from nasa and ee acknowledges support from nsf grant ast-0206154 . the millennium simulation was carried out by the virgo supercomputing consortium at the computing centre of the max- planck society in garching ; semi - analytic galaxy catalogues are publicly available at http://www.mpa - garching.mpg.de / millennium/. purity and completeness are `` success rates '' used when one wants a catalogue of a certain type of object and has obtained , via some method , a catalogue of candidates . a classic definition starts with the number of objects which are both in the candidate set and in the desired target set , i.e. the intersection of these sets . dividing the number in the intersection by the total number of target objects then gives completeness , and dividing by the total number of candidate objects gives purity ( or reliability ) . these definitions go back many years in radio astronomy . for instance @xcite used these definitions to describe how well optical sources were matched to radio sources as a function of search aperture radius . in our case , target objects are halos , defined in terms of true richness or mass , and candidate objects are clusters , defined in terms of observed richness . while these terms have a long history , it is not clear that such definitions are the ` single number ' one wants to characterize the success of a cluster finding algorithm . if the properties of the sample change slowly with e.g. richness or mass for example , we may not wish to impose a hard threshold on richness when computing purity . finding a cluster with 19 members may be just as good as requiring 20 . we shall consider several generalizations of the classic notions of purity and completeness below . we note that there are several choices in all of these definitions : the two catalogues , including their underlying data samples and the search algorithms employed , and the definition of which clusters lie in the intersection . even focussing on the circular overdensity method , as we do here , there is considerable latitude in defining both the target catalogue ( e.g. , specific definition of a halo , use of halo richness or mass as the order parameter ) and the cluster candidate sample ( minimum observed cluster richness , choice of @xmath42 ) . we consider here how purity and completeness vary with @xmath42 . figure [ fig : true_pur_lit ] shows the classic definitions of purity and completeness applied to our catalogues for target halos with true richness @xmath137 and candidate clusters with observed richness @xmath138 . we define a cluster and its halo to be in the overlap of the two catalogues if the halo in the target set contributes the most galaxies to a cluster in the candidate set . a more restrictive definition is to require the halo to contribute more than half of the galaxies in a given cluster ( @xmath139 ) . in both cases , one divides the overlap number by the total number of target halos ( completeness ) and total number of candidate clusters ( purity ) . in fig . [ fig : true_pur_lit ] , these two cases ( all @xmath66 and @xmath94 ) for the overlap set are shown as smooth and dashed lines , respectively . we show fig . [ fig : true_pur_lit ] as a function of _ decreasing _ overdensity threshold , @xmath42 , because this mimics a search region of increasing radial scale . at high @xmath42 , purity is maximized because the cluster sample is selecting the dense cores of the most massive halos . as the threshold is decreased , the enlarged search area and lower intrinsic density within the halo increases the frequency of best matched halos which are below the target richness threshold , lowering the cluster sample purity . in contrast , the completeness grows with lower @xmath42 , as the increasing search scale matches and then exceeds the radial scale used to define the halo population . the number of observed clusters at fixed richness increases rapidly with decreasing @xmath42 , improving the odds of completely matching to the massive halo sample . the purity and completeness curves cross at roughly our chosen threshold , @xmath140 . these measures of purity and completeness are substantially lower than unity because , as shown in fig.[fig : truefound ] , there is substantial scatter between @xmath141 and @xmath142 for matched halo - cluster pairs . this should not be taken as a failure of the algorithm , because the source of impurity and incompleteness the scatter is largely understood . there are a number of ways to take this scatter into account , and the optimal method depends strongly on the intended use of the catalogue . different target / candidate sets will vary in purity and completeness , driven by the form of the scatter . to characterize this one could , for example , use @xmath143 , @xmath144 to calculate completeness , and @xmath145 , @xmath146 to calculate purity . if one approximates the scatter as poisson an obvious choice for @xmath147 would be @xmath148 . for @xmath149 using @xmath150 raises the fractions for completeness and purity by @xmath151 ( the largest change is at @xmath42 for low completeness or purity , the smallest at high completeness or purity ) , while using @xmath152 roughly doubles the effect . the purity and completeness curves still cross around @xmath140 ( slightly lower for @xmath24 ) but at a higher fraction ( for @xmath24 it goes from @xmath153 to @xmath154 to @xmath155 for shifts by 5 and 10 , respectively ) . note that the _ a priori _ arbitrary choice of @xmath156 should be motivated by some understanding of how the sample properties change with the property being used to define the sample . extending the candidate sample when defining completeness and the target sample when defining purity can be taken further . we can take the target halo @xmath141 above some cut and consider _ all _ clusters above the minimum richness threshold ( which could be as low as one ) to define completeness , for example . in our case the minimum richness threshold is @xmath69 . the differential form of completeness for @xmath157 and @xmath158 is shown in fig . [ fig : ffound ] in the main text . for purity one can again reverse the limits . if one goes to @xmath159 , all clusters will get matched to at least one halo , and values close to this will have similar results . a possibly more useful definition of purity could be that the best matched halo contributes at least 0.5 of its partner cluster s members . this definition of purity corresponds to the clean fraction shown in fig . [ fig : mlike ] . fig . [ fig : true_pur ] shows the purity and completeness for these less restrictive sample definitions , taking @xmath160 and @xmath161 . the solid increasing line and the dashed decreasing line ( with decreasing @xmath42 ) are directly analogous to their counterparts in fig . [ fig : true_pur_lit ] . at high surface density @xmath42 , the @xmath69 cluster sample is incomplete with respect to @xmath162 halos because the cores of some @xmath70 halos fall below the cluster richness limit of 8 members . as mentioned in the text , high redshift high mass halos are more likely to be disturbed than their lower redshift counterparts and thus to fall below the overdensity threshold . the purity of the overall cluster sample purity is very high at @xmath2 , but declines at higher redshift where halo blending is more severe . as the threshold @xmath42 is lowered , the fraction of halo galaxies lying above the projected threshold increases , but the potential for confusion by projection also increases . for @xmath163 , the @xmath164 cluster sample is essentially 100% complete for @xmath70 halos at all redshifts . the overall purity of the @xmath165 cluster sample is substantially lower , dropping to values below 0.5 at @xmath5 . another way of choosing the two samples is pursued by @xcite , who take fig . [ fig : truefound ] , combining both blends and clean clusters , and identify strong outliers ( see their fig . 2 ) . they take a slice in a fixed @xmath142 range to define purity ( number which are not outliers over total in slice ) and a slice in a fixed @xmath141 range to define completeness ( number which are not outliers over total in slice ) . the issue of how outliers are defined and which slices in @xmath142 , @xmath141 are taken will affect the detailed results , and , again , the intended use of the catalogue needs to be taken into account before deciding the optimal choices . the difficulty in finding the best definition lies in trying to get two numbers ( purity and completeness ) to characterize an entire joint distribution . the full distribution of candidate properties as a function of target halo properties is the key information required to compute expectations for an observational catalogue the scatter is not a problem if its shape is sufficiently well understood and the required accuracy for understanding this distribution depends upon the specific use of the catalogue . for instance , high purity ( but not necessarily high completeness ) might be of interest if one is interested in high mass clusters for individual x - ray followup , while if one wants a sample of clusters for cosmological parameters , a scatter in mass can be included in the analysis , but high completeness is desirable to beat down statistics . if one can correct for scatter perfectly , obtaining high purity is then not crucial ; purity serves only to quantify the size of the correction being applied to the data or the model . the errors that are most important to avoid , and how well characterized our selection function is , determines the best line of attack in the tradeoff between purity and completeness . 99 abell , g.o . , 1958 , apjs 3,211 barkhouse , w.a . , et al , 2006 , apj , 645 , 955 bernardi , m. , hyde , j.b . , sheth , r.k . , miller , c.j . , nichol , r.c . , 2007 , aj 133 , 1741 blindert , k. , yee , h. k. c. , gladders , m. d. , & ellingson , e. 2004 , iau colloq . 195 : outskirts of galaxy clusters : intense life in the suburbs , 215 blindert et al , 2007 , to appear bower , r. , lucey , j.r . , & ellis , r.s . , 1992 , mnras , 254 , 601 bruzual , g. , charlot , s. , 2003 , mnras 344 , 1000 condon , j. j. , balonek , t.j . , jauncey , d.l . , 1975 , aj , 80 , 887 croton et al , 2006 , mnras 365 , 11 dai , x. , kochanek , c.s . , morgan , n.d . , 2007 , apj 658 , 917 dalton , g.b . , efstathiou , g. , maddox , s.j . , & sutherland , w.j . , 1992 , apjl 390 , l1 davis m. , efstathiou g. , frenk c.s . , white s.d.m . , 1985 , apj , 292 , 371 de lucia , g. , et al , 2004 , apjl 610 , 7 de lucia , g. , to appear , talk at berkeley nov . 2006 de lucia , g. , et al , 2007 , mnras 374 , 809 eke , v.r . , et al , 2004 , mnras 348 , 866 ellingson , e. , lin , h. , yee , h.k.c . , carlberg , r.g . , 2001 , apj 547 , 609 ellis , r.s . , 1997 , ara&a 35 , 389 gal , r. r. , lubin , l. m. , squires , g. k. , 2005 , aj , 129 , 1827 gal , r.r . , 2006 , preprint [ astro - ph/0601195 ] gilbank , d.g . , bower , r.g . , castander , f.j . , ziegler , b.l . , 2004 , mnras 348 , 551 gilbank , d.g . , yee , h.k.c . , ellingson , e. , gladders , m.d.,barrientos , l.f . , blindert , k. , 2007 , apj in press , preprint [ arxiv/0705.0782 ] gladders , m.d . , lopez - cruz , o. , yee , h.k.c . , kodama , t. , 1998 , apj 501 , 571 gladders , m.d . , yee , h.k.c . , 2000 , aj 120 , 2148 , see also http://www.astro.utoronto.ca/ gladders / rcs/ gladders , m.d . , yee , h.k.c . , 2005 , apjs 157 , 1 gladders , m.d . , et al , 2007 , apj 655 , 128 hicks , a. , ellingson , e. , bautz , m. , yee , h. , gladders , m. , 2005 , adspr 36 , 706 hicks , a. , et al , 2007 , to appear kaiser , n. , wilson , g. , luppino , g. , kofman , l. , gioia , i. metzger , m. , dahle , h. , 1998 , preprint [ astro - ph/9809268 ] kitzbichler , m.g . , white , s.d.m . , 2007 , mnras 376 , 2 kochanek , c.s . , white , m. , huchra , j. , macri , l. , jarrett , t.h . , schneider , s.e . , mader , j. , 2003 , apj 585 , 161 koester b.p.,mckay , t.a . , et al , 2007 , preprint [ astro - ph/0701268 ] lemson , g. & the virgo consortium , 2006 , preprint [ astro - ph/0608019 ] lopez - cruz , o. , 1997 , phd thesis , university of toronto lopez - cruz , o. , barkhouse , w.a . , yee , h.k.c . , 2004 , apj 614 , 679 lubin , l.m . , brunner , r. , metzger , m.r . , postman , m. , oke , j.b . , 2000 , apjl , 531 , 5 lubin , l.m . , mulchaey , j.s . , postman , m. , 2004 , apjl 601 , 9 lumsden , s.l . , nichol , r.c . , collins , c.a . , guzzo , l. , 1992 , mnras 258,1 miller , c.j . , et al , 2005 , aj 130 , 968 nakata , f. , et al , 2005 , mnras 357 , 1357 padmanabhan , n. , et al , 2006 , preprint [ astro - ph/0605302 ] postman , m. , lubin , l.m . , gunn , j.e . , oke , j.b . , hoessel , j.g . , schneider , d.p . , christensen , j.a . , 1996 , aj 111 , 615 reblinsky , k. , & bartelmann , m. , 1999 , a & a 345,1 reiprich , t.h . , boehringer , h. , 2002 , apj 567 , 716 rozo , e. , wechsler , r.h . , koester , b.p . , evrard , a.e . , mckay , t.a . , 2007 , preprint [ astro - ph/0703574 ] spergel et al , 2006 , preprint [ astro - ph/0603449 ] springel , v. , et al , 2005 , nature 435 , 629 stanek , r. , evrard , a.e . , boehringer , h. , schuecker , p. , nord , b. , 2006 , apj 648 , 956 van breukelen , c. , et al . , 2007 , preprint[arxiv:0708.3838 ] van haarlem , m.p . , frenk , c.s . , white , s.d.m . , 1997 , mnras , 287 , 817 weinmann , s.m . , van den bosch , f.c . yang , x. , mo , h.j . , 2006a , astro - ph/0607585 , to appear in in the proceedings of the xlist rencontres de moriond , xxvith astrophysics moriond meeting : `` from dark halos to light '' , eds . l.tresse , s. maurogordato and j. tran thanh van ( editions frontieres ) weinmann , s.m . , van den bosch , f.c . , yang , x. , mo , h.j . , 2006b , mnras 373 , 1159 white , r. , et al 1999 , aj 118 , 2014 white , m. , a&a 367 ( 2001 ) 27 white , m. , kochanek , c.s . , 2002 , apj 574 , 24 wilson , g. , 2006 , preprint [ astro - ph/0604289 ] , http://spider.ipac.caltech.edu/staff/gillian/sparcs yan , r. , white , m. , coil , a.l . , 2004 , apj 607 , 739 yang x. , mo h.j . , van den bosch f.c . , jing y.p . , 2005 , mnras , 356 , 1293 yee , h.k.c . , ellingson , e. , 2003 , apj 585 , 215

energetics and quantized conductance in jellium - modeled nanowires are investigated using the local - density - functional - based shell correction method , extending our previous study of uniform - in - shape wires [ c. yannouleas and u. landman , j. phys . chem . b * 101 * , 5780 ( 1997 ) ] to wires containing a variable - shaped constricted region . the energetics of the wire ( sodium ) as a function of the length of the volume - conserving , adiabatically shaped constriction , or equivalently its minimum width , leads to formation of self - selecting magic wire configurations , i.e. , a discrete configurational sequence of enhanced stability , originating from quantization of the electronic spectrum , namely , formation of transverse subbands due to the reduced lateral dimensions of the wire . these subbands are the analogs of shells in finite - size , zero - dimensional fermionic systems , such as metal clusters , atomic nuclei , andhe clusters , where magic numbers are known to occur . these variations in the energy result in oscillations in the force required to elongate the wire and are directly correlated with the stepwise variations of the conductance of the nanowire in units of . the oscillatory patterns in the energetics and forces , and the correlated stepwise variation in the conductance are shown , numerically and through a semiclassical analysis , to be dominated by the quantized spectrum of the transverse states at the narrowmost part of the constriction in the wire . energetics , forces , and quantized conductance in jellium - modeled metallic nanowires +

cond-mat9806107

understanding of the physical origins and systematics underlying the variations of materials properties with size , form of aggregation , and dimensionality are some of the main challenges in modern materials research , of ever increasing importance in the face of the accelerated trend toward miniaturization of electronic and mechanical devices . @xcite interestingly , it has emerged that concepts and methodologies developed in the context of isolated gas - phase clusters and atomic nuclei are often most useful for investigations of finite - size solid - state structures . in particular , it has been shown most recently @xcite through first - principles molecular dynamics simulations that as metallic ( sodium ) nanowires are stretched to just a few atoms in diameter , the reduced dimensions , increased surface - to - volume ratio , and impoverished atomic environment , lead to formation of structures , made of the metal atoms in the neck , which can be described in terms of those observed in small gas - phase sodium clusters ; hence they were termed @xcite as supported _ cluster - derived structures ( cds)_. the above prediction of the occurrence of `` magic - number '' cds s in nanowires , due to characteristics of electronic cohesion and atomic bonding in such structures of reduced dimensions , are directly correlated with the energetics of metal clusters , where magic - number sequences of cluster sizes , shapes and structural motifs due to electronic and/or geometric shell effects , have been long predicted and observed . @xcite these results lead one directly to conclude that other properties of nanowires , derived from their energetics , may be described using methodologies developed previously in the context of clusters . indeed , in a previous letter , @xcite we showed that certain aspects of the mechanical response ( i.e. , elongation force ) and electronic transport ( e.g. , quantized conductance ) in metallic nanowires can be analyzed using the local - density - approximation ( lda ) -based shell correction method ( scm ) , developed and applied previously in studies of metal clusters . @xcite specifically , we showed that in a jellium - modelled , volume - conserving , and uniform in shape nanowire , variations of the total energy ( particularly terms associated with electronic subband corrections ) upon elongation of the wire lead to _ self - selection _ of a sequence of stable `` magic '' wire configurations ( mwc s , specified by a sequence of the wire s radii ) , with the force required to elongate the wire from one configuration to the next exhibiting an oscillatory behavior . moreover , we showed that due to the quantized nature of electronic states in such wires , the electronic conductance varies in a quantized step - wise manner ( in units of the conductance quantum @xmath2 ) , correlated with the transitions between mwc s and the above - mentioned force oscillations . in this paper , we expand our lda - based treatment to wires of variable shape , that is allowing for a constricted region . from this investigation , we conclude that the above self - selection principles and the direct correlations between the oscillatory patterns in the energetic stability , forces , and stepwise variations of the quantized conductance maintain for the variable - shaped wire as well , with the finding that underlying these oscillatory patterns and correlations are the contributions from the narrowmost region of the wire . furthermore , this finding is analyzed and corroborated through a semiclassical analysis . prior to introducing the model studied in this paper , it is appropriate to briefly describe certain previous theoretical and experimental investigations , which form the background and motivation for this study . atomistic descriptions , based on realistic interatomic interactions , and/or first - principles modelling and simulations played an essential role in discovering the formation of nanowires , @xcite and in predicting and elucidating the microscopic mechanisms underlying their mechanical , spectral , electronic and transport properties . these predictions @xcite [ particularly those pertaining to generation of nanowires through separation of the contact between two materials bodies ; size - dependent evolution of the wire s mechanical response to elongation transforming from multiple - slips for wider wires to a succession of stress accumulation and fast relief stages leading to a sequence of structural instabilities and order - disorder transformations localized in the neck region when its diameter shrinks to about 15 ; consequent oscillations of the elongation force and the calculated high value of the resolved yield stress ( @xmath3 4 gpa for au nanowires ; which is over an order of magnitude that of the bulk ) , as well as anticipated electronic quantization effects on transport properties @xcite ] have been corroborated in a number of experiments using scanning tunneling and force microscopy , @xcite break junctions , @xcite and pin - plate techniques @xcite at ambient environments , as well as under ultrahigh vacuum and/or cryogenic conditions . particularly pertinent to our current study are experimental observations of the oscillatory behavior of the elongation forces and the correlations between the changes in the conductance and the force oscillations ; see especially the simultaneous measurements of force and conductance in gold nanowires in ref . , where in addition the predicted `` ideal '' value of the critical yield stress has also been measured ( see also ref . ) . the lda - jellium - based model introduced in our previous paper @xcite and extended to generalized wire shapes herein , while providing an appropriate solution within the model s assumptions ( see section ii ) , is devoid by construction of atomic crystallographic structure and does not address issues pertaining to nanowire formation methods , atomistic configurations , and mechanical response modes [ e.g. , plastic deformation mechanisms , interplanar slip , ordering and disordering mechanisms ( see detailed descriptions in refs . and , and a discussion of conductance dips in ref . ) , defects , mechanical reversibility , @xcite and roughening of the wires s morphology during elongation @xcite ] , nor does it consider the effects of the above on the electron spectrum , transport properties , and dynamics . @xcite nevertheless , as shown below , the model offers a useful framework for linking investigations of solid - state structures of reduced dimensions ( e.g. , nanowires ) with methodologies developed in cluster physics , as well as highlighting certain nanowire phenomena of mesoscopic origins and their analogies to clusters . in this context , we note that several other treatments related to certain of the issues in this paper , but employing free - electron models , have been pursued most recently . @xcite in both of these treatments an infinite confining potential on the surface of the wire is assumed and only the contribution from the kinetic energy of the electrons to the total energy is considered , neglecting the exchange - correlation and hartree terms , and electrostatic interactions due to the positive ionic ( jellium ) background . a comprehensive discussion of the limitations of such free - electron models in the context of calculations of electronic structure and energetics ( e.g. , surface energies ) of metal surfaces can be found in ref . . in section ii.a . , we outline the lda - based shell correction method , describe the jellium model for variable - shaped nanowires , and derive expressions for the energetics of such nanowires ( density of states , energy , and force ) . numerical results pertaining to energetics , force , and electronic conductance , calculated as a function of elongation for variable - shaped sodium nanowires , are given in section ii.b . , including a discussion on the main finding that the contribution from the narrowmost part of the constriction underlies the properties of these quantities and the correlations between them . these correlations between the energetic and transport properties and their dependence on the narrowmost part of the nanowire are further analyzed in section iii , using a semiclassical treatment . we summarize our results in section iv . consider a jellium nanowire with circular symmetry about the axis of the wire ( @xmath4 axis ) . the wire may contain a constricted region ( see fig . 1 ) , that is a section of length @xmath5 where the cross - sectional radius @xmath6 varies along the axis as @xmath7 with @xmath8 ( the @xmath9 plane passes through the middle of the wire ) and @xmath10 . @xmath11 is the uniform radius outside the constricted section , and @xmath12 . in this paper , we take a parabolic shape @xmath13 for the description of the constricted region [ a wire of uniform cross section throughout corresponds to @xmath14 . we also assume that elongation of the wire occurs in the constricted region while maintaining its volume constant ( this is supported by md simulations ) , namely by requiring that @xmath15 for given values of @xmath16 and @xmath17 [ hereafter we will denote the pair of parameters ( @xmath16,@xmath17 ) by @xmath18 ; we further assume that @xmath19 . for the parabolic shape assumed in this paper , the smallest cross - sectional radius is determined for any given value of @xmath20 from eqs . ( [ az ] ) and ( [ vol ] ) as @xmath21~ , \label{a0}\ ] ] i.e. , @xmath22 for @xmath23 , and @xmath24 ( i.e. , breakage of the wire ) for @xmath25 . the shell correction method we employ is based on the lda theory . in the shell correction method @xcite ( scm ) , the total lda energy , @xmath26 , for any configuration of the wire ( specified by @xmath5 and @xmath18 ) is separated as , @xmath27 where @xmath28 varies smoothly as a function of the system size ( @xmath5 ) while @xmath29 varies in an oscillatory manner with @xmath5 , as a result of the quantization of the electronic states . @xmath29 is usually called a shell correction in the nuclear @xcite and cluster @xcite literature ; we continue to use here the same terminology with the understanding that the electronic levels in the nanowire form subbands , which are the analog of electronic shells in clusters where the size of the system is usually given by specifying the number of atoms @xmath30 . the scm method , which has been shown to yield results in excellent agreement with experiments @xcite and self - consistent lda calculations @xcite for a number of cluster systems , is equivalent to a harris functional ( @xmath31 ) approximation to the kohn - sham lda with the input density , @xmath32 , obtained through variational minimization of an extended thomas - fermi ( etf ) energy functional , @xmath33 $ ] . the harris functional is given by the following expression , @xmath34 = e_i + \sum_{i=1}^{\text{occ } } \epsilon_i^{\text{out } } \nonumber \\ & & - \int \ ! \left\ { \frac{1}{2 } v_h [ \rho^{\text{in } } ( { \bf r } ) ] + v_{\text{xc } } [ \rho^{\text{in } } ( { \bf r } ) ] \right\ } \rho^{\text{in } } ( { \bf r } ) d{\bf r } \nonumber \\ & & + \int \ ! { \cal e}_{\text{xc } } [ \rho^{\text{in } } ( { \bf r } ) ] d{\bf r}~ , \label{enhar}\end{aligned}\ ] ] where @xmath35 is the hartree ( electronic ) repulsive potential , @xmath36 is the repulsive electrostatic energy of the ions , and @xmath37 d { \bf r}$ ] is the exchange - correlation ( xc ) functional @xcite [ the corresponding xc potential is given as @xmath38 / \delta \rho({\bf r})$ ] ] . @xmath39 are the eigenvalues ( non - self - consistent ) of the single - particle hamiltonian , @xmath40 with the mean - field potential given by @xmath41 = v_h [ \rho^{\text{in } } ( { \bf r } ) ] + v_{\text{xc } } [ \rho^{\text{in } } ( { \bf r } ) ] + v_i ( { \bf r})~ , \label{mfpot}\ ] ] @xmath42 being the attractive potential between the electrons and ions . in electronic structure calculations where the corpuscular nature of the ions is included ( i.e. , all - electron or pseudo - potential calculations ) , @xmath32 may be taken as a superposition of atomic - site densities . in the case of jellium calculations , we have shown @xcite that an accurate approximation to the ks - lda total energy is obtained by using the harris functional with the input density , @xmath32 , in eq . ( [ enhar ] ) evaluated from a variational extended - thomas - fermi ( etf)-lda calculation . the etf - lda energy functional , @xmath43 $ ] , is obtained by replacing the kinetic energy term , @xmath44 $ ] , in the usual lda functional , namely in the expression , @xmath45=t[\rho ] \nonumber \\ & & + \int \left\ { \frac{1}{2 } v_h [ \rho({\bf r } ) ] + v_i({\bf r } ) \right\ } \rho({\bf r } ) d\/ { \bf r } \nonumber \\ & & + \int { \cal e}_{\text{xc } } [ \rho({\bf r})]d\/ { \bf r } + e_i~ , \label{enlda}\end{aligned}\ ] ] by the etf kinetic energy , given to the 4th - order gradients as follows , @xcite @xmath46 = \frac{2m_e}{\hbar^2 } \int t_{\text{etf}}[\rho ] d { \bf r } = \nonumber \\ & & = \int \ ! \left\ { \frac{3}{5 } ( 3\pi^2)^{2/3 } \rho^{5/3 } + \frac{1}{36 } \frac{(\nabla \rho)^2}{\rho } + \frac{1}{270 } ( 3 \pi^2)^{-2/3 } \rho^{1/3 } \right . \nonumber \\ & & \times \left . \left [ \frac{1}{3 } \left ( \frac{\nabla \rho } { \rho } \right)^4 - \frac{9}{8 } \left ( \frac{\nabla \rho } { \rho } \right)^2 \frac{\delta \rho } { \rho } + \left ( \frac{\delta \rho } { \rho } \right)^2 \right ] \right\ } d { \bf r}~. \label{t4th}\end{aligned}\ ] ] the optimal etf - lda total energy is then obtained by minimization of @xmath43 $ ] with respect to the density . in our calculations , we use for the trial densities parametrized profiles @xmath47 with @xmath48 as variational parameters ( the etf - lda optimal density is denoted as @xmath49 ) . the single - particle eigenvalues , @xmath50 , in eq . ( [ enhar ] ) are obtained then as the solutions to the single - particle hamiltonian of eq . ( [ hin ] ) with @xmath51 replaced by @xmath52 [ given by eq . ( [ mfpot ] ) with @xmath53 replaced by @xmath54 . hereafter , these single - particle eigenvalues will be denoted by @xmath55 . in our approach , the smooth contribution in the separation ( [ etot ] ) of the total energy is given by @xmath56 $ ] , while the shell correction , @xmath57 , is simply the difference @xmath58 - e_{\text{etf } } [ \widetilde{\rho } ] \nonumber \\ & & = \sum_{i=1}^{\text{occ } } \widetilde{\epsilon}_i - \int \ ! \widetilde{\rho}({\bf r } ) v_{\text{etf } } ( { \bf r } ) d\/ { \bf r } - t_{\text{etf } } [ \widetilde{\rho}]~. \label{dsh}\end{aligned}\ ] ] the volume density of the positive background is given by @xmath59 , where @xmath60 is the wigner - seitz radius characteristic to the material , and thus the number of positive charges in the constriction is @xmath61 since the nanowire contains a constricted region of variable cross - sectional radius @xmath6 [ see eq . ( [ az ] ) ] , we define a linear ( i.e. , density per unit length of the nanowire ) background density @xmath62 , which when integrated over the total length of the wire yields @xmath63 [ see eq . ( [ numpos ] ) ] . correspondingly , the variational electronic volume density @xmath64 , and in our calculations it takes the form , @xmath65^{\gamma(z)}}~ , \label{rho}\ ] ] with @xmath66 , @xmath67 , and @xmath68 as @xmath4-dependent variational parameters . in the etf calculation , @xmath49 is determined variationally at a given @xmath4 as the one associated with a uniform cylinder of radius @xmath6 ( adiabatic assumption ) , under the normalization condition for local charge neutrality , namely , @xmath69 = \rho^+_l ( z ; l , { \cal o})$ ] [ which fixes the 4th parameter @xmath70 in eq . ( [ rho ] ) ] . the optimized @xmath49 allows then calculation of the smooth contribution for any length of the constriction , @xmath71 in eq . ( [ etot ] ) . the calculation of the shell - correction term , @xmath72 , in eq . ( [ etot ] ) proceeds by evaluating first the density of states in the nanowire . assuming an adiabatic separation of the `` fast '' transverse and the `` slow '' longitudinal variables , @xcite the electronic wave functions in the classically allowed regions may be written as @xmath73 where @xmath74 is the local wave number along the axial ( @xmath4 ) direction of the nanowire @xmath75 \right]^{1/2}~ , \label{kperp}\ ] ] and @xmath76 is the ( transverse ) local eigenvalue spectrum at @xmath4 . to calculate this spectrum for a wire of a configuration specified by @xmath77 for any value of @xmath4 , the eigenvalues of a cylindrical wire with a ( uniform ) radius @xmath6 are calculated from the two - dimensional schrdinger equation @xmath78 \psi + v_{\text{etf}}(r ; z , l , { \cal o } ) \psi \nonumber \\ & & = \widetilde{\epsilon}_{nm } ( z , l , { \cal o } ) \psi~. \label{scheq}\end{aligned}\ ] ] the linear ( per unit length ) , one - dimensional density of states at @xmath4 , @xmath79 , is given by @xmath80~ , \label{elden}\end{aligned}\ ] ] where spin degeneracy has been included , and @xmath81 is the heaviside step function . from eq . ( [ kperp ] ) , we obtain @xmath82^{-1/2 } \nonumber \\ & & \times \theta [ \epsilon - \widetilde{\epsilon}_{nm}(z ; l , { \cal o } ) ] ~. \label{elden2}\end{aligned}\ ] ] we may now define an integrated density of states in the constriction @xmath83 the total number of states up to energy @xmath84 in the constricted region of the wire is given by @xmath85 } \nonumber \\ & & \times \theta [ \epsilon - \widetilde{\epsilon}_{nm}(z ; l , { \cal o } ) ] ~. \label{totnum}\end{aligned}\ ] ] since the total number of electrons in the constricted region is @xmath63 [ see eq . ( [ numpos ] ) ] , the fermi energy , @xmath86 , for a wire with a configuration specified by ( @xmath87 ) is given from eq . ( [ totnum ] ) , i.e. , @xmath88 using the above and eq . ( [ dsh ] ) , the shell - correction term , @xmath89 - e_{\text{etf } } [ \widetilde{\rho } ; l , { \cal o}]~ , \label{harr}\ ] ] may be calculated as @xmath90 \nonumber \\ & & -2 \pi \int_{-l/2}^{l/2 } dz \int_0^\infty dr r \widetilde{\rho } ( r , z ; l , { \cal o } ) v_{\text{etf } } ( r , z ; l , { \cal o } ) \nonumber \\ & & -2 \pi \int_{-l/2}^{l/2 } dz \int_0^\infty dr r t_{\text{etf } } [ \widetilde{\rho } ( r , z ; l , { \cal o})]~ , \label{dshw}\end{aligned}\ ] ] where @xmath52 is the etf potential ( hartree , exchange - correlation , and electron attraction to the positive background ) and @xmath91 is the volume density of the etf kinetic - energy functional [ see eq . ( [ t4th ] ) ] . in actual calculations , we invert the order of integration in the first term of eq . ( [ dshw ] ) , which then takes the form @xmath92 \nonumber \\ & & \times \sqrt { \frac{2m_e}{\hbar^2 } [ \epsilon_f - \widetilde{\epsilon}_{nm } ( z ; l , { \cal o } ) ] } \ ; \theta [ \epsilon_f - \widetilde{\epsilon}_{nm}(z ; l , { \cal o } ) ] ~. \label{sumi}\end{aligned}\ ] ] note that eq . ( [ fermi ] ) implies a common fermi level for the whole constriction for a given @xmath5 ( i.e. , @xmath93 is not a local property ) . therefore , eq . ( [ sumi ] ) is not equivalent to integration of the corresponding uniform wire result derived by us in ref . over the @xmath4-coordinate , since there @xmath93 varies with the wire s cross - sectional radius . having calculated the smooth and shell - correction contributions to the total energy , as a function of @xmath5 , the total elongation force may be evaluated as the derivative of the total energy with respect to @xmath5 , i.e. , @xmath94 , and the contributions to it from the smooth and shell - correction terms are given by @xmath95 , and @xmath96 . in this section , we report results for the elongation of a sodium nanowire ( @xmath97 a.u . ) , starting with an initial cylindrical constriction of length @xmath98 a.u . and radius @xmath99 a.u . in fig . 2 , we show electronic - potential profiles , @xmath100 $ ] , for a particular constriction with @xmath101 ( @xmath102 ) . we display here the potential profiles calculated at the narrowmost part of the constriction [ @xmath103 a.u . ] and at its end [ i.e. , at @xmath104 ( in this paper , all the subsequent numerical results we will discuss relate to constrictions with the same set of @xmath18-parameters , namely , @xmath98 a.u . and @xmath99 a.u . ) . we found that for other values of @xmath4 ( i.e. , for @xmath105 ) , the potential assumes profiles intermediate between the two profiles shown here , namely , the depth of the potential well remains practically unaltered , while its width follows the enlargement of the jellium - background radius @xmath6 , from @xmath106 to @xmath107 . from the three components , @xmath35 , @xmath108 , and @xmath109 , which contribute to the total @xmath52 [ see eq . ( [ mfpot ] ) ] , we found that for all these potential profiles , calculated for different values of @xmath4 along the constriction , the xc contribution is the dominant one , amounting to approx . @xmath110 ev , while the total electrostatic contribution , @xmath111 , is much smaller , resulting in a characteristic `` winebottle '' profile familiar from lda studies of spherical clusters . @xcite fig . 2 also displays the transverse local eigenvalues @xmath76 associated with the two potential profiles . naturally , a wider potential profile yields a larger number of such eigenvalues below the fermi level . to illustrate the nature of the electronic spectrum in the nanowires , and its dependencies on the characteristics of the wire , i.e. , shape and length , we show in fig . 3(a ) densities of states , @xmath112 , calculated for a variable - shaped wire for two wire lengths ( and consequently two minimal cross - sectional radii of the constricted region ) . the density of states for a uniform wire with a radius equal to that of the unconstricted region of the variable - shaped ones [ shown in fig . 3(a ) ] is displayed in fig . two `` classes '' of features are noted for the variable - shaped wires : ( i ) those associated with the narrowmost constricted region ( marked by numbers ) whose radius , @xmath106 , varies upon elongation , and ( ii ) those associated with the maximal radius of the constriction ( and with the unconstricted part of the wire ) which remains constant throughout the elongation of the wire . identification of the latter class of features ( several of which are marked by arrows ) is facilitated through comparison with the density of states for the corresponding uniform wire [ fig . we observe here that , for the broader ( and thus shorter ) wire [ lower curve in fig . 3(a ) ] , six of the features ( peaks ) in the density of states coming from the spectrum of transverse energy levels at the narrowmost region of the constriction are located below the fermi level , @xmath93 [ all the peaks in the density of states occur at the energies of the transverse levels ; e.g. , compare the location of the peaks in the lower curve in fig . 3(a ) with the corresponding spectrum on the left side of fig . 2 ] . on the other hand , for the much narrower ( and thus longer ) constricted wire , only one of these peaks is below @xmath93 [ see upper curve in fig . 3(a ) ] . when plotting the density of states at @xmath93 versus the elongation ( or equivalently the minimal radius of the constricted region ) , these variations lead to an oscillatory pattern , as peaks in the density of states are shifted above the fermi level , one after the other as the wire is being elongated . these variations are also portrayed in the energetics of the wire ( shown in fig . 4 ) , and in the stepwise behavior of the quantized conductance through the wire versus length ( see fig . 5 below ) . from fig . 4 , we observe that the magnitude of the smooth etf contribution , @xmath113 , to the total energy , @xmath114 , of the wire is dominant , with the shell - correction contribution , @xmath57 , exhibiting an oscillatory pattern , with local minima at a set of wire lengths ( and correspondingly a set of minimal cross - sectional radii ) which we term `` magic wire configurations '' ( mwc s ) , i.e. , wire configurations with enhanced energetic stability . when added to the smooth contribution , these shell - correction features lead to local minima of the total energy toward the end of the elongation ( and consequently , narrowing ) process , while for thicker wires ( i.e. , @xmath115 in fig . 4 ) they are expressed as inflection points of the total energy ( in this context , see the total - force curve , @xmath116 , in fig . 5 , where the local minimum in @xmath114 corresponds to the point with @xmath117 marked by an arrow ) . we note here that the occurrence of local minima in the total energy results from a balance between @xmath57 and @xmath113 , with the latter increasing ( that is acquiring less negative values ) as the constriction elongates due to the increasing contribution from the surface of the constriction . comparison of the magnitudes of the shell corrections in a variable - shaped wire and in a uniform one [ i.e. , one with @xmath118 in eq . ( [ az ] ) , whose case was discussed in ref . ] shows that the amplitudes of the oscillations in the latter case are much larger ( over an order of magnitude ) . the reason for this difference is that in the constant - radius wire the quantization into the transverse subbands is uniform along the wire , while in the variable - shaped case the subband spectrum is different in various parts of the constriction . while the oscillatory pattern is dominated by the spectrum at the narrowmost region ( see also section iii below ) , the amplitudes are influenced by the transverse - mode spectra from other parts of the constriction . consequently , the number of local minima in the total energy , @xmath114 , ( and thus the number of wire configurations , i.e. , lengths , for which the total force , @xmath116 , vanishes ) is larger for a uniform wire than for a variable - shaped one . additionally , we suggest that for materials with relatively smaller surface energies a larger number of local minima may occur . from the total energy , and the smooth and shell - correction contributions to it , we obtain the total `` elongation force '' ( ef ) , @xmath116 , and the corresponding components of it , @xmath119 and @xmath120 . these results are displayed in fig . 5 , along with the conductance of the wire evaluated , in the adiabatic approximation ( i.e. , no mode mixing @xcite ) and neglecting tunneling effects ( assuming unit transmission coefficients for all the conducting modes ) , using the landauer expression , @xcite @xmath121~ , \label{land}\ ] ] where @xmath122 , and the spectrum of the transverse modes is evaluated ( for each constriction length ) at the narrowmost part of the constriction , @xmath9 . tunneling contributions ( see e.g. , ref . ) , mode - mixing and non - adiabaticity may affect the sharpness of the conductance steps , and/or introduce some interference related features , particularly near the transitions between the conductance plateaus . these effects , which can be included in more elaborate evaluations of the conductance , @xcite do not modify the conclusions of our study . also included in this figure is a plot describing the variation of the minimal cross - sectional radius @xmath106 with the length of the constriction [ see eq . ( [ a0 ] ) ] . as evident from fig . 5 , the oscillations in the force resulting from the shell - correction contributions are prominent . in @xmath120 , we observe that the locations of the zeroes of the force situated at the right of the force maxima occur for values of @xmath123 which coincide with the locations of local minima in the shell - correction contribution to the energy of the wire ( i.e , for a sequence of minimal cross - sectional radii corresponding to mwc s ) . in the total force , @xmath116 , only one of these points ( where @xmath117 ) remains [ i.e. , the one corresponding to the local minimum in the total energy towards the end of the elongation process ( see fig . 4 ) ] , for the reasons discussed above in connection with the energetics of the wire . nevertheless , the oscillations in the total force correlate well with those in the total energy of the wire , which as discussed above originate from the subband spectrum at the narrowmost part of the constriction ( see also section iii ) . also , the locations of the local maxima in the total force correlate with the stepwise variations in the conductance signifying the sequential decrease in the number of transverse subbands ( calculated at the narrowmost section of the wire ) below @xmath93 ( i.e. , conducting channels ) as the constricted part of the wire elongates ( and thus narrows ) . additionally , we note that the magnitude of the total force is comparable to measured ones ( i.e. , in the nanonewton range ) . the magnitude of the total force in sodium nanowires ( not measured to date ) is expected to be smaller than that found for gold nanowires , @xcite due mainly to differences in the electron densities and surface energies of the materials . as discussed above , the total energy of the wire is characterized by local minima and inflection points occuring for a set of wire lengths , or equivalently a set of minimal cross - sectional radii of the constriction , and are reflected in the oscillatory patterns of the elongation force . these features correspond to the oscillatory shell - correction contributions and originate from the spectrum of transverse modes at the narrowmost part of the constriction . moreover , these patterns correlate with the locations of the quantized conductance steps , which are determined by the transverse - mode spectrum at the narrowmost region ( i.e. , the number of conducting modes below @xmath93 , and their degeneracies ) . to further investigate the origins of these correlations , we present in this section a semiclassical analysis of the density of states , energetics , forces , and conductance in a free - electron nanowire modeled via an infinite confining potential on the surface of the wire . as in the above ( see fig . 1 ) , we model the constricted region of the wire as a section with a slowly ( adiabatically ) varying shape . dividing the constriction into thin cylindrical slices , the solution of the schrdinger equation for each slice is of the form , @xmath124 where @xmath125 is a normalization constant , @xmath126 is the electron momentum along the axis of the wire , @xmath127 is the bessel function of order @xmath128 , and @xmath129 . consider first a uniform cylindrical wire with a constant cross - sectional radius @xmath130 . with the infinite wall boundary condition assumed here , the single - particle electronic energy levels in the wire are expressed in terms of the roots of the bessel functions , @xmath131 , as @xmath132 here we remark that in the semiclassical approximation the electron performs a complicated trajectory inside the wire . all the semiclassical trajectories are tangent to the caustic surfaces of a set of concentric cylinders inside the wire . @xcite quantization of the electronic states leads to selection of only a certain subset of trajectories associated with a ceratin set of radii @xmath133 of the caustic surfaces , corresponding to allowed values of the azimuthal quantum numbers , @xmath128 , i.e , @xmath134 ; this description is closely related to the semiclassical periodic orbit theory . @xcite in the course of developing semiclassical methods , keller and rubinow @xcite have demonstrated that the debye asymptotic expansion @xcite of the bessel functions ( @xmath135 ) provides an accurate approximation to the eigenfunction @xmath127 , i.e. , @xmath136~. \label{bes2}\end{aligned}\ ] ] this approximation is valid in the region between the caustic cylindrical surface and the boundary surface of the wire ; in the region inside the caustic surface ( @xmath137 ) the solution decays exponentially . in this approximation , the equation for the asymptotic values of the bessel - function zeroes has the form , @xmath138 first we calculate the density of states whose evaluation involves , after integration over @xmath126 , double sums over the quantum numbers @xmath139 and @xmath128 ; @xmath140 ... , @xmath141 ... [ see eq . ( [ elden2 ] ) ] . applying sequentially the poisson summation formula to both sums and separating the oscillatory terms ( note that in our semiclassical approximation @xmath142 ) in complete analogy with refs . , we obtain for the density of states ( per unit length ) , @xmath143 \nonumber \\ & & + \frac{2 \sqrt{2 } } { \pi a \epsilon_a^{3/4 } \epsilon^{1/4 } } \sum_{m=1}^{\infty } \frac{1}{m^{1/2 } } \sin \left [ 2 \pi m k a \ ! + \ ! \frac{\pi}{4 } \right ] ~ , \label{sdosc}\end{aligned}\ ] ] where @xmath144 , and @xmath145 is the electron wave vector . the two terms in eq . ( [ sdosc ] ) correspond to the contribution from the point where the phase is stationary and from the end - points in the sum ( integral ) over @xmath128 ( see discussion in ref . ) . while the second oscillatory term in eq.([sdosc ] ) has a smaller amplitude than the first one [ by a factor of @xmath146 , it corresponds to an important class of electronic states , with @xmath147 , localized near the surface of the wire ( the so - called whispering gallery states @xcite ) . until now we discussed a uniform wire with a constant cross - sectional radius . in a wire with a variable shape , the cross - sectional radii depend on @xmath4 , as discussed in connection with eq . ( [ az ] ) . substituting the @xmath4-dependence of the radii in eq . ( [ sdosc ] ) , i.e. , replacing @xmath130 by @xmath6 , we need to perform an integration over @xmath4 [ see eq . ( [ eldt ] ) ] . this integration involves evaluation of integrals of the form , @xmath148 where for the first term in eq . ( [ sdosc ] ) @xmath149 and @xmath150 , and for the second one @xmath151 and @xmath152 . the fast oscillatory character of the exponential factor [ i.e. , @xmath153 for all @xmath4 ] relative to the slow variation of @xmath155 allows us to use the standard stationary phase method , @xcite obtaining @xmath156^{1/2 } g(0 ) \ ; \re \ { \exp [ i \alpha k a(0 ) + i \pi/4 ] \ } \nonumber \\ & & + \frac{2 } { \alpha k a^\prime ( l/2 ) } \ ; g(l/2 ) \ ; \re \left\ { -i \exp[i \alpha k a(l/2 ) ] \right\}~ , \label{inte2}\end{aligned}\ ] ] where @xmath9 is the stationary ( extremum ) point , the second term is the contribution from the end - points of the integral , and primes denote differentiation with respect to @xmath4 . using the above , and after simple algebraic manipulations , we obtain for the oscillatory part of the density of states , @xmath157^{1/2 } \left [ \frac{\pi } { k a^{\prime\prime } ( 0 ) } \right]^{1/2 } \right . \nonumber \\ & & \times \frac{2 m_e a(0)}{\hbar^2 } \cos \left [ 2 m k a(0 ) \sin \left ( \frac{\pi q } { m } \right ) + \frac{\pi}{2 } \left ( m+\frac{1}{2 } \right ) \right ] \nonumber \\ & & \left . + \frac{1}{m^2 } \frac{2 m_e a(l/2 ) } { \hbar^2 k a^\prime ( l/2 ) } \sin \!\ ! \left [ 2 m k a(l/2 ) \sin \!\ ! \left ( \frac{\pi q}{m } \right ) + \frac{\pi m}{2 } \right ] \right\ } \nonumber \\ & & + \frac{2 \sqrt{2}}{\pi \epsilon^{1/4 } } \sum_{m=1}^{\infty } \left\ { \frac{1}{m } \left [ \frac{1}{k a^{\prime\prime } ( 0 ) } \right]^{1/2 } \frac{1}{a(0 ) } \right . \nonumber \\ & & \times \left ( \frac{2 m_e a^2(0)}{\hbar^2 } \right)^{3/4 } \cos [ 2 \pi m k a(0 ) ] \nonumber \\ & & - \frac{1}{\pi m^{3/2 } } \frac{1}{k a^\prime ( l/2 ) a(l/2 ) } \left ( \frac{2 m_e a^2 ( l/2)}{\hbar^2 } \right)^{3/4 } \nonumber \\ & & \times \left . \cos \left [ 2 \pi m k a(l/2 ) + \frac{\pi}{4 } \right ] \right\}. \label{sdosc2}\end{aligned}\ ] ] the density of states of the wire contains oscillatory contributions from the narrowmost cross - section of the wire [ first and third term in eq . ( [ sdosc2 ] ) ] and from the wires - end cross - sections ( second and fourth terms ) . the amplitudes of the latter oscillations is smaller . having obtained the expression for the oscillatory part of the density of states , we can calculate now the semiclassical approximation to the grand - canonical thermodynamic potential @xmath158 [ see appendix a ; at zero temperature , @xmath159 . restricting ourselves for brevity to the largest contribution [ that is , to the first term in eq . ( [ sdosc2 ] ) corresponding to the main contribution from the narrowmost part of the wire ] , we get for the oscillatory part of @xmath158 , @xmath160~. \label{omosc}\end{aligned}\ ] ] from this expression , the oscillating part of the force as a function of the length of the constricted region [ i.e. , @xmath161 in general depends on @xmath5 , see e.g. , eq . ( [ a0 ] ) ] is given by @xmath162 which upon substitution of ( [ omosc ] ) yields , @xmath163 } { \sqrt{\pi a^{\prime\prime } ( 0 ) } \ ; a(0 ) } \sum_{m=2}^{\infty } \sum_{q=1}^{m/2 } \frac{1}{m^{5/2 } } \sin^{-1/2 } ( \pi q / m ) \nonumber \\ & & \times \cos \!\ ! \left[2 m k_f a(0 ) \sin ( \pi q / m ) + \frac{\pi}{2 } \left ( m - \frac{1}{2 } \right ) \right]~. \label{fsem}\end{aligned}\ ] ] the expression for the conductance of the wire following the landauer formula involves evaluation of the number of transverse states in the narrowmost part of the wire . following ref . , @xmath164 ^ 2 } { 4 } \left \ { 1 - \frac{2}{k_f a(0 ) } + \frac{8}{\sqrt{\pi } } \frac{1 } { [ k_f a(0)]^{3/2 } } \right . \nonumber \\ & & \times \sum_{m=2}^{\infty } \sum_{q=1}^{m/2 } \frac{1}{m^{3/2 } } \sin^{1/2 } ( \pi q / m ) \nonumber \\ & & \times \left . \cos \!\ ! \left [ 2 m k_f a(0 ) \sin(\pi q / m ) + \frac{\pi}{2 } \left ( m - \frac{1}{2 } \right ) \right ] \right\}~ , \label{gsem}\end{aligned}\ ] ] which can be expressed as a function of the length of the constriction [ see e.g. , eq . ( [ a0 ] ) ; we remark here that our semiclassical treatment is valid for any adiabatic wire shape ] . the non - oscillating contribution ( coming from the first two terms in the curly brackets ) describes the sharvin @xcite conductance of the constriction and the weyl @xcite semiclassical corrections , and the third term describes conductance quantum oscillations as a function of @xmath161 . from a comparison of the expression for the oscillatory contribution to the force [ eq.([fsem ] ) ] with the oscillatory contribution to the conductance [ eq . ( [ gsem ] ) ] , the direct correlation between the two is immediately evident , and both depend on the spectrum of transverse modes ( conducting channels ) at the narrowmost part of the wire . this is in agreement with the results shown in fig . 5 obtained through the lda - scm method . in this paper , we extended our investigations @xcite of energetics , conductance , and mesoscopic forces in a jellium modelled nanowire ( sodium ) using the local - density - functional - based shell correction method to variable - shaped wires , i.e. , containing a constricted region modeled here by a parabolic dependence of the cross - sectional radii in the constriction on @xmath4 ( see fig . 1 ) . the results shown above , particularly , the oscillations in the total energy of the wire as a function of the length of the variable - shaped constricted region ( and correspondingly its narrowmost width ) , the consequent oscillations in the elongation force , the corresponding discrete sequence of magic wire configurations , and the direct correlation between these oscillations and the stepwise quantized conductance of the nanowires , originate from quantization of the electronic states ( i.e. , formation of subbands ) due to the reduced lateral ( transverse ) dimension of the nanowires . these results are in correspondence with our earlier lda - scm investigation of jellium - modeled uniform nanowires . @xcite moreover , in the current study of a wire with a variable ( adiabatic ) shaped constriction , we found that the oscillatory behavior of the energetic and transport properties are governed by the subband quantization spectrum ( termed here electronic shells ) at the narrowmost part of the constriction . this characteristic is supported and corroborated by our semiclassical analysis ( section iii ) . we reiterate here that such oscillatory behavior , as well as the appearance of `` magic numbers '' and `` magic configurations '' of enhanced stability , are a general characteristic of finite - size fermionic systems and are in direct analogy with those found in simple - metal clusters ( as well as in @xmath0he clusters @xcite and atomic nuclei@xcite ) , where electronic shell effects on the energetics @xcite ( and most recently shape dynamics @xcite of jellium modelled clusters driven by forces obtained from shell - corrected energetics ) have been studied for over a decade . while these calculations provide a useful and instructive framework , we remark that they are not a substitute for theories where the atomistic nature and specific atomic arrangements are included @xcite in evaluation of the energetics ( and dynamics ) of these systems ( see in particular refs . , where first - principles molecular - dynamics simulations of electronic spectra , geometrical structure , atomic dynamics , electronic transport and fluctuations in sodium nanowires have been discussed ) . indeed , the atomistic structural characteristics of nanowires @xcite ( including the occurrence of cluster - derived structures of particular geometries@xcite ) , which may be observed through the use of high resolution microscopy , @xcite influence the electronic spectrum and transport characteristics , as well as the energetics of nanowires and their mechanical properties and response mechanisms . in particular , the mechanical response of materials involves structural changes through displacement and discrete rearrangenent of the atoms . the mechanisms , pathways , and rates of such structural transformations are dependent on the arrangements and coordinations of atoms , the magnitude of structural transformation barriers , and the local shape of the wire , as well as possible dependency on the history of the material and the conditions of the experiment ( i.e. , fast versus slow extensions ) . further evidence for the discrete atomistic nature of the structural transformations is provided by the shape of the force variations ( compare the calculated fig . 3(b ) in ref . and fig . 3 in ref . with the measurements shown in figs . 1 and 2 in ref . ) , and the interlayer spacing period of the force oscillations when the wire narrows . while such issues are not addressed by models which do not include the atomistic nature of the material , the mesoscopic ( in a sense universal ) phenomena described by our model are of interest , and may guide further research in the area of finite - size systems in the nanoscale regime . such further investigations include the occurrence of magic configurations ( i.e. , sequences of enhanced stability specified by number of particles , size , thickness or shape ) in clusters , dots , wires , and thin films of normal , as well as superconducting metals , and the effect of magnetic fields which can influence the energetics in such systems ( e.g. , leading to magnetostriction effects ) through variations of the subband spectra , in analogy with magnetotransport phenomena in nanowires @xcite . several directions for improving the model ( while remaining within a jellium framework ) are possible . these include : ( i ) consideration of more complex shapes . for example , in our current model the elongation is distributed over the entire constriction throughout the process , while a more realistic description should include a gradual concentration of the elongation , and consequent shape variation , to the narrower part of the constriction as found through molecular - dynamics simulations ; @xcite ( ii ) use of a stabilized - jellium description @xcite of the energetics of the nanowire in order to give it certain elements of mechanical stability . in this context , note also that from the total energy shown in fig . 4(c ) , and the corresponding total force [ fig . 5(c ) ] , it is evident that in our current model , except for the region of large elongation close to the breaking point ( i.e. , @xmath165 ) , the wire is unstable against spontaneous collapse ( that is shortening ) , i.e. , there are no energetic barriers against such process , while both experiments @xcite and md simulations @xcite show that compression of such wires requires the application of an external force . improvements of the model in these directions are most desirable in light of the aforementioned experimental @xcite and md - simulations @xcite observations that the total oscillating forces for elongation and compression of nanowires are of opposite signs ( i.e. , negative and positive , respectively ) , while our current ( equilibrium model ) is limited to certain aspects of the tensile part of an elongation - compression cycle ; ( iii ) inclusion of bias voltage effects in calculations of the energetics and conductance of nanowires . @xcite while such effects may be expected to have little influence ( particularly on the energetics ) at small voltages , they could be of significance at larger ones . work in these directions is in progress in our laboratory . this research was supported by a grant from the u.s . department of energy ( grant no . fg05 - 86er45234 ) and the afosr . useful comments by w.d . luedtke are gratefully acknowledged . calculations were performed at the georgia institute of technology center for computational materials science . in this appendix , we discuss briefly a semiclassical treatment of temperature effects on the oscillatory behavior of the force and conductance in nanowires . the grand - canonical thermodynamic potential at finite temperature , @xmath166 , is given by @xmath167~ , \label{omeg}\ ] ] where @xmath168 denotes @xmath169 , and @xmath170 is the chemical potential . from eq . ( [ omeg ] ) , the finite temperature expressions for @xmath171 , @xmath172 , and @xmath173 differ from those given for the zero - temperature limit in eqs . ( [ omosc ] ) , ( [ fsem ] ) , and ( [ gsem ] ) , respectively , by a multiplicative factor in the sums of these equations . this factor is given by @xcite note that the temperature dependence given in eq . ( [ tdep ] ) is valid for systems with @xmath179 , and leads to reduction of the oscillation amplitudes when @xmath180 , where @xmath181 $ ] is an effective energy - level spacing of the electrons contributing to the oscillatory parts of the thermodynamic potential , force , and conductance . see articles in _ large clusters of atoms and molecules _ , edited by t.p . martin ( kluwer , dordrecht , 1996 ) . see articles in _ proceedings of the 8th international symposium on small particles and inorganic clusters , copenhagen , 1996 _ , edited by h.h . anderson , z. phys . d * 40 * , 1 - 578 ( 1997 ) . see articles in _ atomic and nanometer - scale modifications of materials : fundamentals and applications _ , edited by ph . avouris ( kluwer , dordrecht , 1993 ) . see articles in _ nanowires _ , edited by p.a . serena and n. garcia ( kluwer , dordrecht , 1997 ) . u. landman , r.n . barnett , and w.d . luedtke , z. phys . d * 40 * , 282 ( 1997 ) . barnett , and u. landman , nature * 387 * , 788 ( 1997 ) . de heer , rev . phys . * 65 * , 611 ( 1993 ) . c. yannouleas and u. landman , in ref . , martin , phys . rep . * 273 * , 199 ( 1996 ) . c. yannouleas and u. landman , j. phys . chem . * 101 * , 5780 ( 1997 ) . c. yannouleas and u. landman , phys . b * 48 * , 8376 ( 1993 ) ; chem . phys . lett . * 210 * , 437 ( 1993 ) . u. landman , w.d . luedtke , n. burnham , and r.j . colton , science * 248 * , 454 ( 1990 ) . u. landman , w.d . luedtke , b.e . salisbury , and r.l . whetten , phys . lett . * 77 * , 1362 ( 1996 ) . u. landman , w.d . luedtke , and j. gao , langmuir * 12 * , 4514 ( 1996 ) . bogachek , a.m. zagoskin , and i.o . kulik , fiz . nizk . temp . * 16 * , 1404 ( 1990 ) [ sov . j. low temp . * 16 * , 796 ( 1990 ) ] . j.i . pascual , j. mendez , j. gomez - herrero , j.m . baro , n. garcia , and v.t . binh , phys . lett . * 71 * , 1852 ( 1993 ) . l. olesen , e. laegsgaard , i. stensgaard , f. besenbacher , j. schiotz , p. stoltze , k.w . jacobsen , and j.n . norskov , phys . rev . lett . * 72 * , 2251 ( 1994 ) . j.i . pascual , j. mendez , j. gomez - herrero , j.m . baro , n. garcia , u. landman , w.d . luedtke , e.n . bogachek , and h .- cheng , science * 267 * , 1793 ( 1995 ) . smith , science * 269 * , 371 ( 1995 ) . g. rubio , n. agrait , s. vieira , phys . lett . * 76 * , 2302 ( 1996 ) . a. stalder and u. durig , appl . phys . lett . * 68 * , 637 ( 1996 ) . krans , j.m . van ruitenbeek , v.v . fisun , i.k . yanson , and l.j de jongh , nature * 375 * , 767 ( 1995 ) . costa - kramer , n. garcia , p. garcia - mochales , and p.a . serena , surface science * 342 * , 11 144 ( 1995 ) . van ruitenbeek , m.h . devoret , d. esteve , and c. urbina , to be published . in this study , a uniform wire ( i.e. , with the cross - section independent of the location on the symmetry axis ) has been investigated . electron - density spillout effects were approximated in an ad hoc manner . stafford , d. baeriswyl , and j. brki , to be published . in this study , a wire of non - uniform shape has been studied using a free - electron model and assuming a constant bulk value of the wire s fermi energy for all wire configurations . the last assumption implies neglect of the strong screening effects in metals as discussed in ref . , where , within the context of a free - electron model , it is noted that this can lead to substantial deviations ( that are largest for small cross sectional radii ) of the electronic density from bulk values . we also note that the expression given in this paper for the surface energy is at variance ( i.e. , a factor of five too large ) with the correct result for the infinite - barrier free - electron model given in ref . 266 - 267 ; we thank dr . lang for pointing out this discrepancy ) . ref . includes also a critical discussion pertaining to limitations of the free - electron model . lang , in _ solid state physics _ , edited by h. ehrenreich , f. seitz , and d. turnbull ( academic press , new york , 1973 ) , vol . 28 , p. 225 . c. yannouleas and u. landman , phys . b * 51 * , 1902 ( 1995 ) ; phys . rev . * 78 * , 1424 ( 1997 ) ; j. chem . phys . * 107 * , 1032 ( 1997 ) . c. yannouleas and u. landman , j. chem . phys . * 105 * , 8734 ( 1996 ) ; phys . b * 54 * , 7690 ( 1996 ) . strutinsky , nucl . a * 95 * , 420 ( 1967 ) ; _ ibid . _ * 122 * , 1 ( 1968 ) ; . bohr and b.r . mottelson , _ nuclear structure _ , vol . ii ( benjamin , reading , ma , 1975 ) . here we use the gunnarsson - lundqvist xc functional [ see o. gunnarsson and b.i . lundqvist , phys . b * 13 * , 4274 ( 1976 ) ] . hodges , can . j. phys . * 51 * , 1428 ( 1973 ) . l.i . glazman , g.b . lesovik , d.e . khmelnitskii , and r.i . shekhter , pisma zh . fiz . * 48 * , 218 ( 1988 ) [ jetp lett . * 48 * , 238 ( 1988 ) ] . a. yacoby and y. imry , phys . b * 41 * , 5341 ( 1990 ) . c. yannouleas and r.a . broglia , phys . b * 44 * , 5793 ( 1991 ) ; c. yannouleas , r.a . broglia , m. brack , and p .- f . bortignon , phys . lett . * 63 * , 255 ( 1989 ) ; m. brack , phys . b * 39 * , 3533 ( 1989 ) ; w. ekardt , phys . rev . b * 29 * , 1558 ( 1984 ) . r. landauer , philos . mag . * 21 * , 863 ( 1970 ) . y. imry , in _ directions in condensed matter physics _ , edited by g. grinstein and g. mazenko ( world scientific , singapore , 1986 ) , p. 101 . bogachek , a.g . scherbakov , and u. landman , phys . b * 53 * , r13 246 ( 1996 ) . lang , phys . b * 52 * , 5335 ( 1995 ) . m. brandbyge , k.w . jacobsen , and j.k . norskov , phys . b * 55 * , 2637 ( 1997 ) . todorov and a.p . sutton , phys . lett . * 70 * , 2138 ( 1993 ) . j.b . keller and s.i . rubinow , ann . of phys . * 9 * , 24 ( 1960 ) ; in this context , see also r. landauer , _ phase integral approximations in quantum mechanics _ , ph . d. thesis , harvard ( 1950 ) . m. brack and r.k . bhaduri , _ semiclassical physics _ ( addison - wesley , reading , ma , 1997 ) . e. jahnke , f. emde , and f. loesch , _ tables of higher functions _ ( mcgraw - hill , new york , 1960 ) . dingle , proc . ( london ) ser . a * 212 * , 47 ( 1952 ) . bogachek and g.a . gogadze , zh . . fiz . * 63 * , 1839 ( 1972 ) [ sov jetp * 36 * , 973 ( 1973 ) ] . a. erdelyi , _ asymptotic expansions _ ( dover , new york , 1956 ) . bogachek , m. jonson , r.i . shekhter , and t. swahn , phys . b * 50 * , 18 341 ( 1994 ) ; see in particular eqs . ( 15 ) and ( 17 ) , with @xmath182 , @xmath183 . v. sharvin , zh . . fiz . * 48 * , 984 ( 1965 ) [ sov . jetp * 21 * , 655 ( 1965 ) ] . falko and g.b lesovik , solid state commun . * 84 * , 835 ( 1992 ) ; j.a . torres , j.i . pascual , and j.j . saenz , phys . b * 49 * , 16 581 ( 1994 ) ; e.n . bogachek , a.g . scherbakov , and u. landman , phys . b * 56 * , 1065 ( 1997 ) . c. yannouleas and u. landman , submitted for publication . t. kizuka , k. yamada , s. degachi , m. naruse , and n. tanaka , phys . b * 55 * , r7398 ( 1997 ) . perdew , h.q . tran , and e.d . smith , phys . b * 42 * , 11 627 ( 1990 ) . lang , phys . b * 45 * , 13 599 ( 1992 ) ; _ ibid . _ * 49 * , 2067 ( 1994 ) . such functional dependence on the temperature of the amplitude of oscillations of thermodynamic and transport properties in metals is a typical one ; see e.g. , a.a . abrikosov , _ fundamentals of the theory of metals _ ( elsevier , amsterdam , 1988 ) .

photometric redshifts are used to determine the rest frame luminosity function ( lf ) of both early - type and late - type galaxies to -17.6 for the cluster cl1601 + 42 at = 0.54 . the total lf shows a steep faint - end slope -1.4 , indicating the existence of a numerous population of dwarf galaxies . luminous galaxies , with are mostly red , early - type galaxies , with a lf best described by a gaussian . faint galaxies are predominantly blue , late - type galaxies , well fitted by a schechter function with -1.7 . compared to clusters at lower redshift , the steepening of the faint end starts at brighter magnitudes for cl1601 + 42 , which may indicate a brightening of todays dwarf population relative to the giant population with increasing redshift . early - type galaxies are centrally concentrated , and dominate the core region , implying that the radial gradient of early - type galaxies seen in local clusters is already established at . bright , late - type galaxies are rare , consistent with a decrease in star formation in field galaxies as they are accreted on to the cluster , while faint , blue galaxies are evenly distributed across the cluster , except for a depletion in the core region . the blue fraction is 0.15 , which is somewhat lower than the butcher - oemler average at . the value of is found to increase with limiting magnitude and with radius from the centre . = = = = = = = = # 1 # 1 # 1 # 1 @mathgroup@group @mathgroup@normal@groupeurmn @mathgroup@bold@groupeurbn @mathgroup@group @mathgroup@normal@groupmsamn @mathgroup@bold@groupmsamn = `` 019 = ' ' 016 = `` 040 = ' ' 336 = " 33e = = = = = = = = # 1 # 1 # 1 # 1 = = = = = = = = [ firstpage ] galaxies : clusters : individual : cl1601 + 42 galaxies : distances and redshift galaxies : luminosity function , mass function cosmology : observations

astro-ph0110419

butcher & oemler ( 1978a ; 1984 , hereafter bo84 ) found that the fraction of blue galaxies increases dramatically between redshifts @xmath1 = 0 and @xmath7 0.5 , now usually referred to as the butcher oemler effect ( hereafter bo effect ) . this trend has been confirmed by e.g. rakos & schombert , ellingson et al . , kodama & bower and margoniner et al . . the increase of blue galaxies is explained by a higher rate of star formation in cluster galaxies at earlier epochs . the cause of this higher star formation rate is , however , not fully understood . butcher & oemler ( 1978a ) noted that the excess of blue galaxies at higher redshifts compared to todays clusters approximately matches the excess of s0 galaxies in local clusters . if such a change in the morphological mix is caused by an evolution of the blue galaxies into s0 galaxies , once the gas in the star forming blue galaxies is exhausted or tidally stripped , this could lead to the observed bo effect . the evolution of the fraction of s0 s between high and low redshift clusters is discussed in e.g. dressler et al . and fasano et al . . bower and kauffmann explain the bo effect as a natural consequence of the hierarchical clustering model for structure evolution in the universe . in this picture structures , such as clusters , are built up from bottom up , via repeated mergers of subunits . bower shows that the infall rate of field galaxies to clusters increases with look - back time , leading to a higher fraction of blue galaxies in the clusters . the trend found is not as strong as in the bo84 observations , but can be fitted if the blue fraction in the field increases with redshift . kauffmann shows that high redshift clusters assemble during a much shorter time interval than comparable clusters today . this in turn implies a higher rate of infall of galaxies from the field at higher redshift , producing the bo trend . however , the strong increase in the blue fraction of the observed clusters is explained only in high - density ( @xmath8 ) cdm models , or mdm models . low - density cdm models have difficulty in explaining the strong trend at @xmath9 @xcite . recent simulations by diaferio et al . find that a bo trend is found also in a low - density model ( @xmath10 ) , but significantly weaker than in observations . it is not clear if star formation is triggered in the infalling galaxies , possibly creating a burst , or if star formation in the field galaxies is simply quenched over some period of time . both these scenarios seem compatible with the observed bo effect . the results from bo84 also show that open clusters have a larger fraction of blue galaxies than the more centrally concentrated , compact clusters . ( a formal definition of open and compact clusters is given in section 4.1 . ) this can be explained in the context of the hierarchical scenario , if the smaller concentration in open clusters is a reflection of a more recent assembly of its members . the presence of substructure in a cluster can be a sign of recent subcluster - subcluster merger . from a study of three clusters with signs of substructure at @xmath11 , metevier , romer & ulmer find that at least two of these have an exceptionally large fraction of blue galaxies . this supports a hierarchical scenario where mergers induce star formation in the cluster galaxies . furthermore , results from margoniner et al . indicate that poor clusters in general have higher blue fractions than rich clusters at comparable redshifts . it is therefore clear that the bo effect is not a simple redshift effect , but must be viewed as a consequence of several factors . these include richness , central concentration , epoch of formation and presence of substructure . a further prediction of the hierarchical cdm model is that the number density of galaxies should be dominated by low mass systems . this is reflected in the luminosity function ( hereafter lf ) , defined as the number density of galaxies per unit luminosity , or per unit magnitude . for the hierarchical scenario , the lf is predicted to have a steep faint - end slope , reflecting the increasing number of low mass systems with faint magnitudes @xcite . therefore , studies of the lf , both in the field and in clusters , is one of the most direct probes for testing the hierarchical model . detailed studies of the spatial and morphological distribution of cluster galaxies are important for understanding the internal properties of clusters , as well as the evolution of these over time . to do this observationally , it is necessary to isolate the galaxies in a given field belonging to the cluster , from the foreground or background field galaxies , projected on to the cluster . spectroscopy is costly , since a cluster field may consist of thousands of galaxies . more important , for clusters at higher redshifts spectroscopy is limited to only the most luminous galaxies . the traditional way of correcting for field contamination at faint magnitudes is to use a cluster - free field , preferentially close to the cluster , and calculate the number of galaxies as a function of magnitude to be subtracted from the cluster image . this method is , however , increasingly uncertain at higher redshifts , where the ratio of field galaxies to cluster galaxies increases ( e.g. driver at al . 1998a ) . in section 3.7 we show that the field contamination can exceed 80 per cent of the total number of galaxies for clusters at redshifts @xmath12 . a further important disadvantage is that it is not possible to assign membership to a specific galaxy with any confidence . an alternative and more powerful way to determine cluster membership is to use photometric redshifts based on broad band photometry to estimate redshifts . the main advantage with this method is that a large number of redshifts can be determined in a relatively short time , and to magnitudes considerably fainter than traditional spectroscopy . an important bonus is that the photometric redshift technique also provides an estimate of the hubble type for the observed galaxies , or more precisely , their broad spectroscopic characteristics . this information can be used , together with redshift information , to determine the fraction , as well as the spatial distribution , of different types of galaxies . we later discuss this technique in more detail . the trade - off with this method is that uncertainties in the estimated redshifts are considerably larger compared to spectroscopy . the accuracy has , however , improved and typical rms deviations between spectroscopic redshifts and broad band photometric redshifts are @xmath13 0.05 at @xmath14 @xcite and @xmath15 0.060.1 up to @xmath16 @xcite . this accuracy is sufficient for determination of cluster membership and absolute magnitudes , but is clearly not useful for dynamic studies . in this paper we demonstrate the use of photometric redshifts for a population study of the cluster cl1601 + 42 . with information on cluster membership and spectroscopic galaxy type , we calculate the luminosity function separately for early - type and late - type galaxies within the cluster , as well as their spatial distribution . previously , photometric redshifts have been used to determine cluster membership for a few clusters only . the most extensive study of photometric redshifts applied to clusters of galaxies is presented by brunner & lubin . they in paricular , discuss how the method can be used to either maximize completeness , or to minimize contamination from field galaxies . lubin & brunner present results on the distribution of galaxy morphologies in two distant clusters , where cluster membership is determined by photometric redshifts . kodama , bell & bower use photometric redshifts for determining cluster members in abell 370 at @xmath17 . connolly et al . include photometric redshifts when investigating a supercluster structure at @xmath1 = 0.54 , and connolly , szalay & brunner use the method for measuring the evolution of the angular correlation function in the hdf . finally , bzecourt et al . use photometric redshifts when analysing weak lensing of the cluster abell 2219 at @xmath1 = 0.22 . for studies of the lf and population content of clusters this method has , however , only had limited use . the cluster cl1601 + 42 is an intermediate rich cluster at @xmath1 = 0.54 . it was chosen because there already exist both high resolution hst imaging of the core of the cluster @xcite , and a fairly large number of spectroscopic redshifts of galaxies in the cluster field @xcite . the cluster is included in the morphs sample of ten clusters at @xmath18 ( smail et al . the available information is valuable both for testing the photometric redshifts technique , as well as for determining the morphological properties of the cluster . the plan of this paper is as follows . in section 2 we present the data , in section 3 we describe the photometric redshift technique used . results are presented in section 4 , followed by a discussion in section 5 . finally , conclusions are given in section 6 . we assume in this paper a flat cosmology with @xmath19 and @xmath20 , and a hubble constant @xmath21 km s@xmath22 mpc@xmath22 . the absolute magnitude for other choices of the hubble constant are related by : @xmath23 . at the cluster redshift , @xmath1 = 0.54 , the absolute magnitudes for different cosmologies are related by : @xmath24 . magnitudes are given in the standard vega based system . cl1601 + 42 is located at r.a . 16@xmath2503@[email protected] and dec . @xmath28 45 18(j2000.0 ) . the redshift has been determined spectroscopically to @xmath29 @xcite , and the velocity dispersion is 1210 km s@xmath22 @xcite . the cluster is a weak x ray source with @xmath30 ( 0.33.5 kev ) = 0.35 @xmath31 10@xmath32 ergs @xcite . the galactic extinction is consistent with @xmath33 @xcite . the cluster was observed with the 2.56 m nordic optical telescope ( not ) and the andalucia faint object spectrograph and camera ( alfosc ) during four observational runs 19971999 ( table [ table1 ] ) . alfosc contains a loral 2k@xmath34 chip with a field of [email protected] and an image scale of [email protected]/pixel . for the observations during 1997 we used the not ccd#6 and for the 199899 observations the not ccd#7 cl1601 + 42 was observed in five filters , @xmath37 , @xmath38 , @xmath39 , @xmath40 and @xmath41 . in the @xmath40 band we also observed a neighbouring blank field , which we refer to as the background field . the position of this field is r.a . 16@xmath2508@[email protected] and dec . @xmath42 34 00(j2000.0 ) . all observations were performed under photometric conditions . the seeing in the images varies between 0.75 and 1.0 , depending on filter . a complete log of the observations is given in table [ table1 ] . [ figure1 ] shows a colour image of cl1601 + 42 , constructed from the @xmath38 , @xmath40 and @xmath41 bands . the data were reduced using the iraf and midas packages . bias subtraction was made in a standard manner . because of the large number of science frames in each filter , flat - fields were constructed directly from these images after removing objects . because the @xmath41band science frames show a fringe pattern we used short exposure , twilight frames for the flat - field . after flat - fielding the science @xmath41images , a fringe frame was created , which was subtracted from the science images . the flat - fielded images in each filter were corrected for atmospheric extinction , aligned , and finally combined . the final size of the field is 5.5 @xmath35 5.6 arcmin . the images were calibrated using standard stars from landolt and christian et al . . we estimate the photometric errors in the zero point calibration to @xmath43 in @xmath37 , @xmath38 , @xmath39 and @xmath40 , and @xmath44 in @xmath41 . [ cols="<,^,<,>,^,^ " , ] [ table3 ] as a comparison , we calculate the lf using the standard background subtraction method in a single filter . as background we use our @xmath40band image of a blank field , and subtract this from the cluster image , correcting for the slightly different field sizes . with this method we reach @xmath45 , i.e. , about one magnitude fainter than for the photometric redshifts . for the two faintest magnitude bins we have corrected for incompleteness as described in section 2.5 . this results in @xmath46 cluster galaxies to @xmath45 , while the number of galaxies to @xmath47 is @xmath48 , where the errors are statistical only . including field - to - field variance in the background counts results in @xmath49 at @xmath47 . to compare this with the number found by the photometric method , we correct the latter for the galaxies lost due to the 1.5@xmath50 cut in redshift range , and include field - to - field variance . this results in @xmath51 at @xmath47 , which is within the errors found by the subtraction method . the lf to @xmath45 using the subtraction method is shown in fig . [ figure6 ] . a schecter fit yields a faint - end slope @xmath52 , and the turnoff magnitude is @xmath53 , close to the values derived using photometric redshifts . the faint end fit , as defined by eq . ( 3 ) , for both @xmath54 and @xmath55 results in @xmath56 which shows that there is no dramatic change at the faint end when going slightly deeper in magnitude . comparing the error - bars in figure [ figure4 ] and figure [ figure6 ] clearly illustrates the gain with the photometric method , especially at faint magnitudes . the larger error - bars in figure [ figure4 ] around @xmath57 21 are due to the smaller statistics in these bins for the photometric method , probably caused by the use of different background fields in the two methods . the result also indicates that the subtraction method works reasonably well to redshifts @xmath58 , even though the number of cluster galaxies vary somewhat between the methods . this result may seem contrary to driver et al . , who from simulations find that the method is unreliable for clusters at @xmath59 . their conclusions are , however , valid for a specific observational set - up . we reach a completeness limit @xmath60 mag fainter than the limit used by driver et al . , which should increase the reliability of the subtraction method to a higher redshift . a possible source of uncertainty in the background subtraction method comes from gravitational lensing . background galaxies in the cluster image can have their magnitudes amplified by the cluster , leading to an excess of background galaxies in the cluster image , compared to the cluster free image used for subtraction . for the rich cluster a963 , which show signs of arcs , driver et al . estimate that the increase in background counts is @xmath61 10 per cent . when investigating the coma cluster , bernstein et al . argue that this increase will be cancelled by the reduction of the solid angle being viewed behind the lensing cluster , making no correction necessary . considering this , together with the fact this cl1601 + 42 is both intermediate in richness and concentration and has no detected arcs , we find that gravitational lensing should not affect our results . effects from gravitational lensing should not enter into the photometric method , since we do not subtract galaxies behind the cluster , but instead reject them using the redshift information . another factor to account for in the subtraction method is crowding , i.e. overlapping object . especially faint objects , with small areas , can be hidden behind brighter galaxies and stars . since there are in total more objects in the cluster image than in the background image , this can lead to a higher fractional loss of faint galaxies in the cluster image , resulting in a lf with too few faint objects . using similar arguments as in nslund et al . , we estimate the loss by adding the total area covered by objects in the cluster image and the background image , respectively . we find that the correction in the number of faint galaxies in the cluster lf is less than two per cent . since this correction factor is much less than statistical errors , as well as field - to - field variations , we do not include any corrections due to crowding in our results . for the photometric method , only foreground galaxies contribute to the crowding . the effect should be of equal importance in the cluster field and the field used for subtraction , and therefore cancel . using the cluster galaxies selected by photometric redshifts , we plot the radial distribution of the cluster members with @xmath40 @xmath62 25 in fig . [ figure7 ] . the left panel shows the projected surface density , and the right panel shows the fraction of late - type and early - type galaxies as a function of radius . the most interesting result is that the cluster outside of @xmath63 mpc shows a roughly constant fraction of late - type galaxies , @xmath61 70 per cent of the total . inside this radius the fraction of early - type galaxies increases rapidly to @xmath61 80 per cent at the centre . although , this effect is present also in low redshift clusters ( e.g. whitmore , gilmore & jones 1993 ) , our result shows that this population gradient is set up already at @xmath64 0.54 . in fact , the gradient is even steeper than in local clusters . in fig . [ figure8 ] we plot the radial distribution for bright ( @xmath65 ) and faint ( @xmath66 ) galaxies separately . while the bright galaxies show a steep cluster - centric distribution , the faint galaxies have a slower increase , which flattens at the centre , consistent with a homogeneous distribution of faint galaxies in the cluster . the figure also shows the radial distribution separately for bright and faint late - type galaxies . we discuss the radial distribution further in section 5.2 . bo84 showed that the fraction of blue galaxies , @xmath6 , in cores of rich compact clusters increases dramatically from @xmath67 for clusters with @xmath68 , to @xmath69 at @xmath4 . to determine @xmath6 in cl1601 + 42 we adopt the definition in bo84 . first , galaxies brighter than @xmath70 within a circular radius , @xmath71 , containing 30 per cent of the cluster galaxies are selected . from this sample @xmath6 is defined as the fraction of these galaxies with rest frame colours @xmath72 at least 0.2 mag bluer than the linear locus of the colour - magnitude ( cm ) relation . in order to compare our results with those of bo84 we use in this section their adopted cosmology , i.e @xmath73 km s@xmath22 mpc@xmath22 and @xmath74 . we find @xmath75 arcmin , or @xmath76 mpc for the cosmology above . this is somewhat smaller than the typical radius of @xmath77 mpc given by kodama & bower for a sample of rich clusters at @xmath78 , but similar to the radius @xmath79 mpc used by van dokkum et al . for the cluster ms1054 - 03 at @xmath80 . our value of @xmath71 is also similar in angular size , @xmath81 arcmin , to that used by bo84 for their highest redshift clusters . the blue fraction we find is @xmath82 . this value for cl1601 + 42 is significantly higher than the local value @xmath83 , but lower that the average value @xmath84 at @xmath29 , taken from the fit presented by bo84 . note , however , that bo84 did not use k corrections that depend on the spectral type of the galaxies @xcite . including this , @xmath6 decreases by @xmath85 , and the k - corrected average bo value at @xmath86 would therefore be @xmath87 . even after this correction , we find that a significant difference between the blue fraction in cl1601 + 42 and the bo84 sample remains . finally , a possible concern here are systematic effects , which can discriminate against selecting blue late - type galaxies , and therefore lead to an underestimate of @xmath6 . from the mc simulations described in section 3.3 , as well as comparing with available spectroscopic redshifts , we find that systematic effect can cause a relatively higher loss of blue galaxies when selecting cluster galaxies by using a fixed redshift interval . this may lead to an underestimate of the blue fraction by @xmath88 . the small effect here is due to the relatively bright limiting magnitude used to select the bo - galaxies , i.e. @xmath89 ( corresponding to @xmath90 ) . here systematic effects are not expected to affect the photometric redshifts more than marginally . however , when we in section 5.3 determine the blue fraction to fainter magnitudes , the underestimate may reach @xmath91 . cl1601 + 42 was observed with wfpc2 on hst during cycle 4 with the f702w filter . the total exposure was 16800 s @xcite . the image covers the central 5 square arcmin of the cluster , and detailed morphological information are presented by smail et al . for all 145 objects with @xmath92 . five of these are classified as stars , whereas no classification was possible for a further six objects . of the five objects classified as stars we get the same identification for three , while the other two are found in the lowest redshift bin , consistent with these also being stars . of the six objects without hst classification we identify three as stars , one as possible star and two as galaxies . this leaves 134 galaxies with classified morphology . we are able to determine photometric redshifts for 129 of these objects . of the 68 galaxies classified as early - type galaxies by the photometric redshift method , @xmath93 per cent have morphologies classified as e to sa , and the rest are sab to irr . conversely , of the 61 galaxies photometrically classified as late - type galaxies , @xmath94 per cent are morphologically classified as sab to irr . even though there may be some uncertainty in the visual classification into hubble types ( smail et al . set @xmath95 as the limit where no more than 20 per cent of the galaxies differ by more than one hubble type between two different classifiers ) , the main reason for this relatively low correspondence is the lack of a direct relation between spectroscopic and morphological types , as also shown by e.g. , dressler et al . and poggianti et al . . this can be seen in fig . [ figure9 ] , which shows the rest frame @xmath72 colour for the galaxies within the redshift range of the cluster , divided into two broad morphological classes . the lower panel shows the colours of the galaxies with early hubble types ( e sa ) , and the upper panel later types , as determined from the hst data by smail et al . most early - type galaxies are red and have a small dispersion in colour . there is , however , a tail of galaxies with relatively blue colours . the distribution of later types have a broader distribution , but are on average @xmath96 mag bluer than the earlier types . the figure clearly shows that there is relatively large population of objects , morphologically classified as late - type galaxies , that have red colours . luminosity functions for galaxies fainter than @xmath97 have previously been determined for clusters to @xmath98 . comparing the lfs for coma at @xmath99 @xcite , a2554 at @xmath100 ( smith et al . 1997 ) , a963 at @xmath101 @xcite and ms2255.7 + 2039 at @xmath102 ( nslund et al . 2000 ) reveals that both nearby and distant clusters show a steep faint - end slope ( @xmath103 ) , indicating no qualitative difference with redshift . our results for cl1601 + 42 show that this conclusion is true up to @xmath4 . for the highest redshift cluster with a deep lf , ms2255.7 + 2039 at @xmath102 , nslund et al . derive @xmath104 ( in our cosmology ) and @xmath105 , when fitting their data to a single schechter function . this agrees well with our results in table [ table2 ] . the increased fraction of faint galaxies in the outer part of cl1601 + 42 , shown in the left panel of fig . [ figure8 ] , has the consequence that the faint - end slope of the lf should depend on radius . consequently , we find that the faint - end slope for galaxies within a radius of 80 arcmin ( 0.7 mpc ) is @xmath106 , while between 80160 arcmin ( 0.71.4 mpc ) @xmath107 . when comparing the lf s between clusters , it is therefore important to remember that the slope of the lf in the core region can be much shallower than for the cluster as a whole . a steepening of the slope with radius is consistent with studies of local clusters . adami et al . e.g. , find a faint - end slope @xmath108 for the core region of coma , while for the whole cluster trentham finds a considerably steeper faint - end slope , @xmath109 . also , five of seven clusters at @xmath110 examined by driver , couch & phillipps show this trend . the faint - end slopes for these seven clusters vary between -0.9 and -1.4 , which is more shallow than the slopes for the clusters quoted in the beginning of this section . the flatter slope compared to cl1601 + 42 may be understood as a result of the effect above , since the clusters in the sample of driver et al . only cover a radius half the linear size of our cluster . it is interesting to compare the rest frame @xmath38band lfs of cl1601 + 42 and the coma cluster ( trentham 1998 ) , shown in fig . [ figure10 ] . coma is here normalised to cl1601 + 42 at @xmath111 . note that we use similar methods as trentham to correct the magnitudes at faint levels ( see section 2.5 ) . also , both investigations use a 3@xmath50 detection and similar surface brightness limits . we have for cl1601 + 42 added two data points derived from the one colour subtraction method ( open circles ) . these points are in agreement with a continuation of the steep slope found at intermediate magnitudes . the figure shows that at intermediate magnitudes , @xmath112 , coma has a gaussian lf peaking at @xmath113 , before the faint galaxies start to dominate at @xmath114 , and the slope of the lf turns steep . however , for cl1601 + 42 the steep part sets in already at @xmath111 , and no plateau is present , as in coma . if the apparent lack of a plateau in cl1601 + 42 is a general property at this redshift , it could indicate a general brightening of the faintest cluster members , because of more active star formation , with increasing redshift . this brightening would shift the faint , steep end towards brighter magnitudes at higher redshift , and therefore decrease the extent of the plateau-range , while preserving the slope . this is consistent with coma having @xmath115 at @xmath116 , and cl1601 + 42 @xmath117 at @xmath118 . the plateau between the turnoff at bright magnitudes , and the up - turn at faint , is present in both ms2255 at @xmath102 and a963 at @xmath101 . this indicates that the star formation activity of the faint galaxies indicated by cl1601 + 42 at @xmath29 , is not as important in the two clusters at lower redshift . there is some controversy regarding the colour dependence of the faint - end slope of the lf . de propris et al . determine the lf for four abell clusters in both @xmath38 and @xmath41 . although the slopes are very steep ( @xmath119 ) , there is no difference between the two colours . these results are in contrast to the results of wilson et al . ( 1997 ) , who determine the lf in @xmath38 and @xmath41 for a665 and a1689 , both at @xmath120 , and find @xmath121 for the @xmath39band and @xmath122 for the @xmath41band . comparing the slope of the lf s of the different colours for cl1601 + 42 ( table [ table2 ] ) , there is no significant trend when going from red to blue filters . this is consistent with the dominance of the late - type population at faint magnitudes . if a majority of the galaxies have similar colours , i.e. k - corrections , then one expects a horizontal shift of the faint - end lf between the different bands , while the steepness remains unchanged . to investigate if there is a trend at even shorter wavelengths in cl1601 + 42 , we calculate @xmath123 at rest frame @xmath124 ) and @xmath125 ) , corresponding to the observed @xmath38 and @xmath37bands , and find @xmath126 , and @xmath127 , for @xmath128 -19.0 and @xmath129 -19.1 , respectively . while the optical bands do not show any trend in the faint - end slope , the steepening of the slope in the @xmath130-bands suggests that the faintest cluster galaxies have an increasing fraction of young , @xmath130 bright stars . in a recent paper valotto , moore & lambas discuss systematic effects when determining the lf for clusters selected in two dimensions , such as the abell and edcc ( edinburgh durham cluster catalogue , lumsden et al . 1992 ) catalogues . from n - body simulations valotto find a bias towards selecting clusters which are contaminated by background clusters , or groups , projected on to the cluster . the over - density in the background , along the cluster line - of - sight , will result in an over - abundance of faint galaxies , which remains after the subtraction of a cluster - free background field . for a composite lf , calculated from simulations where the clusters are selected in two dimensions , similar to the abell clusters , valotto et al . find that the selection process therefore can introduce an artificial steepening of the lf . by selecting cluster members with photometric redshifts , the background contamination should be minimized in our sample . the steep faint - end slope that we find should therefore not be much affected by the selection effects described by valotto et al . furthermore , the similar slope we derive from the background subtraction method indicates that projection effects are not important for cl1601 + 42 . from deep studies of both near and intermediate redshift clusters , it is clear that a single schechter function gives a rather poor representation of the lf ( e.g. driver et al . 1994 ; wilson et al . 1997 ; nslund et al . for the virgo cluster binggeli et al . find that ellipticals with @xmath131 , as well as the different types of spiral galaxies , are better represented by gaussian lfs . the mean magnitude is for the spirals shifted towards fainter magnitudes for the later types . the sum of all types of spirals can also be well represented by a gaussian lf . at magnitudes m@xmath132 dwarf ellipticals ( de ) and irregulars the up - turn of the lf at the faintest magnitudes is mostly due to the de population , where a schechter function is a better representation of the lf . the total lf should therefore rise as a gaussian at bright magnitudes and then flatten to a plateau at intermediate magnitudes , before it again turns steep at the faint end . the conclusions above may seem to contradict other studies showing a flat lf . garilli , maccagni & andreon find @xmath133 for a composite lf calculated from 65 clusters at @xmath134 , while paolillo et al . find @xmath135 for 39 clusters at @xmath136 , when fitting their data to a single schechter function . paolillo et al . do , however , point out that the lf below m@xmath137 in these clusters , which probes the dwarf population , is systematically above the schechter function fit , indicating a upturn of the faint - end slope at these magnitudes . the flat slopes found in these investigations are therefore not in conflict with the steeper slopes found in deeper studies . above we argued that the lack of a plateau in the lf of cl1601 + 42 , as compared to clusters at lower redshift , can be explained by a uniform shift in magnitude for the fainter galaxies . because these galaxies are predominantly blue ( fig . [ figure5 ] ) , the proposed shift can be interpreted as a result of a higher star formation level in this population of galaxies compared to the bright population . for cluster galaxies at @xmath1 = 0.24 , which have had their star formation truncated , smail et al . ( 1998 ) show that late types galaxies both fade more and become redder than early type galaxies . this is consistent with the shift of the late type population in cl1601 + 42 . the total reddinging and fading between @xmath1 = 0.24 and @xmath1 = 0 is , however , fairly modest in these models . a more substantial fading is expected if dynamical processes are involved . wilson et al . argue that dwarf irregulars in clusters should fade rapidly , either due to a trunction process , such as ram pressure stripping , or by disruption due to tidal interactions . they estimate the fading since @xmath138 to @xmath139 mag . such a fading would make part of the progenitors of todays dwarfs with @xmath140 -15 detectable in cl1601 + 42 , and could therefore contribute to the faint - end slope . further , simulations of galaxy harassment @xcite show that high redshift ( @xmath141 ) cluster galaxies of type sc or later can be turned into low surface brightness spheroidal systems in todays local clusters . the fading of these harassed galaxies should be @xmath142 mag . the dwarf ellipticals present at @xmath4 should , on the other hand , only fade marginally by @xmath143 0.5 mag . whatever the mechanism , it seems that at least a fraction of the galaxies contributing to the steep slope at the faint end in local clusters have faded considerably compared to earlier epochs . the gradual disappearance of the plateau at high redshift could therefore be explained as a fading of the faint , late - type population , relative to the bright population . this scenario requires that the number of recently accreted galaxies is higher at @xmath4 than today , i.e. that there is an increase in the infall rate with redshift . as shown by the press schechter formalism , n - body simulations , as well as observations , this is indeed the case @xcite . a bluening with redshift , i.e. an increased star formation rate , of the infalling field population can also contribute to the observed effect . diaferio et al . show that there is a significant increase in star formation rate in field galaxies over the redshift range @xmath144 . therefore , the brightening is likely to be due to a combination of the above two effects . the brightening of the blue cluster population also relates to the bo effect , which shows an increase of star formation in a large fraction of cluster galaxies at higher redshifts . we return to this in section 5.3 . we now focus on the evolution of the bright cluster population . smail et al . calculate composite lfs for ten clusters in the morphs sample divided into three redshift bins centred on @xmath145 , @xmath146 and @xmath29 . ( the last bin includes cl1601 + 42 . ) the galaxies are here divided into ellipticals and spirals ( sa and later ) , based on visual classification into morphological types . the limiting magnitude for this sample is , however , too bright to detect the dwarf population , or for determining the faint - end slope . instead , when calculating the turnoff magnitude , smail et al . fix the faint - end slope at @xmath147 for the ellipticals , and at @xmath148 for the spirals . they find that there is a slow brightening of the elliptical population from @xmath149 to @xmath29 by @xmath150 . the spiral population does , however , not show any clear evolution with redshift . a change of cosmology affects these results somewhat . adopting @xmath151 and @xmath152 , instead of @xmath153 as in smail et al . , increases the brightening of the ellipticals to @xmath154 . there will in this case also be a marginal brightening for the spiral galaxies . a somewhat larger evolutionary trend is found by kodama & bower , who estimate that the bright spiral population have faded by about one magnitude since @xmath141 . the fading of the bright population is therefore smaller than of the faint population , which is in turn consistent with the disappearance of the plateau . as we have just discussed , galaxies are in a hierarchical scenario accreted from the field on to the clusters ( e.g. , diaferio et al . 2001 ) . when field galaxies , which are predominately spirals with ongoing star formation , are accreted , star formation is halted after a possible burst of star formation . the quenching can either be abrupt , or there could be a more gradual decline in star formation ( e.g. barger et al . 1996 , poggianti et al . 1999 , balogh et al . 1999 , balogh , navarro & morris 2000 , diaferio et al . 2001 ) . the decrease in star formation should set up a radial gradient in the clusters , where blue galaxies are expected to be more numerous in the outer parts , as is confirmed by observations and simulations ( diaferio et al . also , the higher accretion rate at earlier epochs , as well as higher star formation rate in the general field , implies that the fraction of blue galaxies should be higher in high@xmath1 clusters . there is therefore a connection between the shape of the lf and the existence of a radial colour gradient . in the right panel of fig . [ figure8 ] we showed the radial distribution of late - type galaxies , divided into a bright subsample , with @xmath65 , and a faint subsample with @xmath66 . from the figure it is obvious that bright , blue galaxies are rare across the whole cluster . the flat , or at most slightly increasing , projected surface density is consistent with a decline in the spatial distribution towards the core , even though the small statistics do not make this significant . the faint blue galaxies show a more dramatic trend . at radii @xmath155 mpc this is the most numerous population in the cluster . here , the surface density is consistent with a volume density that is either uniform , or decreasing with radius . the drop in surface density at the innermost point is consistent with an absense of faint blue galaxies at the core of the cluster . an explanantion may be provided by galaxy harassment , which should be increasingly efficient in the denser core region . the increasing infall rate at high redshifts , predicted by hierachical clustering , should result also in a change with redshift of the radial gradient of the fraction of early - type galaxies in the cluster . ellingson et al . show that galaxies with an old population dominate to larger radii in low redshift ( @xmath156 ) clusters , giving a flatter radial gradient , compared to high redshift ( @xmath157 ) clusters . this indicates that high@xmath1 clusters have a higher population of recently accreted , star forming galaxies at radii closer to the core , causing a steeper radial gradient . as the number of recently accreted star forming galaxies declines with time , the gradient becomes less steep . with similar magnitude limits and the same cosmology as in ellingson et al . , we find that the radial gradient of cl1601 + 42 is at least as steep as the slope they find from their high redshift sample , and significantly steeper than that of the low redshift clusters . to check whether the blue fraction , @xmath6 , decreases towards the core as a result of a decreasing star formation rate in the accreted field galaxies as they fall deeper into the cluster potential , we plot in fig . [ figure11 ] @xmath6 as function of radius ( here given in terms of r@xmath158 mpc ) . from this figure it is clear that there is a trend for an increasing value of @xmath6 at larger radii . this is in agreement with both bo84 and kodama & bower , and consistent with the infall scenario . as is discussed by kodama & bower , the change of @xmath6 with radius further illustrates the necessity of a correct determination of @xmath71 for a meaningful comparison of the blue fraction between clusters . it is also interesting to calculate the blue fraction as a function of the magnitude cut , as is done in kodama & bower for their seven clusters at @xmath159 . when varying the magnitude cut by @xmath160 mag from the fiducial value m@xmath161 they find that @xmath6 remains almost constant . only for two of the most distant clusters is there a trend for @xmath6 to increase at fainter magnitudes , which kodama & bower , however , attribute to increasing photometric errors with magnitude . instead , they argue that @xmath6 should depend only weakly with magnitude . in fig . [ figure12 ] we plot @xmath6 as a function of the magnitude cut . we find a clear trend for an increasing value of @xmath6 as fainter galaxies are included , similar to the trend for two of the highest redshift clusters in kodama & bower , but contradicting the general claim by kodama & bower that the blue fraction should be constant with magnitude . as illustrated by fig . [ figure5 ] , the increasing value of @xmath6 at faint magnitudes in cl1601 + 42 is a direct consequence of the different shapes of the red and blue lfs . the steepness of the blue lf causes the blue fraction to increase at fainter magnitude cuts , and decrease at bright . because we are more than one magnitude brighter that our limit , the increase in photometric errors is @xmath162 mag over the magnitude range where we calculate @xmath6 . it is therefore unlikely that the strong trend of @xmath6 with magnitude in cl1601 + 42 can be explained by increasing photometric errors at fainter magnitudes , as proposed by kodama & bower for the clusters showing this trend in their investigation . note also that systematic effects when selecting the bo - galaxies may lead to an underestimate of the blue fraction in the faintest bin by @xmath91 , as discussed in section 4.4 . this means that the trend shown in figure [ figure12 ] could be even stronger . the dependence on the magnitude cut makes the calculated blue fraction dependent on the choice of cosmology . e.g. , a constant cut in absolute magnitude , in the bo84 case @xmath70 , corresponds to a fainter apparent magnitude in a @xmath163dominated cosmology than in a cosmology without cosmological constant . this should lead to a relatively higher blue fraction at increasing redshifts for the @xmath163 cosmology . using @xmath151 and @xmath152 , instead of the cosmology used by bo84 ( @xmath164 , @xmath165 ) changes the cut in apparent magnitude by @xmath166 at @xmath29 . this increases the blue fraction from @xmath167 to @xmath168 . the difference between the extrapolated @xmath6 at @xmath29 from bo84 , and the @xmath6 found for cl1601 + 42 should therefore decrease in a @xmath163cosmology . we finally notice that the blue fraction for the field , derived from the eis data , is @xmath169 at the cluster redshift ( @xmath170 ) , and @xmath171 at lower redshift ( @xmath172 ) . this is close to the blue fraction @xmath173 calculated for the field by bo84 . the similarity between the blue fraction in the low and intermediate redshift bins does not reflect the strong increase with redshift in the star formation in field galaxies found by diaferio et al . . this is somewhat surprising , even though the relatively large errors in @xmath6 do not rule out such trend . the fraction of late - type galaxies for the field is @xmath63 . this fraction is consistent with the rough distribution of field galaxy types used by smail et al . . colours of galaxies do not only reflect star formation , but are also dependent on metallicity , reflected in the cm relation ( e.g. faber 1977 ; arimoto & yoshii 1987 ) . in principle , a low metallicity could cause faint old , but blue , galaxies in our sample to be classified as star forming late - type galaxies . to estimate the importance of this effect for the blue population at @xmath174 , we calculate the cm relation for the bright population with @xmath175 , and extrapolate this to fainter magnitudes . the slope of the @xmath72 vs. @xmath39 cm relation we find for the early - type galaxies in cl1601 + 42 is @xmath176 , for @xmath177 ( @xmath178 ) . over the full interval @xmath179 the mean bluening is then @xmath180 . this is small compared to the mean off - set in colours between early - types and late - types , which is @xmath181 at @xmath182 . in addition , if part of the bluest old galaxies at faint magnitudes ( @xmath183 ) are identified as late , star forming galaxies due to a low metallicity , the mean colour of the remaining early - type galaxies in this magnitude range will be redder than expected from an extrapolation of the cm relation . we do not see this effect , and therefore conclude that the bluening of early - type galaxies at faint magnitudes due to metallicity should not contribute more than marginally to the faint blue population we detect . the blue colours should therefore reflect the star formation . further support for this conclusion comes from ferreras & silk who determine rest frame @xmath184 ( @xmath61 2000 ) to @xmath39 colours of early - type galaxies in cl0939 + 4713 at @xmath185 . comparing the observed colours to synthetic models with different star formation histories , metallicities and dust models , they find that the bluening at faint magnitudes is consistent with at least a fraction of the stars in the galaxies being young . low metallicities alone are insufficient to explain the observed colours . ( for further discussions on metallicity effects in cluster galaxies see e.g. , couch et al . 1998 ; dressler et al . 1999 ; poggianti et al . 1999 . ) using photometry in five broad - band filters we determine photometric redshifts for the galaxies in the field of cl1601 + 42 at @xmath186 . cluster membership can therefore be assigned with a higher reliability for the individual galaxies , which , compared to statistical methods , significantly reduces the amount of background subtraction needed . as a by - product we also obtain information about the spectral properties of the galaxies , making assumptions about e.g. k corrections unnecessary . using this method , we calculate the luminosity function , the radial distribution of galaxies within the cluster , as well as the distribution of early - type and late - type galaxies . with this knowledge we can directly understand e.g. , the dependence of the butcher - oemler effect on such parameters as radius and magnitude cut . we find a steep faint - end slope of the rest frame @xmath38band cluster luminosity function , with @xmath187 . this is consistent with the slope at faint magnitudes for both local and @xmath188 clusters . the steepness of the lf is found to be correlated to the radial distance from the cluster centre , with a more shallow slope near the core , compared to the outer parts of the cluster . this is closely related to the difference found in the radial distribution between bright and faint galaxies , where bright galaxies are more centrally concentrated . most surveys reporting flatter slopes , @xmath189 , only cover the core region of the cluster , or do not probe the faint population beyond the upturn in the lf . when we divide the cluster galaxies spectroscopically into early - type and late - type galaxies , we find that these populations have very different lfs . early - type galaxies , including ellipticals and red spirals , are best represented by a gaussian . the late - type galaxies , including blue spirals , irregulars and dwarfs , are best fitted by a schechter function with a steep faint - end slope , @xmath109 . we find that the plateau region of the lf at intermediate magnitudes , found in most local and low@xmath1 clusters , is absent in cl1601 + 42 , and we suggest that this can be due to a general brightening of the blue population of galaxies , caused by a higher star formation rate . if the progenitors of todays dwarf galaxies in local clusters have faded by @xmath61 23 magnitudes since @xmath4 , this class of galaxies could contribute to the steep slope seen in cl1601 + 42 . we find significant differences in the radial distributions when we divide our sample into bright and faint galaxies . bright galaxies , which are predominantly early - types , are highly concentrated to the centre . this shows that the high density of early - type galaxies in the core of local clusters is established already at @xmath4 . the faint galaxies have a more shallow radial decline at increasing radii . at the core the surface density flattens . bright , blue , late - type galaxies are rare in the cluster . this is consistent with the hierarchical scenario where in - falling spiral galaxies have their star formation quenched as they approach the centre of the cluster . faint blue galaxies are more numerous , and are present across the whole cluster . there is , however , a decrease in the surface density at the core , which is consistent with a region close to the core depleted of faint , blue star forming galaxies . a possible mechanism could be the destruction of small galaxies through galaxy harassment. the fraction of blue galaxies in cl1601 + 42 is @xmath190 , which is higher than the local value , and qualitatively in agreement with the bo effect . the fraction is , however , lower than predicted from the average @xmath191 relation found by bo84 . the blue fraction is found to depend on the distance to the core centre , in accordance with other studies . the increasing value of @xmath6 with increasing radius is consistent with the hierarchical scenario , where clusters are formed by accretion of blue field galaxies . @xmath6 is also found to increase with the limiting magnitude . this is a direct consequence of the different lf s for the early - type and late - type galaxies . the radial dependence and the shape of the lf , especially its faint population , are therefore connected . finally , we note that although this study goes deeper than most previous investigations at this redshift , the conclusions are only based on one cluster , and a large sample including clusters of varying richness and concentration is needed to access the robustness of our conclusions . we hope to address this in future studies . we are grateful to claes - ingvar bjrnsson and gran stlin for comments on the paper , to the referee for several valuable comments and suggestions , and to the eis project at eso for making their data publicly available . this work is supported by the swedish research council and the swedish board for space sciences . nordic optical telescope is operated on the island of la palma jointly by denmark , finland , iceland , norway , and sweden , in the spanish observatorio del roque de los muchachos of the instituto de astrofisica de canarias . abell g.o . , 1958 , apjs , 3 , 211 adami c. , ulmer m.p . , durret f. , nichol r.c . , mazure a. , holden b.p . , romer a.k . , 2000 , a&a , 353 , 930 arimoto n. , yoshii y. , 1987 , a&a , 173 , 23 balogh m.l . , morris s.l . , yee h.k.c . , carlberg r.g . , ellingson e. , 1999 , apj , 527 , 54 balogh m.l . , navarro j.f . , morris s.l . , 2000 , apj , 540 , 113 barger a.j . , aragn salamanca a. , ellis r.s . , couch w.j . , smail i. , sharples r.m . , 1996 , mnras , 279 , 1 belloni p. , bruzual a.g . , thimm g.j . , rser h.j . , 1995 , a&a , 297 , 61 benitez n. , 2000 , apj , 536 , 571 bernstein g.m . , nichol r.c . , tyson j.a . , ulmer m.p . , wittman d. , 1995 , aj , 110 , 1507 bzecourt j. , hoekstra h. , gray m.e . , abdelsalam h.m . , kuijken k. , ellis r.s . , 2000 , a&a submitted , astro / ph0001513 binggeli b. , sandage a. , tammann g.a . , 1988 , ara&a , 26 , 509 bolzonella m. , miralles j.m . , pell r. , 2000 , a&a , 363 , 476 bower r.g . , 1991 , mnras , 248 , 332 bromley b.c . , press w.h . , lin h. , kirshner r.p . , 1998 , apj , 505 , 25 brunner r.j . , lubin l.m . , 2000 , aj , 120 , 2851 brunner r.j . , connolly a.j . , szalay a.s . , 1999 , apj , 516 , 563 bruzual g.a . , charlot s. , 1993 , apj , 405 , 538 bruzual g.a . , charlot s. , 1995 , ftp://www.iap.fr/pub/from_users/charlot/bc95/ butcher h. , oemler a. , 1978a , apj , 219 , 18 butcher h. , oemler a. , 1978b , apj , 226 , 559 butcher h. , oemler a. , 1984 , apj , 285 , 426 ( bo84 ) christian c.a . , adams m. , barnes j.v . , butcher h. , hayes d.s . , mould j.r . , siegel m. , 1985 , pasp , 97 , 363 cohen , j. g. , hogg d. , blandford r. , cowie l. , hu e. , songaila a. , shopbell p. , richberget k. , 2000 , apj , 538 , 29 coleman g.d . , wu c.c . , weedman d.w . , 1980 , apjs , 43 , 393 connolly a.j . , csabai i. , szalay a.s . , koo d.c . , kron r.g . , munn j.a . , 1995 , aj , 110 , 2655 connolly a.j . , szalay a.s . , koo d. , romer a.k . , holden b. , nichol r.c . , miyaji t. , 1996 , aj , 473 , 67 connolly a.j . , szalay a.s . , dickinson m. , subbarao m.u . , brunner r.j , 1997 , apj , 486 , l11 connolly a.j . , szalay a.s . , brunner r.j . , 1998 , apj , 499 , l125 couch w.j . , sharples r.m . , 1987 , mnras , 229 , 423 couch w.j . , barger a.j . , smail i. , ellis r.s . , sharples r.m . , 1998 , apj , 497 , 188 da costa , l. , et al . , 1998 , a&a , under revision ( astro - ph/9812105 ) de propris r. , pritchet c.j . , harris w.e . , mcclure r.d . , 1995 , apj , 450 , 534 diaferio a. , kauffmann g. , balogh m.l . , white s.d.m . , schade d. , ellingson e. , 2001 , mnras , 323 , 999 dressler a. , gunn j.e . , 1983 , apj , 270 , 7 dressler a. , et al . , 1997 , apj , 490 , 577 dressler a. , smail i. , poggianti b.m . , butcher h. , couch w.j . , ellis r.s . , oemler a. , 1999 , apjs , 122 , 51 driver s.p . , phillipps s. , davies j.i . , morgan i. , disney m.j . , 1994 , mnras , 268 , 393 driver s.p . , couch w.j . , phillipps s. , smith r. , 1998a , mnras , 301 , 357 driver s.p . , couch w.j . , phillipps s. , 1998b , mnras , 301 , 369 ellingson e. , lin h. , yee h.k.c . , carlberg r.g . , 2001 , apj , 547 , 609 evrard a.e . , 1991 , mnras , 248 , 8 faber s.m . , 1977 , in the evolution of galaxies and stellar populations , p. 157 , eds tinsley , b. m. & larson , r. b. , yale university observatory , new haven . fasano g. , poggianti b.m . , couch w.j . , bettoni d. , kjrgaard p. , moles m. , 2000 , apj , 542 , 673 fernndez soto a. , lanzetta k.m . , yahil a. , 1999 , aj , 513 , 34 ferreras i. , silk j. , 2000 , apj , 541 , l37 fontana a. , dodorico s. , poli f. , giallongo e. , arnouts s. , cristiani s. , moorwood a. , saracco p. , 2000 , aj , 120 , 2206 furusawa h. , shimasaku k. , doi m. , okamura s. , 2000 , apj , 534 , 624 garilli b. , maccagni d. , andreon s. , 1999 , a&a , 342 , 408 griffiths r.e . , et al . , 1994 , 1994 , apj , 435 , l19 gunn j.e . , . , 1983 , apjs , 52 , 121 gwyn s. , 1995 , ms thesis , university of victoria gwyn s. , hartwick f.d.a . , 1996 , apj , 468 , 77 henry j.p . , briel u. , gunn j.e . , soltan a. , 1982 , apj , 262 , 1 hogg d.w . , et al . 1998 , aj , 115 , 1418 jarvis j.f . , tyson j.a . , 1981 , aj , 86 , 476 kauffmann g. , 1995 , mnras , 274 , 153 kinney a.l . , calzetti d. , bohlin r.c . , mcquade k. , storchi bergmann t. , schmitt h.r . 1996 , apj , 467 , 38 kodama t. , bell e.f . , bower r.g . , 1999 , mnras , 302 , 152 kodama t. , bower r.g . , 2001 , mnras , 321 , 18 koo d.c . , 1999 , in photometric redshifts and high redshift galaxies , eds . r. weymann , l. storrie lombardi , m. sawicki & r. brunner , san francisco : asp conference series , vol 191 landolt a.u . , 1992 , aj , 104 , 340 lavery r.j . , henry j.p . , 1988 , apj , 330 , 596 lavery r.j . , pierce m.j . , mcclure r.d . , 1992 , aj , 104 , 1067 lin h.,yee h.k.c . , carlberg r.g . , morris s.l . , sawicki m. , patton d.r . , wirth g. , shepherd c.w . , 1999 , apj , 518 , 533 liu c.t . , green r.f . 1998 , aj , 116 , 1074 loveday j. , peterson b.a . , efstathiou g. , maddox s.j . , 1992 , apj , 390 , 338 lubin , l. m. , brunner , r. j. 1999 , in photometric redshifts and high redshift galaxies , eds . r. weymann , l. storrie lombardi , m. sawicki & r. brunner , san francisco : asp conference series , vol 191 lumsden s.l . , nichol r.c . , collins c.a . , guzzo l. , 1992 , mnras , 258 , 1 madau p. , 1995 , apj , 441 , 18 margoniner v.e . , de carvalho r.r . , , djorgovski s.g . , 2001 , apj , 548 , 143 marzke r.o . , huchra j.p . , geller m.j . , 1994,apj , 428,43 metevier a.j . , romer a.k . , ulmer m.j . , 2000 , aj , 119 , 1090 mobasher b. , rowan robinson m. , georgakakis a. , eaton n. 1996 , mnras , 282 , l7 moore b. , lake g. , katz n. , 1998 , apj , 495 , 139 nslund m. , fransson c. , huldtgren m. 2000 , a&a , 356 , 435 oke j.b . , gunn j.e . , hoessel j.g . , 1996 , aj , 111 , 29 paolillo m. , andreon s. , longo g. , puddu e. , gal r.r . , scaaramella r. , djorgovski s.g . , de carvalho r. , 2001 , a&a , 367 , 59 peebles p.j.e . , 1975 , apj , 196 , 647 peebles p.j.e . , 1980 , the large - scale structure of the universe , princeton university press , princeton poggianti b.m . , smail i. , dressler a. , couch w.j . , berger a.j . , butcher h. , ellis r.s . , oemler a. , 1999 , apj , 518 , 576 puschell j.j . , owen f.n . , laing r.a . , 1982 , apj , 275 , 57 rakos k.d . , schombert j.m . , 1995 , apj , 439 , 47 sawicki m.j . , lin h. , yee h.k.c . 1997 , aj , 113 , 1 schechter p. , 1976 , apj , 203 , 297 schlegel d.j . , finkbeiner d.p . , davis m. , 1998 , apj , 500 , 525 smail i. , dressler a. , couch w.j . , ellis r.s . , oemler a. , butcher h. , sharples r.m . , 1997 , apjs , 110 , 213 smail i. , edge a.c . , ellis r.s . , blandford r.d . , 1998 , mnras , 293 , 124 smith r.m . , driver s.p . , phillipps s. , 1997 , mnras , 287 , 415 takeuchi t.t . , yoshikawa k. , ishii t.t . , 2000 , apjs , 129 , 1 trentham n. , 1997 , mnras , 286 , 133 trentham n. , 1998 , mnras , 293 , 71 valdes f. , 1982 , faint object classification and analysis system noao , tucson , az valdes f. , 1993 , focas user s guide . noao , tucson , az valotto c.a . , moore b. , lambas d.g . , 2001 , apj , 546 , 157 van dokkum p.g . , franx m. , kelson d.d . , 1998 , apj , 500 , 714 van dokkum p.g . , franx m. , fabricant d. , illingworth g.d . , kelson d.d . , 2000 , apj , 541 , 95 wang y. , bahcall n. , turner e.l . , 1998 , aj , 116 , 2081 white d.m . , frenk c.s . , 1991 , apj , 379 , 52 whitmore b.c . , gilmore d.m . , jones c. , 1993 , aj , 407 , 489 wilson g. , smail i. , ellis r.s . , couch w.j . , 1997 , mnras , 284 , 915 woods d. , fahlman g.g . , 1997 , apj , 490 , 11

the mechanisms for generation of long - lived spin coherence in a two - dimensional electron gas ( 2deg ) have been studied experimentally by means of a picosecond pump - probe kerr rotation technique . cdte/(cd , mg)te quantum wells with a diluted 2deg were investigated . the strong coulomb interaction between electrons and holes , which results in large binding energies of neutral excitons and negatively charged excitons ( trions ) , allows one to address selectively the exciton or trion states by resonant optical excitation . different scenarios of spin coherence generation were analyzed theoretically , among them the direct trion photocreation , the formation of trions from photogenerated excitons and the electron - exciton exchange scattering . good agreement between experiment and theory is found .

0707.4294

the spin coherence of electronic states is one of the key features involved in numerous concepts for spintronics devices ( see , e.g. , refs . it has been studied in semiconductor structures of different dimensionality , including bulk - like thin films , quantum wells ( qws ) and quantum dots . accordingly , the spin coherence time of an electron has been found to vary over a wide range from a few picoseconds up to a few microseconds . an ensemble of electron spins subject to a magnetic field is commonly characterized by three relaxation times @xcite . the longitudinal spin relaxation time @xmath0 , which is related to the relaxation of the spin component parallel to the field , and the transverse relaxation times @xmath1 and @xmath2 . the @xmath1 time describes spin decoherence ( i.e. relaxation of the spin components transverse to the field ) of a _ single _ electron , while the @xmath2 time describes the dephasing of the whole spin ensemble . the coherence time @xmath1 of a single spin is often few orders of magnitude longer than the dephasing time @xmath2 of a spin ensemble . for example , in ( in , ga)as / gaas quantum dots these times are 3 @xmath3s and 0.4 ns , respectively , at @xmath4 = 6 t @xcite . a @xmath2 of 300 ns has been measured in bulk gaas by means of the hanle effect on optically oriented electrons at liquid helium temperature @xcite . for qws the longest spin dephasing times reported so far are 10 ns for gaas/(al , ga)as @xcite and 30 ns for cdte/(cd , mg)te @xcite . they have been measured for structures with a very diluted two - dimensional electron gas ( 2deg ) . electron spin coherence in cdte/(cd , mg)te qws attracts recently an increasing interest @xcite . figure 1 illustrates three typical situations realized in experiments on coherent spin dynamics under resonant optical excitation of the qws . these cases are different in the density of resident 2d electrons , @xmath5 . in the undoped samples [ panel ( a ) , @xmath6 spin oriented excitons are photogenerated . usually hole spin dephasing occurs within several picoseconds and the coherent response is mainly determined by the spin dynamics of the electron in the exciton . this dynamics can not be studied for time scales exceeding the exciton lifetime as the exciton recombination depopulates the photoexcited states . for the dense 2deg [ panel ( c ) , @xmath7 , where @xmath8 is the exciton bohr radius ] , exciton formation is suppressed because of state - filling and screening effects . after photogeneration a hole looses its spin and energy quickly and recombines with an electron from the fermi see . however , the spin oriented electron photogenerated at the fermi level has an infinite lifetime which allows one to study its long - lived spin coherence and spin relaxation . in this case a circularly polarized photon can increase the spin polarization of the 2deg by @xmath9 . in case of a diluted 2deg [ panel ( b ) , @xmath10 the mechanism for generation of the electron spin coherence is not so obvious . indeed the ground state is a singlet trion with an antiparallel orientation of electron spins . being resonantly excited , this state would not contribute to the spin polarization because the hole undergoes fast decoherence and the total spin of the two electrons is @xmath11 . however , generation of electron spin coherence has been observed experimentally under resonant excitation of trions in qws @xcite and quantum dots @xcite . there are two equivalent approaches to explain the generation mechanism in this case . the first one suggests that a coherent superposition of electron and trion states is excited by a circular polarized photon when the system is subject to an external magnetic field @xcite . the second one considers the 2d electrons captured for trion formation : under circular polarized excitation electrons with a specific spin orientation will be extracted from the 2deg and correspondingly spin polarization with the opposite sign will be induced @xcite . in this paper we report on a detailed study of the coherent spin dynamics of electrons and electron - hole complexes in cdte / cd@xmath12mg@xmath13te qws with a low density 2deg , performed by a pump - probe kerr rotation ( kr ) technique . a theoretical analysis of the various mechanisms of spin coherence generation is carried out . the paper is organized as follows . ii is devoted to ( i ) the theoretical analysis of the physical mechanisms leading to kerr and faraday rotation of the probe pulse polarization , and ( ii ) the theoretical description of various scenarios of electron spin coherence generation . the experimental results are presented in sec . iii , which also contains a comparison with the model considerations of sec . the paper is concluded by sec . iv in which we also comment on the specifics of _ p_-doped qws . below we underline the two main aspects of the developed theory , namely , the nature of kerr rotation ( kr ) and faraday rotation ( fr ) signals and the models of spin dynamics of a 2deg and electron - hole complexes , both trions and excitons . in short , the basic features of an experiment designed for time - resolved measurements of the electron spin coherence can be summarized as follows : the sample containing a 2deg is excited along the structure growth axis ( @xmath14-axis ) by an intense pump pulse which induces resonant interband transitions . then a much weaker , linearly - polarized probe pulse with the frequency either coinciding with or different from the pump frequency arrives at the sample . the rotation of the polarization plane of the reflected or transmitted probe pulse is analyzed as a function of the delay between the pump and probe pulses . an external magnetic field @xmath4 is applied in the qw plane , say , along the @xmath15-axis and leads to precessions of the @xmath14 and @xmath16 electron spin components with the larmor frequency @xmath17 , where @xmath18 is the electron in - plane @xmath19-factor along the @xmath15-direction and @xmath20 is the bohr magneton . for 2d heavy holes bound into excitons or trions , the in - plane @xmath19-factor is very small and can be ignored @xcite . in order to describe the kr in the pump - probe experiment under normal - incidence resonant excitation we first consider the amplitude reflection coefficient of an axially - symmetric single qw , which in the vicinity of the exciton or trion resonance is given by @xcite @xmath21 where @xmath22 is the incident light frequency and @xmath23 , @xmath24 and @xmath25 are the exciton ( trion ) resonance frequency , the radiative and the nonradiative damping rates , respectively . taking into account also the cap layer as constituent of the heterostructure , the total amplitude reflection of the light incident on the structure from vacuum reads @xmath26 here @xmath27 is the reflection coefficient at the boundary between the cap layer and vacuum , @xmath28 is the refractive index of the cap layer which for simplicity is assumed to coincide with the background refractive index of the well material , @xmath29 , @xmath30 is the cap - layer thickness and @xmath31 is the light wave vector in the cap - layer . the above equation is valid for a spin - unpolarized system in which case the coefficient @xmath32 is insensitive to the light polarization . for the spin - polarized resident electrons the reflection coefficient for right ( @xmath33 ) and left ( @xmath34 ) circularly polarized light has the form of eq . ( [ unpolarized ] ) but @xmath35 is replaced by @xmath36 with the parameters @xmath37 and @xmath25 dependent on the light helicity . these parameters are , in general , determined by the concentration and spin polarization of the carriers and their complexes as well as by the delay between the pump and probe pulses . when describing the faraday rotation effect we have to analyze the amplitude transmission coefficient @xmath38 . both the two signals measured in reflection ( kerr rotation ) or transmission ( faraday rotation ) geometry are proportional to the difference @xmath39 where @xmath40 and @xmath41 are the time - integrated intensities of the incident and reflected / transmitted probe pulses in time - resolved experiments ( or the stationary intensities under steady - state photoexcitation ) . let us introduce the symmetrized , @xmath42 , and antisymmetrized , @xmath43 , combinations of @xmath44 and assume @xmath43 to be small . then in first approximation one obtains @xmath45\:.\ ] ] if @xmath42 is also small as compared with @xmath46 the right - hand side of the above equation reduces to @xmath47 . the probe at the trion resonance frequency is described by eqs . ( [ r])([signal ] ) , where @xmath48 is the oscillator strength for resonant trion excitation by @xmath49 circularly polarized light @xcite . for heavy - hole optical transitions the values of @xmath48 are proportional to the density of resident electrons with the spin component @xmath50 , respectively . in the case of a dense 2deg the kerr or faraday rotation angle is proportional to @xcite @xmath51}{e_g^{qw } + \varepsilon - \hbar \omega - { \rm i } \hbar \gamma_{eh } } \right\}\:.\ ] ] here @xmath52 is the interband matrix element of the momentum operator , @xmath53 is the band gap renormalized by the quantum confinement of conduction electrons and heavy holes , @xmath54 is the damping rate for an electron - hole pair , @xmath55 is the free electron mass , @xmath56 is the sum of the 2d electron and hole kinetic energies , @xmath57 ; @xmath58 , @xmath59 and @xmath60 is the reduced , electron and heavy - hole effective mass , respectively , @xmath61 is the energy distribution function for electrons with spin @xmath50 , and @xmath62 is the reduced density of states proportional to @xmath63 . in the following we discuss , one after another , physical mechanisms of the pump - probe signal eq . ( [ signal ] ) for three cases of interest : ( i ) a diluted 2deg subject to resonant circular polarized photoexcitation in the singlet trion state , ( ii ) a diluted 2deg with resonant generation of excitons at low temperatures favoring the binding of excitons and resident electrons into trions , ( iii ) exciton generation in a dense 2deg , and ( iv ) a diluted 2deg with photogeneration of carriers with kinetic energy considerably exceeding the exciton binding energy . we start the analysis from the case of kr by a qw with a diluted 2deg and for resonant trion generation . in this case the radiative homogeneous broadenings @xmath64 ( coinciding with the oscillator strength of the trion ) in eq . are proportional to the concentrations of the resident electrons with spin - down and spin - up , see refs . qualitatively , the physical picture looks as follows . according to the selection rules , the absorption of a circularly polarized photon leads to the formation of an electron - hole pair with fixed spin projections : @xmath65 and @xmath66 for right ( @xmath67 ) and left ( @xmath68 ) circularly polarized photons , respectively . at weak and moderate magnetic fields the ground state of negatively charged trions has a singlet electron configuration with antiparallel orientation of electron spins . thus , for resonant excitation only resident electrons with orientation opposite to the photogenerated electrons can contribute to trion formation . this means that the 2deg becomes depleted of electrons with @xmath14-spin component @xmath69 under @xmath67 pumping and of @xmath70 electrons for @xmath68 pumping . an external magnetic field applied in the plane of the structure leads to precession of the spin polarization of resident electrons and , therefore , to a modulation of @xmath64 and oscillations of the kerr signal . this process is shown schematically in fig . [ fig : trion ] . now we turn to an analytical description of this scenario . the kinetic equations describing the spin dynamics of electrons and trions after resonant , pulsed excitation of trions have the form @xmath71 here @xmath72 is the effective trion spin density with @xmath73 being the densities of negatively - charged trions with the heavy - hole spin @xmath74 , @xmath75 and @xmath76 are the corresponding components of the electron - gas spin density , @xmath77 is the lifetime of the trion spin including the trion lifetime @xmath78 and the spin relaxation time @xmath79 , i.e. , @xmath80 , and @xmath81 is the electron spin relaxation time . it can be identified as the ensemble transverse spin relaxation time @xmath2 . under normal incidence of the @xmath67 polarized pump the initial conditions are @xmath82 with @xmath83 being the initial density of photogenerated trions . remember that the magnetic field is directed along the @xmath15 axis , therefore the @xmath15 component of electron spin density is conserved . . ( b ) @xmath68 polarized photon generates a @xmath66 electron - hole pair , which captures a @xmath84 resident electron to form a trion . the 2deg became polarized due to uncompensated @xmath85 electron spin left . ( c ) during trion lifetime , @xmath78 , the 2deg polarization precesses around the magnetic field . trion state does not precess in magnetic field as on the one hand its electronic configuration is singlet and on the other hand in - plane hole g - factor is zero . ( d ) after trion recombination the @xmath84 electron is returned to the 2deg ( we neglect here spin relaxation of the hole in the trion ) . final state of the 2deg with induced polarization is shown . ] let us introduce the complex function @xmath86 . it satisfies the equation @xmath87 with the initial condition @xmath88 . the solution for @xmath89 is readily written as @xmath90 . substituting @xmath89 into eq . ( [ s+ ] ) one finds @xmath91 \:,\ ] ] where @xmath92 $ ] . the real part of @xmath93 is equal to @xmath94 and given by @xmath95\:,\ ] ] where @xmath96,\ ] ] @xmath97 and @xmath98 are the real and imaginary parts of @xmath99 . in the particular case when electron and trion spin relaxation can be ignored , eq . ( [ sza ] ) is reduced to @xmath100\:,\ ] ] where @xmath101 . in s strong magnetic field such that @xmath102 [ with @xmath103 the parameter @xmath104 and @xmath105 , @xmath106 yielding @xmath107 . the high field asymptotics reads @xmath108 for low magnetic fields @xmath109 so that we have @xmath110 the dependence of the phase @xmath111 on magnetic field has a minimum @xmath112 reached at @xmath113 note that eqs . and are valid provided that @xmath81 is magnetic field independent . in principle , this might not be the case in the experiment , where the main contribution to the ensemble dephasing time arises from the spread of @xmath19-factor values , but for typical conditions @xmath114 so that @xmath115 and is almost field independent . at zero magnetic field , the electron spin exhibits no precession , @xmath116 , and one has @xmath117 \:,\ ] ] where @xmath118 . this can be understood as follows . at the moment right after photoexcitation into the trion resonance by a pulsed , say , @xmath67 polarized light , the system contains @xmath83 singlet trions with completely polarized holes @xmath119 , and @xmath83 electrons with uncompensated spin @xmath84 , because the same number of electrons with spin @xmath85 were extracted from the 2deg to form trions . this stage is illustrated by fig . [ fig : trion ] ( b ) . in the absence of spin relaxation , trions decay by emitting @xmath67 photons and giving back lent electrons with spin @xmath85 . as a result the initially generated electron spin polarization is compensated by the returned electrons and tends to zero as the trions vanish . the compensation takes place for the limiting case @xmath120 and also for the special case of @xmath121 , see solid curve in fig . [ fig : comp : theor ] . if the spin relaxation times @xmath81 and @xmath79 coincide , then @xmath122 and the second contribution to @xmath94 proportional to @xmath123 disappears . in this case the electron polarization @xmath94 decays with the trion spin lifetime @xmath77 . spin relaxation of any subsystem brings in an imbalance , namely , the spins of the left over and the returned electrons can not completely compensate each other , see dashed curve in fig . [ fig : comp : theor ] . the two characteristic parts of the dashed line correspond to the fast and slow negative contributions to @xmath94 in eq . ( [ st0 ] ) governed by @xmath124 ps and @xmath125 ns , respectively . in an in - plane external magnetic field the imbalance arises even if spin relaxation is absent or @xmath121 . indeed , as schematically illustrated in fig . [ fig : trion](b ) , both the heavy - holes ( due to their zero in - plain heavy - hole @xmath19-factor ) and the singlet electron pairs bound in trions are not affected by the magnetic field , whereas the spins of the resident electrons precess . the trion recombines radiatively by emission of a @xmath67 photon and the returned electron is spin - up polarized along the @xmath14 axis [ fig . [ fig : trion](c ) ] . as a result , even after the trions have vanished ( @xmath126 ) , the electron spin polarization is nonzero and it oscillates with frequency @xmath127 . in eq . ( [ sta ] ) , the @xmath99-independent term describes the spin precession of resident electrons left after the pump pulse ends while the terms proportional to @xmath99 take into account the electrons left after the trions have recombined and decayed exponentially . at the moment of trion recombination , these electrons become polarized antiparallel to @xmath14 , their spin is added to the total electron spin and rotates as well with frequency @xmath127 in the @xmath128 plane . . the electron spin relaxation time @xmath125 ns , the trion radiative lifetime @xmath129 ps . dashed curve corresponds to the trion spin relaxation time @xmath130 ps , while solid curve corresponds to an unrealistically high @xmath131 ns . the latter situation corresponds to the compensation of the initial spin polarization by the returned electrons , @xmath122 . ] time - resolved kr can originate both from the electron and the trion spin polarizations . according to eq . ( [ sza ] ) the electron contribution to the kerr signal contains both exponential monotonous and oscillatory components , whereas the trion contribution shows only a monotonous behavior , proportional to @xmath132 . it should be stressed that the initial number of photogenerated trions under resonant excitation , @xmath83 , can not exceed @xmath133 , where @xmath5 is the density of the 2deg . thus , the factor @xmath83 in eq . ( [ sza ] ) increases linearly with the pump intensity for small excitation density and then saturates with density increase at the value @xmath133 . in the simplest model @xmath134 where @xmath135 is the generation rate being proportional to the pump power . thus the initial spin polarization of the 2deg shows a saturation behavior , see fig . [ fig : satur : theor ] which can be related to saturation of trion absorbtion . if the pump photon energy is tuned to the exciton transition then , at low temperates where @xmath136 ( with the boltzmann constant @xmath137 , and the trion binding energy @xmath138 ) , the photogenerated excitons tend to bind into trions as long as they find resident electrons with proper spin orientation . the trion thermal dissociation , on the other hand , can be neglected . experimentally , the exciton contribution to the kerr oscillating signal is determined by the larmor spin precession of the electrons in the excitons . in the pump - probe experiment the correlation between the electron and hole spins via the electron - hole exchange interaction is suppressed , see refs . [ ] . therefore , the spin @xmath139 of an electron in an exciton precesses about an in - plane magnetic field with the same frequency as that of a resident electron . note that if the heavy holes forming excitons are unpolarized the excitons can be labelled by the electron spin @xmath140 . moreover , the trions formed from these excitons are unpolarized as well . at zero magnetic field the rates of an electron and an exciton binding into a trion are given by @xmath141 where @xmath142 are the densities of electrons and excitons with electron spin @xmath50 , and @xmath143 is a constant . in addition to this constant the system is characterized by four times , namely , the exciton and trion radiative lifetimes , @xmath144 and @xmath78 , respectively , and the spin relaxation times of resident electrons ( @xmath81 ) and excitons ( @xmath145 ) . in the presence of an in - plane magnetic field @xmath146 , the spins of the resident electrons and the electrons in excitons precess with the same frequency @xmath127 . in the coordinate frame which rotates around @xmath147 at a rate @xmath127 , one can apply the simple equation ( [ second ] ) to describe the spin - dependent decay of resident electrons and photoexcited excitons . at low excitation intensities satisfying the condition @xmath148 ( here @xmath149 is the number of photogenerated excitons ) the total spin of the electron gas after the decay of all excitons can be estimated as @xmath150 is the total lifetime of the exciton spin , including the radiative decay time , the time of exciton binding into a trion , @xmath151 , and the spin relaxation time @xmath152 . in order to simplify the analysis we assume the time of exciton binding into a trion , @xmath153 , to be shorter than the exciton radiative lifetime @xmath144 and the spin relaxation time of the electron in an exciton @xmath154 . in this case , shortly after the pulsed optical excitation all excitons are bound to trions and the qw structure contains @xmath149 trions and @xmath155 resident electrons with a total precessing spin latexmath:[\[\label{se_low } @xmath149 spins of resident electrons contribute to the kerr rotation oscillations . at higher excitation intensity , @xmath157 , and the spin - polarized excitons extract almost at once @xmath133 electrons to form trions . therefore , in absence of electron - in - exciton spin relaxation processes the trion density can not exceed @xmath133 , thus the total spin density of the electron gas is limited by @xmath158 . the electron - in - exciton spin relaxation allows to convert the remaining @xmath159 excitons in trions . obviously , the maximum number of formed trions can not exceed the concentration of background electrons , @xmath5 . the total spin of resident electrons after the excitons and trions have recombined can be estimated as ( provided that the holes are unpolarized ) @xmath160 this equation is valid both for @xmath161 and @xmath162 when @xmath163 , otherwise eq . holds . at @xmath164 , the values of @xmath165 given by eq . ( [ se_low ] ) and eq . ( [ se_exc ] ) coincide and are equal to @xmath158 . when deriving the above equation we have neglected the spin relaxation of the resident electrons assuming @xmath166 . in experiment the exciton radiative lifetime , @xmath144 , may be comparable with the trion formation time , @xmath167 . in this case the above estimation should be taken as a qualitative result predicting a non - monotonous dependence of the kerr signal amplitude on pump intensity . this non - monotonous behavior is illustrated in fig . [ fig : excexc : theor ] . an initial linear growth of @xmath168 followed by a linear decrease the decrease of initial electron spin as function of pump intensity is steeper for smaller values of @xmath169 , i.e. for shorter hole spin relaxation times . it is worth to stress that in this regime the electron spin polarization vanishes at very high pumping whereas , under resonant trion excitation , @xmath165 monotonously increases with increasing pump power and saturates at @xmath158 . , @xmath170 and @xmath171 , respectively . @xmath172 @xmath173 . the cross - over density @xmath174 is shown by the arrow . ] turning to the detection aspects , we find that selective addressing of the exciton resonance also results in temporal oscillations of the probe - pulse kerr rotation . the modulation comes from the photoinduced difference in the resonance frequencies @xmath175 and/or the non - radiative damping rates @xmath176 . both @xmath177 and @xmath178 become nonzero taking into account the exchange interaction between an electron in an exciton and the resident electrons of the 2deg : the first difference is related to the hartree - fock renormalization of the electron energy in the spin - polarized electron gas , and the second one is related to the spin dependence of the electron - exciton scattering @xcite . as a result , rotation of the total spin of the electron ensemble leads to a modulation of the exciton resonance frequency and non - radiative broadening and , thus , to oscillations of the kr angle . we note that an in - plane magnetic field results also in spin precession of the electron in an exciton and the total kerr signal will be a superposition of 2deg and exciton signals . the situation for the probe tuned to the trion resonance is qualitatively the same . the kr amplitude will contain components arising due to spin precession of the 2deg and to the electron - in - exciton spin precession . however , it is expected that detection at trion resonance will be less sensitive to the exciton spin dynamics as compared to detection at the exciton resonance , see sec . we note here that the amplitude of the kr signal induced by the same number of coherent electron spins will be different for detection at the trion or the exciton energies . the difference comes from the different oscillator strengths of these resonances @xcite and from the different efficiencies of signal modulation . an increase of the 2deg density and its fermi level leads to dissociation of trions due to state - filling and screening effects . we consider an intermediate situation where the trions are suppressed , @xmath179 , but the excitons are not , @xmath180 , where @xmath8 is the exciton bohr radius and @xmath181 is the characteristic trion radius . therefore , the scenario of spin coherence generation for electrons involves two subsystems , the 2deg and the spin - polarized excitons resonantly excited by the circular polarized optical pulse . the electrons being initially unpolarized can gain spin polarization due to the electron - electron exchange interaction in electron - exciton flip - flop scattering processes . at zero magnetic field , the flip - flop scattering rates for the resident electrons and those bound in excitons are given by @xmath182 where @xmath183 is a constant different from @xmath143 in eq . ( [ second ] ) . the total number of electrons , @xmath184 , is fixed while the exciton density @xmath185 decays to zero . similarly to the scenario in sec . iii.c , we assume that the holes in the excitons are unpolarized and the spins of both the resident and bound electrons precess with the same frequency . the kerr rotation signal is a result of the difference in shift between the fermi levels for spin - up and spin - down electrons and the concomitant renormalization of the resonance frequencies @xmath175 due to the exchange interaction between the spin - polarized carriers . at weak pump intensities so that @xmath186 and for negligible resident - electron spin relaxation , @xmath187 , one can use , in agreement with eq . , the following estimate for the amount of resident - electron spins oriented by excitons : @xmath188 . with increasing pump intensity the electron spin saturates at @xmath189 $ ] when all 2deg electrons have become spin - polarized . it should be noted that a similar description can be applied for resonant photoexcitation of excitons in the high - temperature regime where @xmath190 , but @xmath191 , so that the trion states are thermally unstable but the exciton bound states are still stable . in the case of non - resonant pumping , the photogeneration of free electrons and holes is followed by their separate , uncorrelated energy relaxation toward the bottoms of the corresponding bands . the energy relaxation is accompanied by carrier spin relaxation . the holes have lost their spin after few scattering events . the total spin of the electron ensemble after energy relaxation can be estimated as @xmath192 where @xmath193 is the number of photogenerated electrons , @xmath194 is the energy relaxation time and @xmath195 is the spin relaxation time of hot electrons . we note that @xmath195 can be much shorter than the spin coherence time @xmath81 of the electron gas in quasi - equilibrium , entering eq . . after the particles have reached the band bottoms they can bind to form excitons and trions . in the diluted electron gas and for moderate pumping densities , the trion formation is more preferable and the subsequent spin dynamics can be described by the model in sec . ii.b . at very strong pumping , when the number of photogenerated electrons exceeds that of the resident electrons , or for a dense 2deg , for which the trions are suppressed , formation of excitons takes place . the spin dynamics in this case can be described by the scenario in sec . ii.d . we have developed a comprehensive model to describe the kerr rotation signal in qws with varying 2deg densities and for different excitation regimes . the signal dynamics reveals spin oscillations of the resident electrons and/or of the electrons forming excitons . the pump power dependence of the kerr rotation amplitude is as follows : at low pumping the amplitude grows linearly with the pump power , whereas in the limit of high pumping the amplitude behavior strongly depends on the excitation energy . for trion resonant excitation it saturates ( as the number of generated trions is limited by half of the resident electron concentration ) , while for resonant excitation of excitons ( at @xmath196 and in the diluted 2deg ) the amplitude exhibits a non - monotonic behavior and eventually vanishes . a theoretical model of excitation of electron spin polarization by short laser pulses tuned to the trion resonance has been also proposed in refs . their approach is based on analysis of the temporal dynamics of the coherent quantum beats between an electron and a trion localized by a quantum dot or by well width fluctuations in a qw plane . our approach and the approach in refs . [ ] are essentially equivalent . indeed , let us consider a localized unpolarized electron . its spin density matrix is diagonal with @xmath197 . after optical excitation with a @xmath198 circular - polarized pulse a superposition state of the localized electron and a trion is formed . after the trion radiative decay the electron remains , but its density matrix is different from the initial one . for example , in the limiting case where the hole spin relaxation time is much shorter than the trion radiative lifetime which , in turn , is shorter than the electron spin relaxation time , the electron spin density matrix after trion recombination remains diagonal with @xmath199 here @xmath200 is a complex coefficient which depends on the optical pulse parameters . the resident electron acquires spin - down polarization under pulsed @xmath67 photoexcitation and a train of such pulses leads to complete spin polarization . the same single electron spin density matrix , eq . , describes an ensemble where the number of spin - up electrons is smaller than that of spin - down electrons . the constant @xmath200 in this case has a transparent physical sense : it characterizes the efficiency of singlet trion formation under polarized excitation of the unpolarized ensemble . one can readily check that this result is equivalent to eq . taken at @xmath201 . the studied cdte / cd@xmath12mg@xmath13te qw heterostructure was grown by molecular - bean epitaxy on an @xmath202-oriented gaas substrate followed by a 2 @xmath3 m cdte buffer layer . it has 5 periods , each of them consisting of a 110 nm thick cd@xmath12mg@xmath13te barrier and a 20 nm thick cdte qw . an additional 110 nm thick barrier was grown on top of this layer sequence to reduce the contribution of surface charges . the barriers contain 15 nm layers doped by in donors . undoped 15 nm spacers separate the modulation - doped layers from the qws . electrons from the barrier donors , being collected into qws , provide there a 2deg with a low density of about @xmath203 @xmath173 . we have observed very similar experimental results for structures with single cdte / cd@xmath204mg@xmath205te qws of 12 nm and 8 nm widths , respectively . this confirms the general character of the reported data . a time - resolved pump - probe kerr rotation technique was used to study the coherent spin dynamics of the electrons @xcite . we used a ti : sapphire laser generating 1.5 ps pulses at a repetition frequency of 75.6 mhz . the laser beam was split in pump and probe beams and the time delay between the pump and probe pulses was varied by a mechanical delay line . the pump beam was circular polarized by means of an elasto - optical modulator operated at 50 khz . the probe beam was linearly polarized , and rotation of the polarization plane was measured by a balanced photodetector . the time - resolved kerr rotation signal allows to follow the evolution of the spin coherence of carriers and their complexes generated by the pump pulses . from an analysis of the decay of the kerr rotation amplitude the spin dephasing time of the electron ensemble @xmath2 can be extracted . the details of this analysis through which the mechanisms of spin dephasing of the 2deg can be understood have been reported elsewhere @xcite . the kerr rotation technique has been used in two regimes . for degenerate kerr rotation the pump and probe beams have the same photon energy , as they originate from the same laser . this regime will be denoted as one - color experiment here . we performed , however , also a two - color experiment where the pump and probe energy can be tuned independently . for that purpose two synchronized ti : sapphire lasers were used . experiments were performed in magnetic fields ( 0 - 7 t ) applied in the plane of the structure , i.e. in the voigt geometry . the sample temperature was tuned in a range 1.9 - 100 k. further , time - resolved photoluminescence was used to study recombination dynamics of excitons and trions . the same laser as described above was used for excitation , and time resolved emission spectra have been recorded by a synchro - scan streak camera connected to a 0.5 meter spectrometer . the time resolution in this experiment was about 5 ps . mg@xmath13te qw measured under nonresonant cw excitation with a photon energy of 2.33 ev . the exciton ( x ) and trion ( t ) resonances are separated by 2 mev , which is the trion binding energy . ( b ) reflectivity spectrum of the same structure . the oscillator strength of the exciton resonance is ten times larger than the trion one . ( c ) kinetics of the photoluminescence measured by a streak camera under 1.5 ps pulsed excitation resonant to the trion energy ( curve 2 ) and detuned from it by 0.8 mev and 27 mev to higher energies . the dashed line shows the laser pulse . ] a typical photoluminescence ( pl ) spectrum of the qw measured at a temperature @xmath206 k is shown in fig . [ fig : a2 ] ( a ) . it shows the exciton and trion recombination lines separated by 2 mev , corresponding to the trion binding energy . the full width at half maximum of the exciton line is about 0.5 mev and is mainly due to exciton localization on the qw width fluctuations . the reflectivity spectrum of the same qw in the energy range of the exciton and trion resonances is given in fig . [ fig : a2 ] ( b ) . following the procedure described in ref . [ ] we have evaluated the oscillator strengths of the resonances and found that the exciton oscillator strength is ten times larger than that of the trion resonance . this fact should be taken into account when the intensity of the kerr rotation signals measured at the exciton and trion energies are compared : firstly , the probe response is proportional to the oscillator strength and , secondly , the number of photogenerated carriers is proportional to it . for interpretation of the results of the spin dynamics experiments , information on the recombination dynamics of excitons and trions under resonant excitation is necessary . we performed corresponding measurements under linearly polarized excitation using the streak camera for detection . the results for pumping into the exciton and trion resonances are very similar to each other . the typical recombination kinetics for the trion is given in fig . [ fig : a2 ] ( c ) . for resonant excitation at the trion energy ( curve 2 , detuning @xmath207 ) about 80@xmath208 of the pl intensity decays with a time of 30 ps and the rest decays with a time of 100 ps . when the excitation energy is detuned by 0.8 mev above the trion resonance a redistribution of the two exponential decays with 30 and 100 ps time occurs in favor of the longer decay component . such a behavior is typical for qw emission @xcite . the shorter decay time , 30 ps , can be attributed to the radiative recombination time of trions and excitons generated in the radiative cone where their wave vectors is transferred to the photons . for excitons scattered out of the radiative cone , a longer time of about 100 ps is required to be returned to the cone via emission of acoustical phonons . the exciton luminescence lifetime is slowed down to 100 ps in this case . the recombination of trions is not restricted to the radiative cone conditions as the electron left in the system can take the finite momentum to satisfy the wave vector conservation . therefore we may expect the fast decay of trion pl of about 30 ps even for nonresonant photoexcitation . in most cases , however , for these experimental conditions the trions are formed from photogenerated excitons , which dominate the absorption due to their larger oscillator strength . as a result the decay of the trion pl in this case is determined not by trion recombination but by trion formation , and will monitor the exciton lifetime . for small detuning exemplified by curve 3 , the fast 30 ps process coexists with the longer ( 100 ps ) one . when the excitation energy is tuned to the band - to - band absorption ( curve 4 , @xmath209 mev ) , the pl decay is extended to 250 ps as additional time is required for the free carriers to be bound to excitons . note that the exciton binding energy in the studied qw is @xmath210 mev . below we present the experimental results for the two - color and the degenerate pump - probe experiment . further , we show the experimental data for the pump power , temperature and magnetic field dependencies of the kerr rotation signals . the two - color kerr rotation technique enables independent tuning of the energies of the pump and probe beams . this allows to perform experiments with either constant excitation or detection conditions , which simplifies identification of the studied relaxation processes . figure [ fig : a3 ] shows kerr rotation signals detected at the trion and exciton resonances for three pump energies : ( a ) resonant with the trion , ( b ) resonant with the exciton , and ( c ) non - resonantly excited 72 mev above the exciton energy . the sample is subject to an in - plane magnetic field @xmath211 t. all signals show damped oscillations with a frequency of 23 ghz , which is the larmor precession frequency of the electron spins . this frequency corresponds to the zeeman splitting of the conduction band electrons with a @xmath19 factor value of 1.64 which is in good agreement with literature data for the electron @xmath19 factor in cdte/(cd , mg)te qws @xcite . another common feature of all signals shown in fig . [ fig : a3 ] is the appearance of long - living spin beats which are observed beyond delays of 2.7 ns . the typical recombination times of excitons and trions do not exceed 30 - 100 ps for resonant and quasi - resonant excitation and 250 ps for nonresonant excitation [ see fig . [ fig : a2 ] ( c ) and sec . a. ] , therefore we identify the long - living signal with the coherent spin precession of the resident electrons in the 2deg . one can see that this coherence is excited efficiently for all pump energies and can be detected by probing both the trion and exciton resonances . k , @xmath211 t and various pump excitation energies : ( a ) resonant with trion at 1.5982 ev , ( b ) resonant with exciton at 1.6005 ev , and ( c ) nonresonant at 1.6718 ev , which is 72 mev above the exciton resonance . pump density 56 w@xmath212@xmath213 and probe density 8 w@xmath212@xmath213 . ] some of the signals shown in fig . [ fig : a3 ] contain a short - living part right after the pump pulse with a typical decay time of 50 - 70 ps . this part is especially pronounced for the `` pump x / probe x '' condition ( i.e. pump and probe are degenerate with the exciton resonance ) , see panel ( b ) . this fast component is related to the exciton contribution to the kerr rotation signal , see sec . c. and discussion below . to extract the times and relative amplitudes of the short and long - living components in the spin beat signals each trace has been fitted with a bi - exponential decay function : @xmath214 where @xmath215 and @xmath4 are constants describing the amplitudes of the fast ( @xmath216 ) and slow ( @xmath217 ) components , respectively , @xmath22 is the larmor frequency which is taken to be the same for both components , and @xmath218 is the initial phase . the parameters extracted from these fits are collected in table 1 . all signals , except the one for `` pump t / probe t '' , are symmetric with respect to the abscissa . the `` pump t / probe t '' signal shows an initial relaxation of the centre - of - gravity of the electron beats with a time constant of about 75 ps , which can be attributed to the spin relaxation time of the hole in the trions , see the second term in eq . and ref . @xcite . a detailed analysis of hole spin coherence in cdte/(cd , mg)te qws will be reported elsewhere . .decay times @xmath219 and amplitude ratios @xmath220 extracted from bi - exponential fits to the experimental data using eq . in fig . [ fig : a3 ] . @xmath211 t and @xmath221 k. [ cols="^,^,^,^",options="header " , ] the decay times and relative amplitudes in table 1 are given for @xmath211 t and @xmath221 k. it is important to note that these parameters depend strongly on pump intensity , magnetic field strength and lattice temperature . these dependencies and the underlying physical mechanisms will be discussed in detail below . here we focus our attention on the pump energy dependence of these parameters . we first concentrate on the relative amplitudes as they are a key to understanding the generation of 2deg spin coherence and the role of the trions in this process . for the `` pump t / probe t '' regime only the long - living electron signal of a 2deg is observed [ see panel ( a ) of fig . [ fig : a3 ] ] . this is in line with the model expectations as in this case only trions are photogenerated . as these trions are in the singlet ground state with antiparallel electron spins , they do not contribute to the kr signal . moving the pump energy in resonance with the exciton and further to a band - band transition leads to appearance of the fast decaying component for the signal probed at the trion energy . this can be attributed to the spin dynamics of the exciton which is excited either resonantly or non - resonantly , see sec . ii c , e. the probe has a finite spectral width ( about 1 mev ) and tuned to the trion resonance slightly overlaps with the exciton resonance . the shortening of the electron spin dephasing time from 5.7 down to 3.5 - 3.6 ns is attributed to heating of the 2deg by photocarriers with excess kinetic energy ( for a detailed discussion see sections iii.d and iii.e ) . the kerr rotation signal probed at the exciton energy has two contributions : ( i ) given by the coherent precession of the electrons in excitons and ( ii ) given by the spin precession of the 2deg . the former decays with the exciton recombination time . this fast exciton component is clearly seen in panels ( b , c ) and is absent for pumping at the trion resonance . for nonresonant excitation its relative amplitude does not exceed 50% . in this case electrons and holes are photogenerated 72 mev above the exciton resonance , therefore they have a high probability to scatter and relax separately to the bottoms of their bands , where they are bound into trions and excitons . the relative amplitudes of the fast and long - lived signals reflect the probability of trion and exciton formation . it may be expected that trion formation is preferable because the 2deg density exceeds by at least an order of magnitude the concentration of photocarriers which is in line with the experimental findings . a very different ratio of the relative amplitudes , 90% for the fast decay and 10% for the long - living dynamics , is seen for the signal when resonantly pumped at the exciton energy and detected at the exciton [ see panel ( b ) in fig . [ fig : a3 ] ] . there are at least two factors which favor an exciton population in comparison with a trion one under resonant pumping of the excitons . first , the photogeneration leads to formation of excitons with very low kinetic energy , therefore they remain within the radiative cone and quickly recombine ( during 30 - 50 ps ) prior to becoming bound to trions @xcite . second , a part of excitons is localized , so that they are not mobile and can not reach the sites in the qw where the background electrons are localized . consequently , the formation of trions out of this exciton reservoir is suppressed . moreover , the ratio between the contributions of the excitons and the 2deg to the kr signal is spectrally dependent : detection at the exciton resonance is more sensitive to the spin precession of the electron in the exciton , see sec . ii . c. the results of the two - color experiments presented in fig . [ fig : a3 ] are in good agreement with the model expectations , namely : ( i ) the signal oscillating with the electron larmor frequency is contributed by resident electrons of the 2deg and by electrons precessing in the excitons , ( ii ) trion formation resulting either from resonant photoexcitation of the trions or from capture of excitons or free carriers is a very efficient mechanism for spin coherence generation in a diluted 2deg . the degenerate pump - probe technique is simpler in technical realization and therefore is a more common method to address spin coherence . in this case the pump and probe pulses are generated by the same laser beam without additional spectral selection . examples of degenerate kerr signals can be found already in fig . [ fig : a3 ] , for example the `` pump t / probe t '' and `` pump x / probe x '' traces measured for coinciding energies of the two lasers . k , pump density is @xmath222 w@xmath212@xmath213 and probe density is @xmath223 w@xmath212@xmath213 . ] the degenerate pump - probe signal presented in fig . [ fig : exp2 ] was measured with the use of one laser , as will be the case for the most experiments presented in the rest of this paper . it is consistent with spin dynamics studied by two - color technique . namely , only long - lived spin beats of resident electrons are observed when the laser is resonant with the trion and two component decay is seen for the laser hitting exciton . the experimental conditions were modified as compared with fig . [ fig : a3 ] in order to achieve longer electron spin coherence times . namely , the pump density was reduced to @xmath222 w@xmath212@xmath213 and a weaker magnetic field @xmath224 t was applied @xcite . the delay time range between pump and probe in fig . [ fig : exp2 ] covers 6 ns . under these conditions the spin dephasing time of the resident electrons reaches 13.7 ns for pumping in the trion resonance and 4.2 ns for pumping in the exciton resonance . moreover the spin coherence does not fully decay during the time interval of 13.2 ns between the pump pulses as is clearly seen by the beats at negative delays in fig . [ fig : exp2 ] . in this part we investigate the modifications of kerr rotation signal occurring when the pump density is increased in a wide range of pump powers @xmath225 from 1 to 320 w@xmath212@xmath213 . degenerate pump - probe resonant either with the exciton or the trion transition energies is used . k. for clarity of presentation up - scaling factors have been used for delays exceeding 0.5 ns . also the initial parts of the signals for pump densities exceeding 20 w@xmath212@xmath213 have been shifted in time to negative delays . for all signals a coherent signal at zero delay formed due to the temporary overlap of the probe with the pump has been removed . ] results for excitation at the exciton transition energy are collected in fig . [ fig : exp3 ] . with increase of the pump density a very different behavior is observed for the fast - decaying exciton component and the long - living 2deg component . one can clearly see that the exciton part is rather weak at low powers , evidencing the high probability for an exciton to become bound to a resident electron and form a trion . at higher excitation power the fast - decaying component becomes more pronounced and even dominates in the power range @xmath226 w@xmath212@xmath213 where the concentration of photogenerated excitons exceeds the 2deg density . moreover the absolute amplitude of the 2deg signal at longer delays is decreased for pump powers exceeding 20 w@xmath212@xmath213 . this result will be discussed below together with the data plotted in fig . [ fig : exp3 ] . ( circles ) and the time of the fast decay @xmath216 ( triangles ) related to the exciton recombination dynamics . @xmath221 k. ] relaxation times obtained from the fit with eq.([st100 ] ) are given in fig . [ fig : exp3 t ] . the fast decaying component falls in the range of 25 - 40 ps and is independent of the pump density . it is clearly determined by the radiative recombination of excitons resonantly excited in the light cone . the decay time of the long - living component related to the spin dephasing of the 2deg decreases from 2.1 ns at 15 w@xmath212@xmath213 down to 1.4 ns at 240 w@xmath212@xmath213 . a possible reason for this behavior can be the heating of the resident electrons by the photoexcitation , leading to a reduction of @xmath2 . k. ] kerr rotation signals for resonant pumping into the trion state are shown in fig . [ fig : exp4 ] . their shape does not change markedly for varying pump densities . only the long - living electron spin dephasing is observed , which becomes shorter for higher pump densities . the corresponding dephasing times are plotted in fig . [ fig : exp4 t ] . two characteristic regions are seen in fig . [ fig : exp4 t ] : a strong decrease from 14 ns down to 4 ns at low densities , and a much slower decrease for pump densities exceeding 100 w@xmath212@xmath213 . the decrease of the dephasing time can be understood in terms of delocalization of the resident electrons caused by their heating due to interaction with photogenerated carriers . thus , the electron localization in the studied samples favors longer spin dephasing times . the change of the behavior of spin dephasing times at @xmath227 w@xmath212@xmath213 can be attributed to saturation of the trion absorption : a further increase of the pumping intensity does not change strongly the number of photogenerated carriers . . @xmath221 k. ] and [ fig : exp4 ] . results for the pump resonant with the excitons and trions are shown , respectively . the amplitude of the fast decaying component evaluated for zero delay and multiplied by a factor 0.15 is given by the triangles . the amplitudes of the long - lived 2deg coherence measured at a delay of 0.5 ns are shown by open and closed circles for trions and excitons , respectively . the inset highlights the low density regime . @xmath221 k. ] measured at a delay of 0.5 ns under resonant pumping of the excitons ( closed circles ) and the trions ( open circles ) . model calculations are shown by the lines ( the dashed curve has been calculated for trion resonant excitation according to eq . . the solid lines have been calculated for excitation at the exciton frequency according to eqs . , . in the latter case , the fitting parameter was the ratio of @xmath228 which is found to be 10 . ] kerr rotation amplitudes of the 2deg measured at the trion and exciton resonances as functions of pump density at @xmath211 t are shown by the circles in fig . [ fig : exp5 ] . the signal has been measured at a delay of 0.5 ns , when the contribution of the fast decaying component is vanishingly small . at low excitation density both dependencies show to a good approximation a linear behavior as is illustrated by the inset . at higher excitation density the kerr rotation amplitudes demonstrate pronounced non - linear behaviors : for the `` pump / probe t '' configuration saturation is observed while for the `` pump / probe x '' configuration the kr amplitudes decreases with increase of the pump power . both these results are in agreement with the theoretical predictions of secs . [ trion ] and [ exciton ] , respectively , as illustrated by the figs . [ fig : satur : theor ] and [ fig : excexc : theor ] . the amplitude of the fast decaying component related to the electron - in - exciton precession in given in fig . [ fig : exp5 ] by triangles . note the scaling factor 0.15 , which shows that this component dominates over other signals at moderate and high pump densities ( see also fig . [ fig : exp3 ] ) . in order to make a quantitative comparison we replot the experimental points in fig . [ fig : exp5 ] using a different vertical scale , see fig . [ fig : exp5 m ] . namely , we normalize the kerr signals to their maximum values ( which , according to our theoretical predictions , correspond to the electron spin density being equal to @xmath158 ) . this allows us to compare the efficiency of spin coherence generation . one can see that in the low pumping regime the spin coherence generation efficiency per absorbed photon is practically the same for the laser tuned either to the exciton or the trion resonance . this is in a good agreement with our theory : each absorbed photon creates a trion either directly or via an intermediate excitonic state , thus an electron with a given spin orientation is removed from the 2deg . the strong pumping regime is different for exciton and trion excitation . in the case of the laser tuned to the trion resonance the spin of the 2deg saturates ( the small decrease of the kerr rotation amplitude can be attributed to heating of the 2deg ) while for the laser tuned to the exciton resonance a strong decrease of the spin coherence generation efficiency is seen . the curves in fig . [ fig : exp5 m ] are the results of theoretical calculations based on the models outlined in secs . [ trion ] and [ exciton ] . the dashed curve corresponds to trion resonant excitation , while the solid line is for exciton resonant excitation . for the dashed curve the only fitting parameter was the saturation level . for the solid line the only fitting parameter was the ratio between the electron - in - exciton spin relaxation time and the exciton radiative lifetime , @xmath169 . the best fit corresponds to a @xmath229 . the electron spin coherence in cdte/(cd , mg)te qws is robust against temperature increases and can be clearly traced up to 100 k. kerr rotation signals measured at the trion and exciton resonances in a temperature range from 1.9 up to 100 k are presented in fig . [ fig : exp6 ] . the signals are normalized to their values at zero time delay in order to highlight the trends in signal decay with increasing delay . one can see , that the decay of the spin beats is thermally accelerated both for resonant pumping in the trion and in the exciton resonance , but in the former case the decrease is slower . the decay times are shown in fig . [ fig : exp6 t ] . the kerr rotation signals for the `` pump / probe trion '' configuration were fitted by a single exponential decay , and the resulting decay times are given in the figure by the open circles . a relatively long @xmath2 time of 440 ps is measured at @xmath230 k at the trion resonance . the signals for the `` pump / probe exciton '' configuration were fitted by double exponential decays , see eq.([st100 ] ) , for temperatures below 60 k. above @xmath231 k only the fast component with a decay time in the @xmath232 ps range is apparent . this time can be assigned to exciton recombination and we may conclude that for high - temperatures the exciton spin coherence dominates over the 2deg signal . this can be explained by a reduction of the trion formation from excitons when the 2deg electrons have elevated kinetic energies . w@xmath212@xmath213 and probe density @xmath170 w@xmath212@xmath213 . ] . the data evaluated from a single exponential fit to the `` pump / probe trion '' data are shown by the open circles and a line as a guide to the eye . the data for the `` pump / probe exciton '' configuration are given by the closed symbols : by the circles for @xmath217 and by the triangles for @xmath216 . @xmath233 t , pump density @xmath234 w@xmath212@xmath213 and probe density @xmath170 w@xmath212@xmath213 . ] we turn now to the analysis of the kerr rotation amplitude . its temperature dependence for trion resonant pumping is plotted in fig . [ fig : exp7 ] . the amplitude of the first maxima after the pump pulse , which closely coincides with the amplitude obtained by extrapolation to zero delay is plotted . in the main panel the kerr rotation amplitude is shown as a function of temperature on a linear scale . the signal rapidly looses about @xmath235 of its intensity by a temperature increase from 1.9 up to 20 k and then gradually decreases in intensity up to 100 k , which is the maximum temperature at which signal could be recorded in experiment . the insert shows the data in a form which allows extraction of characteristic activation energies in the temperature dependence . in the temperature range from 12 to 45 k we obtain an activation energy of @xmath236 mev , which equals to the trion binding energy . the strong temperature dependence of the measured signals can be explained by two mechanisms : ( i ) delocalization of electrons and ( ii ) dissociation of trions . at very low temperatures @xmath237 mev , both electrons and photogenerated trions are localized in the quantum well plane by alloy fluctuations and/or interface roughnesses . the kerr rotation signal at fixed pumping is proportional to the number of photogenerated trions . increase of the temperature up to @xmath238 mev is accompanied by electron delocalization which in turn leads to a decrease of the interaction between photogenerated electron - hole pairs and electrons . thus , the number of photogenerated trions decreases . further , the increase of the temperature leads to thermal dissociation of trions , whose number is proportional to @xmath239 , where @xmath240 mev . thus , the trion contribution to the kr signal becomes weaker . at temperatures strongly exceeding @xmath241 the main channel for spin coherence formation is due to exciton - electron spin - flip scattering , see sec . d. ( a ) . the inset shows a logarithmic plot of the amplitude on inverse temperature . the line corresponds to an activation energy of @xmath236 mev . ] the total spin of the electron ensemble precesses around the external magnetic field . the decay of the total spin with increasing delay is governed by the spin dephasing processes . the spins of the electrons forming a trion are in a singlet state and therefore do not precess around the magnetic field as the resident electrons do . recombination of trions leads to return of the partially @xmath14-polarized electrons to the 2deg . their spin orientation differs from that of the precessing electrons , which results in a shift of the initial phase of the measured kr signal , see eq . . this effect is illustrated in fig . [ fig : aa2 ] , where the full arrow shows the total spin of the 2deg , while the open arrow shows the spin of the returned electrons . one can see that after trion recombination the total spin of the electron gas has been rotated by a larger angle as compared with the rotation of the resident electrons spin . this induces a phase - shift of the oscillating kerr signal . we have analyzed the initial phase by extrapolation of the spin beats at longer delays back toward zero delay . the experiments were performed at low excitation density with pump / probe at the trion resonance . the results are given in fig . [ fig : exp8 ] . a pronounced minimum is seen at @xmath242 t in qualitative agreement with the model calculations shown by the solid line , which was calculated with a trion lifetime @xmath243 ps . we also assumed an electron spin dephasing time @xmath244 ns ( which corresponds to the experimental value at @xmath245 t ) . in experiment this time is longer at smaller fields . however , the calculated curve is insensitive to the choice of this value as far as @xmath81 remains the longest time in the system . the only free parameter in the calculations was the spin relaxation time of the hole in a trion @xmath246 . the two curves correspond to @xmath247 ps ( dashed ) and @xmath248 ps ( solid ) . it is seen , that in accordance with eqs . and , the depth of the mininum and its field position is controlled by @xmath246 . the model results demonstrate qualitative agreement with the experimental data . w@xmath212@xmath213 and probe density @xmath249 w@xmath212@xmath213 . the model calculations have been performed according to eq.([phasenew ] ) . ] as we have already noted above , the studied sample contains a diluted 2deg with the fermi energy smaller than the typical localization potential caused by well width fluctuations . as a result , at low temperatures of both the lattice and the electron gas , a major fraction of electrons is localized . we have concluded also from the pump density and temperature dependencies that the dephasing time @xmath2 is strongly sensitive to electron localization , see figs . [ fig : exp4 t ] and [ fig : exp6 t ] . @xmath2 can be extremely long for the localized electrons and shortens with electron delocalization . to have a deeper insight into the effects of electron localization on the spin coherence we have done two - color pump - probe measurements in the regime where the longest dephasing times have been achieved , i.e. using a low pump density in a weak magnetic field of 0.25 t. figure [ fig : exp9 ] shows the results of such an experiment , where the pump beam is resonant with the exciton transition and the kerr rotation signal is detected at either the trion or the exciton energy . under these experimental conditions the fast decaying component is very small for probing at the exciton energy . the excitation conditions are identical for both signals in fig . [ fig : exp9 ] , therefore one can expect to detect the same dephasing times for the 2deg , irrespective of the probe energy . therefore , it is rather surprising to observe that the dephasing times of the electron coherence differ by a factor of two : @xmath250 and 5.3 ns for probing at the trion and the exciton resonance , respectively . k. pump density @xmath170 w@xmath212@xmath213 and probe density @xmath251 w@xmath212@xmath213 . ] to explain this difference we suggest that different fractions of the resident electrons contribute to the kerr rotation signal measured at the trion or the exciton energies . this can be understood if we take into account that different mechanisms lead to kerr rotation signal at the exciton and trion energy . for detection at the trion resonance the effect is contributed by the variation of the trion oscillator strength , which is directly proportional to the concentration of electrons with a specific spin orientation , see sec . a. as the in - plain localization is important to achieve stability of the trions in qws one may expect that the trions are much more dependent on the localized electrons , which have a longer dephasing time . the kerr rotation signal at free exciton energy monitors 2deg spin beats mostly due to the spin - dependent exciton - electron scattering , see sec . ii . c. and ref . possibility for this scattering to occur implies existence of free electrons . therefore , at the exciton energy we address free or quasi - free resident electrons which have shorter dephasing times . we have demonstrated experimentally the possibility to excite spin coherence of a 2deg by resonant pumping into the trion or the exciton states , as well as by non - resonant excitation . it was shown that the time - resolved kerr rotation signal detected experimentally at the trion and exciton frequencies contains two components : a fast contribution ( which vanishes @xmath252 ps after the pump pulse ) and a long - living one ( with typical decay times of the order of nanoseconds ) . the fast component is related with electrons - in - excitons , while the long one is due to the resident electrons of the 2deg . experimentally we can clearly separate these contributions . a theoretical model has been developed to describe various scenarios of spin coherence generation in qws with a 2deg . it is based on a classical approach to spins and accounts for resonant excitation in the trion and exciton resonances and also for nonresonant photoexcitation . a comprehensive set of experimental results is consistently described in the frame of this model . the suggested model can be generalized to the description of spin coherence generation in a 2d hole gas . in this case the in - plane component of the heavy - hole @xmath19-factor is negligible and , therefore , the regime of weak magnetic fields is realized . it can be expected that due to the stronger spin orbit interaction the free hole spin relaxation is faster than that of free electrons . on the other hand , in a low density hole gas most of the carriers are localized , so that the role of the spin orbit interaction is diminished and the hole spin relaxation times can reach up to 600 ps in gaas/(al , ga)as qws @xcite , for example . it is also worthwhile to note here , that the results of our model consideration can be also applied to for describing the excitation of spin coherence in other low - dimensional semiconductor structures such as in an ensemble of singly - charged quantum dots . we acknowledge fruitful discussions with al . l. efros , a. shabaev , i. v. ignatiev and i. a. yugova . this work was supported by the bmbf `` nanoquit '' program , the deutsche forschungsgemeinschaft ( grant no . ya 65/5 - 1 ) and by the russian foundation for basic research . eazh stays in dortmund were financed by the deutsche forschungsgemeinschaft via grants 436rus17/79/04 , 436rus17/93/05 and 436rus17/77/06 . an eli research visit to dortmund was supported by the gambrinus guest - professor program of the universitt dortmund . mmg was partially supported by the `` dynasty '' foundation icfpm .

we give a criterion which ensures that a group generated by cartan involutions in the automorph group of a rational quadratic form of signature is `` thin '' , namely it is of infinite index in the latter . it is based on a graph defined on the integral cartan root vectors , as well as vinberg s theory of hyperbolic reflection groups . the criterion is shown to be robust for showing that many hyperbolic hypergeometric groups for are thin .

1305.0729

let @xmath2 and consider the @xmath1 hypergeometric differential equation @xmath3 where @xmath4 and @xmath5 . assuming , as we do , that @xmath6 , @xmath7 , and the @xmath8 s and @xmath9 s are distinct , the @xmath10-functions @xmath11 where @xmath12 denotes omit @xmath13 , are linearly independent solutions to ( [ deq ] ) . here @xmath1 is the hypergeometric function @xmath14 and @xmath15 . equation ( [ deq ] ) is regular away from @xmath16 and its monodromy group @xmath17 is generated by the local monodromies @xmath18 ( @xmath19 ) gotten by analytic continuation of a basis of solutions along loops about @xmath20 , and @xmath21 respectively , see @xcite for a detailed description . the local monodromies of equations that come from geometry are quasi - unipotent which is one reason for our restricting @xmath8 and @xmath9 to be rational . we restrict further to such @xmath17 s which after a suitable conjugation are contained in @xmath22 . according to @xcite , this happens if the characteristic polynomials of @xmath23 and @xmath24 , whose roots are @xmath25 and @xmath26 respectively , are products of cyclotomic polynomials . in particular for each @xmath27 there are only finitely many such choices for the pair @xmath28 in @xmath29 . @xcite also determine the zariski closure @xmath30 of @xmath17 explicitly in terms of @xmath31 . furthermore the integrality conditions that we are imposing imply that @xmath17 is self dual so that @xmath32 is either finite , @xmath33 ( @xmath10 even ) or @xmath34 . the signature of the quadratic form in the orthogonal case is determined by the relative locations of the roots @xmath28 ( see section [ beuksum ] ) . our interest is whether @xmath17 is of finite or infinite index in @xmath35 $ ] . in the first case we say that @xmath17 is _ arithmetic _ and in the second case that it is _ thin_. this distinction is important in various associated number theoretic problems ( see @xcite ) and this paper is concerned with understanding which case happens and which is typical . in a given example , if @xmath17 is arithmetic one can usually verify that it is so by producing generators of a finite index subgroup of @xmath36 , on the other hand if @xmath17 is thin then there is no general procedure to show that it is so . our main result is a robust certificate for showing that certain @xmath17 s are thin . until recently , other than the cases where @xmath17 ( or equivalently @xmath32 ) is finite , there were few cases for which @xmath17 itself was known . for @xmath37 it is well known that all the @xmath17 s are arithmetic and we show that the same is true for @xmath38 . for @xmath39 brav and thomas @xcite showed very recently that the dwork family @xcite @xmath40 as well as six other hypergeometrics with @xmath41 which correspond to families of calabi - yau three - folds , are thin . in fact they show that the generators @xmath23 and @xmath42 of the above @xmath17 s play generalized ping - pong on certain subsets of @xmath43 , from which they deduce that @xmath17 is a free product and hence by standard cohomological arguments that @xmath17 is thin . on the other hand , venkataramana shows in @xcite that for @xmath10 even and @xmath44 @xmath17 is arithmetic ( in @xmath45 ) . in particular , there are infinitely many arithmetic @xmath17 s . in @xcite many more examples with @xmath46 and for which @xmath17 is arithmetic are given . another example for which @xmath47 can be shown to be thin is @xmath48 , @xmath49 , see @xcite . in this case @xmath50 is orthogonal and has signature @xmath51 and @xmath36 splits as a product of @xmath52 s . all of our results are concerned with the case that @xmath32 is orthogonal and is of signature @xmath0 over @xmath53 . we call these hyperbolic hypergeometric monodromy groups . there is a unique ( up to a scalar multiple ) integral quadratic form @xmath54 for which @xmath55 , or what is the same thing an integral quadratic lattice @xmath56 with @xmath57 . in section [ quadform ] we determine a commensurable quadratic sublattice explicitly which facilitates many further calculations . in this hyperbolic setting @xmath58 acts naturally as isometries of hyperbolic @xmath59-space @xmath60 and we will use this geometry as a critical ingredient to provide a certificate for @xmath17 being thin . our first result is the determination of the @xmath28 s for which @xmath32 is hyperbolic , see theorem [ theo ] . firstly , these only occur if @xmath10 is odd and for @xmath61 they are completely described by seven infinite parametric families . for @xmath62 there are sporadic examples which are listed in tables 2 and 3 of section [ numeric ] . our determination of the seven families is based on a reduction to @xcite s list of families of @xmath32 s which are finite ( i.e. those @xmath32 s for which @xmath63 and have signature ( @xmath64 ) ) . for @xmath38 , if @xmath17 is not finite then it is hyperbolic and as we noted all @xmath65 of these hyperbolic groups are arithmetic . this is verified separately for each case , there being no difficulty in deciding whether a finitely generated subgroup of @xmath66 is thin or not ( the latter is a double cover of @xmath67 , see the appendix . for @xmath68 the hyperbolic monodromies behave differently . our certificate of thinness applies in these cases and it is quite robust as exemplified by [ theorem1 ] the two families of hyperbolic monodromies @xmath17 with @xmath68 and odd * @xmath69 * @xmath70 are thin . in particular infinitely many of the @xmath17 s are thin and as far as we know these give the first examples in the general monodromy group setting of thin monodromy groups for which @xmath71 is high dimensional and simple . the normalized @xmath1 s corresponding to ( i ) and ( ii ) above ( see @xcite for the normalization ) are @xmath72 respectively . the second has integral coefficients while the first does not , hence this arithmetic feature of @xmath73 is not reflected in the arithmeticity of the corresponding @xmath47 ( see the end of section [ familyclassification ] for more about the integrality of the coefficients ) . our certificate for thinness applies to a number of the families and many of the sporadic examples . the full lists that we can handle thus far are recorded in theorems [ thinfamilythm ] and [ secondthin ] of section [ mindistgraph ] as well as table [ table3 ] of section [ numeric ] . there remain some families for which the method does not apply ( at least not directly ) . in any case we are led to [ allthinconj ] all but finitely many hyperbolic hypergeometric @xmath17 s are thin . we turn to a brief description of our methods . in general if there is an epimorphism @xmath74 with @xmath75 of infinite index in @xmath76 , then clearly @xmath17 must be of infinite index in @xmath36 . in principle , @xmath77 could be infinite which would imply that @xmath17 is thin since the latter is essentially generated by involutions , however we know very little about these cohomology groups for the full @xmath78 . instead we use variations of the quotient @xmath79 where @xmath80 is the vinberg reflective subgroup of @xmath78 . namely , @xmath80 is the group generated by the elements of @xmath78 which induce reflections of hyperbolic space @xmath60 ( that is the induced action on one of the two sheeted hyperboloids in @xmath10-space , see section [ cartaninvol ] ) . vinberg @xcite and nikulin @xcite have shown that except for rare cases @xmath76 is infinite , see section [ cartaninvol ] for a review of their results which we use . to apply this we need to understand the image of @xmath17 in @xmath76 . the key observation is that the element @xmath42 in @xmath17 , which is a linear reflection of @xmath10-space , induces a cartan involution of @xmath60 , that is it is an isometry of @xmath60 which is an inversion in a point @xmath81 . the reflection subgroup @xmath82 of @xmath17 is the group generated by the cartan involutions @xmath83 , @xmath84 and @xmath85 is cyclic . thus essentially up to commensurability the question of whether @xmath17 is thin is a special case of deciding whether a subgroup @xmath86 of @xmath78 generated by cartan involutions is thin or not . we approach this by examining the image of such a @xmath86 in @xmath87 ( when the latter is infinite ) . at this point the study is about @xmath78 , @xmath87 , and a general such @xmath86 ( and @xmath10 need nt be odd ) . we define a graph @xmath88 , the `` distance graph , '' associated with root vectors of @xmath54 which is central to the analysis . we assume that @xmath54 is even ( that is @xmath89 for @xmath90 ) and for @xmath91 or @xmath92 let @xmath93 be the corresponding root vectors ( in the case that @xmath94 for all @xmath95 which comes up in some cases we also allow @xmath96 to be @xmath97 ) . the root vectors define linear reflections lying in @xmath78 given by @xmath98 . by our choice of the signature of @xmath54 , for @xmath99 the map @xmath100 induces a cartan involution on @xmath60 while for @xmath101 it induces a hyperbolic reflection on @xmath60 . assume that the cartan involutions generating @xmath86 come from root vectors in @xmath102 . @xmath88 has for its vertices the set @xmath102 and we join @xmath103 to @xmath104 if @xmath105 , this corresponds to @xmath103 and @xmath104 having the smallest distance allowed by discreteness , as points in @xmath60 . the graph @xmath88 is a disjoint union of its connected components @xmath106 and these satisfy ( see section [ graphprop ] for a detailed statement ) * the components @xmath107 consist of finitely many isomorphism types . * [ [ ( ii ) ] each type is either a singleton or an infinite homogeneous graph corresponding to a transitive isometric action of a coxeter group with finite stabilizer . * if @xmath107 is a connected component of @xmath88 , then the group generated by the corresponding cartan roots , @xmath108 is up to index @xmath109 contained in the reflection group @xmath110 . with this the certificate for showing that @xmath86 is thin is clear : according to vinberg and nikulin , @xmath111 is infinite except in rare cases ( and nikulin has a classification of thee ) , hence if ( iii ) is satisfied for the generators of @xmath86 then @xmath86 must be thin . the calculations connected with the minimal distance graph can be carried out effectively in general and even explicitly for some of our families . the process in the general case invokes an algorithm for computing the fundamental polyhedral cell for a discrete group of motions of hyperbolic space which is generated by a finite number of reflections . such an algorithm is provided in section [ thinguys ] . the above leads to the cases discussed in sections [ mindistgraph ] and [ numeric ] for which @xmath17 is shown to be thin . to end we note that it is possible that a much stronger version of vinberg s theorem in the following form is valid : for all but finituely many rational quadratic forms @xmath54 ( at least if @xmath10 is large enough ) the full `` weyl subgroup '' @xmath112 generated by _ all _ reflections in @xmath113 ( that is , both those inducing hyperbolic reflections and cartan involutions on @xmath60 ) is infinite index in @xmath114 ( see nikulin @xcite ) . if this is true then conjecture [ allthinconj ] would follow easily from our discussion . in section [ cartaninvol ] we give an example in dimension @xmath115 with @xmath116 being thin and @xmath117 arithmetic , so there is no general commensurability between these groups . finally , we note that when our analysis of the thinness of @xmath47 succeeds , it comes with a description of @xmath47 as a subgroup ( up to commensurability ) of a geometrically finite subgroup of @xmath116 and this opens the door to determine the group structure of @xmath47 itself . we leave this for the future . we begin by reviewing @xcite which forms the basis of our analysis . the setup and notation is as in the introduction . the starting point for studying @xmath17 is the following theorem of levelt . [ levthm ] for @xmath118 let @xmath119 and @xmath120 be as above . define the complex numbers @xmath121 to be the coefficients of the polynomials @xmath122 then @xmath17 is the group generated by @xmath123 note that @xmath17 can be conjugated into @xmath22 if and only if the polynomials @xmath124 and @xmath125 above factor as cyclotomic polynomials , and so the roots of @xmath124 and @xmath125 are reciprocal , meaning they are left invariant under the map @xmath126 . since we are interested in integral monodromy groups , we will assume this throughout the article . also , given a group @xmath127 as above , one obtains another hypergeometric group @xmath128 by taking @xmath129 to be @xmath130 and @xmath131 modulo @xmath21 , respectively . the group @xmath128 is called a _ scalar shift _ of @xmath17 , and as pointed out in remark 5.6 in @xcite , we have that @xmath132 , up to a finite center . thus when classifying all possible pairs @xmath133 such that @xmath17 is integral and fixes a quadratic form of signature @xmath0 in section [ familyclassification ] we will do so up to scalar shift . note that for any given @xmath10 there are only finitely many such groups @xmath17 . we also assume that @xmath47 is _ irreducible _ , meaning that it fixes no proper subspace of @xmath134 , and _ primitive _ , meaning that there is no direct sum decomposition @xmath135 with @xmath136 and @xmath137 for all @xmath138 such that @xmath47 simply permutes the spaces @xmath139 . finally , we define the reflection subgroup @xmath140 of @xmath47 to be the group generated by the reflections @xmath141 . by theorem 5.3 of @xcite we have that the primitivity of @xmath47 implies the irreducibility of @xmath140 . with the notation above , beukers - heckman show in @xcite that @xmath142 falls into one of three categories . given a hypergeometric monodromy group @xmath17 , let @xmath143 where @xmath144 and @xmath145 are as in theorem [ levthm ] . note that in the cases we consider @xmath146 . then @xmath17 belongs to one of the following categories . * a finite group ( beukers - heckman list such cases completely in @xcite , and we summarize these cases in theorem [ finiteh ] below ) * if @xmath10 is even , @xmath47 is infinite , and @xmath147 then @xmath148 and @xmath149 * if @xmath10 is odd and @xmath47 is infinite ; or if @xmath10 is even , @xmath47 is infinite , and @xmath150 , then @xmath151 for some rational , unique up to scalar multiple quadratic form @xmath152 in @xmath10 variables and @xmath153 as noted in the introduction , in this article we study hypergeometric groups @xmath47 which fall into category ( 2 ) and such that @xmath47 fixes a quadratic form of signature @xmath0 . we should mention that beukers - heckman show that the signature of @xmath154 in category ( 2 ) is given by @xmath155 where @xmath156 and @xmath157 and the @xmath119 and @xmath120 are ordered as described in the introduction : @xmath6 , @xmath7 . although we concern ourselves with the case where @xmath47 is infinite , irreducible , and primitive , we state below beukers - heckman s classification of _ all _ possible finite groups @xmath17 . in section [ familyclassification ] we will need this classification to classify all of the primitive , irreducible hypergeometric groups @xmath17 which we consider i.e. integral orthogonal of signature @xmath0 . the precise statement of the theorem is taken from @xcite . [ finiteh ] let @xmath158 and @xmath159 and let @xmath17 be as before . define @xmath160 and @xmath161 and let @xmath162 and @xmath163 . let @xmath164 corresponding to @xmath17 be as in theorem [ levthm ] . if @xmath17 is finite then the corresponding polynomials @xmath164 are as in one of the cases below . \(1 ) the case where @xmath140 is primitive : in this case the corresponding polynomials @xmath124 and @xmath125 belong to one of two infinite families or to one of @xmath165 sporadic examples . one such infinite family corresponds to @xmath166 where @xmath167 , @xmath168 , @xmath169 . the other infinite family is obtained by replacing @xmath170 with @xmath171 above . \(2 ) the case where @xmath140 acts reducibly on @xmath134 : in this case theorem 5.3 of @xcite implies that @xmath140 is primitive , and we have that there is some primitive @xmath172th root of unity @xmath173 with @xmath174 such that @xmath175 . then @xmath133 gives the pair @xmath176 where @xmath177 and @xmath178 are as in case ( 1 ) or correspond to one of the sporadic examples mentioned there . \(3 ) the case where @xmath47 is imprimitive and @xmath140 is irreducible : in this case the corresponding polynomials @xmath124 and @xmath125 belong to one of two infinite families . one such family is @xmath179 where @xmath180 , @xmath181 , and @xmath182 . the other such family is obtained by replacing @xmath170 with @xmath171 above . in this section we classify all almost interlacing cyclotomic sequences ( see definition [ almostinterlacing ] ) . this classification will be used in the next section to classify all hyperbolic hypergeometric monodromies in dimension @xmath61 . two sequences @xmath183 and @xmath184 are called interlacing cyclotomic sequences if the following three conditions hold : * @xmath185 for every @xmath186 . * @xmath187 and @xmath188 are products of cyclotomic polynomials . * @xmath189 or @xmath190 . define @xmath191 and @xmath192 . condition 3 in the above definition is equivalent to the following condition : * @xmath193 . [ almostinterlacing ] two sequences @xmath194 and @xmath195 are called almost interlacing cyclotomic sequences if the following three conditions hold : * @xmath185 for every @xmath186 . * @xmath187 and @xmath196 are products of cyclotomic polynomials . * @xmath197 . note that , by the expression for the signature in ( [ sig ] ) , it is precisely when @xmath8 and @xmath9 are almost interlacing that @xmath17 is hyperbolic ( signature @xmath0 ) . beukers and hekman classify all interlacing cyclotomic sequences in @xcite . we use their classification in order to classify all almost interlacing cyclotomic sequences . we start with the following lemma : [ interlacingtypes ] let @xmath198 and @xmath199 be almost interlacing cyclotomic sequences with @xmath200 . then @xmath10 is odd and @xmath201 and @xmath202 or @xmath203 and @xmath204 . moreover , if the first possibility happens then one of the following 4 options holds : * the sequences @xmath205 and @xmath206 are interlacing cyclotomic sequences where : @xmath207 * the sequences @xmath208 and @xmath209 are interlacing cyclotomic sequences where : @xmath210 * the sequences @xmath208 and @xmath209 are interlacing cyclotomic sequences where : @xmath211 * the sequences @xmath212 and @xmath213 are interlacing cyclotomic sequences where : @xmath214 the proof is divided into easy steps : * the sequences @xmath198 and @xmath199 are cyclotomic so for every @xmath215 , @xmath216 and @xmath217 * for every @xmath218 denote @xmath219 and @xmath220 . the equality @xmath221 implies that exactly one of the following possibilities holds : * * @xmath222 * * @xmath223 * * @xmath224 * * @xmath225 * * @xmath226 for some @xmath227 . * the symmetry of step @xmath228 shows that options ( 3 ) and ( 4 ) are not possible and that : * * if option ( 1 ) happens then @xmath229 . * * if option ( 2 ) happens then @xmath230 . * * if option ( 3 ) happens then @xmath10 is odd , @xmath231 and @xmath232 . * denote @xmath233 . changing the roles of the @xmath119 s and the @xmath234 s we see that @xmath10 is odd and that either @xmath235 and @xmath202 or @xmath236 and @xmath237 . * from now on we will assume that @xmath235 so @xmath238 and @xmath202 . * since there are exactly @xmath10 @xmath234 s the assumptions of step ( e ) imply : @xmath239 furthermore , the symmetry of step ( a ) implies that if @xmath240 then also @xmath241 while if @xmath242 then @xmath243 . * the symmetry of step ( a ) implies that either @xmath244 or @xmath245 . thus , there are only 4 options : * * @xmath246 @xmath247 * * @xmath248 @xmath249 * * @xmath250 @xmath251 * * @xmath252 @xmath253 in this section we combine lemma [ interlacingtypes ] with the classification of finite hypergeometric monodromy groups in theorem [ finiteh ] to produce seven infinite families of primitive integer hypergeometric groups which fix a quadratic form of signature @xmath0 . each family is either a two or three parameter family . as will be clear in the proof of the following theorem , the @xmath10-dimensional two parameter families listed below are derived from the roots of @xmath254 where @xmath255 is taken to be @xmath256 or @xmath10 , while the @xmath10-dimensional three parameter families are derived from the roots of @xmath257 where @xmath258 and @xmath255 is taken to be @xmath256 or @xmath10 . these families describe all primitive hypergeometric hyperbolic monodromy groups in dimension @xmath61 up to scalar shift . in addition to groups in these families , there are several sporadic groups in dimensions @xmath259 which are listed in table [ table3 ] of section [ sporadic ] along with all primitive hyperbolic @xmath17 in dimension @xmath259 . [ theo ] let @xmath167 be odd and let @xmath260 be as defined in ( [ polystate0 ] ) and ( [ polystate ] ) . let @xmath261 and @xmath262 where @xmath263 for @xmath118 and let @xmath17 be as in theorem [ levthm ] . if @xmath133 belongs to one of the following families or is a scalar shift of a pair in one of these families , then @xmath264 . if @xmath61 , this list of families completely describes ( up to scalar shift ) the groups @xmath265 where @xmath54 is a quadratic form in @xmath10 variables of signature @xmath0 . * @xmath266 : @xmath267 where @xmath268 is an odd integer . * @xmath269 : @xmath270 where @xmath271 is an integer and @xmath272 is odd . * @xmath273 : @xmath274 where @xmath275 is an odd integer . * @xmath276 : @xmath277 where @xmath278 , @xmath279 , @xmath280 , @xmath281 are the roots of @xmath282 , and @xmath283 are the roots of @xmath284 . * @xmath285 : @xmath286 where @xmath278 , @xmath280 , @xmath287 , @xmath288 are the roots of @xmath289 , and @xmath290 are the roots of @xmath291 . * @xmath292 : @xmath293 where @xmath278 , @xmath280 , @xmath287 , @xmath288 are the roots of @xmath289 , and @xmath290 are the roots of @xmath291 . * @xmath294 : @xmath295 where @xmath278 , @xmath296 , @xmath297 are the roots of @xmath298 and @xmath299 are the roots of @xmath300 . from the previous section , we have that every family of pairs @xmath133 for which @xmath301 can be derived from a family of pairs @xmath302 for which @xmath128 is finite , but not necessarily primitive or irreducible . there are several cases to consider . * case 1 : * @xmath303 . without loss of generality , assume @xmath304 for some @xmath96 and @xmath305 . by lemma [ interlacingtypes ] , letting @xmath306 @xmath307 @xmath308 @xmath309 we have that @xmath301 . suppose @xmath128 is finite , imprimitive , and irreducible . from theorem [ finiteh ] there are two infinite families of @xmath302 , the second of which gives pairs that are scalar shifts by @xmath310 of the pairs coming from the first family . since we are only interested in determining pairs up to scalar shift , we consider just one of these families , which has corresponding cyclotomic polynomials @xmath311 and @xmath312 from ( [ polystate0 ] ) where @xmath313 is odd . up to scalar shift , this yields the family @xmath314 : @xmath315 where @xmath313 is odd . note that , unlike @xmath302 , the pair @xmath133 is no longer imprimitive , and still irreducible . suppose @xmath128 is finite and reducible or primitive and irreducible . then for every @xmath316 , theorem [ finiteh ] gives two infinite families of @xmath302 , of which we consider just the first , as the other can be obtained via scalar shift of the first . the corresponding cyclotomic polynomials in this case are @xmath282 and @xmath284 from ( [ polystate ] ) , where @xmath317 , with @xmath278 and @xmath318 . let @xmath279 and denote the roots of @xmath282 by @xmath281 . let @xmath280 and denote the roots of @xmath284 by @xmath283 . up to scalar shift , this yields the family @xmath319 : @xmath320 note that the pair @xmath133 is primitive and irreducible . * case 2 : * @xmath321 . as before , we consider the imprimitive irreducible , and the primitive irreducible or reducible cases . suppose @xmath128 is finite , imprimitive , and irreducible . from theorem [ finiteh ] , we have that up to scalar shift the only family of @xmath302 in this case corresponds to the roots of @xmath322 and @xmath323 in ( [ polystate0 ] ) : @xmath324 where @xmath272 is odd . according to lemma [ interlacingtypes ] , there are two ways to obtain from @xmath302 a pair @xmath133 for which @xmath17 is signature @xmath0 is obtained as above , giving the two families @xmath325 where @xmath326 and @xmath327 where @xmath272 is odd . however , since these two families are shifts of each other by @xmath310 , we record them under one family @xmath328 . suppose @xmath128 is finite and reducible or primitive irreducible . from theorem [ finiteh ] , we have that for every @xmath329 that there is only one family up to scalar shift of @xmath302 for which @xmath128 is reducible . namely , let @xmath317 , with @xmath278 and @xmath330 . denote the roots of @xmath289 in ( [ polystate ] ) by @xmath288 . let @xmath280 , @xmath287 , and denote the roots of @xmath291 in ( [ polystate ] ) by @xmath290 . then the family @xmath331 is the only family up to scalar shift in this case such that @xmath128 is reducible . according to lemma [ interlacingtypes ] , there are two ways to obtain a pair @xmath133 for which @xmath17 is signature @xmath0 . one way is to remove the term @xmath280 and add two @xmath310 s to @xmath332 , and add the term @xmath333 to @xmath334 , yielding the family @xmath335 : @xmath336 another way is to add the term @xmath310 to @xmath334 , take away the term @xmath287 from @xmath332 , and insert two @xmath337 into @xmath332 , yielding the family @xmath338 : @xmath339 * case 3 : * @xmath340 . suppose @xmath128 is finite , imprimitive , and irreducible . from theorem [ finiteh ] , we have that up to scalar shift the only family of @xmath302 in this case is obtained from the roots of the cyclotomic polynomials @xmath341 and @xmath342 in ( [ polystate0 ] ) : @xmath343 where @xmath313 is odd . the corresponding @xmath133 for which @xmath17 is signature @xmath0 is obtained by adding two @xmath310 s to @xmath334 , and adding two @xmath333 s to @xmath332 , yielding the family @xmath344 : @xmath345 where @xmath313 is odd . suppose @xmath128 is finite and reducible or primitive irreducible . from theorem [ finiteh ] , we have that for every @xmath346 that there is only one family up to scalar shift of @xmath302 for which @xmath128 is reducible . namely , let @xmath317 , with @xmath278 and @xmath347 . denote the roots of @xmath298 in ( [ polystate ] ) by @xmath297 . denote the roots of @xmath300 in ( [ polystate ] ) by @xmath299 . then the family @xmath348 is the only family up to scalar shift in this case such that @xmath128 is reducible . the corresponding @xmath133 for which @xmath17 is signature @xmath0 is obtained by adding two @xmath310 s to @xmath334 , and adding two @xmath333 s to @xmath332 , yielding the family @xmath349 : @xmath350 this exhausts all of the possibilities and thus we have the statement in the theorem as desired . since from the previous section we have that any primitive hypergeometric hyperbolic monodromy group is obtained by permuting the coordinates of some interlacing pair @xmath302 , and since there are no sporadic such interlacing pairs in dimensions @xmath351 , we have that the families @xmath352 and @xmath349 completely describe the primitive hyperbolic hypergeometric @xmath17 s in dimension @xmath61 . we end this subsection with a remark about the integrality of the coefficients of @xmath1 for hyperbolic hypergeometrics . writing @xmath353 , where @xmath177 and @xmath178 are as in theorem [ levthm ] , as @xmath354 with @xmath355 positive integers and @xmath356 . the coefficients of the corresponding hypergeometric function are @xmath357 if these are to be integers then @xmath358 must be at least @xmath21 . note that @xmath359 is the multiplicity of the @xmath96 s for which @xmath360 . then @xmath361 where @xmath362 ( see @xcite , @xcite ) . it is easy to check using landau s criterion @xcite and our characterization of hyperbolic hypergeometric monodromies , that for the latter @xmath363 iff @xmath364 . from the description of the families of hyperbolic hypergeometric monodromies in theorem [ theo ] we see that @xmath365 , @xmath366 , @xmath367 , and @xmath368 are the infinite families with integral coefficients while the sporadic ones can be read off from tables 2 and 3 , namely the entries with @xmath369 . this can be used to give the list of integral factorials ( [ star ] ) which correspond to hyperbolic hypergeometric monodromies . this is the analogue of @xcite and @xcite who classify the integral @xmath370 s in ( [ star ] ) which correspond to finite monodromy groups ( which in turn correspond to @xmath371 ) . in this section we calculate the quadratic forms preserved by the primitive hyperbolic hypergeometric monodromy groups described in the previous section . this will be a necessary ingredient in our minimal distance graph method described in section [ mindistgraph ] . recall that the monodromy group @xmath142 is generated by two matrices @xmath372 where the characteristic polynomials of @xmath23 and @xmath24 , @xmath373 and @xmath374 respectively , are products of cyclotomic polynomials . denote @xmath375 the eigenvalues of @xmath42 are @xmath21 and @xmath376 and their geometric multiplicities are @xmath59 and @xmath21 respectively . the first @xmath59 elements @xmath377 of the standard basis are eigenvectors for the eigenvalue @xmath21 while @xmath103 is an eigenvector for the eigenvalue @xmath376 . [ lemtec1 ] let @xmath378 . then @xmath379 if and only if @xmath380 for every @xmath381 where @xmath382 . the only if part follows from comparing the first @xmath59 columns of both sides of @xmath383 . for the if part it is suffices to check that also the last columns of @xmath384 and @xmath385 are equal . the equalities of the first @xmath59 columns imply that @xmath386 for every @xmath218 . the characteristic polynomial @xmath387 of @xmath23 is also the characteristic polynomial of @xmath388 since it is a a product of cyclotomic polynomials . thus , @xmath389 @xmath390 [ lemtec2 ] assume that @xmath378 satisfies @xmath379 . then @xmath391 if and only if @xmath392 is euclidean - orthogonal to @xmath393 for every @xmath381 . in particular , if @xmath379 and @xmath391 then @xmath394 is euclidean - orthogonal to @xmath103 for every @xmath218 . lemma [ lemtec1 ] and its analog with respect to @xmath24 implies that @xmath391 if and only if @xmath395 for every @xmath381 . as the first row of @xmath396 is @xmath397 while the other rows are zeros , @xmath392 must be orthogonal to @xmath103 ( under the usual euclidean scalar product ) . for every @xmath398 @xmath399 so by induction on @xmath400 we see that @xmath54 satisfies the required condition if and only if @xmath401 which is equivalent to @xmath54 being euclidean - orthogonal to @xmath402 for every @xmath403 . finally , if @xmath379 , @xmath391 then also @xmath404 . thus , for @xmath218 we have @xmath405 so @xmath406 . [ prop form ] let @xmath23 , @xmath24 , @xmath42 , @xmath407 , @xmath408 and @xmath103 be as above . assume that the group @xmath409 is primitive . then : * the vectors @xmath410 are linearly independent and @xmath411 preserves the lattice @xmath56 spanned by them . * there exists a unique ( up to a scalar product ) non - zero integral quadratic form @xmath412 such that @xmath23 and @xmath24 belong to its orthogonal group . * if the quadratic form is normalized to have @xmath413 then @xmath414 equals to minus the @xmath415-coordinate of @xmath416 for every @xmath417 . * if the coefficients of @xmath418 satisfy @xmath419 then the matrix representing @xmath54 w.r.t @xmath410 is given by @xmath420 we start by proving the existence part of @xmath421 . by lemma [ lemtec2 ] there exists a non - zero @xmath378 such that @xmath422 and @xmath391 . assume first that @xmath54 is anti - symmetric so @xmath423 . lemma [ lemtec2 ] implies that @xmath406 for every @xmath218 . since @xmath424 span @xmath425 , the set @xmath426 is a non - trivial proper subspace of @xmath425 . this subspace is preserved by @xmath47 , a contradiction to the irreducibility of @xmath47 . thus , @xmath54 is not anti - symmetric and the quadratic form @xmath427 has the desired properties . lemma [ lemtec2 ] implies that @xmath428 for some @xmath429 so @xmath430 holds . since @xmath47 is primitive , every non - trivial normal subgroup of it is irreducible . thus in order to prove @xmath431 it is enough to show that the normal subgroup @xmath432 preserves the @xmath433-lattice spanned by @xmath434 . property @xmath430 implies that @xmath435 for every @xmath417 . since @xmath24 preserves the quadratic form , @xmath436 for every @xmath417 . thus , @xmath140 preserves the @xmath437-lattice spanned by @xmath438 and the proof of @xmath431 is complete . the uniqueness part of @xmath421 now follows from @xmath431 together with lemma [ lemtec2 ] . for @xmath218 the @xmath439-th and the @xmath440-th coordinates of bottom row of @xmath441 are equal to 1 and all the other coordinates of this row are zero . thus , the last coordinate of @xmath442 is @xmath443 if @xmath444 and @xmath445 if @xmath446 . from @xmath430 we get @xmath447 for every @xmath448 . property @xmath449 follows from the fact the @xmath450 depends only on @xmath451 since @xmath24 preserves the quadratic form . we now record the invariant quadratic form for two cases that we consider in section [ thinguys ] . [ corol form 1 ] let @xmath452 be the normalized quadratic form preserved by @xmath453 . if @xmath186 then @xmath454 [ corol form 3 ] let @xmath452 be the normalized quadratic form preserved by @xmath455 . if @xmath186 then @xmath456 as noted in the introduction we make crucial use of vinberg s @xcite , @xcite as well as nikulin s @xcite results concerning the size of groups generated by hyperbolic reflections . before reviewing this arithmetic theory we begin with quadratic forms and hyperbolic space over the reals . a non - degenerate real quadratic form @xmath457 with corresponding symmetric bilinear form @xmath458 is determined by its signature . our interest is in the case that @xmath54 has signature @xmath459 , which after a real linear change of variable can be brought to the form @xmath460 the null cone @xmath42 of @xmath54 is @xmath461 as depicted in figure [ conefig ] . @xmath462 consists of @xmath463 components , the outside , @xmath464 , which consists of one component , and the inside , @xmath465 , which consists of @xmath109 components : @xmath466 and @xmath467 . the quadrics @xmath468 where @xmath469 are two sheeted hyperboloids and either sheet , say the one with @xmath466 and @xmath470 , can be chosen as a model of hyperbolic @xmath59 dimensional space ( we assume that @xmath180 ) , @xmath60 . this is done by restricting the line element @xmath471 to @xmath60 . the quadrics @xmath468 with @xmath472 are one sheeted hyperboloids . if @xmath473 , the real orthogonal group of @xmath54 , then @xmath474 preserves the quadrics , however it may switch the sheets of the two sheeted hyperboloid . if it preserves each of these , then @xmath474 acts on @xmath60 isometrically , while if @xmath474 switches the sheets then @xmath475 acts isometrically on @xmath60 . in either case we obtain an induced isometry . of particular interest to us are linear reflections of @xmath476 given by @xmath103 s with @xmath477 ( i.e. not on the null cone ) . for such a @xmath103 denote by @xmath100 the linear transformation @xmath478 it is easy to see that @xmath479 , that it is an involution , and that it preserves the components of the complement of the light cone if @xmath480 , and switches them if @xmath481 . in the first case , the induced isometry of @xmath60 is a hyperbolic reflection . the fixed point set of @xmath100 in @xmath476 is @xmath482 and this intersects @xmath60 in an @xmath483-dimensional hyperbolic hyperplane and the induced isometry is a ( hyperbolic ) reflection about this hyperplane . in the second case , @xmath481 , the induced isometry @xmath484 fixes the line @xmath485 and this line meets @xmath60 in a point @xmath486 . the induced isometry of @xmath60 in this case inverts geodesics in @xmath486 and is a cartan involution . these two involutive isometrics of @xmath60 are quite different . in the case of a discrete group of motions generated by hyperbolic reflections , there is a canonical fundamental domain , namely the connected components of the complement of the hyperplanes corresponding to all the reflections . this is the starting point to vinberg s theory of reflective groups . for groups generated by cartan involutions there is no apparent geometric approach . we turn to the arithmetic theory . let @xmath487 be a quadratic lattice , that is a rank @xmath10 @xmath433-module equipped with a nondegenerate integral symmetric bilinear form @xmath458 , whose associated quadratic form @xmath488 is rational . the dual lattice @xmath489 of @xmath56 is given by @xmath490 it contains @xmath56 as a finite sublattice and the invariant factors of @xmath56 are defined to be the invariants of the torsion module @xmath491 . the product of these invariants is equal to @xmath492 , where @xmath493 is the discriminant of @xmath56 ( that is @xmath494 where @xmath73 is an integral matrix realization of @xmath54 ) . we assume that @xmath495 is hyperbolic and of signature @xmath459 . the @xmath433 analogue of the null cone is @xmath496 and that of the quadrics is @xmath497 for @xmath498 , @xmath499 . we assume throughout that @xmath54 is isotropic over @xmath500 which means that @xmath501 ( and is in fact infinite ) . the group of integral automorphs of @xmath56 is denoted by @xmath113 or @xmath78 . a primitive vector @xmath103 in @xmath502 is called a @xmath96-root if @xmath503 . in this case the linear reflection @xmath100 of @xmath29 given by ( [ eq3.2 ] ) is integral and lies in @xmath113 . for @xmath504 or @xmath505 , @xmath506 is always in @xmath489 and we will have occasions where @xmath507 yields root vectors as well . as discussed above , the root vectors with @xmath472 induce hyperbolic reflections on @xmath508 ( which we choose to be one of the components of @xmath509 ) while for @xmath469 they induce cartan involutions . the vinberg reflective group @xmath510 is the subgroup of @xmath113 generated by all hyperbolic reflections coming from root vectors with @xmath472 . it is plainly a normal subgroup of @xmath113 and the main result in @xcite asserts that if @xmath511 then @xmath512 . vinberg s proof is based on studying the fundamental domain in @xmath60 of @xmath510 and relating it to the cusps of @xmath513 which correspond to null vectors @xmath514 . in particular it uses the assumption that @xmath54 is isotropic . nikulin s results @xcite are concerned with the case that @xmath56 is even , that is @xmath515 for all @xmath90 , and the subgroup @xmath116 generated by all the @xmath109-root vectors in @xmath56 ( note he chooses @xmath54 to have signature @xmath516 so that his @xmath92-root vectors are our @xmath109-root vectors ) . @xmath116 is a subgroup of @xmath510 and it is also normal in @xmath113 . for these he gives a complete classification of all @xmath56 s for which @xmath517 is finite . there are only finitely many such and for @xmath68 the list is quite short . @xmath56 is called two elementary if @xmath491 is @xmath518 for some @xmath519 . if @xmath56 is not two elementary and @xmath68 and odd ( the last condition on @xmath10 is what is of interest to us for our applications ) then @xmath520 unless @xmath56 is isomorphic to @xmath521 and @xmath76 is one of @xmath522 @xmath523 or @xmath56 is isomorphic to @xmath524 here the positive definite ( @xmath109-reflective ) lattices @xmath144 , @xmath525 , @xmath526 are the standard ones ( see @xcite ) while @xmath527 $ ] and @xmath528 $ ] . as a consequence unless @xmath56 is two elementary or its invariant factors are among the list below , @xmath529 . [ cols="<,<",options="header " , ] h. park @xcite has recently shown that all the @xmath80 s in table 3 are thin except for the second entry , which is arithmetic . moreover , in that case @xmath140 is still contained in a thin reflection subgroup of @xmath80 . in particular , all the @xmath140 s in table 3 are thin . ] while our certificate for being thin succeeds in many cases as demonstrated in sections [ mindistgraph ] and [ numeric ] , there are families such as @xmath294 for which we found almost no paths between the basis vectors in the minimal distance graph . if indeed the corresponding components of @xmath88 are singletons for members of this family , then clearly a variation of the minimal distance graph is needed if this approach is to work . a crude , and it appears mostly reliable , test for the size of @xmath142 in @xmath530 is to simply count the number of elements in @xmath47 in a large ball . that is , let @xmath531 be a large number and let @xmath532 if @xmath47 is arithmetic then it is known ( @xcite ) that @xmath533 thus to probe the size of @xmath47 we generate elements in @xmath47 using our defining generators and try to determine @xmath534 for @xmath531 quite large . the difficulty is that we do nt know which elements of @xmath113 are in @xmath47 and the above procedure ( by going to large generations of elements in @xmath47 gotten from the generating set ) only gives a lower bound for @xmath534 . in any case we can compute in this way a lower bound for @xmath535 for @xmath531 large . if this is essentially @xmath483 then this suggests that @xmath47 is arithmetic , while if this number is less than @xmath483 it suggests that @xmath47 is thin ( if @xmath47 is geometrically finite then this quantity should be an approximation to the hausdorff dimension of the limit set of @xmath47 ( see for example @xcite ) . we have run this crude test for many of our hyperbolic hypergeometric monodromies and it gives the correct answer in the cases where we have a rigorous treatment . for example , in figures [ arith1 ] and [ arith2 ] we give a plot of @xmath536 versus @xmath537 for the arithmetic examples @xmath538 and @xmath539 . the similar plots for four examples with @xmath540 , and for which our certificate failed are given in figures [ num1 ] , [ num2 ] , [ num3 ] , and [ num4 ] below . based on these , it appears that these @xmath47 s are all thin . further such experimentation is consistent with conjecture [ allthinconj ] in the introduction and even that for @xmath68 , every hyperbolic hypergeometric monodromy group is thin . versus @xmath537 graph for @xmath17 where @xmath541 , @xmath542.,height=207 ] versus @xmath537 graph for @xmath17 where @xmath541 , @xmath543.,height=207 ] versus @xmath537 graph for @xmath17 where @xmath544 , @xmath545.,height=207 ] versus @xmath537 graph for @xmath17 where @xmath546 , @xmath547.,height=207 ] versus @xmath537 graph for @xmath17 where @xmath548 , @xmath549.,height=207 ] versus @xmath537 graph for @xmath17 where @xmath550 , @xmath545.,height=207 ] in this appendix , we consider the six @xmath463-dimensional hyperbolic hypergeometric monodromy groups and show that they are all arithmetic . our method is to determine the preimage of the groups in the spin double cover @xmath66 of @xmath551 and work in the setting of @xmath52 . once we pull back to @xmath52 , we rely on one of the following strategies . if the quadratic form @xmath54 that is fixed by hyperbolic hypergeometric monodromy group @xmath47 is isotropic over @xmath500 then one can find @xmath552 such that @xmath553 is some scalar multiple of @xmath554 below . moreover , under such a change of variable , the preimage of @xmath555 in the spin double cover of @xmath556 can be made to sit inside @xmath557 , and in this setting showing that the group is finite index is usually straightforward . out of the six @xmath463-dimensional hyperbolic hypergeometric monodromy groups , four fix an isotropic form @xmath54 , and in those cases we show that the preimage of the group in the spin double cover of @xmath556 contains a principle congruence subgroup of @xmath558 if the form @xmath54 fixed by @xmath47 is anisotropic , we again pull back to the spin double cover @xmath559 of @xmath556 in @xmath66 and , after generating several elements of @xmath559 , construct a region in the upper half plane which contains a fundamental domain for @xmath559 acting on @xmath560 ( this method works for the isotropic case as well ) . specifically , a dirichlet fundamental domain for @xmath559 is given by @xmath561 where @xmath562 is any base point and @xmath563 is a closed half space . one can prove that @xmath564 is compact ( and hence @xmath559 is arithmetic ) if the intersection @xmath565 is compact even when taken over a _ finite _ number of elements @xmath566 . note that this method uses the same idea as described in section [ numeric1 ] : generating elements of @xmath559 to say something about thinness , but in this case it can actually prove whether the group is thin or not . note that @xmath576 where @xmath577\ ] ] and conjugation by @xmath578 gives an isomorphism of the subgroup @xmath579 with the subgroup @xmath580 where @xmath581 , b'=\left[\begin{array}{ccc } 1&0&0\\ 1&1&0\\ 1&2&1 \end{array}\right].\ ] ] the preimage of this group in @xmath557 under the spin homomorphism in ( [ spin2 ] ) is the group @xmath582 where @xmath583 , y=\left[\begin{array}{cc } 1&1\\ 0&1 \end{array}\right].\ ] ] this group contains the generators @xmath584 , \left[\begin{array}{cc } \minus 15&4\\ \minus 4&1 \end{array}\right ] , \left[\begin{array}{cc } 5&\minus 4\\ 4&\minus 3 \end{array}\right ] , \left[\begin{array}{cc } 9&\minus 16\\ 4&\minus 7 \end{array}\right ] , \left[\begin{array}{cc } 13&\minus 36\\ 4&\minus 11 \end{array}\right]\ ] ] of @xmath585 . hence @xmath17 is itself arithmetic . in addition , our second strategy of finding a region which contains a fundamental domain for @xmath47 yields the compact region shown in figure [ fundy1 ] . @xmath587 , a=\left[\begin{array}{ccc } 0&0&\minus 1\\ 1&0&\minus 2\\ 0&1&\minus 2 \end{array}\right ] , b=\left[\begin{array}{ccc } 0&0&1\\ 1&0&\minus 3\\ 0&1&3 \end{array}\right].\ ] ] note that @xmath588 where @xmath589\ ] ] and conjugation by @xmath578 gives an isomorphism of the subgroup @xmath579 with the subgroup @xmath580 where @xmath590 , b'=\left[\begin{array}{ccc } 1&\minus 2&1\\ 0&1&\minus 1\\ 0&0&1 \end{array}\right].\ ] ] the preimage of this group in @xmath557 under the spin homomorphism in ( [ spin2 ] ) is the group @xmath582 where @xmath591 , y=\left[\begin{array}{cc } 1&0\\ \minus 1&1 \end{array}\right].\ ] ] this group contains the generators @xmath592 , \left[\begin{array}{cc } \minus 8&3\\ \minus 3&1 \end{array}\right ] , \left[\begin{array}{cc } 4&\minus 3\\ 3&\minus 2 \end{array}\right]\ ] ] of @xmath593 . hence @xmath17 is itself arithmetic . in addition , our second strategy of finding a region which contains a fundamental domain for @xmath47 yields the compact region shown in figure [ fundy2 ] . @xmath595 , a=\left[\begin{array}{ccc } 0&0&\minus 1\\ 1&0&\minus 1\\ 0&1&\minus 1 \end{array}\right ] , b=\left[\begin{array}{ccc } 0&0&1\\ 1&0&\minus 3\\ 0&1&3 \end{array}\right].\ ] ] note that @xmath596 where @xmath597\ ] ] and conjugation by @xmath578 gives an isomorphism of the subgroup @xmath579 with the subgroup @xmath598 where @xmath599 , b'=\left[\begin{array}{ccc } 1&\minus 2&\minus 2\\ 2&\minus 1&\minus 2\\ \minus 2&2&3 \end{array}\right].\ ] ] the preimage of this group in @xmath557 under the spin homomorphism in ( [ spin1 ] ) is the group @xmath582 where @xmath600 , y=\left[\begin{array}{cc } 0&1\\ \minus 1&2 \end{array}\right]\ ] ] and @xmath601 generate the principal congruence subgroup @xmath602 . hence @xmath17 is itself arithmetic . in addition , our second strategy of finding a region which contains a fundamental domain for @xmath47 yields the compact region shown in figure [ fundy3 ] . @xmath604 , a=\left[\begin{array}{ccc } 0&0&\minus 1\\ 1&0&0\\ 0&1&0 \end{array}\right ] , b=\left[\begin{array}{ccc } 0&0&1\\ 1&0&\minus 3\\ 0&1&3 \end{array}\right].\ ] ] note that @xmath596 where @xmath605\ ] ] and conjugation by @xmath578 gives an isomorphism of the subgroup @xmath579 with the subgroup @xmath606 where @xmath607 , b'=\left[\begin{array}{ccc } 1/2&\minus 1&\minus 1/2\\ 1&1&1\\ 1/2&1&3/2 \end{array}\right].\ ] ] the preimage of this group in @xmath557 under the spin homomorphism in ( [ spin1 ] ) is the group @xmath582 where @xmath608 , y=\left[\begin{array}{cc } 1&1\\ 0&1 \end{array}\right]\ ] ] and @xmath609 are a well - known generating set for all of @xmath557 . hence @xmath17 is itself arithmetic . in addition , our second strategy of finding a region which contains a fundamental domain for @xmath47 yields the compact region shown in figure [ fundy4 ] . @xmath610 , a=\left[\begin{array}{ccc } 0&0&\minus 1\\ 1&0&\minus 2\\ 0&1&\minus 2 \end{array}\right ] , b=\left[\begin{array}{ccc } 0&0&1\\ 1&0&\minus 1\\ 0&1&1 \end{array}\right].\ ] ] in this case @xmath54 is anisotropic over @xmath500 , and hence we turn to our strategy of approximating the fundamental domain to prove that @xmath47 is arithmetic : we produce the compact region shown in figure [ fundy5 ] which contains the fundamental domain of @xmath47 . @xmath611 , a=\left[\begin{array}{ccc } 0&0&\minus 1\\ 1&0&\minus 2\\ 0&1&\minus 2 \end{array}\right ] , b=\left[\begin{array}{ccc } 0&0&1\\ 1&0&\minus 2\\ 0&1&2 \end{array}\right].\ ] ] in this case @xmath54 is anisotropic over @xmath500 , and hence we turn to our strategy of approximating the fundamental domain to prove that @xmath47 is arithmetic : we produce the compact region shown in figure [ fundy6 ] which contains the fundamental domain of @xmath47 . 1 f. beukers , g. heckman , _ monodromy for hypergeometric function @xmath1_. invent . math . * 95 * , pp . 325 - 354 ( 1989 ) j. bober , _ integer ratios of factorials , hypergeometric functions , and related step functions_. ph.d . thesis , university of michigan ( 2009 ) c. brav , h. thomas , _ classical and derived monodromy of the quintic threefold_. preprint at http://people.maths.ox.ac.uk/brav/monodromy.pdf j.w.s . cassels , _ rational quadratic forms_. dover publications , inc . , mineola , ny ( 1978 ) j.h . conway , n.j.a . sloane , _ sphere packings , lattices and groups_. springer - verlag ( 1998 ) b. dwork , _ @xmath612-adic cycles_. publ . ihes , * 37 * , pp.27 - 115 ( 1969 ) j. elstrodt , f. grunewald , f and j. mennicke , _ groups acting on hyperbolic space . harmonic analysis and number theory . _ springer monographs in mathematics . springer - verlag , berlin ( 1998 ) e. fuchs , _ the ubiquity of thin groups . _ proceedings of msri conference on thin groups and super - strong approximation . lax , r. phillips , _ the asymptotic distribution of lattice points in euclidean and non - euclidean spaces_. jnl . 46 * , pp . 280 - 350 ( 1982 ) e. landau , _ sur les conditions de divisibilite dun produit de factorielles par un autre _ , nouv . , * 3 * , 19 , pp . 344 - 362 ( 1900 ) a.h.m levelt , _ hypergeometric functions . _ ph.d . thesis , university of amsterdam ( 1961 ) g. margulis , _ discrete subgroups of semi - simple lie groups _ , springer ( 1991 ) j. mckee , c. smyth , _ single polynomials that correspond to pairs of cyclotomic polynomials with interlacing zeros . eur . j. math . , * 11 * , pp . 882 - 899 ( 2013 ) v. nikulin , _ on factor groups of the automorphism groups of hyperbolic forms modulo subgroups generated by 2-reflections_. soviet math * 20 * , pp . 1156 - 1158 ( 1979 ) v. nikulin , _ self correspondences of @xmath613-surfaces via moduli of sheaves and arithmetic hyperbolic reflection groups_. arxiv : 0810.2945 h. park , ph.d . thesis , princeton university ( august 2013 ) . f. rodriguez - villegas , _ hypergeometric families of calabi - yau manifolds_. fields inst . commun . * 38 * , ams ( 2003 ) p. sarnak , _ notes on thin matrix groups_. arxiv : 1212.3525 p. sarnak , _ reciprocal geodesics_. clay math . * 7 * , pp . 217 - 237 ( 2007 ) s. singh , t.n . venkataramana , _ arithmeticity of certain symplectic hypergeometric groups_. arxiv : 1208.6460 ( 2012 ) t.n . venkataramana , _ arithmeticity of the braid group at roots of unity_. arxiv : 1204.4778 ( 2012 ) e. vinberg , _ on groups of unit elements of certain quadratic forms_. math . ussr sbornik * 16 * , no.1 , pp . 17 - 35 ( 1972 ) e. vinberg , _ the absence of crystallographic reflection groups in lobachevsky spaces of large dimension_. trudy moscow 47 * , pp . 68 - 102 ( 1984 )

this work presents the differential charmonia production cross sections in high energy p+p collisions calculated using nrqcd formalism . the nrqcd formalism , factorizes the quarkonia production cross sections in terms of short distance qcd cross sections and long distance matrix elements ( ldmes ) . the short distance cross sections are calculated in terms of perturbative qcd and ldmes are obtained by fitting the experimental data . measured transverse momentum distributions of ,(2s ) and j/ in p + collisions at 1.8 , 1.96 tev and in p+p collisions at 7 , 8 and 13 tev are used to constrain ldmes . the feed - down contribution to each state from the higher states are taken into account . the formalism provides a very good description of the data in a wide energy range . the values of ldmes are used to predict the charmonia cross sections in p+p collisions at 13 and 5 tev in kinematic bins relevant for the lhc detectors .

1606.08265

the quarkonia ( @xmath4 ) have provided useful tools for probing both perturbative and nonperturbative aspects of quantum chromodynamics ( qcd ) eversince the discovery of j/@xmath1 resonance @xcite . the quarkonia states are qualitatively different from most other hadrons since the velocity @xmath5 of the heavy constituents is small allowing a non - relativistic treatment of bound states . the quarkonia are the richest probes to study the bulk strongly interacting matter produced in heavy ion collisions @xcite . the ratios of excited to ground state quarkonia yields are considered as robust signals of quark gluon plasma ( qgp ) and measuring such ratios has become the most important goals of pb+pb experiments at lhc @xcite . the heavy quarks due to their high mass ( @xmath6 1.6 gev / c@xmath7 , @xmath8 4.5 gev / c@xmath7 ) , are produced in initial partonic collisions with sufficiently high momentum transfers . thus the heavy quark production can be treated perturbatively @xcite . the formation of quarkonia out of the two heavy quarks is a nonperturbative process and is treated in terms of different models @xcite . most notable models for quarkonia production are the color - singlet model ( csm ) , the color - evaporation model ( cem ) , the non - relativistic qcd ( nrqcd ) factorization approach , and the fragmentation - function approach . in the csm @xcite , it is assumed that the @xmath4 pair that evolves into the quarkonium is in a color - singlet state and has the same spin and angular - momentum as the quarkonium . the production rate of quarkonium state is related to the absolute values of the color - singlet @xmath4 wave function and its derivatives , evaluated at zero @xmath4 separation . these quantities can be extracted by comparing calculated quarkonium decay rates in the csm with the experimental measurements . the csm was successful in predicting quarkonium production rates at relatively low energy @xcite but , at high energies , very large corrections appear at next - to - leading order ( nlo ) and next - to - next - to - leading order ( nnlo ) in @xmath9 @xcite . the nrqcd factorization approach comprise the color - singlet model , but also includes color - octet states . in the cem @xcite , it is assumed that the produced @xmath4 pair evolves into a quarkonium if its invariant mass is less than the threshold for producing a pair of open - flavor heavy mesons . the nonperturbative probability for the @xmath4 pair to evolve into a quarkonium state is fixed by comparison with the measured production cross section of that quarkonium state . the cem calculations provide good descriptions of the cdf data for j/@xmath1 , @xmath1(2s ) , and @xmath0 production at @xmath10 tev @xcite but it fails to predict the quarkonium polarization . in the nrqcd factorization approach @xcite , the probability for a @xmath4 pair to evolve into a quarkonium is expressed as matrix elements of nrqcd operators in terms of the heavy - quark velocity @xmath5 in the limit @xmath11 . this approach takes into account the complete structure of the @xmath4 fock space , which is spanned by the state @xmath12}$ ] with spin @xmath13 , orbital angular momentum @xmath14 , total angular momentum @xmath15 , and color multiplicity @xmath16 = 1 ( color - singlet ) , 8 ( color - octet ) . the @xmath4 pairs which are produced at short distances in color - octet ( co ) states , evolve into physical , color - singlet ( cs ) quarkonia by emitting soft gluons nonperturbatively . in the limit @xmath17 , the csm is recovered in the case of s - wave quarkonia . the short distance cross sections can be calculated within the framework of perturbative qcd ( pqcd ) . the long distance matrix elements ( ldme ) corresponding to the probability of the @xmath18 state to convert to the quarkonium can be estimated by comparison with the experimental measurements . the leading order ( lo ) nrqcd gives a good description of j/@xmath1 yields at tevatron rhic and lhc energies @xcite . the nlo corrections to color - singlet j/@xmath1 production have been investigated in refs . the color - singlet j/@xmath1 production is found to be enhanced by 2 - 3 order of magnitude in high @xmath19 region @xcite . the nlo corrections to j/@xmath1 production via s - wave color octet ( co ) states ( @xmath20}\,^3s_{1}^{[8]}$ ] ) are studied in ref . @xcite and the corrections to @xmath21 distributions of both j/@xmath1 yield and polarization are found to be small . in refs . @xcite , nlo corrections for @xmath22 hadroproduction are also studied . several nlo calculations are performed to obtain the polarization and yield of j/@xmath1 . the j/@xmath1 polarization presents a rather confusing pattern @xcite . the works of ref . @xcite and ref . @xcite present nlo - nrqcd calculations of j/@xmath1 yields . in both the works , the set of co ldmes fitted to @xmath21 distributions measured at hera and cdf are used to describe the @xmath21 distributions from rhic and the lhc . the fitted ldmes of ref . @xcite and ref . @xcite are incompatible with each other . a recent work @xcite gives calculations for both the yields and polarizations of charmonia at the tevatron and the lhc where the ldmes are obtained by fitting the tevatron data only . with the lhc running for several years we now have very high quality quarkonia production data in several kinematic regions up to very high transverse momentum which could be used to constrain the ldmes . in this paper , we use cdf data @xcite along with new lhc data @xcite to constrain the ldmes . the feed - down contribution to each state from the higher states are taken into account . these new ldmes are then used to predict the j/@xmath1 and @xmath1(2s ) cross - section at 13 tev and 5 tev for the kinematical bins relevant to lhc detectors . the nlo calculations are still evolving and thus we use lo calculations in this work . the values of fitted ldmes with lo formulations are always useful for straightforward predictions of quarkonia cross section and for the purpose of a comparison with those obtained using nlo formulations . the uncertainties in the ldmes due to nlo are quantified . the nrqcd formalism provides a theoretical framework for studying the heavy quarkonium production . the dominant processes in the production of heavy mesons @xmath1 are @xmath24 , @xmath25 and @xmath26 . we represent these processes by @xmath27 , where @xmath16 and @xmath28 are the light incident partons . the invariant cross - section for the production of a heavy meson @xmath1 can be written in a factorized form as @xmath29 where @xmath30 is the distribution function ( pdf ) of the incoming parton @xmath31 in the incident proton , which depends on the momentum fraction @xmath32 and the factorization scale @xmath33 . the parton level mandelstam variables @xmath34 , @xmath35 , and @xmath36 can be expressed in terms of @xmath37 , @xmath38 as @xmath39 where @xmath40 being the total energy in the centre - of - mass , @xmath41 is the rapidity and @xmath21 is the transverse momentum of the @xmath4 pair . the mass of heavy meson is represented by m and @xmath42 is the transverse mass defined as @xmath43 . writing down @xmath44 and solving for @xmath45 we obtain @xmath46 the double differential cross - section upon @xmath21 and @xmath41 then is obtained as @xmath47 where the minimum value of @xmath37 is given by @xmath48 the parton level cross - section @xmath49 is defined as @xcite @xmath50 the short distance contribution @xmath51 corresponds to the production of a @xmath18 pair in a particular color and spin configuration can be calculated within the framework of perturbative qcd ( pqcd ) . the long distance matrix elements ( ldme ) @xmath52 corresponds to the probability of the @xmath18 state to convert to the quarkonium wavefunction and can be estimated by comparison with experimental measurements . the short distance invariant differential cross - section is given by @xmath53 where @xmath54 is the feynman squared amplitude . we use the expressions for the short distance cs cross - sections given in refs . @xcite and the co cross - sections given in refs . @xcite which we reproduce in the appendix a. the cteq6 m @xcite parametrization is used for parton distribution functions . the ldmes scale with a definite power of the relative velocity @xmath5 of the heavy quarks inside @xmath4 bound states . in the limit @xmath55 , the production of quarkonium is based on the @xmath56}$ ] and @xmath57}$ ] ( @xmath15 = 0,1,2 ) cs states and @xmath58}$ ] , @xmath59}$ ] and @xmath60}$ ] co states . the differential cross section for the direct production of @xmath61 can be written as the sum of these contributions , @xmath62_{1 } ) ) \,m_{l}(q{{\bar{q}}}([^3s_1]_{1})\rightarrow j/\psi ) + d\sigma(q{{\bar{q}}}([^1s_0]_{8 } ) ) \,m_{l}(q{{\bar{q}}}([^1s_0]_{8})\rightarrow j/\psi)\\ & + d\sigma(q{{\bar{q}}}([^3s_1]_{8 } ) ) \,m_{l}(q{{\bar{q}}}([^3s_1]_{8})\rightarrow j/\psi ) + d\sigma(q{{\bar{q}}}([^3p_0]_{8 } ) ) \,m_{l}(q{{\bar{q}}}([^3p_0]_{8})\rightarrow j/\psi)\\ & + d\sigma(q{{\bar{q}}}([^3p_1]_{8 } ) ) \,m_{l}(q{{\bar{q}}}([^3p_1]_{8})\rightarrow j/\psi ) + d\sigma(q{{\bar{q}}}([^3p_2]_{8 } ) ) \,m_{l}(q{{\bar{q}}}([^3p_2]_{8})\rightarrow j/\psi)\\ & + \cdot\cdot\cdot \label{eq : dsigmaj } \end{split}\ ] ] the dots represent contribution of terms at higher powers of @xmath5 . the contributions from the co matrix elements in eq . [ eq : dsigmaj ] are suppressed by @xmath63 compared to the cs matrix elements . for the case of the @xmath64-wave bound states @xmath22 ( @xmath65 , @xmath66 and @xmath67 ) , the color - singlet state @xmath68_{1}$ ] and the color - octet state @xmath69_{8}$ ] contribute to the same order in @xmath5 ( @xmath70 ) because of the angular momentum barrier for the @xmath71wave states , and hence both need to be included . the @xmath72 differential cross section thus can be written as @xmath73_{1 } ) ) \,m_{l}(q{{\bar{q}}}([^3p_j]_{1})\rightarrow \chi_{cj } ) + d\sigma(q{{\bar{q}}}([^3s_1]_{8 } ) ) \,m_{l}(q{{\bar{q}}}([^3s_1]_{8})\rightarrow \chi_{cj})\\ & + \cdot\cdot\cdot \label{eq : dsigmachi } \end{split}\ ] ] the prompt @xmath61 production at lhc energies consists of direct @xmath61 production from the initial parton - parton hard scattering and the feed - down contributions to the @xmath61 from the decay of heavier charmonium states @xmath74 , @xmath65 , @xmath66 and @xmath67 . the relevant branching fractions are given in the table [ table : charmoniabfs ] @xcite . the prompt @xmath74 has no significant feed - down contributions from the higher mass states . . [ cols="^,^,^,^,^",options="header " , ] [ table : ldmejpsi ] table [ table : ldmechic0 ] shows @xmath65 ldmes extracted in present analysis along with the results from other analysis . the value of charm quark mass as well as the pdfs used in the calculations are also shown in the table . the short distance calculations are at lo except the last row in the table . we have made major extension in fitting the @xmath72 ldme . all the earlier calculations @xcite use only cdf data to fit the @xmath72 ldme . we use cdf data @xcite along - with the data from lhc @xcite to constrain the co ldme of @xmath72 . the new high energy lhc data require larger value of @xmath75_{8})\rightarrow \chi_{c0}))$ ] to fit the data . table [ table : ldmepsi2s ] shows @xmath1(2s ) ldmes extracted in present analysis along with the results from other works . all the calculations are at lo in @xmath76 . the calculations in ref . @xcite use only cdf data to fit the ldmes while the ref . @xcite uses cdf and lhc data at mid - rapidity . in our analysis we use data from cdf @xcite and lhc data in mid rapidity @xcite as well as lhc data in forward rapidity @xcite , both the datasets covering a much wider @xmath19 range . our value of matrix element @xmath77_{8}\rightarrow \psi(2s)))$ ] is similar with other analysis . the value of linear - combination , @xmath78_{8},[^3p_0]_{8})\rightarrow \psi(2s))$ ] , varies significantly from 0.0022 to 0.01246 between different analysis . as it can be seen from table [ table : ldmepsi2s ] the high energy lhc data require larger value of @xmath78_{8},[^3p_0]_{8})\rightarrow \psi(2s))$ ] . table [ table : ldmejpsi ] shows j/@xmath1 ldmes extracted in present analysis along with the results from other works . all the calculations except ref . @xcite are at lo in @xmath76 . the calculations in ref . @xcite use only cdf data to fit the ldmes while ref . @xcite uses cdf , rhic and lhc data at mid - rapidity . in our analysis , we use data from cdf @xcite and lhc data in mid rapidity @xcite as well as lhc data in forward rapidity @xcite , both the datasets covering a much wider @xmath19 range . the value of the matrix element @xmath77_{8}\rightarrow j/\psi))$ ] is different in different analysis . the large error present on @xmath77_{8}\rightarrow j/\psi))$ ] in ref . @xcite is significantly improved by our simultaneous fitting of several datasets . the value of linear - combination , @xmath78_{8},[^3p_0]_{8})\rightarrow j/\psi)$ ] , varies significantly from 0.0114 to 0.06384 ( our value ) between different analysis at lo . the nlo analysis @xcite does not fit the linear combination but fit both @xmath78_{8}\rightarrow j/\psi))$ ] and @xmath79_{8}\rightarrow j/\psi))$ ] ) ldmes independently and their values are given as [email protected] and [email protected] respectively . the value of @xmath78_{8},[^3p_0]_{8})\rightarrow j/\psi)$ ] is very small for ref . @xcite because of the negative value of @xmath79_{8}\rightarrow j/\psi))$ ] . we use our newly constrained co ldmes shown in equation [ eq : coldme_jpsi ] to predict the j/@xmath1 cross - section at 13 tev and 5 tev for the kinematical bins relevant to lhc detectors . figure [ fig : sigmajpsi ] shows the nrqcd calculations of production cross section of @xmath61 in p+p collisions as a function of transverse momentum at @xmath40 = 13 tev . figure [ fig : sigmajpsi5tev ] is same as fig . [ fig : sigmajpsi ] but at @xmath40 = 5 tev . both the figures give calculations in the kinematic bins relevant for ( a ) cms , atlas and ( b ) alice , lhcb detectors at lhc . for the j/@xmath1 meson all the relevant contributions from higher mass states are also shown . we have presented nrqcd calculations for the differential production cross sections of prompt j/@xmath1 and prompt @xmath1(2s ) in p+p collisions . for the j/@xmath1 meson , all the relevant contributions from higher mass states are estimated . measured transverse momentum distributions of @xmath1(2s ) , @xmath0 and j/@xmath1 in p + @xmath2 collisions at @xmath3 1.8 , 1.96 tev and in p+p collisions at 7 , 8 and 13 tev are used to constrain ldmes . the calculations for prompt j/@xmath1 and prompt @xmath1(2s ) are compared with the measured data at tevatron and lhc . the formalism provides a very good description of the data in wide energy range . the values of ldmes are used to predict the charmonia cross sections in p+p collisions at 13 and 5 tev in kinematic bins relevant for lhc detectors . we compare the ldmes for charmonia obtained in this analysis with the results from earlier works . the high energy lhc data require a smaller value of the ldme @xmath77_{8}\rightarrow \psi))$ ] and a larger value for the combination @xmath78_{8},[^3p_0]_{8})\rightarrow \psi)$ ] of ldmes . we acknowledge the fruitful discussions on this topic with rishi sharma . here we list the leading order qcd cross sections for the resonance production used in our calculations . we write the formulas in terms of the invariants @xmath81 , @xmath82 , @xmath83 . to order @xmath84 one only has the gluon fusion processes , g g @xmath85 which gives resonance with very small @xmath86 , and hence are not used in our calculations . to order @xmath87 , on the other hand , one has typically two - by - two scattering processes . we reproduced relevant cross sections from refs . @xcite below : * g q @xmath88 q or ( q@xmath89 @xmath90 * q @xmath91 @xmath88 g @xmath92 we define two new variables as a combination of @xmath81 , @xmath82 and @xmath83 which can be used to define the g g @xmath88 g cross sections . @xmath93 @xmath94 @xmath95}{(q - m^2p)^3}\\ \frac{d\sigma}{d{{\hat{t}}}}(^3p_0 ) = ~&\frac{4\pi { { \alpha_{s}}}^3 ( r_{1}^\prime)^{2}}{m^3{{\hat{s}}}^2}\frac{1}{q(q - m^2p)^4}[9m^4p^4(m^8 - 2m^4p+p^2)\\ ~&-6m^2p^3q(2m^8 - 5m^4p+p^2)\\ ~&-p^2q^2(m^8 + 2m^4p - p^2)\\ ~&+2m^2pq^3(m^4-p)+6m^4q^4 ] \end{split}\ ] ] @xmath96 we list below short distance squared amplitudes for @xmath97 scattering processes which mediate color - octet quarkonia production . these expressions are averaged over initial spins and colors of the two incident partons . the helicity levels of outgoing @xmath98 and @xmath99 pairs are labeled by the subscript @xmath100 . * q @xmath91 @xmath101 $ ] g @xmath102}}g ) |^2 & = { 5 ( 4\pi { { \alpha_{s}}})^3 \over 27 m } { { { \hat{t}}}^2+{{\hat{u}}}^2 \over { { \hat{s}}}({{\hat{s}}}-m^2)^2}\\ \mathop{{\bar{\sum_{h=0 } } } } | { { \cal a}}(q{{\bar{q}}}\to \qqbaroctettrips g ) |^2 & = { 8 ( 4\pi { { \alpha_{s}}})^3 \over 81 m^3 } { m^2 { \hat{\rm s}}\over ( { \hat{\rm s}}-m^2)^4 } \bigl[4({\hat{\rm t}}^2+{\hat{\rm u}}^2)-{\hat{\rm t}}{\hat{\rm u}}\bigr]\\ \mathop{{\bar{\sum_{|h|=1 } } } } | { { \cal a}}(q{{\bar{q}}}\to \qqbaroctettrips g ) |^2 & = { 2 ( 4\pi { { \alpha_{s}}})^3 \over 81 m^3 } { { \hat{\rm s}}^2+m^4 \over ( { \hat{\rm s}}-m^2)^4 } { { \hat{\rm t}}^2+{\hat{\rm u}}^2 \over { \hat{\rm t}}{\hat{\rm u } } } \bigl[4({\hat{\rm t}}^2+{\hat{\rm u}}^2)-{\hat{\rm t}}{\hat{\rm u}}\bigr ] \end{split}\ ] ] @xmath103 @xmath104 * g q @xmath101 $ ] q @xmath105}}q ) |^2 & = -{5 ( 4\pi { { \alpha_{s}}})^3 \over 72 m } { { \hat{\rm s}}^2+{\hat{\rm u}}^2 \over { \hat{\rm t}}({\hat{\rm t}}-m^2)^2}\\ \mathop{{\bar{\sum_{h=0 } } } } | { { \cal a}}(gq \to \qqbaroctettrips q ) |^2 & = -{(4\pi { { \alpha_{s}}})^3 \over 54 m^3 } { m^2 { \hat{\rm t}}\bigl[4({\hat{\rm s}}^2+{\hat{\rm u}}^2)-{\hat{\rm s}}{\hat{\rm u}}\bigr ] \over \bigl[({\hat{\rm s}}-m^2)({\hat{\rm t}}-m^2)\bigr]^2}\\ \mathop{{\bar{\sum_{|h|=1 } } } } | { { \cal a}}(gq \to \qqbaroctettrips q ) |^2 & = -{(4\pi { { \alpha_{s}}})^3 \over 108 m^3 } \\ & \times { ( { \hat{\rm s}}^2+{\hat{\rm u}}^2 + 2 m^2 { \hat{\rm t}})({\hat{\rm s}}-m^2)^2- 2 m^2 { \hat{\rm s}}{\hat{\rm t}}{\hat{\rm u}}\over { \hat{\rm s}}{\hat{\rm u}}\bigl[({\hat{\rm s}}-m^2)({\hat{\rm t}}-m^2)\bigr]^2 } \\ & \times \bigl[4({\hat{\rm s}}^2+{\hat{\rm u}}^2)-{\hat{\rm s}}{\hat{\rm u}}\bigr ] \end{split}\ ] ] + @xmath106 \over({\hat{\rm t}}-m^2)^4 ( { \hat{\rm s}}-m^2)^2}\\ \mathop{{\bar{\sum_{|h|=1 } } } } | { { \cal a}}(gq \to \qqbaroctetpone q ) |^2 & = -{20 ( 4\pi { { \alpha_{s}}})^3 \over 27 m^3 } { m^2 { \hat{\rm s}}{\hat{\rm u}}({\hat{\rm t}}^2+{\hat{\rm t}}{\hat{\rm u}}+{\hat{\rm u}}^2 ) \over ( { \hat{\rm t}}-m^2)^4 ( { \hat{\rm s}}-m^2)^2 } \end{split}\ ] ] + @xmath107\\ \mathop{{\bar{\sum_{|h|=1 } } } } | { { \cal a}}(gq \to \qqbaroctetptwo q ) |^2 & = -{4 ( 4\pi { { \alpha_{s}}})^3 \over 9 m^3 } { m^2 { \hat{\rm s}}{\hat{\rm u}}\over ( { \hat{\rm t}}-m^2)^4 } \\ & \times { ( { \hat{\rm s}}-m^2)^2 ( { \hat{\rm s}}^2+m^4 ) - ( { \hat{\rm s}}+m^2)^2 { \hat{\rm t}}{\hat{\rm u}}\over ( { \hat{\rm s}}-m^2)^4}\\ \mathop{{\bar{\sum_{|h|=2 } } } } | { { \cal a}}(gq \to \qqbaroctetptwo q ) |^2 & = -{2 ( 4\pi { { \alpha_{s}}})^3 \over 9 m^3 } { m^4\over { \hat{\rm t}}({\hat{\rm t}}-m^2)^4 } \\ & \times \bigl [ { \hat{\rm s}}^2+{\hat{\rm u}}^2 + 2 { \hat{\rm s}}^2 { \hat{\rm t}}{\hat{\rm u}}{({\hat{\rm s}}-m^2)(2 { \hat{\rm t}}+{\hat{\rm u } } ) - { \hat{\rm u}}^2 \over ( { \hat{\rm s}}-m^2)^4 } \bigr ] \end{split}\ ] ] * g g @xmath101 $ ] g ( the @xmath108}}\ g$ ] squared amplitudes are expressed in terms of the variables @xmath109 and @xmath110 . ) @xmath111}}g ) |^2 & = { 5 ( 4\pi { { \alpha_{s}}})^3 \over 16 m } \bigl [ { \hat{\rm s}}^2 ( { \hat{\rm s}}-m^2)^2 + { \hat{\rm s}}{\hat{\rm t}}{\hat{\rm u}}(m^2 - 2{\hat{\rm s } } ) + ( { \hat{\rm t}}{\hat{\rm u}})^2 \bigr ] \\ & \times { ( { \hat{\rm s}}^2-m^2 { \hat{\rm s}}+ m^4)^2-{\hat{\rm t}}{\hat{\rm u}}(2 { \hat{\rm t}}^2 + 3 { \hat{\rm t}}{\hat{\rm u}}+ 2 { \hat{\rm u}}^2 ) \over { \hat{\rm s}}{\hat{\rm t}}{\hat{\rm u}}\bigl [ ( { \hat{\rm s}}-m^2 ) ( { \hat{\rm t}}-m^2 ) ( { \hat{\rm u}}-m^2 ) \bigr]^2}\\ \mathop{{\bar{\sum_{h=0 } } } } | { { \cal a}}(g g \to \qqbaroctettrips g ) |^2 & = -{(4\pi { { \alpha_{s}}})^3 \over 144 m^3}{2 m^2 { \hat{\rm s}}\over ( { \hat{\rm s}}-m^2)^2 } ( { \hat{\rm t}}^2+{\hat{\rm u}}^2 ) { \hat{\rm t}}{\hat{\rm u}}\\ & \times { 27({\hat{\rm s}}{\hat{\rm t}}+{\hat{\rm t}}{\hat{\rm u}}+{\hat{\rm u}}{\hat{\rm s}})-19m^4 \over \bigl[({\hat{\rm s}}-m^2)({\hat{\rm t}}-m^2)({\hat{\rm u}}-m^2)\bigr]^2 } \end{split}\ ] ] + @xmath112\\ & \times { 27({\hat{\rm s}}{\hat{\rm t}}+{\hat{\rm t}}{\hat{\rm u}}+{\hat{\rm u}}{\hat{\rm s}})-19m^4 \over \bigl[({\hat{\rm s}}-m^2)({\hat{\rm t}}-m^2)({\hat{\rm u}}-m^2)\bigr]^2 } \end{split}\ ] ] + @xmath113}\\ & \times \bigl\{{\hat{\rm s}}^2 { \hat{\rm z}}^4 ( { \hat{\rm s}}^2-{\hat{\rm z}}^2)^4 + m^2 { \hat{\rm s}}{\hat{\rm z}}^2 ( { \hat{\rm s}}^2-{\hat{\rm z}}^2)^2 ( 3{\hat{\rm s}}^2 - 2{\hat{\rm z}}^2)(2{\hat{\rm s}}^4 - 6 { \hat{\rm s}}^2 { \hat{\rm z}}^2 + 3 { \hat{\rm z}}^4)\\ & + m^4 \bigl [ 9{\hat{\rm s}}^{12 } - 84 { \hat{\rm s}}^{10 } { \hat{\rm z}}^2 + 265 { \hat{\rm s}}^8 { \hat{\rm z}}^4 - 382 { \hat{\rm s}}^6 { \hat{\rm z}}^6 + 276 { \hat{\rm s}}^4 { \hat{\rm z}}^8 - 88 { \hat{\rm s}}^2 { \hat{\rm z}}^{10}+ 9 { \hat{\rm z}}^{12 } \bigr]\\ & - m^6 { \hat{\rm s}}\bigl [ 54 { \hat{\rm s}}^{10 } - 357 { \hat{\rm s}}^8 { \hat{\rm z}}^2 + 844 { \hat{\rm s}}^6 { \hat{\rm z}}^4 - 898 { \hat{\rm s}}^4 { \hat{\rm z}}^6 + 439 { \hat{\rm s}}^2 { \hat{\rm z}}^8- 81 { \hat{\rm z}}^{10 } \bigr ] \\ & + m^8 \bigl [ 153 { \hat{\rm s}}^{10 } - 798 { \hat{\rm s}}^8 { \hat{\rm z}}^2 + 1415 { \hat{\rm s}}^6 { \hat{\rm z}}^4 - 1041 { \hat{\rm s}}^4 { \hat{\rm z}}^6 + 301 { \hat{\rm s}}^2 { \hat{\rm z}}^8 - 18 { \hat{\rm z}}^{10 } \bigr ] \\ & -m^{10 } { \hat{\rm s}}\bigl [ 270 { \hat{\rm s}}^8 - 1089 { \hat{\rm s}}^6 { \hat{\rm z}}^2 + 1365 { \hat{\rm s}}^4 { \hat{\rm z}}^4 - 616 { \hat{\rm s}}^2 { \hat{\rm z}}^6 + 87 { \hat{\rm z}}^8 \bigr ] \\ & + m^{12 } \bigl [ 324 { \hat{\rm s}}^8 - 951 { \hat{\rm s}}^6 { \hat{\rm z}}^2 + 769 { \hat{\rm s}}^4 { \hat{\rm z}}^4 - 189 { \hat{\rm s}}^2 { \hat{\rm z}}^6 + 9 { \hat{\rm z}}^8 \bigr ] \\ & - 9 m^{14 } { \hat{\rm s}}\bigl [ ( 6 { \hat{\rm s}}^2 - { \hat{\rm z}}^2 ) ( 5 { \hat{\rm s}}^4 - 9 { \hat{\rm s}}^2 { \hat{\rm z}}^2 + 3 { \hat{\rm z}}^4 ) \bigr]\\ & + 3 m^{16 } { \hat{\rm s}}^2 \bigl[51 { \hat{\rm s}}^4 - 59 { \hat{\rm s}}^2 { \hat{\rm z}}^2 + 12 { \hat{\rm z}}^4\bigr]\\ & - 27 m^{18 } { \hat{\rm s}}^3 \bigl[2{\hat{\rm s}}^2 - { \hat{\rm z}}^2 \bigr]\\ & + 9 m^{20 } { \hat{\rm s}}^4 \bigr\ } \end{split}\ ] ] + @xmath114}\\ & \times { \hat{\rm s}}{\hat{\rm z}}^2 \bigl[({\hat{\rm s}}^2-{\hat{\rm z}}^2)^2 - 2 m^2 { \hat{\rm s}}{\hat{\rm z}}^2 - m^4 ( { \hat{\rm s}}^2 + 2 { \hat{\rm z}}^2 ) + m^8 \bigr ] \\ & \times \bigl[({\hat{\rm s}}^2-{\hat{\rm z}}^2)^2 - m^2 { \hat{\rm s}}(2{\hat{\rm s}}^2 - { \hat{\rm z}}^2 ) + m^4 { \hat{\rm s}}^2 \bigr ] \end{split}\ ] ] + @xmath115}\\ & \times m^2 \bigl\ { 2({\hat{\rm s}}^2-{\hat{\rm z}}^2)^2 ( { \hat{\rm s}}^6 - 4 { \hat{\rm s}}^4{\hat{\rm z}}^2 + { \hat{\rm s}}^2 { \hat{\rm z}}^4 - { \hat{\rm z}}^6 ) \\ & - m^2 { \hat{\rm s}}(2 { \hat{\rm s}}^2-{\hat{\rm z}}^2 ) ( 5 { \hat{\rm s}}^6 - 17 { \hat{\rm s}}^4{\hat{\rm z}}^2 + 9 { \hat{\rm s}}^2 { \hat{\rm z}}^4 - { \hat{\rm z}}^6 ) \\ & + m^4 ( 21{\hat{\rm s}}^8 - 49 { \hat{\rm s}}^6 { \hat{\rm z}}^2 + 21 { \hat{\rm s}}^4 { \hat{\rm z}}^4 - 4 { \hat{\rm s}}^2 { \hat{\rm z}}^6 + { \hat{\rm z}}^8 ) \\ & - m^6 { \hat{\rm s}}(24 { \hat{\rm s}}^6 - 30 { \hat{\rm s}}^4 { \hat{\rm z}}^2 + 6 { \hat{\rm s}}^2 { \hat{\rm z}}^4 - { \hat{\rm z}}^6 ) \\ & + m^8 { \hat{\rm s}}^2 ( 16 { \hat{\rm s}}^4 - 9 { \hat{\rm s}}^2 { \hat{\rm z}}^2 + 2 { \hat{\rm z}}^4)\\ & - m^{10 } { \hat{\rm s}}^3 ( 6 { \hat{\rm s}}^2 - { \hat{\rm z}}^2 ) \\ & + m^{12 } { \hat{\rm s}}^4 \bigr\ } \end{split}\ ] ] + @xmath116}\\ & \bigl\ { { \hat{\rm s}}^2 ( { \hat{\rm s}}^2-{\hat{\rm z}}^2)^4 - m^2 { \hat{\rm s}}{\hat{\rm z}}^2 ( { \hat{\rm s}}^2-{\hat{\rm z}}^2)^2(11 { \hat{\rm s}}^2 + 2 { \hat{\rm z}}^2 ) \\ & + m^4 \bigl [ { \hat{\rm s}}^8 - 12 { \hat{\rm s}}^6 { \hat{\rm z}}^2 + 41 { \hat{\rm s}}^4 { \hat{\rm z}}^4 - 20 { \hat{\rm s}}^2 { \hat{\rm z}}^6 + { \hat{\rm z}}^8 \bigr ] \\ & - m^6 { \hat{\rm s}}\bigl [ 4{\hat{\rm s}}^6 - 26 { \hat{\rm s}}^4 { \hat{\rm z}}^2 - { \hat{\rm s}}^2 { \hat{\rm z}}^4 - 5 { \hat{\rm z}}^6 \bigr ] \\ & + m^8 \bigl [ 29 { \hat{\rm s}}^6 - 114 { \hat{\rm s}}^4 { \hat{\rm z}}^2 + 108 { \hat{\rm s}}^2 { \hat{\rm z}}^4 - 10 { \hat{\rm z}}^6 \bigr ] \\ & - m^{10 } { \hat{\rm s}}\bigl [ 65 { \hat{\rm s}}^4 - 104 { \hat{\rm s}}^2 { \hat{\rm z}}^2 -33 { \hat{\rm z}}^4 \bigr ] \\ & + m^{12 } \bigl [ 54 { \hat{\rm s}}^4 - 20 { \hat{\rm s}}^2 { \hat{\rm z}}^2 + 7 { \hat{\rm z}}^4 \bigr ] \\ & - m^{14 } { \hat{\rm s}}\bigl[23 { \hat{\rm s}}^2 + 5 { \hat{\rm z}}^2\bigr]\\ & + 7 m^{16 } { \hat{\rm s}}^2 \bigr\ } \end{split}\ ] ] + @xmath117}\\ & \times \bigl\{2 { \hat{\rm s}}^2 ( { \hat{\rm s}}^2-{\hat{\rm z}}^2)^2 ( { \hat{\rm s}}^6 - 4 { \hat{\rm s}}^4 { \hat{\rm z}}^2 + { \hat{\rm s}}^2 { \hat{\rm z}}^4 - { \hat{\rm z}}^6 ) \\ & - m^2 { \hat{\rm s}}\bigl [ 10 { \hat{\rm s}}^{10 } - 37 { \hat{\rm s}}^8 { \hat{\rm z}}^2 + 19 { \hat{\rm s}}^6 { \hat{\rm z}}^4 + 11 { \hat{\rm s}}^4 { \hat{\rm z}}^6 - { \hat{\rm s}}^2 { \hat{\rm z}}^8 - 4 { \hat{\rm z}}^{10 } \bigr ] \\ & + m^4 \bigl [ 25 { \hat{\rm s}}^{10 } - 61 { \hat{\rm s}}^8 { \hat{\rm z}}^2 + 27 { \hat{\rm s}}^6 { \hat{\rm z}}^4 - 34 { \hat{\rm s}}^4 { \hat{\rm z}}^6 + 23 { \hat{\rm s}}^2 { \hat{\rm z}}^8 - 2 { \hat{\rm z}}^{10 } \bigr ] \\ & - m^6 { \hat{\rm s}}\bigl [ 42 { \hat{\rm s}}^8 - 77 { \hat{\rm s}}^6 { \hat{\rm z}}^2 + 41 { \hat{\rm s}}^4 { \hat{\rm z}}^4 - 22 { \hat{\rm s}}^2 { \hat{\rm z}}^6 + 17 { \hat{\rm z}}^8 \bigr ] \\ & + m^8 \bigl [ 53 { \hat{\rm s}}^8 - 88 { \hat{\rm s}}^6 { \hat{\rm z}}^2 + 69 { \hat{\rm s}}^4 { \hat{\rm z}}^4 - 68 { \hat{\rm s}}^2 { \hat{\rm z}}^6 + 3 { \hat{\rm z}}^8 \bigr ] \\ & - m^{10 } { \hat{\rm s}}\bigl [ 54 { \hat{\rm s}}^6 - 85 { \hat{\rm s}}^4 { \hat{\rm z}}^2 + 60 { \hat{\rm s}}^2 { \hat{\rm z}}^4 - 9 { \hat{\rm z}}^6 \bigr ] \\ & + m^{12 } { \hat{\rm s}}^2 \bigl [ 43 { \hat{\rm s}}^4 - 47 { \hat{\rm s}}^2 { \hat{\rm z}}^2 + 20 { \hat{\rm z}}^4 \bigr ] \\ & - m^{14 } { \hat{\rm s}}^3 \bigl[22 { \hat{\rm s}}^2 - 9 { \hat{\rm z}}^2\bigr]\\ & + 5 m^{16 } { \hat{\rm s}}^4 \bigr\ } \end{split}\ ] ] + @xmath118}\\ & \times \bigl\ { 2 { \hat{\rm s}}^2 \bigl [ { \hat{\rm s}}^{12 } - 8 { \hat{\rm s}}^{10 } { \hat{\rm z}}^2 + 22 { \hat{\rm s}}^8 { \hat{\rm z}}^4 - 24 { \hat{\rm s}}^6 { \hat{\rm z}}^6 + 10 { \hat{\rm s}}^4 { \hat{\rm z}}^8 - 3 { \hat{\rm s}}^2 { \hat{\rm z}}^{10 } + { \hat{\rm z}}^{12 } \bigr ] \\ & - m^2 { \hat{\rm s}}\bigl [ 16 { \hat{\rm s}}^{12 } - 102 { \hat{\rm s}}^{10 } { \hat{\rm z}}^2 + 210 { \hat{\rm s}}^8 { \hat{\rm z}}^4 - 153 { \hat{\rm s}}^6 { \hat{\rm z}}^6 + 36 { \hat{\rm s}}^4 { \hat{\rm z}}^8 - 6 { \hat{\rm s}}^2 { \hat{\rm z}}^{10 } + 4 { \hat{\rm z}}^{12 } \bigr ] \\ & + m^4 \bigl [ 60 { \hat{\rm s}}^{12 } - 306 { \hat{\rm s}}^{10 } { \hat{\rm z}}^2 + 482 { \hat{\rm s}}^8 { \hat{\rm z}}^4 - 271 { \hat{\rm s}}^6 { \hat{\rm z}}^6 + 77 { \hat{\rm s}}^4 { \hat{\rm z}}^8 - 18 { \hat{\rm s}}^2 { \hat{\rm z}}^{10 } + 2 { \hat{\rm z}}^{12 } \bigr ] \\ & - m^6 { \hat{\rm s}}\bigl [ 140 { \hat{\rm s}}^{10 } - 573 { \hat{\rm s}}^8 { \hat{\rm z}}^2 + 710 { \hat{\rm s}}^6 { \hat{\rm z}}^4 - 344 { \hat{\rm s}}^4 { \hat{\rm z}}^6 + 91 { \hat{\rm s}}^2 { \hat{\rm z}}^8 - 18 { \hat{\rm z}}^{10 } \bigr ] \\ & + m^8 \bigl [ 226 { \hat{\rm s}}^{10 } - 741 { \hat{\rm s}}^8 { \hat{\rm z}}^2 + 737 { \hat{\rm s}}^6 { \hat{\rm z}}^4 - 310 { \hat{\rm s}}^4 { \hat{\rm z}}^6 + 77 { \hat{\rm s}}^2 { \hat{\rm z}}^8 - 4 { \hat{\rm z}}^{10 } \bigr ] \\ & - m^{10 } { \hat{\rm s}}\bigl [ 264 { \hat{\rm s}}^8 - 686 { \hat{\rm s}}^6 { \hat{\rm z}}^2 + 541 { \hat{\rm s}}^4 { \hat{\rm z}}^4 - 177 { \hat{\rm s}}^2 { \hat{\rm z}}^6 + 25 { \hat{\rm z}}^8 \bigr ] \\ & + m^{12 } \bigl [ 226 { \hat{\rm s}}^8 - 452 { \hat{\rm s}}^6 { \hat{\rm z}}^2 + 261 { \hat{\rm s}}^4 { \hat{\rm z}}^4 - 55 { \hat{\rm s}}^2 { \hat{\rm z}}^6 + 2 { \hat{\rm z}}^8 \bigr ] \\ & - m^{14 } { \hat{\rm s}}\bigl [ 140 { \hat{\rm s}}^6 - 201 { \hat{\rm s}}^4 { \hat{\rm z}}^2 + 71 { \hat{\rm s}}^2 { \hat{\rm z}}^4 - 6 { \hat{\rm z}}^6 \bigr ] \\ & + m^{16 } { \hat{\rm s}}^2 \bigl [ 60 { \hat{\rm s}}^4 - 53 { \hat{\rm s}}^2 { \hat{\rm z}}^2 + 8 { \hat{\rm z}}^4 \bigr ] \\ & - 2 m^{18 } { \hat{\rm s}}^3 \bigl [ 8 { \hat{\rm s}}^2 - 3 { \hat{\rm z}}^2 \bigr]\\ & + 2 m^{20 } { \hat{\rm s}}^4 \bigr\ } \end{split}\ ] ] p. nason , s. dawson and r. k. ellis , `` the one particle inclusive differential cross - section for heavy quark production in hadronic collisions , '' nucl . phys . b * 327 * , 49 ( 1989 ) . [ erratum - ibid . b * 335 * , 260 ( 1990 ) ] . b. gong , x. q. li and j. x. wang , `` qcd corrections to j/@xmath1 production via color octet states at tevatron and lhc , '' phys . b * 673 * , 197 ( 2009 ) , [ phys . lett . * 693 * , 612 ( 2010 ) ] . y. q. ma , k. wang and k. t. chao , `` qcd radiative corrections to @xmath22 production at hadron colliders , '' phys . d * 83 * , 111503 ( 2011 ) . m. butenschoen and b. a. kniehl , `` j / psi polarization at tevatron and lhc : nonrelativistic - qcd factorization at the crossroads , '' phys . lett . * 108 * , 172002 ( 2012 ) . k. t. chao , y. q. ma , h. s. shao , k. wang and y. j. zhang , `` @xmath61 polarization at hadron colliders in nonrelativistic qcd , '' phys . * 108 * , 242004 ( 2012 ) . b. gong , l. p. wan , j. x. wang and h. f. zhang , `` polarization for prompt j/@xmath1 and @xmath1(2s ) production at the tevatron and lhc , '' phys . lett . * 110 * , 042002 ( 2013 ) . m. butenschoen and b. a. kniehl , `` reconciling @xmath61 production at hera , rhic , tevatron , and lhc with nrqcd factorization at next - to - leading order , '' phys . lett . * 106 * , 022003 ( 2011 ) . v. khachatryan _ et al . _ [ cms collaboration ] , `` measurement of j/@xmath1 and @xmath1(2s ) prompt double - differential cross sections in pp collisions at @xmath40=7 tev , '' phys . lett . * 114 * , no . 19 , 191802 ( 2015 ) . s. chatrchyan _ et al . _ [ cms collaboration ] , `` measurement of the relative prompt production rate of @xmath67 and @xmath66 in @xmath135 collisions at @xmath136 tev , '' eur . j. c * 72 * , 2251 ( 2012),[arxiv:1210.0875 [ hep - ex ] ] . g. aad _ et al . _ [ atlas collaboration ] , `` measurement of the differential cross - sections of prompt and non - prompt production of @xmath61 and @xmath137 in @xmath135 collisions at @xmath138 and @xmath139 tev with the atlas detector , '' arxiv:1512.03657 [ hep - ex ] . r. aaij _ et al . _ [ lhcb collaboration ] , `` measurement of @xmath61 production in @xmath135 collisions at @xmath140 , '' eur . j. c * 71 * , 1645 ( 2011 ) . r. aaij _ et al . _ [ lhcb collaboration ] , `` measurement of forward @xmath61 production cross - sections in @xmath135 collisions at @xmath141 tev , '' jhep * 1510 * , 172 ( 2015 ) . h. f. zhang , l. yu , s. x. zhang and l. jia , `` global analysis of the experimental data on @xmath127 meson hadroproduction , '' phys . d * 93 * , no . 5 , 054033 ( 2016 ) addendum : [ phys . d * 93 * , no . 7 , 079901 ( 2016 ) ] , [ arxiv:1410.4032 [ hep - ph ] ] .

we have determined li abundances for eleven metal - poor turn - off stars , among which eight have [ fe / h ] , based on lte analyses of high - resolution spectra obtained with the hds on the subaru telescope . the li abundances for four of these eight stars are determined for the first time by this study . effective temperatures are determined by a profile analysis of h and h . while seven stars have li abundances as high as the spite plateau value , the remaining four objects with [ fe / h ] have(li )(li / h) , confirming the existence of extremely metal - poor turn - off stars having low li abundances , as reported by previous work . the average of the li abundances for stars with [ fe / h] is lower by 0.2 dex than that of the stars with higher metallicity . no clear constraint on the metallicity dependence or scatter of the li abundances is derived from our measurements for the stars with [ fe / h] . correlations of the li abundance with effective temperatures , with abundances of na , mg and sr , and with the kinematical properties are investigated , but no clear correlation is seen in the extremely metal - poor star sample .

0904.1448

over the course of the past few decades , li abundances have been measured for many metal - poor main - sequence turn - off stars , in hopes of placing constraints on the nature of big bang nucleosynthesis ( bbn ) . the first such investigation , by @xcite , revealed that warm metal - poor main - sequence stars exhibited a constant li abundance . this so - called spite plateau was interpreted as a result of the synthesis of light nuclei in the first several minutes of the evolution of the universe ( it is conventionally assumed that li is not destroyed at the surface of such stars because of their shallow surface convection zones ) . since then , measurements of li abundances have been obtained over wide ranges of effective temperature and metallicity , confirming that the scatter of the li abundances in most stars with @xmath7@xmath8 k and [ fe / h]@xmath9 is small , if present at all ( * ? ? ? * and references therein ) . = @xmath10 , and @xmath11 for elements a and b. lithium abundances are conventionally presented as @xmath3(li ) instead of @xmath12 . ] however , it has been recognized that the spite plateau value of a(li ) ( @xmath13 ) is 0.30.4 dex lower than the li abundance predicted by standard bbn models @xcite , adopting the baryon density determined by recent measurements of the cosmic microwave background ( cmb ) radiation with the wmap satellite @xcite . this discrepancy indicates the existence of poorly understood processes that deplete li in metal - poor turn - off stars ( e.g. * ? ? ? * ; * ? ? ? * ) , astration of li in the gas that formed these stars ( perhaps by massive zero - metallicity progenitors ; piau et al . 2006 ) , problems in the measured abundances of li in metal - poor stars , systematic uncertainties in the standard bbn model predictions , or exotic processes in the early universe arising from nonstandard particle physics ( cyburt et al . 2008 and references therein ) . recent measurements of very metal - poor turn - off stars also suggest a decreasing trend of li abundances with decreasing metallicity @xcite . in particular , @xcite investigated the li abundances for a sample that includes eight stars with [ fe / h ] @xmath6 . they showed that the li abundances of these stars are lower than those of stars with higher metallicity , investigating very carefully the scatter and trend of li abundances with metallicity and effective temperature . another unsolved problem is the low li abundance found in he 13272326 , a hyper metal - poor turn - off star having [ fe / h]@xmath14 @xcite . only an upper - limit on the li abundance has been determined , @xmath3(li)@xmath15 @xcite . although the star has probably evolved to the subgiant branch @xcite , the effective temperature of this object ( 6180 k ; frebel et al . 2005 ) is still sufficiently high that depletion of li by surface convection is not expected . another possible explanation for the apparent depletion of li is mass accretion from an evolved companion star @xcite , in particular because this star is highly carbon - enhanced ( [ c / fe ] @xmath16 ; aoki et al . 2006 ) . however , no signature of binarity has been found yet for this object @xcite . thus , concerning the @xmath17li abundances in very metal - poor stars , we are confronted with three problems : ( 1 ) the discrepancy between the observed spite plateau value and the prediction of standard bbn models adopting the baryon density determined by cmb measurements ; ( 2 ) a possible trend of the li abundance as a function of metallicity in extremely metal - poor stars ; and ( 3 ) the low li abundance in he 13272326 ( see also the summary by piau et al . although possible connections between the above three problems are still unknown , we have obtained measurements of li abundances for several extremely metal - poor turn - off stars in a search for hints to solving these li puzzles . in this paper we report measurements of li abundances for very and extremely metal - poor stars . our sample includes eight stars with [ fe / h ] @xmath6 , among which four stars are studied for the first time . the sample selection and high - resolution spectroscopy are described in [ sec : obs ] . section [ sec : ana ] reports the determination of stellar parameters and details of the measurement of li abundances . uncertainties and comparisons with previous work are also discussed in this section . we discuss the implications of our measurements in [ sec : disc ] , and consider possible correlations with the derived stellar atmospheric parameters , other elemental abundances , and the kinematics of the sample . high - resolution spectra of very and extremely metal - poor main - sequence turn - off stars were obtained in the course of three different observing programs , using the subaru telescope high dispersion spectrograph ( hds ; noguchi et al . table 1 lists the objects and details of the observations . the spectra of the first five objects were obtained in an observing program for extremely metal - poor stars in 2005 @xcite . the two objects from the sdss sample were observed in a program for carbon - enhanced metal - poor ( cemp ) stars @xcite . although these two stars were selected as candidate cemp turn - off stars having [ fe / h ] @xmath18 , they turned out to show no clear carbon excess in our high - resolution spectroscopy , thus were good targets for studying the li abundances in the extremely low metallicity range . the other four bright stars were observed with very high signal - to - noise ( s / n ) ratios in order to measure li isotope ratios ( p.i . s. inoue ) . lithium abundances for seven objects in our sample have been recently measured with high - resolution spectroscopy by other authors . the stars cs 22948093 and cs 22965054 were studied by @xcite . the five bright objects are well - known stars for li studies . bd+26@xmath193578 was recently studied by @xcite . li abundances of g 6412 and g 6437 were measured by @xcite and @xcite . @xcite also determined li abundances for hd 84937 , bd+26@xmath193578 and cd @xmath20 . duplications of targets with previous studies provide the opportunity to examine the consistency between independent abundance measurements . the li abundances of the other four stars ( bs 16545089 , he 11480037 , sdss 0040 + 16 and sdss 1033 + 40 ) are reported for the first time by the present study . the spectral resolution for the last four stars in table 1 is @xmath21 or @xmath22 . the other objects were observed with @xmath23 with 2 @xmath24 2 ccd on - chip binning . an exception is sdss 1033 + 40 , which was observed with @xmath25 to collect sufficient photons , using a wider slit , under relatively poor seeing conditions . the spectra cover 41006800 , although the coverage is slightly different between the individual observing programs . data reduction was carried out with standard procedures using iraf . photon counts at 6700 are listed in table [ tab : obs ] . spectra of this wavelength region are shown in figure [ fig : sp ] . equivalent widths of fe , na , mg , and sr lines were measured by fitting gaussian profiles , in order to determine atmospheric parameters and chemical compositions . since the @xmath266708 line consists of a doublet , gaussian fitting is not appropriate for the measurement of equivalent widths . instead , we determined the li abundances by fitting synthetic spectra ( [ sec : liana ] ) . for comparisons of equivalent widths with previous studies , we measured the equivalent widths of this line from the synthetic spectra ( [ sec : ana ] ) that provided the best fits to the observed ones . the equivalent widths of the interstellar d lines , which can be used to estimate the interstellar reddening ( [ sec : ana ] ) , were measured by direct integration of the absorption features . the heliocentric radial velocities listed in table [ tab : obs ] were measured from the same clean lines that were used to determine iron abundances . the uncertainty given in table [ tab : obs ] is @xmath27 , where @xmath28 is the standard deviation of the measurements and @xmath29 is the number of lines used . in addition , one should take into account possible small systematic errors due to the instability of the instrument , which are at most 0.4 km s@xmath30 @xcite . the effective temperatures ( @xmath7 s ) of our program stars were determined from profile fits to hydrogen balmer lines , a technique used as well by @xcite and @xcite . we employed the h@xmath1 and h@xmath2 lines [ @xmath7(h@xmath1 ) and @xmath7(h@xmath2 ) ] . although h@xmath31 is also covered by our observations , it lies at the edge of the ccd , making continuum rectification difficult , and thus has not been used . the method for continuum rectification of the observed spectra and analysis used follows exactly that described in @xcite , which may be consulted for details . the most important aspects of the analysis are that the synthetic profiles are computed assuming lte line formation using 1d lte plane - parallel marcs models @xcite , with convection described by mixing length theory with parameters @xmath32 and @xmath33 . the most important line - broadening mechanisms for the wings are stark broadening and self broadening , which are described by calculations of @xcite and @xcite , respectively . the fitting is accomplished by minimisation of the @xmath34 statistic , comparing the observed and synthetic profiles . essentially the same method was used by @xcite , although with some differences in the rectification and profile comparison techniques . we note that the assumption of lte for formation of the balmer line wings in cool stars has recently been shown to be questionable on the basis of theoretical non - lte calculations @xcite , the role of hydrogen collisions being a major uncertainty . those calculations admit that the temperatures from lte balmer line wings could be systematically too cool by of order 100 k if hydrogen collisions are inefficient , although lte is not ruled out . due to the fact that there is no strong evidence favoring any particular hydrogen collision model , we chose to calculate in lte as this temperature scale is well studied , and it is computationally most practical . however , we emphasize that lte is not a safe middle ground , and will lead to temperatures systematically too cool should departures from lte exist in reality . often , h@xmath1 is given higher weight than higher series lines such as h@xmath2 for reasons discussed by @xcite . however , in extremely metal - poor turn - off stars this is no longer obvious since blending by metal lines becomes unimportant , and h@xmath2 becomes in fact almost insensitive to gravity , while h@xmath1 is quite gravity sensitive ( see table 4 of @xcite ) . further , the calculations by @xcite suggest that non - lte effects , if they exist , will be largest in h@xmath1 . thus , in combining the temperatures from h@xmath1 and h@xmath35 , we have in fact given h@xmath2 double the weight of h@xmath1 . the derived effective temperatures are dependent on the gravity assumed in the calculation . the analysis is iterated for the gravity , which is determined from the analysis of the high - resolution spectrum ( [ sec : ana ] ) for each object . the results and uncertainties of the @xmath7 determinations are listed in table [ tab : param ] . the uncertainties listed in the tables are estimated from the quality of the spectrum and the fit . for comparison purposes , we also estimated the effective temperatures from the @xmath36 colors [ @xmath7@xmath37 , using the effective temperature scales of @xcite , @xcite , and @xcite . for the estimates using the scale of @xcite , we assume [ fe / h]@xmath38 for stars having [ fe / h]@xmath39 following @xcite . the photometry data were collected from @xcite , the 2mass catalogue @xcite , and the simbad database . the @xmath40 magnitudes of the sdss stars are derived from the @xmath41 magnitude and @xmath42 color , as in @xcite . interstellar reddening is estimated from the dust maps of @xcite and from the interstellar d line , using the scale of @xcite ; results are listed in table [ tab : teff ] . we adopt the reddening estimate obtained from the na d line for the brightest four stars ( hd 84937 , bd+26@xmath193578 , g 6412 and g 6437 ) because these are nearby stars , and the reddening might be overestimated from the dust maps . the interstellar na d lines blend with the stellar lines in the spectrum of he 11480037 , so the reddening value from the dust map is adopted for this object . for the other six objects , the averages of the reddening values derived from the two methods are adopted . the differences between the @xmath43 values based on the two methods are generally 0.01 mag or smaller . the exception is for cs 22965054 , which exhibits the largest @xmath43 among the sample , @xmath44 from the dust map and 0.13 from the na d line . the extinction in each band is obtained from the reddening relation given by @xcite . the photometry data and the derived effective temperatures are listed in table [ tab : teff ] . figure [ fig : teff ] shows the differences between the effective temperatures from balmer lines and those from @xmath45 colors for the three temperature scales . the effective temperatures from the scales of @xcite and @xcite are systematically higher than those from @xcite . the differences between the results from @xcite and from @xcite are 100 - 150 k. the results from @xcite are 100 - 200 k higher still than those from @xcite for stars with [ fe / h]@xmath0 , while the agreement is fairly good for the three stars having higher metallicity ( shown by smaller symbols in the figure ) . these differences between the effective temperature scales were already reported by @xcite and @xcite . the comparisons with the results from balmer line analyses indicate that the effective temperatures from the scale of @xcite agree in general with those from the balmer lines . the star showing the largest discrepancy is sdss 0040 + 16 . this star has a @xmath36 smaller than the color range ( @xmath46 ) for which the formulae of alonso et al . are applicable . the results from the scales of @xcite and @xcite for stars with [ fe / h]@xmath0 are systematically higher than those from balmer lines by 150 k and 250 k , respectively . determination of the li abundance is sensitive to the effective temperatures adopted in the analysis . we discuss the impact of effective temperatures on the metallicity dependence of li abundances in [ sec : disc ] . the estimate of effective temperatures for extremely metal - poor stars from these temperature scales would be rather uncertain , because of the small number of stars in the sample used to produce the scales . @xcite includes a very small number of stars for the ranges of the color and metallicity . @xcite increased the number of such stars , which are used to produce the scale of @xcite . however , the sample is still not large ( 10 stars ) and includes one carbon - enhanced star and two known double - lined spectroscopic binaries . further measurements of effective temperatures by the infrared flux method are desired for this metallicity range . the abundance analyses were made using the grid of atlas9 newodf model atmospheres @xcite , with enhancement of the @xmath1-elements . calculations of synthetic spectra for abundance analyses were made employing a 1d - lte spectral synthesis code that is based on the same assumptions as the model atmosphere program of @xcite . the surface gravities ( @xmath47 ) are estimated from the effective temperatures and @xmath48 isochrones @xcite for [ fe / h]=@xmath49 ( for hd 84937 and bd+26@xmath193578 ) and [ fe / h ] @xmath50 ( for the others ) assuming the ages of these stars to be 12 gyr . for the @xmath7 range of our sample , two possibilities of @xmath47 exist : the subgiant case ( @xmath47@xmath51 ) and the main - sequence case ( @xmath47@xmath52 ) . we performed abundance analyses for and ( see below for details ) , and selected the @xmath47 that provides better agreement of iron abundances from the two species for each star . the derived @xmath47 value is not sensitive to the assumption of the age for selecting isochrones . the results for @xmath47 and the derived iron abundances from the two species are listed in table [ tab : param ] . we re - determined effective temperature when the gravity assumed in the first estimate of effective temperature is different from the derived one . we note that the fe abundance from lines and the li abundance are not sensitive to the gravity adopted in the analysis . the analysis , when adopting the micro - turbulent velocity ( @xmath53 ) of 1.5 km s@xmath30 , results in no statistically significant dependence of the iron abundances on the strengths of the lines used for the analysis for bs 16545089 , he 11480037 , g 6412 , g 6437 and bd+26@xmath193578 , while 1.2 km s@xmath30 is preferable for hd 84937 . since the number of iron lines available in the abundance analyses of the other four stars is too small to estimate their micro - turbulent velocity , we assume @xmath53 of 1.5 km s@xmath30 for these objects . we note that the li abundance derived from the weak @xmath26 6708 line is insensitive to the micro - turbulent velocity adopted in the analysis . in the following discussion , we adopt the iron abundances from lines ( table [ tab : param ] ) as the metallicity . although the iron abundances from lines are considered to be possibly affected by non - lte effects ( * ? ? ? * and references therein ) , our results are insensitive to the gravity adopted in the analysis . we refer to a solar fe abundance of @xmath54(fe)@xmath55 @xcite to derive the [ fe / h ] and [ x / fe ] values . in order to investigate correlations between the abundances of li and other elements , na , mg , and sr abundances are determined by a standard lte analysis from the 5890 and 5896 lines ( d lines ) , 5172 , 5183 , and/or 5528 lines , and 4078 and/or 4215 lines , respectively . the agreement of the abundances from two or three lines for each element is fairly good . the results are listed in table [ tab : res ] . correlations between the abundances of li and these elements are examined in [ sec : disc ] . the na d lines are known to suffer a significant nlte effect that is dependent on the line strengths . the effect is , however , not large ( @xmath56[na / fe]@xmath57[na / fe]@xmath580.1 dex ) for the extremely metal - poor stars that have equivalent widths of the d lines less than 50 m @xcite . the effect could be larger ( @xmath59 dex ) for the less metal - poor stars whose equivalent widths are as large as 100 m . the li abundances are derived from the 6708 line by fitting the observed spectra with synthetic spectra . doppler corrections for the observed spectra are determined by the line positions used for the above abundance measurements . in the fitting procedure for the four bright stars ( hd 84937 , bd+26@xmath193578 , g 6412 and g 6437 ) , a small wavelength shift is included as a fitting parameter ( see below ) . the continuum level is estimated from the wavelength range around the li line ( 6706.96707.4 and 6708.26708.7 ) . the list of the li lines provided by @xcite is adopted , neglecting the contribution of @xmath60li . we assume a gaussian profile to account for the broadening by macro - turbulence , including rotation , and by the instrument . we assume a broadening of 7 km s@xmath30 for the stars observed with @xmath61 , and 8 km s@xmath30 for the others , values that sufficiently explain the widths of weak fe lines . we performed the analyses changing the value by @xmath62 km s@xmath30 , and confirmed that the effect of the assumed line broadening on the derived @xmath17li abundance is small ( @xmath63 dex ) . we then search for the li abundance that yields the minimum @xmath64 . we found that the fit is slightly improved by allowing a small wavelength shift for the four brightest stars , while the effect is negligible for the others . hence , in our procedure for fitting of the synthetic spectra to the observations , the free parameter was the li abundance , and for the four brightest stars also a wavelength shift . the number of data points to which the fitting is applied is 18 and 36 for the cases of two pixel binning and of no binning , respectively . the 2@xmath65 range is adopted as the fitting error ( @xmath65[@xmath3(li)]@xmath66 in table [ tab : res ] ) . while the fitting errors for the four bright stars ( g 6412 , g 6437 , hd 84937 and bd+26@xmath193578 ) are on the order of 0.02 dex and are smaller than the other errors ( see below ) , those for the other fainter stars are 0.06 0.18 dex , depending on the s / n of the spectra . the equivalent widths of the li doublet are determined by integrating the flux density of the synthetic spectra calculated for the final li abundance as mentioned in [ sec : obs ] . the values are listed in table [ tab : res ] . in addition to the above fitting errors , the uncertainties in the adopted continuum levels , macro - turbulence , and atmospheric parameters should be considered . the continuum level is estimated for the wavelength range around the li doublet , in which 25 ( in the case of two pixel binning ) or 50 ( in the case of no binning ) data points exist . the uncertainty is 0.1% or smaller for the four bright stars ( @xmath67300 ) , while it is approximately 0.5% for the spectra with @xmath68 . the effects of continuum level shifts of 0.1% and 0.5% are estimated to be 0.01 dex and 0.04 dex , respectively , by means of an analysis of the spectrum of g 6412 , in which the continuum level was changed artificially . the errors due to the uncertainty of the macro - turbulent velocity are on the order of 0.01 dex , as mentioned above . the errors due to the uncertainties in atmospheric parameters are estimated by changing the parameters by @xmath69(@xmath7 ) @xmath70 k , @xmath69(@xmath47 ) @xmath71 dex , and @xmath69(@xmath53 ) = + 0.3 km s@xmath30 for g 6412 . the effect of metallicity changes of 0.2 dex on the abundance analyses is negligible for extremely metal - poor stars . we confirmed that the effects of changes of @xmath47 and @xmath53 on the derived li abundance are negligible . the li abundance changes by + 0.07 dex when the effective temperature is changed by + 100 k. such error estimates are scaled for the uncertainty for @xmath7 for each object listed in table [ tab : param ] . the total abundance errors ( @xmath65[@xmath3(li)]@xmath72 ) are obtained by adding , in quadrature , the fitting errors and the errors due to the uncertainty of the continuum placement , the macro - turbulence , and the effective temperature for each star ( see table [ tab : res ] ) . table [ tab : comp ] compares the results of the present study for stars in common with other recent studies . the @xmath7 and @xmath3(li ) of cs 22948093 determined by this work and by @xcite agree fairly well , while a significant discrepancy of @xmath3(li ) is found for cs 22965054 . this is clearly caused by the different effective temperatures adopted in the two analyses . the abundances of bd+26@xmath193578 were determined by @xcite(where the star is referred to as hd 338529 ) , adopting very similar atmospheric parameters ( @xmath7 , @xmath47 ) = ( 6335 k , 4.04 ) to those of the present work ( 6340 k , 3.9 ) . their fe abundance ( [ fe / h ] @xmath73 and @xmath74(fe ) @xmath75 ) from lines as well as the li abundance agrees well with our result . the @xmath7 s of g 6412 and g 6437 determined by @xcite are about 200 k lower than ours , which result in their lower @xmath3(li ) values . this is , however , partially compensated by the difference of model atmospheres used in the analysis . boesgaard et al . adopted the grid of @xcite , which assumes convective overshooting . we found that the resulting @xmath3(li ) from this grid is systematically higher by 0.06 dex than that obtained from the no overshooting models by conducting analysis with both models . the discrepancy of @xmath3(li ) derived by us and @xcite is explained by the differences of @xmath7 and the model atmosphere grid . the @xmath7 of @xcite is spectroscopically determined by the analysis of lines . a systematic difference between @xmath7 s from the excitation equilibrium and those from other methods ( such as photometric colors ) have been reported by previous studies ( e.g. , * ? ? ? * ; * ? ? ? * ) , which might be attributed to non - lte effects on fe line formation , although the recent work by @xcite reported no discrepancy between the @xmath7 from lte analyses of lines and @xmath7 from the h@xmath1 profile analysis . the @xmath7 s determined by @xcite are systematically lower than ours . the differences in @xmath7 for g 6412 , g 6437 and cd @xmath20 are smaller than 100 k , while the differences are as large as 200 k for hd 84937 and bd+26@xmath193578 . the discrepancy of @xmath3(li ) s between the two studies are basically explained by the differences in the adopted effective temperatures . the above comparisons demonstrate that the differences of derived li abundances are due to differences of the effective temperature and/or the model atmosphere grid adopted in the analysis . these effects are systematic , and do not essentially affect the discussion of the slope or scatter of li abundances if the same technique of effective temperature estimates and similar model atmosphere grids are used . figure [ fig : life ] shows the li abundances as a function of [ fe / h ] for our sample and others . values of @xcite and @xcite are adopted without any correction . the solar fe abundance adopted by @xcite is 7.51 , while that of @xcite is determined by their own analysis of the solar spectrum . @xcite also reported fe abundances from lines , which are systematically higher by 0.08 dex than those from lines according to the authors . hence , a possible shift of [ fe / h ] values by at most 0.1 dex should be taken into consideration in the comparisons of the three works . ] we confirm that the li abundance of the `` reference '' star bd+26@xmath193578 ( [ fe / h]@xmath76 ) determined by our analysis agrees well with the measurement by @xcite . by contrast , stars with [ fe / h ] @xmath6 ( the extremely metal - poor , or emp , stars ) appear to have lower li abundances on average , and also exhibit some scatter . we discuss this point further below . the average of @xmath3(li ) ( i.e. , @xmath77(li)@xmath78 ) of the eight stars with [ fe / h]@xmath6 is 2.03 , with a sample standard deviation ( @xmath65 ) of 0.09 dex . the standard deviation is comparable with , or slightly smaller than , the measurement errors of our analysis ( 0.070.23 dex ) . hence , we do not detect any significant scatter of the li abundances in our sample of emp stars . the @xmath77(li)@xmath78 is lower by 0.24 dex and 0.20 dex than the the average of our two reference stars ( 2.27 ) and the average of the results of the lte analysis by asplund et al . ( 2006 ) for six stars in @xmath79[fe / h]@xmath80 ( 2.23 ) , respectively . the difference of 0.2 dex is significant , compared to the @xmath65 @xmath81=0.09/@xmath82=0.03 dex , where @xmath29 is the number of objects in the extremely metal - poor sample . we note that the li abundances for stars in @xmath83 [ fe / h]@xmath84 are well determined by asplund et al . ( 2006 ) , and the scatter is small ( @xmath850.04 dex and @xmath65 @xmath81=0.02 dex ) . we conclude that the li abundances for stars with [ fe / h]@xmath6 are 0.2 dex lower than those of stars with higher metallicity on average , while no significant scatter or trend with metallicity is detected in our emp sample . a similar conclusion was reached by @xcite ; our new measurements for stars in the lowest metallicity range supports their results . such a difference can obviously be produced by a decreasing trend ( slope ) of @xmath3(li ) as a function of metallicity . the possible slope of @xmath3(li ) was discussed by @xcite in detail . however , no clear physical reason for the slope , which appears only at the lowest metallicity range , has yet been identified . another possibility is that scatter of @xmath3(li ) increases in the range [ fe / h ] @xmath18 , and , as a result , the average decreases . this case would be relatively easily explained by depletion of li , although some reason for the metallicity dependence of the depletion is also required . if the effective temperatures estimated from the @xmath36 colors using the scales of @xcite and @xcite are employed , the li abundances of our stars would be systematically higher . the effective temperatures derived using the scale of @xcite are systematically higher by 220 k and 150 k than the values from the balmer line analysis for emp stars and for stars with [ fe / h]@xmath76 , respectively , resulting in 0.15 dex and 0.10 dex higher li abundances . although the difference of li abundances between the emp stars and less metal - poor stars becomes slightly smaller than the result derived adopting the effective temperature from the balmer line analysis , the difference is still significant . by contrast , if the effective temperatures estimated using the scales of @xcite are adopted , the values are 290 k and 150 k higher than those from the balmer lines , resulting in 0.20 dex and 0.10 dex higher li abundances . in this case , the difference of li abundances between the stars with [ fe / h]@xmath0 and @xmath86 is only 0.1 dex , which is no longer significant compared to our measurement errors . [ [ correlations - with - stellar - parameters - elemental - abundances - and - kinematics ] ] correlations with stellar parameters , elemental abundances , and kinematics ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ although no statistically significant scatter of @xmath3(li ) is found for our full emp sample , within our measurement errors , the existence of _ some _ star - to - star differences in @xmath3(li ) is suggested . for instance , even if the stars with the largest measurement errors are excluded from the evaluation , a similar scatter of @xmath3(li ) remains . a difference of 0.14 dex is found in @xmath3(li ) for the two bright stars g 6412 and g 6437 , as found by @xcite and @xcite . in this subsection , we investigate correlations between @xmath3(li ) and the adopted stellar parameters , in order to search for a hint for understanding the lower li abundances among the emp stars . figure [ fig : liteff ] shows @xmath3(li ) as a function of @xmath7 . no clear correlation can be seen in this figure . it should be noted that the random error of the effective temperature propagates into the derived li abundance , which is represented by the arrows shown in the diagram . an increasing trend of li abundance with decreasing @xmath7 is potentially influenced by this error . our sample includes stars that have already evolved to the subgiant branch . their effective temperatures during the main - sequence stage should be higher than the current values , and might be as high as the hottest stars among the main - sequence sample . the stars having @xmath87 are over - plotted by large circles representing candidate subgiants in the figure . if these stars are excluded , we find that stars cooler than 6150 k exhibit higher and almost constant li abundances , while the warmer stars show some scatter . however , this probably reflects a metallicity effect , as the cooler stars ( excluding subgiants ) are objects having [ fe / h ] @xmath88 studied by @xcite , while the warmer stars have [ fe / h ] @xmath89 . that is , the sample of very metal - poor stars ( [ fe / h ] @xmath90 ) have @xmath7@xmath91 k , or are subgiants . this could be due to a bias in the sample selection caused by the fact that distant stars must be observed to cover lower metallicity ranges , and intrinsically faint stars are not sampled . other than this point , no clear correlation appears between the li abundances and effective temperature , even if subgiants are removed from the plot , or if the highest temperature among the sample ( @xmath92 k ) is assigned for them as their @xmath7 during their main - sequence phase . figure [ fig : limgsr ] shows li abundances as functions of [ na / fe ] , [ mg / fe ] , and [ sr / fe ] . an anti - correlation between the li abundance and the [ na / fe ] ratio was reported for the globular cluster ngc 6752 by @xcite . such a trend is , however , not seen in our sample ( fig . [ fig : limgsr ] ) : the two low li stars he 11480037 and sdss 0040 + 16 have comparatively high [ na / fe ] ratios , while that of the other low li star , cs 22948093 , is low . if the nlte effects on the na abundances are taken into consideration , the abundances of less metal - poor stars become lower by about 0.3 dex . however , this correction does not result in any correlation between the li and na abundances . no significant scatter is found for [ mg / fe ] in our sample . also , no correlation is found between the abundances of li and the mg abundances . in contrast , a large scatter of measured sr abundances exists , which could reflect the contribution of a neutron - capture process that is efficient in the very early galaxy @xcite , as well as the contribution of the main r - process to higher metallicity objects . no correlation is found between the li and sr abundances , indicating that the li production or depletion is not likely to be related to neutron - capture nucleosynthesis processes . we are also interested in exploring whether the kinematics of the stars with lower @xmath3(li ) exhibit any peculiarities that distinguish them from the rest of the stars . for this exercise , we combined our present sample with that of bonifacio et al . proper motions of sdss stars are listed in the public dr6 database , while those of other stars are taken from the nomad database ( zacharias et al . distances were estimated from the luminosity classifications given by either the present paper or in bonifacio et al . ( 2007 ) , applying the methods described by beers et al . stars with @xmath93 were considered dwarfs , those with @xmath94 were considered main - sequence turnoff stars , and those with @xmath95 were considered subgiants . the adopted surface gravities , metallicities , photometry , distances , radial velocities , and proper motions are listed in table [ tab : kine1 ] . these data were used to derive the full space motions and other orbital information , following the procedures described by carollo et al . ( 2007 ) ; results are listed in table [ tab : kine2 ] . figure [ fig : kine ] shows the derived rotational velocity with respect to the galactic center , @xmath96 , as a function of [ fe / h ] . the stars with low values of @xmath3(li ) are labeled by filled circles . we note that three stars with `` normal '' @xmath3(li ) are on highly prograde orbits ( an additional star , cs 310610032 , with @xmath96 = 205 km s@xmath30 , may be a member of the metal - weak thick disk ) . this is somewhat unexpected , given the essentially zero net rotation of the inner halo , and net retrograde rotation of the outer halo . we plan to study these stars in more detail in the near future . based on the derived space velocities and orbital parameters , an attempt was made to assign approximate population memberships for these stars , listed in the second column of table [ tab : kine2 ] . the assignments take into account the values of the velocity ellipsoids derived for the inner and outer halo , and the thick disk ( including the metal - weak thick disk ; d. carollo et al . , in preparation ) , as well as the derived @xmath97 ( the maximum distance from the galactic plane in the vertical direction ) for each star . we did not attempt to assign a membership to the four highly prograde stars , because , as was mentioned above , they require further investigation ( fi in the second column of table [ tab : kine2 ] ) . note that in some cases it was not possible to uniquely distinguish a single population assignment , so multiple assignments are given . inspection of these assignments indicates no tendency for the stars with low @xmath3(li ) to be associated preferentially with either the inner- or outer - halo populations . this is similar to the conclusions drawn by bonifacio et al . ( 2007 ) based on inspection of radial velocities alone . however , the sample size ( especially of low @xmath3(li ) stars ) and the existing errors on the li abundance determinations preclude a final determination . our measurements of the li doublet for extremely metal - poor stars based on the effective temperatures from the balmer lines showed that the li abundances in the metallicity range [ fe / h ] @xmath6 are lower , on average , than for stars with higher metallicity . the same conclusion is also reached by adopting the effective temperatures from @xmath36 using the scales of @xcite and @xcite , although the li abundances are systematically higher if the latter is adopted . if the temperature scale of @xcite for @xmath36 is adopted , the dependence of the li abundances on metallicity becomes marginal . although no scatter or trend of li abundances is detected within the sample of extremely metal - poor stars , the metallicity dependence of the li abundance could be a key to understanding the li problems found for metal - poor stars . observations to obtain better quality spectra for such extremely metal - poor stars , to improve measurements of the li feature itself , as well as for improved determination of @xmath7 from the h@xmath1 and h@xmath2 profiles , will provide a useful constraint on the possible scenarios proposed to explain the li problems . moreover , measurements of li abundances for even lower metallicity stars ( [ fe / h ] @xmath98 ) are also vital . measurements of li abundances in this metallicity range have been reported so far only for one system , a double - lined spectroscopic binary ( cs 22876032 : * ? ? ? * ; * ? ? ? we plan additional investigations of this metallicity range by high - resolution spectroscopic studies of extremely metal - poor star candidates discovered by the hamburg / eso survey and sdss / segue in the near future . the authors would like to thank the anonymous referee for useful comments for improving this paper . w. a. is supported by a grant - in - aid for science research from jsps ( grant 18104003 ) . b is a royal swedish academy of sciences research fellow supported by a grant from the knut and alice wallenberg foundation . s. b also acknowledges support from the swedish research council . t. c. b. acknowledges support from the us national science foundation under grants ast 04 - 06784 and ast 07 - 07776 , as well as from grants phy 02 - 16783 and phy 08 - 22648 ; physics frontier center / joint institute for nuclear astrophysics ( jina ) . n.c . acknowledges support from the knut and alice wallenberg foundation . colors with those from the balmer line analysis . the differences of the effective temperature from those from the balmer line analysis are shown for the scales of @xcite ( circles ) , @xcite ( squares ) , and @xcite ( triangles ) . , width=453 ] . the results of the present work are plotted by filled circles with error bars . the open triangles and squares indicate the results by @xcite and @xcite , respectively . large open circles are overplotted for subgiant stars ( @xmath47 @xmath99).,width=453 ] , [ mg / fe ] , and [ sr / fe ] for our sample . filled circles indicate stars having [ fe / h ] @xmath102 . the upper limit of the sr abundance of cd @xmath20 ( [ sr / fe ] @xmath103 ) is not plotted in this figure.,width=453 ] as a function of [ fe / h ] for our sample and that of @xcite . the filled circles indicate stars with @xmath3(li ) @xmath104 , while the open ones indicate stars with @xmath3(li ) @xmath105 . a typical error of fe abundance determination is 0.15 dex . note the presence of several stars with rather high @xmath96 . , width=453 ] llcrcr & 28 feb . 2005 & 225 & 16,500 & 2453429.86 & @xmath106 + cs 22948093 & 19 june 2005 & 174 & 5,900 & 2453541.06 & @xmath107 + cs 22965054 & 20 june 2005 & 100 & 5,000 & 2453541.97 & @xmath108 + he 11480037 & 27 feb . 2005 & 160 & 15,000 & 2453428.94 & @xmath109 + cd @xmath20 & 20 june 2005 & 20 & 14,000 & 2453542.12 & @xmath110 + sdss 0040 + 16 & 14 sep . 2006 & 80 & 4,800 & 2453992.90 & @xmath111 + sdss 1033 + 40 & 10 feb . 2007 & 160 & 4,100 & 2454141.98 & @xmath112 + g 6412 & 22 feb . 2003 & 300 & 103,000 & 2453507.76 & @xmath113 + g 6437 & 18 may 2005 & 420 & 211,000 & 2453508.83 & @xmath114 + hd 84937 & 22 feb . 2003 & 180 & 626,000 & 2452692.83 & @xmath115 + bd+26@xmath193578 & 17 may 2005 & 132 & 342,000 & 2453508.00 & @xmath116 + lcccccccc & 6380 & 6290 & 6320 & 150 & 3.9 & 1.5 & 3.96 & 3.88 + cs 22948093 & 6320 & 6410 & 6380 & 150 & 4.4 & 1.5 & 4.02 & 4.02 + cs 22965054 & 6390 & 6270 & 6310 & 200 & 3.9 & 1.5 & 4.61 & 4.56 + he 11480037 & 6100 & 5940 & 5990 & 200 & 3.7 & 1.5 & 3.99 & 3.90 + cd @xmath20 & 6150 & 6190 & 6180 & 150 & 4.4 & 1.5 & 4.05 & 4.17 + sdss 0040 + 16 & 6350 & 6360 & 6360 & 200 & 4.4 & 1.5 & 4.16 & 4.25 + sdss 1033 + 40 & 6380 & 6370 & 6370 & 200 & 4.4 & 1.5 & 4.22 & 4.29 + g 6412 & 6260 & 6280 & 6270 & 100 & 4.4 & 1.5 & 4.08 & 4.20 + g 6437 & 6310 & 6280 & 6290 & 100 & 4.4 & 1.5 & 4.23 & 4.33 + hd 84937 & 6330 & 6270 & 6290 & 100 & 3.9 & 1.2 & 5.30 & 5.35 + bd+26@xmath193578 & 6370 & 6330 & 6340 & 100 & 3.9 & 1.5 & 5.17 & 5.20 + lccccccccc & 14.450 & 0.318 & 1.248 & 0.026 & 0.020 & 0.023 & 6408 & 6659 & 6508 + cs 22948093 & 15.180 & 0.360 & 1.175 & 0.018 & 0.008 & 0.013 & 6497 & 6777 & 6618 + cs 22965054 & 15.069 & 0.497 & 1.620 & 0.089 & 0.131 & 0.110 & 6180 & 6269 & 6263 + he 11480037 & 13.614 & 0.404 & 1.342 & 0.022 & & 0.022 & 6215 & 6381 & 6282 + cd @xmath20 & 12.18 & 0.33 & 1.373 & 0.025 & 0.027 & 0.027 & 6184 & 6328 & 6250 + sdss 0040 + 16 & 15.231 & 0.310 & 1.202 & 0.047 & 0.058 & 0.052 & 6685 & 7044 & 6847 + sdss 1033 + 40 & 15.990 & 0.360 & 1.289 & 0.013 & 0.018 & 0.015 & 6278 & 6449 & 6363 + g 6412 & 11.453 & 0.385 & 1.245 & 0.028 & 0.003 & 0.003 & 6295 & 6484 & 6379 + g 6437 & 11.140 & 0.370 & 1.217 & 0.027 & 0.014 & 0.014 & 6417 & 6632 & 6527 + hd 84937 & 8.28 & 0.41 & 1.218 & 0.037 & 0.007 & 0.007 & 6310 & 6456 & 6469 + bd+26@xmath193578 & 9.37 & 0.35 & 1.226 & 1.106 & 0.010 & 0.010 & 6318 & 6469 & 6474 + bs 16545089 & @xmath1173.49 & 15.3 & 2.02 & 0.12 & 0.17 & @xmath1170.12 & 0.25 & @xmath1170.34 + cs 22948093 & @xmath1173.43 & 12.4 & 1.96 & 0.12 & 0.17 & @xmath1170.22 & 0.13 & 0.13 + cs 22965054 & @xmath1172.84 & 20.3 & 2.16 & 0.12 & 0.19 & 0.05 & 0.26 & 0.45 + he 11480037 & @xmath1173.46 & 19.3 & 1.90 & 0.08 & 0.17 & 0.14 & 0.25 & @xmath1170.64 + cd @xmath20 & @xmath1173.40 & 21.2 & 2.08 & 0.06 & 0.13 & @xmath1170.13 & 0.31 & @xmath1180.9 + sdss 0040 + 16 & @xmath1173.29 & 13.6 & 1.99 & 0.14 & 0.20 & 0.19 & 0.32 & @xmath1170.39 + sdss 1033 + 40 & @xmath1173.24 & 16.5 & 2.09 & 0.18 & 0.23 & @xmath1170.24 & 0.03 & @xmath1170.32 + g 6412 & @xmath1173.37 & 22.4 & 2.18 & 0.02 & 0.07 & @xmath1170.06 & 0.37 & 0.08 + g 6437 & @xmath1173.23 & 16.2 & 2.04 & 0.02 & 0.07 & @xmath1170.14 & 0.24 & 0.01 + hd 84937 & @xmath1172.15 & 24.7 & 2.26 & 0.02 & 0.07 & 0.28 & 0.28 & 0.22 + bd+26@xmath193578 & @xmath1172.28 & 23.7 & 2.28 & 0.02 & 0.07 & 0.22 & 0.35 & 0.04 + lcccccccccc & 6380 & 1.96 & & 6356 & 1.94 & @xcite & & & & + cs 22965054 & 6310 & 2.16 & & 6089 & 2.03 & @xcite & & & & + bd+26@xmath193578 & 6340 & 2.28 & & 6335 & 2.25 & @xcite & & 6150 & 2.15 & @xcite + g 6412 & 6270 & 2.18 & & 6074 & 2.15 & @xcite & & 6222 & 2.14 & @xcite + g 6437 & 6290 & 2.04 & & 6122 & 1.97 & @xcite & & 6240 & 2.09 & @xcite + cd @xmath20 & 6180 & 2.08 & & & & & & 6070 & 1.97 & @xcite + hd 84937 & 6290 & 2.26 & & & & & & 6160 & 2.17 & @xcite + lccrcccrrrrr bs 160230046 & 4.50 & @xmath1172.97 & 14.170 & 0.378 & 0.017 & 0.935 & @xmath1177.5 & @xmath11740.0 & 2.0 & @xmath11740.0 & 1.0 + bs 169680061 & 3.75 & @xmath1173.05 & 13.260 & 0.430 & 0.048 & 0.712 & @xmath11780.7 & @xmath11738.0 & 11.0 & @xmath11736.0 & 8.0 + bs 175700063 & 4.75 & @xmath1172.92 & 14.510 & 0.330 & 0.039 & 1.275 & @xmath117184.4 & 38.0 & 4.0 & @xmath11730.0 & 1.0 + cs 221770009 & 4.50 & @xmath1173.10 & 14.270 & 0.401 & 0.044 & 0.869 & @xmath117208.4 & 11.8 & 5.9 & @xmath11762.5 & 5.8 + cs 228880031 & 5.00 & @xmath1173.30 & 14.900 & 0.413 & 0.014 & 0.968 & @xmath117125.1 & 47.1 & 5.2 & @xmath11720.6 & 5.2 + cs 229480093 & 4.25 & @xmath1173.30 & 15.180 & 0.360 & 0.015 & 1.346 & 364.3 & @xmath11715.6 & 5.1 & @xmath11722.8 & 4.8 + cs 229530037 & 4.25 & @xmath1172.89 & 13.640 & 0.367 & 0.027 & 0.753 & @xmath117163.3 & 31.3 & 6.1 & 37.0 & 6.1 + cs 229650054 & 3.75 & @xmath1173.04 & 15.069 & 0.497 & 0.131 & 1.532 & @xmath117281.5 & 26.0 & 1.0 & 0.0 & 3.0 + cs 229660011 & 4.75 & @xmath1173.07 & 14.555 & 0.422 & 0.013 & 0.830 & @xmath11713.5 & 15.1 & 6.0 & @xmath11742.2 & 6.0 + cs 294990060 & 4.00 & @xmath1172.70 & 13.030 & 0.370 & 0.019 & 0.726 & @xmath11758.7 & 15.4 & 4.9 & 22.4 & 4.9 + cs 295060007 & 4.00 & @xmath1172.91 & 14.180 & 0.382 & 0.045 & 1.189 & 56.4 & @xmath1173.6 & 7.9 & @xmath11730.3 & 7.9 + cs 295060090 & 4.25 & @xmath1172.83 & 14.330 & 0.399 & 0.046 & 1.035 & @xmath11721.3 & @xmath11712.5 & 7.9 & @xmath11739.1 & 7.9 + cs 295180020 & 4.50 & @xmath1172.77 & 14.003 & 0.415 & 0.023 & 0.741 & @xmath11722.2 & 26.1 & 4.8 & @xmath11710.4 & 4.8 + cs 295180043 & 4.25 & @xmath1173.20 & 14.566 & 0.371 & 0.019 & 1.064 & 144.8 & 26.0 & 5.0 & @xmath1176.0 & 1.0 + cs 295270015 & 4.00 & @xmath1173.55 & 14.260 & 0.400 & 0.021 & 1.114 & 50.9 & 48.6 & 12.8 & @xmath11722.5 & 12.8 + cs 303010024 & 4.00 & @xmath1172.75 & 12.950 & 0.420 & 0.064 & 0.646 & @xmath11767.7 & @xmath11933.8 & 1.8 & @xmath11723.1 & 1.7 + cs 303390069 & 4.00 & @xmath1173.08 & 14.750 & 0.360 & 0.009 & 1.572 & 34.9 & 22.0 & 5.0 & @xmath11716.0 & 12.0 + cs 310610032 & 4.25 & @xmath1172.58 & 13.874 & 0.409 & 0.036 & 0.818 & 21.0 & @xmath1171.3 & 5.6 & @xmath11711.6 & 5.7 + bs 165450089 & 3.90 & @xmath1173.49 & 14.450 & 0.318 & 0.023 & 1.356 & @xmath117161.5 & 14.0 & 2.0 & @xmath11722.0 & 3.0 + cs 229480093 & 4.40 & @xmath1173.43 & 15.180 & 0.360 & 0.013 & 1.346 & 369.2 & @xmath11715.6 & 5.1 & @xmath11722.8 & 4.8 + cs 229650054 & 3.90 & @xmath1172.84 & 15.069 & 0.497 & 0.110 & 1.532 & @xmath117281.7 & 26.0 & 1.0 & 0.0 & 3.0 + he 11480037 & 3.70 & @xmath1173.46 & 13.614 & 0.404 & 0.022 & 0.844 & @xmath11710.9 & @xmath11746.0 & 8.0 & @xmath11742.0 & 1.0 + cd @xmath20 & 4.40 & @xmath1173.40 & 12.180 & 0.330 & 0.027 & 0.386 & 136.6 & 202.0 & 2.3 & @xmath117185.8 & 1.6 + sdss 0040 + 16 & 4.40 & @xmath1173.29 & 15.231 & 0.310 & 0.052 & 1.714 & @xmath11749.4 & 24.0 & 4.0 & @xmath11720.0 & 2.0 + sdss 1033 + 40 & 4.40 & @xmath1173.24 & 15.990 & 0.360 & 0.015 & 2.150 & @xmath117132.9 & 7.7 & 6.2 & 14.7 & 6.5 + g 6412 & 4.40 & @xmath1173.37 & 11.453 & 0.385 & 0.003 & 0.217 & 434.7 & @xmath117230.4 & 2.4 & @xmath11780.3 & 1.8 + g 6437 & 4.40 & @xmath1173.23 & 11.140 & 0.370 & 0.014 & 0.210 & 76.7 & @xmath11754.3 & 3.6 & @xmath117399.5 & 1.8 + hd 84937 & 3.90 & @xmath1172.15 & 8.280 & 0.410 & 0.007 & 0.073 & @xmath11714.6 & 373.8 & 1.2 & @xmath117774.7 & 0.4 + bd+26@xmath193578 & 3.90 & @xmath1172.28 & 9.370 & 0.350 & 0.010 & 0.137 & -91.2 & 0.9 & 0.6 & @xmath117172.4 & 1.4 + llcrrrrcrrr bs 160230046 & ih & 2.18 & 9 & @xmath117237 & 30 & @xmath11717 & 0.901 & 8.3 & 0.4 & 1.1 + bs 169680061 & ih & 2.17 & 51 & @xmath117165 & @xmath11744 & 54 & 0.722 & 8.3 & 1.3 & 1.1 + bs 175700063 & ih / oh & 2.05 & 60 & @xmath117322 & @xmath11745 & @xmath117108 & 0.499 & 9.3 & 3.1 & 1.3 + cs 221770009 & ih / oh & 2.20 & @xmath117298 & @xmath11780 & 147 & 126 & 0.840 & 32.8 & 2.8 & 13.9 + cs 228880031 & ih / oh & 2.03 & 196 & @xmath117153 & 50 & 65 & 0.788 & 12.5 & 1.4 & 2.8 + cs 229480093 & ih / oh & 1.92 & @xmath117329 & @xmath117120 & @xmath117201 & 99 & 0.854 & 38.3 & 3.0 & 30.9 + cs 229530037 & fi & 2.16 & 191 & 109 & 89 & 338 & 0.677 & 37.2 & 7.1 & 8.0 + cs 229650054 & ih & 2.06 & 250 & @xmath117189 & 90 & 0 & 0.999 & 16.3 & 0.0 & 7.5 + cs 229660011 & ih & 1.91 & @xmath11719 & @xmath117164 & 7 & 55 & 0.716 & 8.3 & 1.3 & 0.8 + cs 294990060 & fi & 2.16 & 82 & 51 & 53 & 270 & 0.346 & 15.7 & 7.6 & 2.3 + cs 295060007 & ih & 2.15 & @xmath117106 & @xmath117126 & @xmath11756 & 98 & 0.557 & 8.8 & 2.5 & 1.8 + cs 295060090 & ih & 2.10 & @xmath11790 & @xmath117172 & 24 & 51 & 0.758 & 8.6 & 1.1 & 0.8 + cs 295180020 & td / ih & 2.13 & 43 & @xmath11768 & 38 & 152 & 0.304 & 8.8 & 4.7 & 1.2 + cs 295180043 & ih & 2.14 & 85 & @xmath117105 & @xmath117121 & 116 & 0.453 & 10.3 & 3.8 & 4.6 + cs 295270015 & ih & 2.08 & 150 & @xmath117209 & @xmath11781 & 7 & 0.970 & 11.3 & 0.1 & 2.9 + cs 303010024 & td / ih & 2.10 & 60 & @xmath117108 & @xmath11723 & 111 & 0.493 & 8.4 & 2.8 & 0.6 + cs 303390069 & ih & 2.13 & 68 & @xmath117176 & @xmath11722 & 44 & 0.774 & 8.8 & 1.1 & 1.6 + cs 310610032 & td & 2.22 & @xmath11719 & @xmath11715 & @xmath11733 & 205 & 0.073 & 9.1 & 7.9 & 0.9 + bs 165450089 & ih & 2.02 & @xmath117192 & @xmath11787 & @xmath11795 & 132 & 0.655 & 16.3 & 3.4 & 4.4 + cs 229480093 & oh & 1.96 & @xmath117332 & @xmath117120 & @xmath117205 & 99 & 0.857 & 39.8 & 3.0 & 32.4 + cs 229650054 & ih / oh & 2.16 & 250 & @xmath117190 & 90 & 0 & 1.000 & 16.3 & 0.0 & 7.5 + he 11480037 & ih & 1.90 & 71 & @xmath117185 & @xmath117124 & 39 & 0.727 & 9.29 & 1.47 & 5.8 + cd @xmath20 & oh & 2.08 & 135 & @xmath117388 & @xmath117294 & @xmath117169 & 0.607 & 28.9 & 7.0 & 23.7 + sdss 0040 + 16 & ih / oh & 1.99 & 91 & @xmath117269 & @xmath117110 & @xmath11759 & 0.688 & 10.6 & 1.9 & 5.0 + sdss 1033 + 40 & fi & 2.09 & @xmath11786 & 127 & @xmath117101 & 347 & 0.602 & 37.1 & 9.2 & 9.8 + g 6412 & oh & 2.18 & @xmath11748 & @xmath117308 & 391 & @xmath11789 & 0.657 & 40.2 & 8.3 & 39.0 + g 6437 & ih / oh & 2.04 & @xmath117183 & @xmath117332 & @xmath117131 & @xmath117113 & 0.642 & 14.3 & 3.1 & 6.1 + hd 84937 & ih & 2.26 & @xmath117214 & @xmath117203 & 0 & 15 & 0.955 & 13.5 & 0.3 & 0.0 + bd+26@xmath193578 & td / ih & 2.28 & @xmath11751 & @xmath117112 & @xmath11752 & 108 & 0.497 & 8.7 & 2.9 & 0.9 +

motivated by our previous paper , in which we argued for the formation of molecular clouds from large - scale flows in the diffuse galactic interstellar medium , we examine the formation of molecular gas behind shocks in atomic gas using a one - dimensional chemical / dynamical model . in our analysis we place particular emphasis on constraints placed on the dynamical evolution by the chemistry . the most important result of this study is to stress the importance of shielding the molecular gas from the destructive effects of uv radiation . for shock ram pressures comparable to or exceeding typical local interstellar medium pressures , self - shielding controls the formation time of molecular hydrogen but co formation requires shielding of the interstellar radiation field by dust grains . we find that for typical parameters the molecular hydrogen fractional abundance can become significant well before co forms . the timescale for ( co ) molecular cloud formation is not set by the h formation rate on grains , but rather by the timescale for accumulating a sufficient column density or extinction , . the local ratio of atomic to molecular gas ( 4:1 ) , coupled with short estimates for the lifetimes of molecular clouds ( 3 - 5 myr ) , suggests that the timescales for accumulating molecular clouds from atomic material typically must be no longer than about 12 - 20 myr . based on the shielding requirement , this implies that the typical product of pre - shock density and velocity must be . in turn , depending upon the shock velocity , this implies shock ram pressures which are a few times the typical estimated local turbulent gas pressure , and comparable to the total pressures ( gas plus magnetic plus cosmic rays ) . coupled with the rapid formation of co once shielding is sufficient , flow - driven formation of molecular clouds in the local interstellar medium can occur sufficiently rapidly to account for observations . we also provide detailed predictions of atomic and molecular emission and absorption that track the formation of a molecular cloud from a purely atomic medium , with a view toward helping to verify cloud formation by shock waves . however , our predictions suggest that the detection of the pre - co stages will be challenging . finally , we provide an analytic solution for time - dependent formation which may be of use in numerical hydrodynamic calculations .

astro-ph0405329

the formation of stars and planetary systems is one of the fundamental problems in astrophysics . much of the work over the past decades has examined the formation of low mass stellar systems because these objects sometimes form in isolation and are therefore easier to study individually . one of the earliest stages of stellar birth that has been the focus of numerous investigations is the creation of a centrally concentrated molecular core from a portion of a low - density parent giant molecular cloud ( gmc ) . theoretical models account for the condensation as occurring possibly via the slow diffusion of magnetic flux occurring over long timescales ( @xmath4 myr ) ( mouschovias 1999 ; lizano & shu 1989 ) or the dissipation of turbulence on shorter timescales ( stone , ostriker , & gammie 1998 ; mac low et al . 1998 ; myers & lazarian 1998 ; nakano 1998 ) . observations of isolated pre - stellar molecular cores , such as l1544 in the taurus molecular cloud , have provided fertile ground for comparison to these theories ( caselli et al 2002 ; ciolek & basu 2000 ; williams et al 1999 ; tafalla et al 1998 ) . however , it is now recognized that most stars form in groups from small aggregates to large clusters and it is not clear that all theories developed for isolated star formation are easily applicable to the larger scales and simultaneity required for star cluster formation ( ballesteros - paredes , hartmann , & vzquez - semadeni 1999 ) . expanding to larger scales opens the question as to whether the formation of stars , both isolated and clustered , might perhaps be intimately related to the formation of the gmc itself . in a previous paper ( hartmann , ballesteros - paredes , & bergin 2001 @xmath5 hbb01 ) we pointed out that the great majority of molecular cloud complexes in the solar neighborhood appear to be forming young stars , and that the ages of the stellar populations in these clouds are typically @xmath6 2 myr ; stellar associations of ages @xmath7 10 myr are devoid of molecular gas . the calculations indicating that mhd turbulence damps rapidly ( stone et al . 1998 , mac low et al . 1998 , padoan & nordlund 1999 ) also favor the suggestion of short cloud lifetimes , since there is no need for a continuous regeneration of mhd turbulence ; additional support is found through arguments related to to cloud crossing times ( elmegreen 2000 ; hbb01 ) . these results place significant empirical constraints on the mechanism(s ) of nearby molecular cloud formation ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * for a review ) . for example , it is difficult to reconcile for the transient nature of local clouds with models in which complexes are built up by the coalescence of smaller molecular clouds , or theories in which molecular gas is mostly moved around from one place to another ( see elmegreen 1993 and references therein ) , because these processes are likely to take much more than @xmath8 myr to occur . moreover , short cloud lifetimes also place severe constraints on the processes leading toward fragmentation of the gmc and the condensation towards star formation . in hbb01 we suggested that chemical transformations of local gas are essential to understanding the observational constraints . we suggested that most clouds are formed by large scale flows in the diffuse atomic medium , and that they appear as molecular clouds only when the column density becomes high enough to shield the molecular gas from the dissociating effects of the interstellar radiation field ( isrf ) . we further noted that self - gravity is likely to become important for column densities comparable to that needed for shielding ( also franco & cox 1986 ) , which would explain the rapid onset of star formation after molecular cloud formation . finally , we suggested that dispersal of star - forming gas is accompanied by a reduction in shielding , so that the gas may revert to an atomic state some time before it is completely physically removed from the neighborhood . these suggested chemical transformations lessen , but do not necessarily eliminate , the difficulty of making clouds `` fast enough '' . the formation of h@xmath0 from atomic gas , which generally must precede the formation of co , is not instantaneous . in addition , while h@xmath0 formation places important constraints on the problem , it must also be examined in the context of a model that incorporates the effects of h@xmath0 ( and co ) self - shielding from the ultraviolet ( uv ) radiation field with extinction by dust grains , each with its own associated timescale . following the molecular evolution is critical to an understanding of cloud formation because of observational bias ; molecular clouds are essentially defined not by h@xmath0 emission but through co emission . in addition , possible atomic precursors are difficult to identify against the galactic h i background ( ballesteros - paredes et al . 1999 ) . in this paper we explore the formation of molecular gas in plane - parallel shocks , starting with atomic ( neutral ) ( warm ) gas . our model incorporates all the relevant heating and cooling mechanisms appropriate for the interstellar medium ( ism ) , including chemical processes relevant to the transformation of atomic gas to molecular form . koyama & inutsuka ( 2000 ) considered a similar problem , and did include the important effects of shielding from uv radiation , but only examined maximum column densities of standard h i clouds ( @xmath9 ) . thus , they only achieve molecular hydrogen fractions of a few percent , insufficient to follow cloud formation . we explore a range of shock velocities comparable to the flows expected in the diffuse atomic medium ( ballesteros - paredes , hartmann , & vzquez - semadeni 1999 , and references therein ) ; the parameter ranges are also appropriate for models in which gmc formation is induced by galactic spiral density waves or energetic supernova . moreover , by following the primary cooling lines in the atomic shock and the post - shock evolution we can predict fluxes for key transitions of c ii , c i , o i , co and other species spanning a range of initial ram pressures . this should aid in the search for the progenitors of molecular clouds . finally as a by - product of this work we provide an analytic solution for the formation of molecules behind a shock and discuss the relation between gas temperature and extinction , both of which may be of some use to mhd modeling . in 2 we describe the model with results and a parameter study provided in 3 . in 4 we outline the observational possibilities for the detection of forming molecular clouds via both emission and absorption lines . section 5 summarizes the implications of these results for star formation in the local neighborhood and in 6 we present our conclusions . we combine the results of two separate one - dimensional plane - parallel models to examine the full thermal and dynamical evolution of parcel of atomic gas that is heated by a shock , condenses , and cools via radiation while slowly evolving to molecular form . the first model ( labeled as the _ atomic _ code ) examines the initial shock wave propagating through a purely atomic medium . this code , described in @xcite and @xcite , has been extensively used to examine the coupled dynamical and ionization evolution of shocked gas in the ism . the second model ( labeled as the _ molecular _ code ) is a time dependent chemical model that has been used to examine molecular chemical evolution in the dense regions of the galaxy that are currently forming stars @xcite . this model computes the time evolution through the atomic / molecular transition and has been modified to include the same flow equations , dominant heating and cooling processes as @xcite ( but see also * ? ? ? * ) . additional processes more relevant to the atomic / molecular transition have also been included . these will be discussed below . in each code the fluid equations for a steady state plane - parallel flow are used @xcite : @xmath10 @xmath11 @xmath12 where @xmath13 is the density in @xmath14 , @xmath15 is the gas velocity , @xmath16 is the mass density , @xmath17 is the pressure , and @xmath18 is the magnetic field . @xmath19 is the enthalpy per particle , while @xmath20 and @xmath21 are the cooling and heating rates respectively . these equations are solved simultaneously with the chemistry to provide atomic and molecular abundances , gas temperature , and gas density as a function of distance / time . in the following we assume that @xmath22 , but in our dynamical calculations we have accounted for the presence of heavy atoms . in the one - dimensional framework as the shock propagates through the medium there is a build - up of colder and denser material behind the shock . gas and dust in this column will absorb interstellar uv radiation . the majority of absorption is by dust grains , but as molecules begin to form self - shielding of h@xmath0 or co molecules can become important . the visual extinction can be estimated via the time integration of equation ( 1 ) : @xmath23 in this equation @xmath24 is assumed to be the moment of the initial shock in the medium . this calculation has been included in both models to examine the attenuation of x - rays and the uv photons responsible for gas heating and photodissociation . in our models we have assumed that a shock , with speeds up to 50 @xmath25 , propagates into the galactic warm neutral medium with an initial temperature of 1000 k. the pre - shock density is set to the average density of the ism @xmath13 = 1 @xmath14 , but higher pre - shock densities have been explored . in all the models examine a range of shock ram pressures from 1 30 @xmath26 @xmath14 k. since dust is required for the uv shielding we have adopted depleted metal abundances based on observations of @xmath27 oph . abundances and references are provided in table 1 . species are assumed singly ionized or neutral depending upon whether their ionization potential is above ( he , n , o , ne , ar ) or below ( c , mg , si , s , ca , fe , ni ) 13.6 ev . at the start of the calculation h is predominantly neutral with a fractional ionization of 10@xmath28 in the form of h@xmath29 . the choice of magnetic field strength necessitates special comment . as noted by many authors , even a very small component of the interstellar magnetic field tangential to the flow can prevent compression sufficient to make molecular gas . if the ram pressure of the shock is equal to the typical value in the local ism , the equipartition magnetic pressure occurs @xmath30 g ; any field parallel to the shock of this magnitude will prevent further compression . in a plane - parallel steady shock , the parallel field will be linearly proportional to the density , @xmath31 . if we wish to compress the gas from an initial density of , say , 3 @xmath32 to @xmath6300 @xmath32 , the parallel magnetic field must increase by a factor of at least 100 from its initial value . but @xmath33 g is the typical rms field strength in the diffuse ism . thus , unless the initial field parallel to the shock is only @xmath34 1% of the initial total field strength , compression to form a cloud can not occur . this qualitative argument is not changed unless the ram pressure is enormously greater than the typical ism value . hbb01 suggested that the solution to this problem is indicated by the mhd calculations of passot , vzquez - semadeni , & poquet ( 1995 ) ; specifically , that clouds tend to form at bends or `` kinks '' in the magnetic field , allowing material to flow along the field and finally compress in a direction perpendicular to the magnetic field . ( see figures 4 and 5 of hbb01 and related explanation in the text . ) putting it another way ; flows which are _ not _ preferentially along field lines ( bent or not ) _ can not _ form even diffuse clouds , let alone molecular clouds . thus clouds must preferentially occur when conditions are such that the magnetic field does not prevent compression ; flows in other directions , for other conditions , simply do nt make clouds . based on this argument , and the results from the simulations of passot et al . ( 1995 ) , we assume that the _ initial _ magnetic field component parallel to the shock front is not dynamically important , adopting a nominal value below 0.01 @xmath35 g . note that the component of the magnetic field perpendicular to the shock front is unconstrained in this one - dimensional model . further , we note that even with such a small initial parallel magnetic field , the post - shock parallel magnetic field strength is not negligible in some cases , and can approach @xmath36 g . the _ atomic _ and _ molecular _ models include relevant chemical processes specific for the treated phase . the _ atomic _ model self - consistently solves for the time - dependent ionization state of all species included in table 1 . however , this model does not include molecular processes . in particular , the h@xmath37 mechanism to form h@xmath0 is not incorporated , as this process is more important in fast shocks with significantly higher pre - shock densities and greater ion fraction @xcite . this model evolves until the post - shock gas temperature reaches 1000 k at which point the _ molecular _ model adopts the final physical and chemical state from the _ atomic _ code as initial conditions . all important parameters ( density , temperature , abundances , ly @xmath38 flux , heating rates , cooling rates , ionization fraction ) are uniform to within a few percent at this hand - off . the _ molecular _ model self - consistently solves for the time dependent chemistry of molecular formation from a purely atomic medium using a minimized chemical network that focuses on the formation of h@xmath0 , co , and other simple carbon and oxygen - bearing molecules . the system of ordinary differential equations governing the chemical evolution is solved using the dvode algorithm @xcite . in this calculation a conservation equation is included for each atomic pool . the minimized rate network has been used previously to examine the formation of water and molecular oxygen in dense molecular clouds @xcite , and under those conditions has been tested extensively through successful comparisons to a larger network @xcite . the model has been adapted to include the formation of h@xmath0 via surface recombination on grains with a rate provided in table 2 . we have assumed that the dust temperature is low enough to allow for h@xmath0 formation at the rate estimated in the ism by jura ( 1974 ) . the calculations of cazaux & tielens ( 2004 ; see also cazaux & tielens 2002 ) indicate a near - unity efficiency of formation of h@xmath0 , provided the dust temperature does not exceed @xmath39 k. in the ism the average dust temperature derived from cobe data lies between 16 23 k ( reach et al . 1995 ) ; one expects that in regions with the needed dust extinction for shielding ( see below ) , prior to star formation , the dust temperature will lie closer to the lower value . the _ molecular _ model does not treat the ionization state of atomic species beyond the singly ionized phase . this assumption is valid as most highly ionized atoms recombine early in the post - shock phase that is treated by the _ atomic _ model prior the hand - off . one additional simplification in the _ molecular _ model is that it does not include all the atomic species listed in table 1 . instead the most abundant atomic species are included ( h , he , c , o ) with atomic and neutral iron as the sole representative heavy ion . this difference does not affect our results . for the photodissociation rates of molecular species we use the rates compiled by @xcite , except for co and h@xmath0 where we adopt the self - shielding formalisms of @xcite and @xcite . these approximations include the effects of dust absorption and molecular line shielding . the h@xmath0 self - shielding approximation ( eq . 36 ) of @xcite is well matched to more exact models with a turbulent velocity parameter of 3 km s@xmath40 . the purely thermal line width in our calculations is generally well below this value . we do not correct for this difference , and under the assumption of small turbulent widths , our models would under - predict the effects of h@xmath0 shielding . this is not a question for co , because the dissociation lines are broadened due to pre - dissociation @xcite . we have also assumed the radiation field from the shock is not greater than the isrf . any significant ultraviolet radiation from the shock would lengthen timescales for molecular formation . however , our favored solutions involve weaker shocks ( @xmath41 km s@xmath40 ) impinging upon gas with above average density . such shocks would produce smaller amounts of uv radiation that will not significantly alter our primary results . even the faster shocks we consider produce little uv radiation aside from ly @xmath38 , and the large h i column in the atomic part of the flow effectively shields the molecular region from ly @xmath38 photons . we note that the _ atomic _ model does not work well for shock speeds below @xmath42 . however , for many purposes the _ molecular _ model behavior is independent of the results from the _ atomic _ model , as long as the same pressure is maintained . thus , one may extrapolate many of the results of the _ molecular _ model to slower shocks with the same ram pressure . both the _ atomic _ and _ molecular _ models include the relevant heating and cooling processes for each regime . thus both codes contain photoelectric , x - ray , and cosmic ray heating . the _ molecular _ model also includes heating due to the formation of h@xmath0 . each model contains cooling via c ii , c , and o line emission and h@xmath29 recombination . the molecular model also includes cooling via co rotational emission in the ground vibrational state . following the example of @xcite in table 2 we summarize the major thermal processes and references included in these models . in the initial stages of the post - shock evolution , while the excess dynamical energy is lost through radiation , the temperature and compression factor ( @xmath43 ) are determined using the fluid equations ( 2.1 ) and methods outlined by cox ( 1972 ) . at later stages when the dynamical energy of the shock has dissipated the temperature is derived through thermal balance of the heating and cooling terms via standard newton - rhapson techniques ( press et al 1988 ) . pressure balance then provides the compression factor . these solutions are determined in tandem with the chemical evolution . our standard model assumes an initial shock velocity of @xmath44 km s@xmath40 impacting gas with a temperature of @xmath45 k and a density of @xmath46 ( @xmath47 28000 @xmath14 k ) . this shock speed is consistent with that induced by either spiral density waves @xcite or turbulence driven collisions between flows ( ballesteros - paredes et al 1999 ) . a supernova will likely induce a series of shocks with a range of ram pressures covered by this standard solution and the additional runs that explore parameter space . in all calculations presented below the _ atomic _ model is used to examine the gas evolution while the temperature exceeds 1000 k. when gas temperature in the post - shock evolution decays to below this value the _ molecular _ model is used to examine the subsequent evolution . figure [ basic](a - f ) presents the physical and chemical evolution of the post - shock gas . here panels ( a - c ) show the major heating and cooling terms along with the primary physical parameters ( density , temperature , optical extinction ) as a function of time . panels ( e - f ) show the chemical abundances ( relative to total h ) of important molecular and atomic species as a function of time . in this figure t = 0 represents the onset of the shock . after the shock the gas temperature is quite high ( @xmath48 k ) with the primary cooling is via [ c ii ] , [ o i ] , [ si ii ] , and [ fe ii ] line emission ( hydrogen line emission is also an important coolant in the initial stages ) . at early times ( @xmath49 yr ) the heating processes which dominate the cold neutral medium ( cnm ) , photoelectric and x - ray heating , are unimportant . at t = 10@xmath50 yrs the temperature drops sharply at the cooling timescale set by the overall cooling rate . for this case ( @xmath51 and an initial density of 1 @xmath14 ) the cooling timescale is @xmath52 yr ( spitzer 1978 ) . when the gas cools the density rises due to pressure equilibrium . at this point the solution reaches a stable high density atomic phase where photoelectric heating is balanced by [ c ii ] cooling . this stable `` plateau '' is essentially the wolfire et al . ( 1995 ) cnm solution and is roughly at the temperature of the cnm as measured by @xcite . the plateau exists for @xmath53 yr until the onset of h@xmath0 formation which is followed by the formation of co. at this point co begins to dominate the cooling reducing the temperature to @xmath4 k. photoelectric heating dominates throughout most of the post shock evolution until a@xmath54 mag , whereupon cosmic - ray heating is more important . prior to the formation of co there is a period where the abundance of neutral carbon rises and it contributes to the cooling , but it is not the dominant carbon reservoir . the timescale of this process is controlled by the formation time of h@xmath0 on grains and the slow buildup of shielding due to grains and h@xmath0 molecules downstream in the post - shock gas . at the density of the stable plateau ( n @xmath55 @xmath14 ; t = 23 k ) the h@xmath0 formation time is @xmath56 yr . the onset of molecular formation occurs at much later times and therefore _ the evolution is controlled by the shielding of uv radiation . _ in our standard case the shielding of h@xmath0 is dominated by self - shielding with only small contributions from dust absorption . for co the shielding is dominated through the uv photon absorption by dust grains . in the following we will examine different initial ram pressures through changes in the shock speed given the same initial density and temperature as the standard model . we also examine a few cases with a constant shock velocity but varying initial densities , to verify the similarity of results for similar ram pressures for the molecular gas . in figure [ phys_ev ] we present the results from models with @xmath57 ( @xmath45 k ; @xmath46 ) , which have ram pressures ranging from 1.4 36 @xmath58 @xmath14 k. the h@xmath0 and co abundance as a function of time and visual extinction are given in figure [ abun_ev ] . rather than show the detailed evolution ( as in figure [ basic ] ) we present the temporal evolution of salient parameters : density , temperature , and extinction . we note that the cooling time to t @xmath59 k is extremely short due to collisional excitation of h i. thus the much higher immediate post - shock temperatures in the faster shock models do not appear in figure [ phys_ev ] . in figure [ phys_ev ] the cooling timescale to reach the stable temperature plateau decreases with increasing ram pressure . similarly the timescales of molecular formation decrease ( figure [ abun_ev ] ) . both effects are due to the density of the solution plateau increasing with ram pressure . for h@xmath0 , in each case , except for the 10 @xmath60 shock , self - shielding dominates over dust absorption . the reverse is true for co , dust absorption dominates over self - shielding , except for a 50 @xmath60 shock , where self - shielding becomes more important at a@xmath61 mag . in figures [ phys_ev_ic ] and [ abun_ev_ic ] we examine models with the same shock speed ( @xmath62 ) but different initial densities . we also show the @xmath63 , @xmath46 case as it has similar ram pressure as the @xmath62 , @xmath64 solution . here we see similar effects as illustrated by models with increasing shock velocity , all timescales ( cooling , molecular formation ) shorten with increasing density . models with similar ram pressure ( compare p / k = 32000 @xmath14 k with p / k = 29000 @xmath14 k ) have slightly different evolution in the physical and chemical properties seen in fig . 4 and 5 . indeed the model with the slightly smaller ram pressure has faster physical and chemical evolution . in this model the temperature decay occurs more quickly because of the higher initial density , which increases the cooling through collisionally excited lines . subsequently , h@xmath0 forms earlier , and the higher @xmath65 allows for faster co formation . overall the differences between these two cases are not large and are magnified by the log scale . however , this points out that the effects of mass conservation and pressure conservation are not equivalent . solutions with higher densities will have faster dynamical evolution through increased cooling and quicker _ molecular _ evolution due to the decrease in the time to reach full shielding . in principle the primary observational evidence for the cloud formation process would be h i emission from dense ( @xmath66 @xmath14 ) purely atomic gas . however , in practice this is difficult because the excitation characteristics of 21 cm emission , and the pervasive presence of warm h i emission in the galactic plane , makes it difficult to disentangle cold and warm gas along a given line of sight @xcite . some cold h i clouds have been isolated from emission studies @xcite , but their relation to forming molecular clouds is unclear . a better way to detect cold atomic hydrogen is via absorption against warm h i background emission ; and this has received more attention in recent years @xcite in general , the search for molecular cloud formation has typically concentrated on the cold stages of the evolution ( i.e. not the shock itself but on the build up of molecular material ) . thus catching a cloud undergoing incipient formation can be pursued by searching for gas that is dense by galactic ism standards ( @xmath67 @xmath14 ) , with emission from atomic carbon but not associated with co emission . however , any mechanism ( e.g. spiral density wave , supernova shock , turbulent flow collision in ism ) that produces a shock wave which compresses the gas will produce emission and absorption associated with both warm ( t @xmath68 1000 k ) and cold ( t @xmath69 k ) gas . thus the important cold phase can be searched for via traditional tracers ( e.g. [ c i ] , co ) , but traces of the shock may also be evident in atomic emission and absorption . in the following sections we will present our model results which provide detailed predictions of atomic and molecular emission and absorption tracing the formation of a dense condensation of molecular gas . table 3 shows the relative line intensities of the stronger atomic emission lines . they are generally similar to the spectra of slow shocks used to interpret herbig - haro object spectra ( hartigan , morse & raymond 1994 ) , except for the effects of the much higher densities in the hh object shocks . depending on shock speed , the spectral signature may be strongly enhanced h@xmath38 or strongly enhanced [ s ii ] emission . the surface brightness of a slow shock in tenuous gas is quite low , typically a fraction of a rayleigh , and a shock seen face - on would be difficult to detect . however , such a shock seen edge - on would be geometrically brightened in the same manner as emission filaments in supernova remnants . edge - on shocks might be found near star forming regions , but depending on the thickness of the colliding clouds , the shock emission may cease before stars are formed . therefore , the emission might be best seen as an extension of a filamentary molecular cloud . the shocks studied here might also be detected through their cumulative contribution to the diffuse galactic emission . for instance , haffner et al ( 1999 ) report that some heating beyond that supplied by photoionization is required to account for the strengths of [ n ii ] and [ s ii ] lines detected at low spatial resolution by the wisconsin h@xmath38 mapper ( wham ) . minter & spangler ( 1997 ) suggest that dissipation of turbulence might provide that extra heating . if that dissipation occurs in shocks , the shocks in the warm ionized medium ( wim ) would provide the [ n ii ] , while those in the warm neutral medium ( wnm ) would contribute [ o i ] and [ s ii ] . while much of the shock energy is released via cooling lines in the uv , optical , and infrared , the majority of emission in cool post - shock phase lies in far - infrared and millimeter wavelengths . in figure [ intensity ] we provide a plot of line intensity ( erg cm@xmath28 s@xmath40 sr@xmath40 ) as a function of time and visual extinction for our standard model ( @xmath71 and @xmath44 @xmath60 ) . figure [ intensity ] shows only the evolution during the cold post - shock phase , but includes contributions to emission from the atomic shock model . in this figure some transitions ( e.g. [ o i ] ) do not emit during the cold post - shock evolution , while others ( [ c ii ] , [ c i ] , co ) show a slow steady increase in emission with time due to the build up of column behind the shock . in figure [ iratio ] we provide diagnostic plots of line intensity ratios for specific transitions that can be used as suitable tracers of cloud formation . in this plot examples are given for a range of parameter space . because there is little [ o i ] emission in the cold phase the line ratio of [ o i](63 @xmath35m)/[c ii](158 @xmath35 m ) exhibits a factor of @xmath630 drop until it saturates when co has formed and removes carbon ions from the gas . the intensity ratio of [ c i](610 @xmath35m)/co ( 1 - 0 ) also strongly varies ( by nearly 2 orders of magnitude ) over much of the evolution until the ratio again flattens due to the formation of co. a key question for these diagnostics is whether this signature is unique or is indistinguishable with galactic photodissociation region ( pdr ) emission arising from gas with similar conditions . to examine this question we use the pdr models of @xcite . in the pdr model the [ o i](63 @xmath35m)/[c ii](158 @xmath35 m ) ratio ranges from 0.03 @xmath72 0.3 for conditions that mimic those in our models ( n @xmath73 ; g@xmath74 ) . this range is quite similar to that seen in our results ; only for a short time at the start of the post - shock evolution is the ratio significantly greater . thus this ratio is not a strong signpost of molecular cloud formation . however , the [ c i](610 @xmath35m)/co ( 1 - 0 ) ratio does differ from pdr models ( which predict ratios between 25 50 for similar conditions ) . this is essentially due to the increased column of atomic carbon during stages prior to co formation , in pdr models the atomic carbon is created via photodissociation of co. however , it should be noted that the [ c i](610 @xmath35 m ) emission is quite weak at times where the [ c i](610 @xmath35m)/co ( 1 - 0 ) ratio is highly elevated above typical pdr values . best results appear to require deep searches for [ c i](610 @xmath35 m ) emission to a level below @xmath60.1 k km s@xmath40 ( i = 1.22 @xmath75 erg cm@xmath28 s@xmath40 sr@xmath40 ) . in some instances favorable placement may provide a background source to detect cold post - shock gas absorption . for h i , galactic hydrogen emission provides a nearly ubiquitous background . in our models the total column for a given species is determined by the initial shock conditions , the chemistry , and also by the time evolution . thus , there is no single value of the column density . in table 4 we provide a sample of the predicted column densities of atomic and molecular tracers as a function of shock model and for a sampling of total extinction ( a@xmath76 = 0.2 , 0.5 , and 1.0 mag ) . for atomic hydrogen we have divided the column by noting the amount that resides in warm ( t @xmath77 k ) and cold ( t @xmath78 k ) gas . for the weakest shock ( @xmath79 @xmath60 , @xmath71 ) the warm h i column is significant in comparison to the column of cold gas . however , the cold gas column dominates in all other cases . although the warm gas column is not provided for o i and c ii , similar results are seen for these atoms as well . typically the warm column is @xmath6few % or less of the column at any given extinction . in the case of co and c i , these species only form when the gas has cooled . some general trends are seen in table 4 . first , the warm and cold h i column is lower for stronger shocks , while c i and co columns increase with shock velocity / ram pressure . this is due to the faster chemical and physical timescales at the higher post - shock densities . in contrast because neutral oxygen and ionized carbon are the respective atomic reservoirs during much of the evolution , their column densities are relatively insensitive to shock strength , showing only a factor of @xmath80 variation at the highest extinction . * hereafter jt01 ) used uv absorption observations of atomic carbon multiplets to determine the relative populations in fine structure levels along numerous galactic lines of sight . based on excitation ratios they find that the median pressure in the cnm is @xmath81 @xmath14 k and that many lines of sight exceed this median value . in figure [ carbon ] the grey points show the distribution of the 1st atomic carbon fine structure level fractional population ( f1 = @xmath82 ) as a function of the 2nd ( f2 = @xmath83 ) . these numbers have been derived along numerous lines of sight in the ism by jt01 . the cluster of points between f1 = 0.12 0.25 and f2 = 0.0 0.1 show typical cnm ratios near the median pressure . based on excitation analyses , clouds with f1 @xmath68 0.3 and f2 @xmath68 0.1 receive some contribution from over - pressurized gas ( jt01 ) . @xcite theorized that the pressure enhancements are the result of compression due to converging turbulent flows or warm boundary layers in the surfaces of dense clouds moving within a lower density medium . the predicted fine - structure excitation ratios from our models , which are of the shock creation of such over - pressurized gas , are also shown in figure [ carbon ] . we have selected three models , each with an initial density of @xmath71 @xmath14 and shock velocities of 10 , 20 , and 50 @xmath60 , with corresponding to ram pressures of 1.4 , 5.8 , and 36 @xmath84 @xmath14 k. the predicted ratios overlap with the `` high '' pressure gas seen by jt01 , which demonstrates that high c i excitation ratios are an additional signature of the shock . weaker shocks , which evolve on longer timescales , are likely to be more prevalent in the ism and thus these shocks ( @xmath85 @xmath14 k ) appear more representative of observations . we did not model such weak shocks as they will not evolve on the timescales required for fast star formation ( see 5.1 ) . however , the excellent agreement of our models with the high excitation end of this distribution suggests that the jt01 results could indeed be the result of a distribution of weak and strong shocks active in the atomic medium throughout the galaxy . our model calculations raise an obvious question : when should we consider that a `` molecular cloud '' has been formed ? as shown in figures 1 and 3 , @xmath3 can form at considerably earlier times and lower column densities than co , especially for low ram pressures / shock velocities . thus low - pressure clouds could become molecular , in terms of the dominant constituent , before they become co clouds . however , @xmath3 is difficult to detect , and is not usually a criterion for defining a molecular cloud . we consider the problem of @xmath3 detection in the next subsection ; here we concentrate on the detection of co. it is useful to distinguish between what we call the `` accumulation '' timescale , the length of time it takes to accumulate @xmath86 from the atomic gas and the cloud becomes `` detectable '' in co , from other timescales , for instance the evolution of the co abundance . for purposes of discussion we have made the somewhat arbitrary decision that the cloud appears when co j=10 emission can be detected at a level of 1 k km / s which defines the accumulation timescale . in figure [ co_flux ] we provide the intensity of the j=10 transition of co as a function of time with the dashed line denoting an integrated intensity of 1 k km / s . based on this definition , the cloud accumulation timescale ranges from @xmath87 yr for a 50 @xmath60 shock to @xmath88 yr for a 10 @xmath60 shock ( both with an initial density of 1 ) . as noted before , we find that the column density or extinction is the most important parameter in determining the formation of co. based on the right - hand panel of figure [ co_flux ] , the above definition is roughly equivalent to molecular cloud `` formation '' at @xmath86 mag for ram pressures between @xmath89 @xmath14 k. threshold will increase by an amount of @xmath90 . here we list the uv enhancement factor g@xmath91 as a function of time to denote the fact that effects of radiation will diminish as the shock front moves away from the star . ] the above threshold of @xmath86 mag is based on a 1-dimensional calculation and it is useful to discuss how this might change in a more realistic 3-dimensional geometry . chemically the transition from cii / ci / co will be seen at a level where the co photodissociation becomes ineffective . in 1-d this threshold is found at 0.7 mag with little contribution from self - shielding . in a three - dimensional calculation for a clumpy cloud , there will be two , competing , effects : first , uv radiation will propagate more freely into the cloud ; and second , clumping may increase the local column densities and thus promote the transition to molecular gas . our 1-d requirement of @xmath86 would thus translate into an average over solid angle of the extinction of the diffuse ultraviolet radiation field . the time to reach this threshold is no longer a simple calculation of the @xmath92 product , though we expect that the transition to molecular gas will occur roughly when the average density / column density reaches values comparable to those in our calculations . we also note that in addition to clumping , overall contraction of the cloud in the direction perpendicular to the shock front will promote rapid molecular gas formation by increasing the density and the shielding . recent numerical simulations of finite self - gravitating sheets ( burkert , a. , & hartmann , l. 2004 , in preparation ) suggest that global gravitational collapse of flattened clouds is likely and rapid . in hbb01 we showed that the stellar population ages of nearby molecular clouds imply cloud lifetimes of no more than about 3 - 5 myr . generally , slightly older associations are not immediately next to current sites of star formation , indicating that molecular material is not simply being pushed around but that there is some cycling between the atomic and molecular states . the ratio of atomic to molecular gas in the solar neighborhood is estimated to be @xmath93 @xcite . this implies that the timescale for turning atomic gas into molecular material , on average , is about @xmath94 myr . ( if only a fraction of the atomic gas cycles through molecular stages , the timescale for the conversion of atomic to molecular gas in specified regions must be shorter . ) in our terminology the cloud accumulation timescale must therefore be @xmath95 20 myr ; the cloud lifetime would then be defined by the time from when a cloud is detectable in co emission until the gas is dissipated by star formation . thus , the one - dimensional constraint on the accumulation timescale timescale @xmath96 implied by the ratio of atomic to molecular gas , coupled with the requirement that @xmath86 for co formation , constrains the possible average pre - shock parameters , @xmath97 can this constraint reasonably be satisfied ? the one - dimensional rms turbulent velocity of cold h i in the local interstellar medium is @xmath98 ( boulares & cox 1990 ) ; thus it seems reasonable to take a typical shock velocity @xmath99 . accumulation timescales of 10 - 20 myr then require typical pre - shock densities @xmath100 . this is a few times larger than the average density of the ism , but it would not be particularly surprising if molecular clouds were preferentially formed from initially slightly higher densities . moreover , it has been increasingly accepted that the density fluctuations in the ism ( clouds ) are produced primarily by compressions due to a supersonic turbulent velocity field ( von weiszacker 1951 ; sasao 1973 ; elmegreen 1993 ; padoan 1995 ; ballesteros - paredes , vzquez - semadeni & scalo 1999 ) . in such an environment , the production of the density fluctuations are likely the result of a succession of compression events , so that denser structures are formed by compressions within previously compressed , larger ones , rather than a single , very strong one ( vzquez - semadeni 1994 ) . this scenario naturally explains the density probability density function observed in numerical simulations of isothermal and polytropic flows @xcite , and here it naturally provides the necessary conditions ( 24 @xmath14 ) for the pre - shock gas in our calculations . as already pointed out , ram pressure is the most important parameter for co formation ; the ram pressures for the above parameters are a few times the average turbulent gas pressure ( gas plus magnetic plus cosmic rays ) in the ism ( boulares & cox 1990 ) ; again , this does not seem to be an unreasonable constraint for making the highest density regions in the solar neighborhood . once the shielding column density is achieved , co formation is rapid . additional increases in density due to subsidiary shocks and/or gravitational contraction will yield to even more rapid co formation . at a column density corresponding to @xmath101 , comparable to the shielding length estimated here , the characteristic growth time for gravitational contraction in ( subsonic ) gas at @xmath102 k is of the order of 1 myr . thus , once molecular gas is formed , star formation can ensue rapidly , at least in regions where supersonic turbulence has been dissipated ( hbb01 ) . we conclude that our results are consistent with the picture presented in hbb01 of rapid formation of molecular clouds from atomic material , as long as the starting densities are typically a few times the average interstellar density . here we consider whether the presence of some molecular hydrogen in the pre - shock gas would shorten either the accumulation timescale or the timescale for co abundances to rise . indeed , because h@xmath0 does not emit for typical conditions in the atomic cnm it might be possible to hide a significant molecular component . to examine this question we have examined solutions with a substantial h@xmath0 fraction in the pre - shock gas . this is only performed using the _ molecular _ model because the _ atomic _ model does not include chemical processes linked to h@xmath0 . one limitation is that the cooling via h@xmath0 emission and shock dissociation of h@xmath0 are not included . however , the addition of h@xmath0 cooling will only shorten the cooling time to reach the stable plateau solution . as we will show below , the chemical evolution ( i.e. co formation ) is dominated by the shielding timescale which depends primarily on the initial shock parameters . in figure 10 we present the h@xmath0 and co abundances as a function of time and extinction in our standard model with an initial h@xmath0 fraction of 0.0 , 0.125 , and 0.25 . the sharp rise of the h@xmath0 abundance at t = 10@xmath50 yr is due to the non - inclusion of h@xmath0 in the _ atomic _ model . at this time the abundance of co shows a sharp spike which is due to rapid co formation from the pre - existing h@xmath0 . however , co molecules are quickly dissociated , due to the lack of uv shielding . what is striking in these plots is that even an a priori presence of h@xmath0 molecules a necessary requirement for co formation has little effect on the co chemical evolution . this effect is discussed in 3 , because the co formation requires dust shielding the evolution can not proceed until sufficient dust column exists . this result is robust provided the pre - shock gas is exposed to the isrf with a@xmath103 mag . in figure 11 we present solutions with higher initial extinction and an initial h@xmath0 fraction of @xmath104 . the left - hand panels show the evolution of the co concentration , while the right - hand panels present the co j=10 emission , which we have used to define the cloud accumulation timescale ( @xmath105 = time where emission reaches a level of 1 k @xmath60 ) . with this definition even pre - shock gas with a@xmath106 mag will only accumulate on timescales a factor of two shorter than the model starting with unshielded gas . moreover the differences in the emitted j=10 intensity between these solutions are negligible unless @xmath107 mag . in sum , the presence of h@xmath0 in the pre - shock gas would not hasten the timescales for cloud formation , unless the gas is already significantly shielded from uv radiation and co is already effectively in existence . let us now place these results in an observational context . allen and co - workers have argued for a pervasive molecular component on the basic of the relative placement of dust lanes , radio continuum , and h i emission in external spiral galaxies @xcite . in the standard picture of cloud formation h i emission would appear in front of the spiral shock wave as traced by the radio continuum emission . instead these authors found that the h i emission is observed downstream of the shock , which is interpreted as the result of photodissociation of h@xmath0 by young stars . this led @xcite to theorize that there may be a significant reservoir of molecular gas in the low density cnm which would allow for fast cloud and star formation . they argue that this inter - arm gas could be hidden by having the temperature colder than 10 k , producing only weak co emission . however , these results must be placed in the context of studies of h@xmath0 in absorption and co in emission in our own galaxy . the most extensive initial study of h@xmath0 in absorption was performed by the copernicus satellite and @xcite found that the average h@xmath0 fraction within 500 pc of the sun is @xmath108 . more recent fuse observations have demonstrated that h@xmath0 absorption is pervasive , even in sight lines well out of the galactic plane towards background extragalactic sources @xcite . in addition , an analysis of x - ray absorption spectra along similar lines of sight by @xcite finds that the required x - ray absorption column exceeds the h i column estimated by 21 cm emission . this excess is difficult to account for via other atomic components , but could be due to molecular gas . the uv and x - ray observations point to the potential ubiquitous presence of h@xmath0 in the galaxy with fractions of h@xmath0 towards these high galactic latitude lines of sight ranging from @xmath610% @xcite to @xmath109% @xcite of the h@xmath0 fraction estimated in the local solar neighborhood . while there is important evidence for the presence of some h@xmath0 in the low density cnm , the evidence for co is less substantial . for instance , the surface density of molecular gas estimated by co emission towards high galactic latitudes is only 1% of the value derived in the local neighborhood . this is well below that seen for the h@xmath0 fraction and hints that co may not be fully tracing h@xmath0 in the low density ism . this is not terribly surprising , because the column density needed for equivalent co self - shielding is three orders of magnitude larger than that needed for h@xmath0 formation ( lee et al . 1996 ; draine & bertoldi 1996 ) . given the possibility of a significant present of h@xmath0 in the ism one might question whether significant reservoir of h@xmath0 molecular clouds exist that are `` inert '' in the sense that they will never create co. there are two possibilities that can be examined in this regard . first the @xcite results suggest an abundance of weak shocks are active in the galactic ism . these shocks will evolve on slow times for the creation of both h@xmath0 and co. thus in the @xmath110 myr required to make h@xmath0 there is an additional @xmath110 myr before co creation ( see fig . for such a weak shock there will be a significant amount of time spent in the molecular but pre - co phase . the overall lifetime of such systems may be limited by the large distance that would be transversed . for example , a @xmath6 5 km s@xmath40 shock will travel a distance @xmath111500 pc and therefore has an increased likelihood of passing by a massive star that will certainly destroy the co and perhaps h@xmath0 . an alternate picture would emerge if the shock evolution were limited in some fashion ( perhaps by a strong tangential b - field ) . this will halt the density evolution but column will continue to build up behind the shock . thus , the combination of ensuing extinction evolution , h@xmath0 molecules , and cosmic ray ionization eventually leads to co formation . however , if the density evolution were stunted at a values below @xmath112 then the h@xmath0 phase would last longer in a similar fashion as the weak shock case . however , the importance of such inert clouds might be limited by passing hot stellar wind or supernova bubbles which could disrupt or evaporate them . in sum we can not discount the possibility that there might be a hidden reservior of h@xmath0 molecular gas in the ism . in this picture with ubiquitous h@xmath0 , and little co , cloud formation timescales will depend primarily on the timescale to accumulate a sufficient shielding column density , not the h@xmath0 formation time . hence the solution is similar to that found in the traditional case where cloud formation is modeled from the atomic - molecular hydrogen transition . complex mhd numerical simulations of structure formation in the ism , such as those performed by @xcite , typically neglect the chemical considerations when computing the evolution . this is easily understood given the extreme computational complexities . in our simulations we have included the chemistry but with a simpler dynamical prescription . in this process we have identified two areas where this work can have useful impact on numerical simulations of cloud formation and evolution . our results point to the key importance of the shielding of uv photons in the overall physical evolution . in figure 12 we show the temperature evolution as a function of the mass surface density ( visual extinction ) . the initial spikes below 1 g cm@xmath28 are the shock and initial cool down . the following evolution shows the gradual temperature cool down as the shielding column increases . for all shocks the gas temperature , at a given surface density , is nearly identical , to within a factor of @xmath113 , and declines in a linear fashion in all cases except for the strongest shock . this difference is due to the lack of atomic carbon cooling which typically contributes to the cooling at a@xmath114 1 ( in the fast shock quick co formation via self - shielding reduces the influence of c i ) . the similarities of these solutions and the linear decay suggests that tracking the shielding in each cube of a mhd simulation can be used to provide a simple estimate of the gas temperature and hence pressure . another potential aid to mhd simulations is provided in the appendix where we show that there exists an analytical solution for the time - dependence of h@xmath0 formation ( co formation can be treated via equilibrium calculations ) . this analytical solution requires only knowledge of the local gas density and extinction . under those conditions computationally intensive chemical calculations need not be performed , while still keeping the capability to create a realistic simulation that includes the chemical formation of the two most important molecular species : h@xmath0 and co. if this can adapted into mhd models then it provides a method to use simulations to predict maps of molecular emission , which can be readily compared to observations . we have presented a detailed coupled physical and chemical model that examines the shock compression of atomic gas and the slow transition to molecular form . this model includes all of the physical and chemical processes relevant to investigate the formation of molecular clouds via shocks induced by cloud - cloud collisions , spiral density waves , or supernovae within the dynamic interstellar medium . our principle results are as follows : \1 ) we find that the molecular cloud formation timescale is not controlled by the formation rate of h@xmath0 on grains . rather the shielding of molecules from the uv radiation is the limiting parameter . for all but the weakest shocks we find that h@xmath0 self - shields quite efficiently . however , co formation requires shielding of the interstellar radiation by dust grains . thus the cloud formation timescale is effectively set by the time needed to accumulate a column equivalent to a@xmath115 mag in extinction . \2 ) molecular cloud formation times can be as short as @xmath116 myr , as required by our picture of rapid cloud formation from large scale flows , adopting typical velocities in the ism ( @xmath99 ) for starting densities and ram pressures a few times higher than average interstellar values . \3 ) since shielding is required for co formation the a priori presence of h@xmath0 in the low density medium will not appreciably shorten the time required to create molecular clouds . \4 ) we provide detailed predictions of the atomic and molecular emission and absorption that trace the formation of molecular clouds . a subset of these predictions match current conditions observed in over - pressurized gas within the cold neutral medium of the galaxy by jenkins & tripp ( 2001 ) . \5 ) a by - product of this work is an examination of ways to incorporate the effects of chemistry into detailed mhd simulations of structure formation in the galaxy . this includes an analytic solution for the time - dependent formation of molecular hydrogen and a discussion of the overall temperature structure of dense cooling proto - clouds . we thank an anonymous referee for helpful suggestions which improved the manuscript . eab is grateful to e. jenkins for kindly providing his ci excitation data and for discussions with e. ostriker and j. bregman . this work was supported in part by nasa grants nag5 - 9670 and nag5 - 13210 . this research has made use of nasa s astrophysics data system service . for more complex applications it is useful to have an analytical approximation to time dependence of the formation of @xmath3 , following a shock in the low density gas . a reasonable approximation can be derived through the differential equation that describes the @xmath3 evolution : here @xmath118 is the total density , @xmath119 is the temperature , @xmath120 is the temperature dependent grain formation rate of in @xmath121 s@xmath40 , @xmath122 the cosmic ray ionization rate , and @xmath123 the photodissociation rate , which is a function of the visual extinction and the total column density . in equilibrium @xmath124()/@xmath125 and @xmath13(h)/@xmath13 ( ) = ( @xmath122 + @xmath123)/@xmath126 . strictly speaking the total density , formation rate ( temperature dependent ) , photorate ( through the column density build up ) have time - dependencies . however , after the shock the evolution quickly converges to a roughly constant plateau in density and temperature , which is essentially the wolfire et al . ( 1995 ) solution for non - shielded ism gas . this plateau in the physical evolution persists until shielding decreases the photo - electric heating rate and allows co formation . this ultimately lowers the temperature and raises the density by factors of 23 . thus , for a given shock pressure , the density and temperature can be approximated using values from the non - shielded `` wolfire '' solution ( here : @xmath118 , @xmath119 ) . the h@xmath0 photo - rate can be treated in a different fashion , by discretizing eqn . ( a3 ) and lagging the build - up of the h@xmath0 column ( and the determination of the photo - rate ) behind the h@xmath0 density solution . in this fashion the h@xmath0 photodissociation rate from the previous step is used in the current one . these approximations both result in some loss in accuracy , but as will be shown below the difference between the analytical and exact treatments is not large . with these assumptions , ( a3 ) can therefore be solved , where @xmath133 is a constant . using the @xmath134 = 0 boundary condition ( @xmath13(h@xmath135 0 ) , the constant can be determined and , @xmath136 . substituting back , @xmath137 2/@xmath138 + 1/@xmath139 + 1/@xmath140 and @xmath141 : in this expression @xmath138 is the depletion timescale , @xmath139 is the cosmic ray ionization timescale , and @xmath140 is the timescale for photodissociation including self - shielding and dust shielding . this expression can be used to derive a reasonable estimate of the h@xmath0 evolution , with knowledge of the photodissociation rate . using the approximation of draine & bertoldi ( 1996 ) , two factors are required to estimate the h@xmath0 photo - dissociation timescale : ( 1 ) the dust extinction at 1000 and ( 2 ) the h@xmath0 column density . ( 1 ) the extinction can be estimated though the mass conservation equation as described in 2.1 , using @xmath143 @xmath144 , where @xmath145 @xmath146 ( @xmath147 and @xmath148 are the initial pre - shock density and velocity , and @xmath149 is the dust opacity at 1000 ) . _ in simulations the identification of a shock may not be straightforward . the most important quantity is the tracking of the average 2-d/3-d total column , which can be converted to the opacity using @xmath150 . _ ( 2 ) as discussed above , the h@xmath0 column density can be approximated through the buildup of the column behind the shock . figure 13 ( left ) shows the fraction of as a function of time for the exact solution for the 15 km s@xmath40 shock model along with the analytical solution . we also show for comparison , the exact treatment of a limiting case where the is fully shielded from the start . here the analytical solution assumes @xmath151 @xmath14 and @xmath152 k , which is the evolutionary plateau for the 15 km s@xmath40 shock with a pre - shock density of 1 @xmath14 . in this figure the analytical solution lies between the exact solution and the fully shielded calculation . however , the agreement between the analytical and exact treatments is within a factor of two for relevant timescales . the analytical model over - predicts the amount of , because the calculation can not mirror the full treatment of the density and temperature evolution , which leads to increased self - shielding . in the above example , the density at the early stages of the full treatment is @xmath1531400 @xmath14 , thus the analytical example will over - predict the early creation of . in figure 13 ( right ) this effect is demonstrated by presenting the derived column density in the analytical model compared to the exact calculation for the 15 km s@xmath40 shock . the nature of analytical solution will change depending on the initial conditions ( shock velocity , pre - shock density ) . appropriate values of @xmath118 and @xmath119 for different solutions can be easily obtained from the `` plateau '' in figure 2 . however , _ in practice only knowledge of the local density and extinction / shielding need be identified ( gas temperature will track with extinction ) . with these few parameters one can reliably solve for the formation of molecular hydrogen . _ in the case of co its formation requires the tracking of only dust shielding and could be computed via simple equilibrium calculations . ballesteros - paredes , j. 2004a . `` molecular clouds . formation and disruption '' , in `` from observations to self - consistent modeling of the interstellar medium '' , eds . m. avillez & d. breitschwerdt . kluwer academic publishers ( in press ) elmegreen , b. g. 1991 , nato asic 342 , the physics of star formation and early stellar evolution , eds . lada & n.j . kylafis ( dordrecht : kluwer ) , 35 elmegreen , b.g . 1993 , in protostars and planets iii , ed . levy & j.i . lunine , university of arizona press , tucson , 97 lll [ proc ] & & + fuv field & drain field ; g@xmath91 = 1.7 in units of habing field & ( 1 ) + heating rate & formalism of bakes and tielens ( 1994 ) & + attenuation by dust & heating rate attenuated by e@xmath154 ; average attenuation & ( 2 ) + & of uv field by dust between 1000 2000 & + & & + h ionization rate & @xmath155 s@xmath40 & ( 3 ) + h@xmath0 ionization rate & @xmath156 & + heating & @xmath157 erg @xmath14 s@xmath40 & ( 4 ) + & & + ionization and heating & expressions provided by wolfire et al ( 1995 ) ; n@xmath158 ; & + & n@xmath159 from chemical equilibrium calculation & + & & + formation rate & @xmath160 t@xmath161 s@xmath40 ; @xmath162 & ( 5 ) + formation heating & formalism of hollenbach & mckee ( 1979 ) ; except only & + & 4% of formation energy is deposited on grain surface & + & ( takahashi et al 2001 ) & + self - shielding rate & formalism of draine & bertoldi ( 1996 ) & + & & + self - shielding rate & formalism of lee et al . ( 1996 ) & + & & + c ii & impacts with e@xmath163 , h , h@xmath0 ; @xmath164 & ( 6 ) + 157.7 @xmath35 m & a(@xmath165 s@xmath40 & ( 7 ) + o i & impacts with e@xmath163 , h , h@xmath0 , h@xmath29 ; @xmath166 & ( 8) + 145.6 @xmath35 m & a(@xmath167 s@xmath40 & ( 9 ) + 63.2 @xmath35 m & a(@xmath168 s@xmath40 & + 44.0 @xmath35 m & a(@xmath169 s@xmath40 & + c i & impacts with h , h@xmath0 & ( 10 ) + 610 @xmath35 m & a(@xmath170 s@xmath40 & ( 11 ) + 370 @xmath35 m & a(@xmath171 s@xmath40 & + 980 @xmath35 m & a(@xmath172 s@xmath40 & + h@xmath29 & @xmath173(2/3)k(t/300)@xmath174tn@xmath175 erg s@xmath40 h@xmath40 & + co & @xmath176 rotational transitions , collisions with h@xmath0 & ( 12 ) + h@xmath0 & not included due to low temperatures at time of formation & + llllll [ optical ] h@xmath177 & @xmath178 & @xmath179 & @xmath180 & @xmath181 & @xmath182 + h@xmath38 & @xmath183 & @xmath184 & @xmath185 & @xmath186 & @xmath187 + @xmath188 $ ] 6300 & @xmath189 & @xmath190 & @xmath191 & @xmath192 & @xmath193 + @xmath194 $ ] 6723 & @xmath195 & @xmath196 & @xmath197 & @xmath198 & @xmath199 + lccccccccccccc [ colden ] 10/1 & 20.4 & & 20.2 & 20.8 & 20.9 & & 17.0 & 17.4 & 17.8 & & 16.7 & 17.1 & 17.4 + 10/2 & 19.7 & & 20.5 & 20.8 & 20.9 & & 17.0 & 17.4 & 17.8 & & 16.7 & 17.1 & 17.4 + 10/3 & 19.6 & & 20.5 & 20.8 & 20.8 & & 17.0 & 17.3 & 17.7 & & 16.7 & 17.0 & 17.3 + 20/1 & 19.8 & & 20.2 & 20.6 & 20.6 & & 17.0 & 17.5 & 17.7 & & 16.7 & 17.0 & 17.3 + 30/1 & 19.4 & & 20.3 & 20.4 & 20.5 & & 17.0 & 17.5 & 17.7 & & 16.7 & 17.0 & 17.3 + 50/1 & 18.9 & & 20.0 & 20.2 & 20.2 & & 17.0 & 17.3 & 17.6 & & 16.7 & 17.0 & 17.0 + + & & & & & + + & & & & & & & & & + 10/1 & & & 13.9 & 15.0 & 15.9 & & 8.5 & 12.5 & 14.3 + 10/2 & & & 14.6 & 15.3 & 16.3 & & 11.7 & 13.5 & 15.3 + 10/3 & & & 14.8 & 15.6 & 16.5 & & 12.0 & 13.9 & 15.7 + 20/1 & & & 14.9 & 15.7 & 16.5 & & 12.7 & 14.3 & 16.0 + 30/1 & & & 15.3 & 16.0 & 16.5 & & 13.9 & 15.0 & 16.7 + 50/1 & & & 15.8 & 16.3 & 16.3 & & 14.0 & 16.3 & 17.0 + , @xmath147 = 1 , and @xmath200 = 1000 k. in the following panel ( a ) shows the heating rates and temperature ( shown as dotted line with axis labeled to right ) , ( b ) cooling rates , ( c ) density and extinction with labels to left and right respectively , ( d ) h and h@xmath0 abundances , ( e ) carbon and oxygen abundances , ( f ) major electron contributors . , height=755 ] ( top left ) and co ( top right ) abundance . bottom panels show the evolution of the h@xmath0 abundance ( bottom left ) and co abundance ( bottom right ) as a function of visual extinction . all abundances are relative to total h. each panel shows models with different shock speeds which are labeled in the bottom left panel . each model adopts @xmath147 = 1 and @xmath200 = 1000 k. , height=604 ] ( top left ) and co ( top right ) abundance . bottom panels show the evolution of the h@xmath0 abundance ( bottom left ) and co abundance ( bottom right ) as a function of visual extinction . all abundances are relative to total h. the density in @xmath14 and initial shock velocity in k are provided in the bottom left panel . the shock ram pressure in @xmath14 k is given in parenthesis . , height=604 ] s@xmath40 sr@xmath40 ) for selected atomic and molecular transitions arising from the plane - parallel slab as a function of time ( left - panel ) and visual extinction ( right - panel ) in the standard model . for reference 1 k km s@xmath40 is 1.22 @xmath201 for the [ c i ] 610 @xmath35 m line , 1.6 @xmath202 for co ( 1 - 0 ) , and 1.2 @xmath75 for co ( 2 - 1 ) . , height=226 ] 63 @xmath35 m to the [ c ii ] 158 @xmath35 m line intensity shown as a function of time . right : ratio of the intensity of the [ c i ] 610 @xmath35 m line to the co ( 1 - 0 ) line . in both panels several models are shown with a key of velocity / density of the model . thus 20/1 is a model with a shock velocity of 20 km s@xmath40 and an initial density of 1 . , height=226 ] ) and 2nd ( f2 = @xmath83 ) carbon atom fine structure levels . the open grey circles are observed values in the interstellar medium taken from the observations of jenkins & tripp ( 2001 ) . the solid markers denote the range of values predicted in our models with variable shock strength ( ram pressure provided in key ) and with an initial density of 1 . the plotted ratios are the column density of gas in each of the fine - structure levels at a given time ( @xmath203p@xmath0 and @xmath203p@xmath204 ) over the total column and thus accurately reflect what would be observed . the range in values for given shock parameters is due to the time evolution of the post - shock gas . , height=340 ] s@xmath40 sr@xmath40 ) for co j=10 arising from the plane - parallel slab as a function of time . several models are shown with a key of velocity / density of the model . thus 20/1 is a model with a shock velocity of 20 km s@xmath40 and an initial density of 1 . the dashed horizontal line denotes a co j=10 integrated intensity of 1 k km s@xmath40 . , height=226 ] ( top left ) and co ( top right ) abundance . bottom panels show the evolution of the h@xmath0 abundance ( bottom left ) and co abundance ( bottom right ) as a function of visual extinction . all abundances are relative to total h. each panel shows models with different initial h@xmath0 fraction which are labeled in the bottom left panel . each model adopts @xmath148 = 20 @xmath60 and @xmath46 . the sharp rise in the h@xmath0 abundance at t = 10@xmath50 yr is due to the non - inclusion of h@xmath0 in the _ atomic _ model . similarly the rise in co is due to formation from the existing h@xmath0 , but the abundance drops because the molecules are unshielded to uv radiation . , height=604 ] fraction and initial visual extinction in the pre - shock gas . right : predicted line intensities ( erg cm@xmath28 s@xmath40 sr@xmath40 ) for co j=10 arising from the plane - parallel slab as a function of time for the same models as in the right - hand panel . the dashed horizontal line denotes a co j=10 integrated intensity of 1 k km s@xmath40 . , height=415 ]

the subject of this work is the study of-perfect graphs defined as those graphs for which the stable set polytope is achieved in one iteration of lovsz - schrijver psd - operator , applied to its edge relaxation . in particular , we look for a polyhedral relaxation of that coincides with and if and only if is-perfect . an according conjecture has been recently formulated (-perfect graph conjecture ) ; here we verify it for the well - studied class of claw - free graphs . stable set polytope ,-perfect graphs , claw - free graphs

1612.02670

the context of this work is the study of the stable set polytope , some of its linear and semi - definite relaxations , and graph classes for which certain relaxations are tight . our focus lies on those graphs where a single application of the lovsz - schrijver positive semi - definite operator introduced in @xcite to the edge relaxation yields the stable set polytope . the _ stable set polytope _ @xmath2 of a graph @xmath5 is defined as the convex hull of the incidence vectors of all stable sets of @xmath1 ( in a stable set all nodes are mutually nonadjacent ) . two canonical relaxations of @xmath2 are the _ edge constraint stable set polytope _ @xmath6^v : x_i + x_j \ : \le \ : 1 , ij \in e \},\ ] ] and the _ clique constraint stable set polytope _ @xmath7^v : \sum_{i \in q } \ , x_i \ : \le \ : 1,\ q \subseteq v\ \mbox { maximal clique of $ g$ } \}\ ] ] ( in a clique all nodes are mutually adjacent , hence a clique and a stable set share at most one node ) . we have @xmath8 for any graph , where @xmath2 equals @xmath3 for bipartite graphs , and @xmath9 for perfect graphs only @xcite . according to a famous characterization achieved by chudnovsky et al . @xcite , perfect graphs are precisely the graphs without chordless cycles @xmath10 with @xmath11 , termed _ odd holes _ , or their complements , the _ odd antiholes _ @xmath12 as node induced subgraphs ( where the complement @xmath13 has the same nodes as @xmath1 , but two nodes are adjacent in @xmath13 if and only if they are non - adjacent in @xmath1 ) . then , odd holes and odd antiholes are the only _ minimally imperfect graphs_. perfect graphs turned out to be an interesting and important class with a rich structure and a nice algorithmic behavior @xcite . however , solving the stable set problem for a perfect graph @xmath1 by maximizing a linear objective function over @xmath9 does not work directly @xcite , but only via a detour involving a geometric representation of graphs @xcite and the resulting semi - definite relaxation @xmath14 introduced in @xcite . for some @xmath15 , an orthonormal representation of a graph @xmath5 is a sequence @xmath16 of @xmath17 unit - length vectors @xmath18 , such that @xmath19 for all @xmath20 . for any orthonormal representation of @xmath1 and any additional unit - length vector @xmath21 , the orthonormal representation constraint is @xmath22 . @xmath14 denotes the convex set of all vectors @xmath23 satisfying all orthonormal representation constraints for @xmath1 . for any graph @xmath1 , @xmath24 holds and approximating a linear objective function over @xmath14 can be done with arbitrary precision in polynomial time @xcite . moreover , if @xmath14 is a rational polytope , an optimal solution can be obtained in polynomial time . this fact gives a great relevance to the beautiful characterization of perfect graphs obtained by the same authors : @xmath25 for all imperfect graphs , @xmath2 does not coincide with any of the above relaxations . it is , thus , natural to study further relaxations and to combinatorially characterize those graphs where @xmath2 equals one of them . [ [ linear - relaxations - and - related - graphs . ] ] linear relaxations and related graphs . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + a natural generalization of the clique constraints are rank constraints associated with arbitrary induced subgraphs @xmath26 . by the choice of the right hand side @xmath27 , denoting the size of a largest stable set in @xmath28 , rank constraints @xmath29 are valid for @xmath2 . a graph @xmath1 is called _ rank - perfect _ by @xcite if and only if @xmath2 is described by rank constraints only . by definition , rank - perfect graphs include all perfect graphs . by restricting the facet set to rank constraints associated with certain subgraphs , several well - known graph classes are defined , e.g. , _ near - perfect graphs _ @xcite where only rank constraints associated with cliques and the whole graph are allowed , or _ t - perfect _ @xcite and _ h - perfect graphs _ @xcite where rank constraints associated with edges , triangles and odd holes resp . cliques of arbitrary size and odd holes suffice . as common generalization of perfect , t - perfect , and h - perfect graphs , the class of _ a - perfect graphs _ was introduced in @xcite as graphs @xmath1 where @xmath2 is given by rank constraints associated with antiwebs . an _ antiweb _ @xmath30 is a graph with @xmath31 nodes @xmath32 and edges @xmath33 if and only if @xmath34 and @xmath35 . antiwebs include all complete graphs @xmath36 , odd holes @xmath37 , and their complements @xmath38 . antiwebs are @xmath39-perfect by @xcite , further examples of @xmath39-perfect graphs were found in @xcite . a more general type of inequalities is obtained from complete joins of antiwebs , called _ joined antiweb constraints _ @xmath40 associated with the join of some antiwebs @xmath41 and a clique @xmath42 ( note that the inequality is scaled to have right hand side 1 ) . this includes , e.g. , all odd ( anti)wheels ( the join of a single node with an odd ( anti)hole ) . we denote the linear relaxation of @xmath2 obtained by all joined antiweb constraints by @xmath43 . by construction , we see that @xmath44 in @xcite , a graph @xmath1 is called _ joined a - perfect _ if and only if @xmath2 coincides with @xmath43 . besides all a - perfect graphs , further examples of joined a - perfect graphs are _ near - bipartite graphs _ ( where the non - neighbors of every node induce a bipartite graph ) due to @xcite . [ [ a - semi - definite - relaxation - and - rm - ls_-perfect - graphs . ] ] a semi - definite relaxation and @xmath0-perfect graphs . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in the early nineties , lovsz and schrijver introduced the psd - operator @xmath0 ( called @xmath45 in @xcite ) which , applied to @xmath3 , generates a positive semi - definite relaxation of @xmath2 stronger than @xmath14 ( see section [ sec : n+perfect ] for details ) . in order to simplify the notation we write @xmath46 . as in the case of perfect graphs , the stable set problem can be solved in polynomial time for the class of graphs for which @xmath47 by @xcite . these graphs are called _ @xmath0-perfect _ , and all other graphs _ @xmath0-imperfect _ ( note that they are also called @xmath45-(im)perfect , see e.g. @xcite ) . in addition , every subgraph of an @xmath0-perfect graph is also @xmath0-perfect . this motivates the definition of _ minimally _ @xmath0-_imperfect graphs _ as the @xmath0-imperfect graphs whose proper induced subgraphs are all @xmath0-perfect . the two smallest of such graphs ( regarding its number of nodes ) were found by @xcite and @xcite and are depicted in figure [ grafos6 ] . ( on the left ) and @xmath48 ( on the right ) . ] in @xcite , the authors look for a characterization of @xmath0-perfect graphs similar to the characterization ( [ equivalencias ] ) for perfect graphs : they intend to find an appropriate polyhedral relaxation @xmath49 of @xmath2 such that @xmath1 is @xmath0-perfect if and only if @xmath50 . a conjecture has been recently proposed in @xcite , which can be equivalently reformulated as follows @xcite : [ conjecture ] a graph @xmath1 is @xmath0-perfect if and only if @xmath51 . in @xcite it is shown that , for every graph @xmath1 @xmath52 thus , the conjecture states that @xmath0-perfect graphs coincide with joined a - perfect graphs and @xmath43 is the polyhedral relaxation of @xmath2 playing the role of @xmath49 in . conjecture [ conjecture ] has been already verified for near - perfect graphs by @xcite , for _ fs - perfect graphs _ ( where the only facet - defining subgraphs are cliques and the graph itself ) by @xcite , for _ webs _ ( the complements @xmath53 of antiwebs ) by @xcite and for _ line graphs _ ( obtained by turning adjacent edges of a root graph into adjacent nodes of the line graph ) by @xcite , see section [ sec : n+perfect ] for details . [ [ the - rm - ls_-perfect - graph - conjecture - for - claw - free - graphs . ] ] the @xmath0-perfect graph conjecture for claw - free graphs . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the aim of this contribution is to verify conjecture [ conjecture ] for a well - studied graph class containing all webs , all line graphs and the complements of near - bipartite graphs : the class of _ claw - free graphs _ ( i.e. , the graphs not containing as node induced subgraph the complete join of a single node and a stable set of size three ) . claw - free graphs attracted much attention due to their seemingly asymmetric behavior w.r.t . the stable set problem . on the one hand , the first combinatorial algorithms to solve the problem in polynomial time for claw - free graphs @xcite date back to 1980 . therefore , the polynomial equivalence of optimization and separation due to @xcite implies that it is possible to optimize over the stable set polytope of a claw - free graph in polynomial time . on the other hand , the problem of characterizing the stable set polytope of claw - free graphs in terms of an explicit description by means of a facet - defining system , originally posed in @xcite , was open for several decades . this motivated the study of claw - free graphs and its different subclasses , that finally answered this long - standing problem only recently ( see section [ sec : clawfree ] for details ) . the paper is organized as follows : in section 2 , we present the state - of - the - art on @xmath0-perfect graphs ( including families of @xmath0-imperfect graphs needed for the subsequent proofs ) and on claw - free graphs , their relevant subclasses and the results concerning the facet - description of their stable set polytopes from the literature . in section 3 , we verify , relying on the previously presented results , conjecture [ conjecture ] for the studied subclasses of claw - free graphs . as a conclusion , we obtain as our main result : [ thm_main ] the @xmath0-perfect graph conjecture is true for all claw - free graphs . we close with some further remarks and an outlook to future lines of research . in order to introduce the @xmath0-operator we denote by @xmath54 the vectors of the canonical basis of @xmath55 ( where the first coordinate is indexed zero ) , @xmath56 the vector with all components equal to @xmath57 and @xmath58 the convex cone of symmetric and positive semi - definite @xmath59-matrices with real entries . let @xmath60^n$ ] be a convex set and @xmath61 then , the convex set @xmath62 is defined as : @xmath63 where @xmath64 denotes the vector whose @xmath65-th entry is @xmath66 , for every @xmath67 . projecting this lifting back to the space @xmath68 results in @xmath69^n : \left(\begin{array}{c } 1\\ \x \end{array } \right)= y \e_0 , \mbox { for some } y \in m_{+}(k)\right\}}.\ ] ] in @xcite , lovsz and schrijver proved that @xmath70 is a relaxation of the convex hull of integer solutions in @xmath71 and that @xmath72 where @xmath73 and @xmath74 for every @xmath75 . in this work we focus on the behavior of a single application of the @xmath0-operator to the edge relaxation @xmath3 of the stable set polytope of a graph . recall that we write @xmath46 to simplify the notation and that graphs for which @xmath47 holds are @xmath0-perfect . exhibiting one @xmath0-imperfect subgraph @xmath28 in a graph @xmath1 certifies the @xmath0-imperfection of @xmath1 . hereby , characterizing @xmath0-imperfect graphs within a certain graph class turns out to be a way to attack the conjecture for this class . recall that @xmath76 and @xmath48 are the smallest @xmath0-imperfect graphs . in @xcite the authors showed that they are the two smallest members of an infinite family of @xmath0-imperfect graphs having stability number two that will play a central role in some subsequent proofs : [ @xcite ] [ mnpmascero ] let @xmath1 be a graph with @xmath77 such that @xmath78 is an odd antihole for some node @xmath79 . @xmath1 is @xmath0-perfect if and only if @xmath79 is completely joined to @xmath78 . further @xmath0-imperfect graphs can be obtained by applying operations preserving @xmath0-imperfection . in @xcite , the _ stretching _ of a node @xmath79 is introduced as follows : partition its neighborhood @xmath80 into two nonempty , disjoint sets @xmath81 and @xmath82 ( so @xmath83 , and @xmath84 ) . a stretching of @xmath79 is obtained by replacing @xmath79 by two adjacent nodes @xmath85 and @xmath86 , joining @xmath87 with every node in @xmath88 for @xmath89 , and subdividing the edge @xmath90 by one node @xmath91 . in @xcite it is shown : [ @xcite ] [ stretching ] the stretching of a node preserves @xmath0-imperfection . hence , all stretchings of @xmath76 and @xmath48 are @xmath0-imperfect , see figure [ fig_stretchings ] for some examples . in black ) of @xmath76 and @xmath48 . ] using stretchings of @xmath76 and @xmath48 and exhibiting one more minimally @xmath0-imperfect graph , namely the web @xmath92 , @xmath0-perfect webs are characterized in @xcite as follows : [ @xcite ] [ webs ] a web is @xmath0-perfect if and only if it is perfect or minimally imperfect . the proof shows that all imperfect not minimally imperfect webs with stability number 2 contain @xmath48 and all webs @xmath93 different from @xmath94 , some stretching of @xmath76 . furthermore , all other webs contain some @xmath0-imperfect @xmath95 and are , thus , also @xmath0-imperfect . another way to attack the conjecture is from the polyhedral point of view . a graph @xmath1 is said to be _ facet - defining _ if @xmath2 has a full - support facet . observe that verifying conjecture [ conjecture ] is equivalent to prove that the only facet defining @xmath0-perfect graphs are the complete joins of antiwebs . that is why we need to rely on structural results and complete facet - descriptions of stable set polytope of the graphs . using this approach , in @xcite , the authors characterized @xmath0-perfect line graphs by showing : [ @xcite ] [ hipomatch ] a facet - defining line graph @xmath1 is @xmath0-perfect if and only if @xmath1 is a clique or an odd hole . the proof relies on a result due to edmonds & pulleyblank @xcite who showed that a line graph @xmath96 is facet - defining if and only if @xmath97 is a 2-connected hypomatchable graph ( that is , for all nodes @xmath79 of @xmath97 , @xmath98 admits a perfect matching ) . such graphs @xmath97 have an ear decomposition @xmath99 where @xmath100 is an odd hole and @xmath101 is obtained from @xmath102 by adding an odd path ( ear ) between distinct nodes of @xmath102 . in @xcite , it is shown that the line graph @xmath103 is a node stretching of @xmath76 or @xmath48 and , thus , @xmath0-imperfect by @xcite . moreover , it is proved that the only minimally @xmath0-imperfect line graphs are stretchings of @xmath76 and @xmath48 . in several respects , claw - free graphs are generalizations of line graphs . an intermediate class between line graphs and claw - free graphs form _ quasi - line graphs _ , where the neighborhood of any node can be partitioned into two cliques ( i.e. , quasi - line graphs are the complements of near - bipartite graphs ) . quasi - line graphs can be divided into two subclasses : fuzzy circular interval graphs and semi - line graphs . let @xmath104 be a circle , @xmath105 a collection of intervals in @xmath104 without proper containments and common endpoints , and @xmath106 a multiset of points in @xmath104 . a _ fuzzy circular interval graph _ @xmath107 has node set @xmath106 and two nodes are adjacent if both belong to one interval @xmath108 , where edges between different endpoints of the same interval may be omitted . _ semi - line graphs _ are either line graphs or quasi - line graphs without a representation as a fuzzy circular interval graph . it turned out that so - called _ clique family inequalities _ suffice to describe the stable set polytope of quasi - line graphs . given a graph @xmath1 , a family @xmath109 of cliques and an integer @xmath110 , the clique family inequality ( @xmath109 , @xmath111 ) is the following valid inequality for @xmath2 @xmath112 where @xmath113 and @xmath114 ( resp . @xmath115 ) is the set of nodes contained in at least @xmath111 ( resp . exactly @xmath116 ) cliques of @xmath109 . this generalizes the results of edmods @xcite and edmonds & pulleyblank @xcite that @xmath117 is described by clique constraints and rank constraints @xmath118 associated with the line graphs of 2-connected hypomatchable induced subgraphs @xmath119 . note that the rank constraints of type ( [ eq_hypomatch - constraints ] ) are special clique family inequalities . chudnovsky and seymour @xcite extended this result to semi - line graphs , for which @xmath2 is given by clique constraints and rank constraints of type ( [ eq_hypomatch - constraints ] ) . then , semi - line graphs are rank - perfect with line graphs as only facet - defining subgraphs . moreover , in @xcite galluccio and sassano prove that if a rank constraint is facet - defining for a claw - free graph @xmath1 then , either , @xmath1 is a clique or @xmath1 contains the line graph of a minimal 2-connected hypomatchable graph @xmath97 or @xmath1 contains @xmath120 with @xmath121 and @xmath122 . eisenbrand et al . @xcite proved that clique family inequalities suffice to describe the stable set polytope of fuzzy circular interval graphs . stauffer @xcite verified a conjecture of @xcite that every facet - defining clique family inequality of a fuzzy circular interval graph @xmath1 is _ associated with a web in _ @xmath1 . all these results together complete the picture for quasi - line graphs . however , there are claw - free graphs which are not quasi - line . in particular , every graph with stability number 2 is claw - free and the 5-wheel is the smallest claw - free not quasi - line graph . due to cook ( see @xcite ) , all facets for graphs @xmath1 with @xmath77 are @xmath123-valued _ clique - neighborhood constraints_. this is not the case for graphs @xmath1 with @xmath124 . in fact , all the known difficult facets of claw - free graphs occur in this class . some non - rank facets with up to five different non - zero coefficients are presented in @xcite . all of these facets turned out to be so - called _ co - spanning 1-forest constraints _ due to @xcite , where it is also shown that it is possible to build a claw - free graph with stability number three inducing a co - spanning 1-forest facet with @xmath125 different left hand side coefficients , for every positive integer @xmath125 . the problem of characterizing @xmath2 when @xmath1 is a connected claw - free but not quasi - line graph with @xmath126 was studied by galluccio et al . : in a series of results @xcite , it is shown that if such a graph @xmath1 does not contain a clique cutset , then 1,2-valued constraints suffice to describe stab(@xmath1 ) . here , besides 5-wheels , different rank and non - rank facet - defining inequalities of the geared graph @xmath1 shown in fig . [ fig:5wheelstrips ] play a central role . in addition , graphs of this type can be decomposed into strips . a _ strip _ @xmath127 is a ( not necessarily connected ) graph with two designated simplicial nodes @xmath39 and @xmath125 ( a node is _ simplicial _ if its neighborhood is a clique ) . a claw - free strip containing a 5-wheel as induced subgraph is a _ 5-wheel strip_. given two node - disjoint strips @xmath128 and @xmath129 , their _ composition _ is the union of @xmath130 and @xmath131 together with all edges between @xmath132 and @xmath133 , and between @xmath134 and @xmath135 @xcite . as shown in @xcite , this composition operation can be generalized to more than two strips : every claw - free but not quasi - line graph @xmath1 with @xmath126 admits a decomposition into strips , where at most one strip is quasi - line and all the remaining ones are 5-wheel strips having stability number at most 3 . there are only three `` basic '' types of 5-wheel strips ( see fig . [ fig:5wheelstrips ] ) which can be extended by adding nodes belonging to the neighborhood of the 5-wheels ( see @xcite for details ) . note that a claw - free but not quasi - line graph @xmath1 with @xmath126 containing a clique cutset may have a facet - defining subgraph @xmath28 with @xmath136 ( inside a 5-wheel strip of type 3 ) , see @xcite for examples . taking all these results together into account gives the complete list of facets needed to describe the stable set polytope of claw - free graphs . in this section , we verify the @xmath0-perfect graph conjecture for all relevant subclasses of claw - free graphs . the graphs with @xmath137 play a crucial role in this context . relying on the behavior of the stable set polytope under taking complete joins @xcite and the result on @xmath0-(im)perfect graphs @xmath1 with @xmath137 ( theorem [ mnpmascero ] ) , we can prove : [ thm_n+imperfect_alpha2 ] all facet - defining @xmath0-perfect graphs @xmath1 with @xmath137 are odd antiholes or complete joins of odd antihole(s ) and a ( possibly empty ) clique . let @xmath1 be a facet - defining @xmath0-perfect graph with stability number 2 . we first observe that @xmath1 is imperfect ( because it has a full - support facet , but it is different from a clique ) . thus , @xmath1 contains an odd antihole @xmath138 by @xcite . if @xmath139 , we are done . if @xmath140 , every node @xmath141 outside @xmath138 is completely joined to @xmath138 due to theorem [ mnpmascero ] ( otherwise , @xmath141 and @xmath138 induce an @xmath0-imperfect subgraph of @xmath1 , a contradiction to @xmath1 @xmath0-perfect ) . therefore , @xmath1 is the complete join of @xmath138 and @xmath142 . note that @xmath142 is again @xmath0-perfect , facet - defining by chvtal @xcite , and @xmath143 . if @xmath144 , we apply the same argument as for @xmath1 ; if @xmath145 , it is a clique . @xmath146 this shows that all facet - defining @xmath0-perfect graphs @xmath1 with @xmath137 are joined @xmath39-perfect , and we conclude : [ cor_alpha2_2 ] the @xmath0-perfect graph conjecture is true for graphs with stability number 2 . recall that quasi - line graphs divide into the two subclasses of semi - line graphs and fuzzy circular interval graphs . chudnovsky and seymour @xcite proved that the stable set polytope of a semi - line graph is given by rank constraints associated with cliques and the line graphs of 2-connected hypomatchable graphs . together with the result from @xcite ( presented in theorem [ hipomatch ] ) , we directly conclude that the @xmath0-perfect graph conjecture holds for semi - line graphs . based on the results of eisenbrand et al . @xcite and stauffer @xcite , combined with the characterization of @xmath0-imperfect webs from @xcite ( theorem [ webs ] ) , we are able to show : [ thm_n+imperfect_fcig ] all facet - defining @xmath0-perfect fuzzy circular interval graphs are cliques , odd holes or odd antiholes . let @xmath1 be a fuzzy circular interval graph such that it is a facet - defining @xmath0-perfect graph . if @xmath1 is a clique , the result is immediate . otherwise , @xmath1 is the support graph of a clique family inequality with parameters @xmath147 @xmath148 associated with a web @xmath149 with @xmath150 ( @xcite ) . more precisely , if for any node @xmath151 , @xmath152 , there exist @xmath153 and @xmath154 in @xmath155 such that @xmath156 . the clique family inequality associated with @xmath157 is the clique family inequality having parameters @xmath158 and @xmath111 where @xmath159 . by theorem [ webs ] , @xmath149 is @xmath0-perfect if and only if it is an odd hole or an odd antihole . that is , since @xmath1 is @xmath0-perfect then @xmath160 and @xmath161 or @xmath162 . in both cases , @xmath163 follows . consider first the case in which @xmath161 . then the clique family inequality @xmath147 takes the form @xmath164 suppose there exists @xmath165 . then , @xmath79 belongs to @xmath166 consecutive cliques in @xmath167 implying that @xmath79 is connected to exactly @xmath168 consecutive nodes in @xmath169 . observe that @xmath170 since @xmath1 is a claw - free graph . then , if @xmath171 ( resp . @xmath172 ) @xmath1 contains an odd subdivision of @xmath76 ( resp . @xmath48 ) . since @xmath1 is @xmath173-perfect then @xmath174 or , equivalently , @xmath175 . now suppose that @xmath162 . let us call @xmath176 the nodes in @xmath177 . suppose there exists @xmath178 . then @xmath79 belongs to at least @xmath179 consecutive cliques of the family @xmath109 . w.l.o.g we may assume that the @xmath180 of the @xmath181 consecutive cliques are the ones that contain the sets of nodes @xmath182 , @xmath183 , ... @xmath184 . then @xmath79 is connected to at least @xmath185 consecutive nodes in @xmath186 . moreover , since @xmath1 is quasi - line , @xmath79 is connected with at most @xmath187 nodes . it follows that the subgraph of @xmath1 induced by @xmath188 has stability number two , and from theorem [ thm_n+imperfect_alpha2 ] it is @xmath0-imperfect . but from our assumption that @xmath1 is @xmath0-perfect , we conclude @xmath189 or , equivalently , @xmath190 . @xmath146 as a consequence , every @xmath0-perfect fuzzy circular interval graph is a - perfect . this verifies the @xmath0-perfect graph conjecture for fuzzy circular interval graphs . since the class of quasi - line graphs divides into semi - line graphs and fuzzy circular interval graphs , we obtain as direct consequence : [ cor_quasi - line ] the @xmath0-perfect graph conjecture is true for quasi - line graphs . it is left to treat the case of claw - free graphs that are not quasi - line . here , we distinguish two cases according to their stability number . to treat the case of claw - free not quasi - line graphs @xmath1 with @xmath126 , we rely on the decomposition of such graphs into strips , where at most one strip is quasi - line and all the remaining ones are 5-wheel strips @xcite . by noting that 5-wheel strips of type 3 contain @xmath76 and exhibiting @xmath0-imperfect line graphs in the other two cases , we are able to show : [ thm_not_quasi - line_4 ] every facet - defining claw - free not quasi - line graph @xmath1 with @xmath126 is @xmath0-imperfect . let @xmath1 be a facet - defining claw - free not quasi - line graph with @xmath126 . according to @xcite , @xmath1 has a decomposition into strips , where at most one strip is quasi - line and all the remaining ones have stability number at most 3 and contain a 5-wheel each . recall that there are only three types of 5-wheel strips , fig . [ fig:5wheelstrips ] shows the `` basic '' types , which can be extended by adding nodes belonging to the neighborhood of the 5-wheels @xcite . since @xmath1 is not quasi - line , it contains at least one 5-wheel strip @xmath28 . if @xmath28 is of type 3 , then @xmath28 contains @xmath76 , induced by the squared nodes indicated in fig . [ fig:5wheelstrips ] , and we are done . hence , let @xmath28 be of type 1 or 2 . note further that @xmath28 is a proper subgraph of @xmath1 ( by @xmath191 but @xmath126 ) and connected to @xmath192 ( since @xmath1 is facet - defining and , thus , can not have a clique cutset by chvtal @xcite ) . according to the strip composition , there are nodes in @xmath192 playing the role of the two simplicial nodes of @xmath28 ( the two black nodes in fig . [ fig:5wheelstrips ] ) , and they are connected by a path @xmath193 with nodes exclusively in @xmath192 ( again , since @xmath1 can not contain a clique cutset ) . if @xmath28 is of type 1 , then @xmath1 has , as induced subgraph , a node stretching of @xmath48 ( resp . of @xmath76 ) if @xmath193 is even ( resp . odd ) , see the squared nodes in fig . [ fig:5wheelstrip1 ] . -imperfect subgraphs if @xmath28 is of type 1 . ] if @xmath28 is of type 2 , then @xmath1 has , as induced subgraph , a node stretching of @xmath76 ( resp . of @xmath48 ) if @xmath193 is even ( resp . odd ) , see the squared nodes in fig . [ fig:5wheelstrip2 ] . -imperfect subgraphs if @xmath28 is of type 2 . ] hence , in all cases , @xmath1 contains an @xmath0-imperfect line graph and is itself @xmath0-imperfect . @xmath146 for graphs having stability number three , there is no decomposition known yet . relying only on the behavior of the stable set polytope under clique identification @xcite and the result on @xmath0-(im)perfect graphs from theorem [ mnpmascero ] , we can prove : [ thm_not_quasi - line_3 ] every facet - defining claw - free not quasi - line graph @xmath1 with @xmath124 is @xmath0-imperfect . let @xmath1 be a facet - defining claw - free graph with @xmath124 that is not quasi - line . then , there is a node @xmath79 in @xmath1 such that @xmath194 $ ] can not be partitioned into 2 cliques . hence , in the complement @xmath13 of @xmath1 , the subgraph @xmath195 can not be partitioned into two stable sets . thus , @xmath195 is non - bipartite and contains an odd cycle . let @xmath196 be the shortest odd cycle in @xmath195 . then @xmath196 is not a triangle ( otherwise , @xmath138 and @xmath79 induce a claw in @xmath1 ) . hence , @xmath196 is an odd hole ( because it is an odd cycle of length @xmath197 , but has no chords according to its choice ) . therefore , @xmath138 is an odd antihole in @xmath1 ; let @xmath198 denote it nodes and @xmath199 be its non - edges . furthermore , let @xmath200 $ ] and @xmath201 . from now on , we will use @xmath138 to denote both , the node set and the odd antihole when it is clear from the context . [ claim1 ] @xmath202 has stability number 2 and + * either contains an @xmath173-imperfect subgraph * or is the complete join of @xmath79 , @xmath138 and @xmath114 . firstly , note that @xmath80 does not contain a stable set of size 3 ( otherwise , @xmath1 clearly contains a claw ) . hence , @xmath203 follows . if @xmath204 , we are done . if there is a node @xmath205 , then for each such node , either @xmath91 is completely joined to @xmath138 or else the subgraph of @xmath1 induced by @xmath91 and @xmath138 is @xmath173-imperfect due to theorem [ mnpmascero ] . @xmath206 + we are done if @xmath202 is @xmath173-imperfect . hence , assume in the sequel of this proof that @xmath202 is the complete join of @xmath79 , @xmath138 and @xmath114 . since @xmath124 holds , @xmath202 is a proper subgraph of @xmath1 . we partition the nodes in @xmath207 into 3 different subsets : + * @xmath208 containing all nodes from @xmath207 having a neighbor in @xmath114 , * @xmath209 having all nodes from @xmath207 having no neighbor in @xmath114 , but a neighbor in @xmath138 , * @xmath210 containing all nodes from @xmath207 having no neighbor in @xmath211 . [ claim2 ] every node @xmath212 + * either induces together with @xmath138 an @xmath173-imperfect subgraph of @xmath1 , * or is completely joined to @xmath138 . no node @xmath212 can belong to a stable set @xmath213 ( otherwise , any neighbor @xmath205 of @xmath214 induces together with @xmath215 a claw in @xmath1 ) . hence , for every @xmath212 , the subgraph @xmath216 $ ] has stability number 2 and is either @xmath173-imperfect or an odd antiwheel by theorem [ mnpmascero ] . @xmath206 + we are done if some node @xmath212 yields an @xmath173-imperfect graph @xmath216 $ ] . hence , assume in the sequel of this proof that @xmath138 and @xmath208 are completely joined . [ claim3 ] @xmath208 is a clique . otherwise , @xmath1 contains a claw containing some node @xmath217 as central node , @xmath79 and two non - adjacent nodes @xmath218 . @xmath206 + let @xmath219 denote the subgraph of @xmath1 induced by @xmath79 , @xmath80 and @xmath208 . [ claim4 ] we have @xmath220 . we know already that @xmath203 by claim 1 . if @xmath219 contains a stable set @xmath215 of size 3 , then @xmath221 for some @xmath212 . this implies @xmath222 ( recall that we assume that @xmath208 and @xmath138 are completely joined ) . in addition , @xmath223 ( since @xmath79 is adjacent to all nodes in @xmath114 ( so we would have @xmath224 if @xmath225 ) , but @xmath215 can not contain 2 nodes from @xmath208 ( since @xmath208 is a clique by claim 3 ) ) . finally , @xmath215 can not contain 2 non - adjacent nodes @xmath226 ( otherwise , any node @xmath217 induces with @xmath215 a claw in @xmath1 ) . hence , there is no such stable set @xmath215 in @xmath219 . @xmath206 + by @xmath124 and @xmath220 , there is a node in @xmath227 . we conclude that @xmath228 ( otherwise , @xmath208 would constitute a clique cutset of @xmath1 , separating @xmath210 from @xmath28 , a contradiction to @xmath1 facet - defining by chvtal @xcite . [ claim5 ] @xmath114 induces a clique . otherwise , @xmath1 contains a claw induced by some node @xmath229 , a neighbor @xmath217 of @xmath230 , and two non - adjacent nodes @xmath226 . @xmath206 + hence , @xmath202 is in fact the complete join of a clique @xmath231 and @xmath138 . in addition , @xmath208 is a clique and completely joined to @xmath138 , @xmath209 is non - empty , and there is no edge between @xmath42 and @xmath209 . we further obtain : [ claim6 ] every node @xmath229 is completely joined to @xmath208 . otherwise , @xmath1 contains a claw induced by some node @xmath229 , a neighbor @xmath217 of @xmath230 , @xmath79 and a non - neighbor @xmath212 of @xmath230 . @xmath206 + note that , according to theorem [ mnpmascero ] , each node @xmath229 has three possibilities for its connections to @xmath138 : + * either @xmath230 induces together with @xmath138 an @xmath173-imperfect subgraph of @xmath1 , * or @xmath230 is completely joined to @xmath138 , * or @xmath230 belongs to a stable set @xmath232 ( recall that , by theorem [ mnpmascero ] , whenever @xmath233 has stability number 2 , it is either @xmath173-imperfect or an odd antiwheel ) . if a node @xmath229 gives rise to an @xmath173-imperfect subgraph of @xmath1 , we are done . hence , assume in the sequel of this proof that @xmath209 is partitioned into two subsets @xmath234 and @xmath235 containing all nodes @xmath230 that are completely joined to @xmath138 resp . belong to a stable set @xmath232 . we next show : [ claim7 ] @xmath236 . assume to the contrary that we have @xmath237 . then , @xmath209 also induces a clique ( otherwise , there is a claw in @xmath1 induced by @xmath79 , some node @xmath217 and two non - adjacent nodes @xmath238 ) . this implies that @xmath239 $ ] has stability number 2 ( by @xmath220 due to claim 4 , @xmath209 completely joined to @xmath208 due to claim 6 , and @xmath237 clique ) . thus , @xmath210 is non - empty ( because @xmath124 ) . hence , @xmath1 contains a clique cutset @xmath240 , separating @xmath210 from @xmath202 ( recall that every node in @xmath210 has only neighbors in @xmath208 or @xmath209 , but not in @xmath202 ) , a contradiction to @xmath1 facet - defining by chvtal @xcite . therefore , we conclude that @xmath237 can not hold . @xmath206 + having ensured the existence of a stable set @xmath232 in @xmath1 , we next observe : [ claim8 ] @xmath241 . otherwise , @xmath1 contains a claw induced by @xmath242 and any node @xmath212 ( recall : we assume that @xmath208 and @xmath138 are completely joined ( otherwise , @xmath1 is @xmath173-imperfect by claim 2 ) , and have that @xmath208 and @xmath209 are completely joined by claim 6 ) . @xmath206 + this implies particularly that no node outside @xmath202 has a neighbor in @xmath114 . we next study the connections between @xmath138 and @xmath209 in more detail and obtain the following important fact : [ claim9 ] @xmath243 and each node @xmath244 has exactly two consecutive neighbors on @xmath138 . consider some node @xmath244 and the stable set @xmath232 . by construction of @xmath209 , @xmath230 has a neighbor @xmath245 . this node @xmath246 ( and any further neighbor of @xmath230 in @xmath138 ) can not be a common neighbor of @xmath247 and @xmath248 ( otherwise , @xmath246 induces together with @xmath242 a claw in @xmath1 ) . hence , @xmath246 equals either @xmath249 ( which is not adjacent to @xmath248 ) or else @xmath250 ( which is not adjacent to @xmath251 ) . , say that @xmath230 has @xmath249 as neighbor in @xmath138 . then @xmath243 follows because for any @xmath252 , the graph induced by @xmath230 and @xmath138 contains a claw with center @xmath249 and the nodes @xmath253 ( or else @xmath254 or @xmath255 induce with @xmath242 a claw if @xmath230 is adjacent to @xmath254 or @xmath255 ) . moreover , we observe that @xmath230 is also adjacent to @xmath256 ( otherwise , there is a claw with center @xmath249 and the nodes @xmath257 ) . this shows the assertion that each node @xmath244 has exactly two consecutive neighbors on @xmath243 . @xmath206 + we next observe : [ claim10 ] @xmath210 induces a clique . otherwise , @xmath1 contains a stable set of size 4 , consisting of two non - adjacent nodes in @xmath138 and two non - adjacent nodes in @xmath210 ( recall : by definition of @xmath210 , there is no edge between @xmath138 and @xmath210 ) . @xmath206 + hence , so far we have the following : @xmath202 is the complete join of a clique @xmath231 and @xmath243 . @xmath207 is partitioned into two subsets @xmath209 and @xmath210 where + * @xmath209 is non - empty and partitions into two subsets @xmath234 and @xmath235 consisting of all nodes @xmath230 that are completely joined to @xmath243 resp . belong to a stable set @xmath232 and have exactly two consecutive neighbors on the @xmath258 ; * @xmath210 induces a clique and no node in @xmath210 has a neighbor in @xmath202 . we continue to explore the composition of @xmath209 and its connections to @xmath243 : [ claim11 ] if two nodes @xmath238 share a same neighbor @xmath245 , then @xmath230 and @xmath259 are adjacent . otherwise , @xmath1 contains a claw with center @xmath246 and the nodes @xmath260 . @xmath206 [ claim12 ] there are at least two nodes in @xmath235 . by claim 7 , there is a node @xmath244 . then @xmath261 follows ( otherwise , the only two and consecutive neighbors of @xmath230 on @xmath243 ( by claim 9 ) form a clique cutset in @xmath1 , separating @xmath230 from @xmath42 , a contradiction to @xmath1 facet - defining by chvtal @xcite ) . if all nodes from @xmath262 belong to @xmath234 , then @xmath209 induces a clique by claim 11 ( because all share a common neighbor in @xmath138 ) , and @xmath262 together with the two and consecutive neighbors of @xmath230 on @xmath138 form a clique cutset in @xmath1 , separating @xmath230 from @xmath42 , again a contradiction to @xmath1 facet - defining . hence , @xmath235 contains at least two nodes . @xmath206 + using similar arguments , we next show : [ claim13 ] not all nodes in @xmath235 have the same two consecutive neighbors on @xmath138 . otherwise , @xmath1 has a clique cutset ( consisting of @xmath234 and the two common , consecutive neighbors on @xmath138 of all nodes in @xmath235 ) , separating @xmath235 from @xmath42 , again a contradiction to @xmath1 facet - defining . @xmath206 [ claim14 ] if all nodes in @xmath235 share a common neighbor @xmath247 on @xmath138 , then @xmath1 contains @xmath48 as induced subgraph . by assumption , we have only three types of nodes in @xmath209 : nodes @xmath230 with @xmath263 , nodes @xmath259 with @xmath264 , and nodes @xmath265 . @xmath209 induces a clique ( by claim 11 ) and @xmath266 follows ( otherwise , @xmath209 is a clique cutset separating @xmath210 from @xmath42 , again a contradiction to @xmath1 facet - defining ) . claim 12 combined with claim 13 shows that there is at least one node @xmath230 adjacent to @xmath267 and at least one node @xmath259 adjacent to @xmath268 . moreover , there is also at least one node @xmath265 ( otherwise , @xmath1 equals the gear ( induced by @xmath79 , @xmath138 , @xmath230 and @xmath259 , see figure [ fig_n+p_alpha3_claim14 ] ) with possible replications of @xmath79 , @xmath230 , @xmath259 and is not facet - defining , a contradiction ) . hence , @xmath1 contains @xmath48 ( induced by @xmath269 and the nodes @xmath79 , @xmath249 , @xmath254 , @xmath230 , and @xmath259 , see figure [ fig_n+p_alpha3_claim14 ] ) . @xmath206 + and @xmath79 , @xmath230 , @xmath259 , @xmath269 ( removing @xmath269 yields the gear , the bold edges indicate the @xmath48 ) . ] this shows that @xmath1 is either @xmath173-imperfect ( and we are done ) or @xmath235 has two nodes with distinct neighbors on @xmath138 . let us assume the latter . [ claim15 ] if two nodes @xmath270 with distinct neighbors on @xmath138 are adjacent , then @xmath1 contains @xmath76 as induced subgraph . , let @xmath271 and @xmath272 . if @xmath230 is adjacent to @xmath259 , then @xmath273 , @xmath230 , @xmath259 , @xmath274 , @xmath275 induce together with @xmath79 a @xmath76 . @xmath206 + hence , assume that no two nodes in @xmath270 with distinct neighbors on @xmath138 are adjacent ( otherwise , we are done ) . [ claim16 ] if @xmath276 , then @xmath1 contains a node stretching of @xmath48 as induced subgraph . if there is a node @xmath277 , then @xmath278 is adjacent to every node in @xmath235 ( otherwise , there is a node @xmath244 such that @xmath278 together with @xmath242 forms a stable set of size 4 ) . recall that we assume that @xmath235 contains two non - adjacent nodes @xmath230 , @xmath259 with distinct neighbors on @xmath138 . then @xmath278 together with @xmath230 , @xmath259 and @xmath138 induce a node stretching of @xmath48 . @xmath206 + hence , we are done if @xmath276 ( because @xmath1 contains an @xmath173-imperfect subgraph ) . so let us assume @xmath266 from now on . furthermore , assume w.l.o.g . that @xmath230 with @xmath271 and @xmath259 with @xmath272 is a pair of nodes in @xmath235 having distinct neighbors on @xmath138 . since @xmath1 is facet - defining ( and , thus , without clique cutset ) , there is a path connecting @xmath230 and @xmath259 ; let @xmath193 denote the shortest such path . note that @xmath193 has length @xmath279 ( recall : @xmath230 and @xmath259 are supposed to be non - adjacent , otherwise @xmath1 contains a @xmath76 by claim 15 ) . [ claim17 ] if @xmath193 has length @xmath280 , then @xmath1 is @xmath173-imperfect . so , let @xmath193 have length @xmath280 and denote by @xmath281 its only internal node . then @xmath282 follows ( by @xmath266 and because there is no common neighbor of @xmath230 and @xmath259 in @xmath138 ) . we conclude that @xmath283 holds ( otherwise , @xmath1 has a claw induced by @xmath281 and @xmath242 ) and that @xmath281 and @xmath275 are non - adjacent ( otherwise , @xmath1 has a claw induced by @xmath281 and @xmath230 , @xmath259 , @xmath275 ) . in fact , @xmath284 follows by our assumption that no two nodes in @xmath235 with distinct neighbors on @xmath138 are adjacent ( otherwise , @xmath1 contains a @xmath76 by claim 15 ) . since the graph induced by @xmath202 together with @xmath285 , called a 3-gear ( see figure [ fig_alpha3_claim17 ] ) , is not facet - defining , there must be another node @xmath286 in @xmath209 ( recall : we have @xmath266 ) . we are done if there is a node @xmath287 ( because @xmath1 contains a @xmath48 induced by @xmath288 and @xmath286 in this case ) . hence , assume @xmath289 . , let @xmath290 ( note : the case @xmath291 is symmetric , and in all other cases , @xmath1 is still a 3-gear with some replicated nodes , thus not facet - defining ) . then @xmath286 and @xmath259 are adjacent by claim 11 . if @xmath286 is also adjacent to @xmath230 or @xmath281 , then we are done since then @xmath1 contains a @xmath76 by claim 15 . if @xmath286 is neither adjacent to @xmath230 nor to @xmath281 , then @xmath1 contains an @xmath173-imperfect line graph induced by @xmath288 and @xmath292 ( being a node stretching of @xmath48 ) . note that @xmath1 still contains one of the above @xmath173-imperfect subgraphs if @xmath1 contains more nodes than considered so far . @xmath206 + and @xmath79 , @xmath230 , @xmath259 , @xmath281 , called a 3-gear . ] thus , we are done if @xmath193 has length @xmath280 . let us finally assume that @xmath193 has length @xmath293 and @xmath230 and @xmath259 have no common neighbor ( recall : @xmath193 is a shortest path connecting them ) . then @xmath294 follows ( because each node @xmath287 shares a common neighbor with @xmath230 and @xmath259 on @xmath138 and is , thus , adjacent to both @xmath230 and @xmath259 by claim 11 , a contradiction to the choice of @xmath193 as shortest path connecting them ) . moreover , there is a neighbor @xmath295 of @xmath230 in @xmath235 ( otherwise , @xmath296 forms a clique cutset separating @xmath230 from @xmath42 , a contradiction to @xmath1 facet - defining by chvtal @xcite ) . in addition , @xmath297 is different from @xmath298 and from @xmath299 ( otherwise , @xmath230 and @xmath259 would be a pair of nodes with distinct neighbors on @xmath138 and connected by a path of length 2 , hence @xmath1 is @xmath173-imperfect by claim 17 and we are done ) . we clearly have @xmath300 ( otherwise , @xmath295 adjacent to @xmath259 follows by claim 11 and we have again a pair of nodes with distinct neighbors on @xmath138 and connected by a path of length 2 , hence @xmath1 is @xmath173-imperfect by claim 17 and we are done ) . if there is no node in @xmath235 having @xmath301 as neighbors on @xmath138 , then @xmath296 forms still a clique cutset separating @xmath230 from @xmath42 , again a contradiction . hence , let @xmath302 . then , there is no node in @xmath235 having @xmath299 as neighbors on @xmath138 ( this node @xmath286 would be adjacent to @xmath295 by claim 11 , so that @xmath295 would be a common neighbor of @xmath230 and @xmath286 , leading to an @xmath173-imperfect subgraph of @xmath1 by claim 17 and we are done ) . similarly , there is no node in @xmath235 having @xmath298 as neighbors on @xmath138 . this finally implies that @xmath298 is a clique cutset separating @xmath259 from @xmath42 , a contradiction to @xmath1 facet - defining by chvtal @xcite ) . that all further nodes of @xmath1 are either replicates of @xmath230 , @xmath259 or @xmath295 ( and @xmath298 remains a clique cutset in all cases ) finishes the proof . @xmath146 hence , the only facet - defining subgraphs @xmath28 of @xmath0-perfect claw - free not quasi - line graphs @xmath1 with @xmath124 have @xmath303 and are , by theorem [ thm_n+imperfect_alpha2 ] , cliques , odd antiholes or their complete joins . we conclude that @xmath0-perfect facet - defining claw - free not quasi - line graphs @xmath1 with @xmath124 are joined @xmath39-perfect and , thus , the @xmath0-perfect graph conjecture is true for this class . this together with theorem [ thm_not_quasi - line_3 ] shows that the only facet - defining subgraphs @xmath28 of @xmath0-perfect claw - free not quasi - line graphs @xmath1 with @xmath126 have @xmath303 and are , by theorem [ thm_n+imperfect_alpha2 ] , cliques , odd antiholes or their complete joins . thus , every @xmath0-perfect claw - free not quasi - line graph @xmath1 with @xmath126 is joined a - perfect and , thus , the @xmath0-perfect graph conjecture holds true for this class . combining corollary [ cor_alpha2_2 ] with the above results shows that all @xmath0-perfect claw - free but not quasi - line graphs are joined @xmath39-perfect and we obtain : [ cor_not_quasi - line ] the @xmath0-perfect graph conjecture is true for all claw - free graphs that are not quasi - line . finally , we obtain our main result ( theorem [ thm_main ] ) as direct consequence of corollary [ cor_quasi - line ] and corollary [ cor_not_quasi - line ] : the @xmath0-perfect graph conjecture is true for all claw - free graphs . the context of this work was the study of @xmath0-perfect graphs , i.e. , graphs where a single application of the lovsz - schrijver psd - operator @xmath0 to the edge relaxation yields the stable set polytope . hereby , we are particularly interested in finding an appropriate polyhedral relaxation @xmath49 of @xmath2 that coincides with @xmath304 and @xmath2 if and only if @xmath1 is @xmath0-perfect . an according conjecture has been recently formulated ( @xmath0-perfect graph conjecture ) ; here we verified it for the well - studied class of claw - free graphs ( theorem [ thm_main ] ) . from the presented results and proofs , we can draw some further conclusions . first of all , we can determine the subclass of joined @xmath39-perfect graphs to which all @xmath0-perfect claw - free graphs belong to . in @xcite , it is suggested to call a graph @xmath1 _ @xmath306-perfect _ if the only facets of @xmath2 are associated with cliques and minimally imperfect graphs . according to @xcite , @xmath1 is _ joined @xmath306-perfect _ if @xmath2 is given only by facets associated with cliques , minimally imperfect graphs and their complete joins . theorem [ hipomatch ] together with the results from section [ results ] provide the complete list of all facet - defining @xmath0-perfect claw - free graphs : note that galluccio and sassano provided in @xcite a complete characterization of the rank facet - defining claw - free graphs : they either belong to one of the following three families of rank - minimal graphs * cliques , * partitionable webs @xmath307 ( where @xmath308 and @xmath309 stand for stability and clique number , resp . ) , * line graphs of minimal 2-connected hypomatchable graphs @xmath97 ( where @xmath310 is not hypomatchable anymore for any edge @xmath311 ) , or can be obtained from them by means of two operations , sequential lifting and complete join . our results show : an @xmath0-perfect claw - free graph @xmath1 has , besides cliques , only odd holes and odd antiholes as rank - minimal subgraphs ; cliques are the only subgraphs in @xmath1 that can be sequentially lifted to larger rank facet - defining subgraphs , where complete joins can only be taken of odd antiholes . note further that , besides verifying the @xmath0-perfect graph conjecture for claw - free graphs , we obtained the complete list of all minimally @xmath0-imperfect claw - free graphs . in fact , the results in @xcite imply that the following graphs are minimally @xmath0-imperfect : * graphs @xmath1 with @xmath77 such that @xmath78 is an odd antihole for some node @xmath79 , not completely joined to @xmath78 , * the web @xmath92 , * @xmath0-imperfect line graphs ( which are all node stretchings of @xmath76 or @xmath48 ) . finally , the subject of the present work has parallels to the well - developed research area of perfect graph theory also in terms of polynomial time computability . in fact , it has the potential of reaching even stronger results due the following reasons . recall that calculating the value @xmath312 can be obtained with arbitrary precision in polynomial time for every graph @xmath1 , even in the weighted case , by @xcite . thus , the stable set problem can be solved in polynomial time for a strict superset of perfect graphs , the @xmath0-perfect graphs , by @xmath313 . hence , our future lines of research include to find * new families of graphs where the conjecture holds ( e.g. , by characterizing the minimally @xmath0-imperfect graphs within the class ) , * new subclasses of @xmath0-perfect or joined a - perfect graphs , * classes of graphs @xmath1 where @xmath2 and @xmath304 are `` close enough '' to have @xmath314 . in particular , the class of graphs @xmath1 with @xmath314 can be expected to be large since @xmath304 is a much stronger relaxation of @xmath2 than @xmath14 . in all cases , the stable set problem could be approximated with arbitrary precision in polynomial time in these graph classes by optimizing over @xmath304 . finally , note that @xmath315 with @xmath316 clearly gives an even stronger relaxation of @xmath2 than @xmath304 . however , already approximating with arbitrary precision over @xmath317 can not be done in polynomial time anymore for all graphs @xmath1 by @xcite . hence , @xmath0-perfect graphs or their generalizations satisfying @xmath314 are the most promising cases in this context . s. bianchi , m. escalante , g. nasini , and l. tunel , `` near - perfect graphs with polyhedral @xmath318 , '' _ electronic notes in discrete math . _ , vol . 37 , pp . 393398 , 2011 . s. bianchi , m. escalante , g. nasini , and l. tunel , `` lovsz - schrijver psd - operator and a superclass of near - perfect graphs '' , _ electronic notes in discrete mathematics _ 44 , pp . 339344 , 2013 . s. coulonges , a. pcher , and a. wagler , `` characterizing and bounding the imperfection ratio for some classes of graphs , '' _ math . programming a _ , vol . 118 , pp . 3746 , 2009 . edmonds , `` maximum matching and a polyhedron with ( 0,1 ) vertices , '' _ bur . standards _ , 69b , pp . 125130 , 1965 . a. galluccio , c. gentile , and p. ventura . the stable set polytope of claw - free graphs with stability number at least four . i. fuzzy antihat graphs are w - perfect , journal of combinatorial theory series b 107 , 92 - 122 , 2014 a. galluccio , c. gentile , and p. ventura . the stable set polytope of claw - free graphs with stability number at least four . striped graphs are g - perfect , journal of combinatorial theory series b 108 , 1 - 28 , 2014 t.m . liebling , g. oriolo , b. spille , and g. stauffer . on non - rank facets of the stable set polytope of claw - free graphs and circulant graphs . mathematical methods of operations research , 59(1):2535 , 2004 . g. oriolo , u. pietropaoli , and g. stauffer , `` a new algorithm for the maximum weighted stable set problem in claw - free graphs '' , in a. lodi , a. panconesi and g. rinaldi ( eds ) , ipco 2008 , vol . 5035 of lecture notes in computer science , springer , bertinoro , italy , may 26 - 28 , pp . 7796 , 2008 . u. pietropaoli , a. wagler , `` some results towards the description of the stable set polytope of claw - free graphs '' , proc . of alio / euro workshop on applied combinatorial optimization , buenos aires , 2008

the spatial distribution of stars in the nearby s0 galaxy ngc 5102 is investigated using images obtained with wircam and megacam on the canada - france - hawaii telescope . with the exception of gaps between detector elements , the entire galaxy is surveyed in and , while the and data extend out to r kpc , which corresponds to almost 7 disk scale lengths . a modest population of main sequence ( ms ) stars with m and ages myr are detected throughout the disk , with the majority located in the southern half of the galaxy . the stellar disk in the northern half of the galaxy is warped , following structure that is also seen in hi . objects with photometric properties that are consistent with those of bright agb stars are seen throughout the disk , and the ratio of c stars to bright m giants is consistent with an overall increase in the star formation rate within the past 1 gyr . star - forming activity during the interval gyr was more centrally concentrated than during the past myr . the structure of the disk changes near r kpc ( 5.5 disk scale lengths ) , in the sense that the radial surface density profile defined by red supergiants ( rsgs ) and bright agb stars levels off at larger radii . rsgs and bright agb stars are traced out to a radius of 14 kpc ( 15.6 scale lengths ) along the southern portion of the major axis , while a tentative detection is also made of bright agb stars at a projected distance of kpc along the south east minor axis . a large clump of agb stars that subtends arcmin is identified to the west of the galaxy center . it is argued that this is the remnant of a companion galaxy that triggered past episodes of elevated star - forming activity . = 1.0 cm

0908.2263

the stellar contents of nearby galaxies constitute a fossil record that can be used to gain insight into the processes that shape galaxy evolution and define basic global properties , such as morphology . there are five galaxy groups within @xmath12 mpc ( classically referred to as the local , sculptor , m81 , centaurus , and maffei groups ) , which in turn contain sub - systems centered on the largest members ( e.g. karachentsev 2005 ) . the brightest members of the nearest groups show a diverse range of properties , although a common feature shared by the most massive members of the local , m81 , and centaurus groups is that they show signs of having experienced cosmologically recent interactions . this commonality underlines the important role that local environment and galaxy - galaxy interactions play in the evolution of even relatively commonplace galaxies . located at a distance of only 3.2 mpc , the s0 galaxy ngc 5102 is a key laboratory for probing the evolution of gas - poor disks . there is only modest star - forming activity in ngc 5102 today , and this is restricted to the south east quadrant of the disk , where hii regions are found ( van den bergh 1976 , mcmillan , ciardullo , & jacoby 1994 ) . davidge ( 2008a ) found small numbers of red supergiants ( rsgs ) and blue main sequence ( ms ) stars in a field along the north west minor axis , and estimated that the integrated sfr during the past ten million years was 0.02 m@xmath13 year@xmath14 . the moderately deep color - magnitude diagram ( cmd ) of the northern disk of ngc 5102 discussed by karachentsev et al . ( 2002 ) shows that while there is a spray of blue objects , there is neither a well - defined bright ms nor a rsg sequence in this part of ngc 5102 , again consistent with a low recent sfr . there are indications that the specific sfr in ngc 5102 within the past few hundred myr was much higher than it is today , at least in some parts of the galaxy . the deep balmer absorption lines ( gallagher , faber , & balick 1975 ; rocca - volmerange & guiderdoni 1987 ) and blue colors ( van den bergh 1976 ; pritchet 1979 ; bica 1988 ; kraft et al . 2005 ) that are seen in the central regions of ngc 5102 are perhaps the most overt signatures that at least 10% ( e.g. serra & trager 2007 ) of the central stellar mass formed during intermediate epochs . applying spectral synthesis techniques to the ultraviolet visible spectrum of the center of ngc 5102 , kraft et al . ( 2005 ) conclude that at least 50% of the central stellar mass in ngc 5102 formed within the past 3 gyr . there is also evidence of recent episodes of large - scale star formation in other parts of the galaxy . van den bergh ( 1976 ) detected filamentary h@xmath15 emission , which mcmillan et al . ( 1994 ) subsequently found to be part of a complex web of emission features , the dominant structure of which is a supershell . the size and expansion velocity of the supershell suggest an age of @xmath16 myr . finally , the numbers of bright asymptotic giant branch ( agb ) stars in the western portion of ngc 5102 suggests that @xmath17 of the total stellar mass in this field may have formed during the past gyr ; assuming that the stellar content of this area is representative of the galaxy as a whole then the galaxy - wide sfr was at least 1.4 m@xmath13 year@xmath14 throughout much of the past gyr ( davidge 2008a ) . elevated levels of star - forming activity are usually associated with interactions , and so it is puzzling that ngc 5102 has no obvious close companions . the distance and radial velocity of ngc 5102 are consistent with it being a companion of cen a ( karachentsev et al . 2002 ) , although it is located near the outer periphery of cen a satellites . indeed , ngc 5102 is relatively isolated when compared with other galaxies associated with cen a and m83 ( e.g. figure 1 of karachentsev et al . 2002 ) . while sampling relatively faint objects , the photometric observations discussed by karachensev et al . ( 2002 ) and davidge ( 2008a ) cover only a small fraction of ngc 5102 , missing the areas near the hii regions where the richest concentrations of young stars should lurk . these data are also restricted to visible and red wavelengths , and so are biased against stars that may have their photometric properties affected by line blanketing , have very low effective temperatures , and/or are highly obscured . these observational deficiencies , coupled with the evidence that there was much more extensive star formation throughout ngc 5102 within the past @xmath18 gyr , provide a clear motivation for conducting a detailed census of the bright stellar content throughout the galaxy . in the present study , observations obtained with the megacam and wircam imagers on the 3.6 meter canada - france - hawaii telescope ( cfht ) are used to survey the stellar content in and around ngc 5102 . the data obtained with these instruments are complementary . the brightest ngc 5102 stars in the megacam observations are young ms stars and rsgs , while those in the wircam observations are highly evolved oxygen - rich and carbon - rich agb stars . observations of highly - evolved cool stars at infrared wavelengths benefit from the enhanced contrast with respect to the comparatively blue body of unresolved stars in a galaxy , while observations of bright ms stars at visible wavelengths exploit the low number density of blue background galaxies in the magnitude range of interest , making bright ms stars potentially important probes of structure in areas of low stellar mass density ( e.g. davidge 2008b ) . the paper is structured as follows . the observations and data reduction procedures are described in 2 , while the cmds obtained from these data are discussed in 3 . the statistics of m giants and c stars in the wircam data are used to investigate the recent star - forming history of ngc 5102 in 4 . the spatial distribution of stars spanning a range of ages and evolutionary states are examined in 5 , while a summary and discussion of the results follows in 6 . ngc 5102 was observed with megacam ( boulade et al . 2003 ) during the night of january 16 , 2008 . the detector in megacam is a mosaic of 36 ccds that covers a @xmath19 degree@xmath20 field with 0.185 arcsec pixel@xmath14 . there is a 13 arcsec gap between most ccds , although there are two 70 arcsec gaps that separate ccd banks . the nucleus of ngc 5102 , where individual stars can not be resolved , was centered on the detector mosaic , and thus fell in a gap between detectors . four 300 sec exposures were recorded each in @xmath0 and @xmath1 , and the raw images were reduced with the elixer pipeline . the elixer processing consisted of bias subtraction , the division by a flat - field frame , the subtraction of a fringe frame from the @xmath1 images , and the insertion into the image header of photometric calibration information based on standard star observations that were made during the megacam run . to create final images , the elixer - processed data in each filter were aligned , median - combined , and trimmed to the area common to all frames . stars in the final images have fwhm @xmath11 arcsec . ngc 5102 was observed with wircam ( puget et al . 2004 ) on the night of february 17 2008 . the detector in wircam is a mosaic of four @xmath21 hgcdte arrays , which together image @xmath22 arcmin@xmath20 with 0.3 arcsec pixel@xmath14 . gaps between the arrays have a width of @xmath23 arcsec . twenty 45 sec exposures were recorded in @xmath2 , and eighty 20 sec exposures were recorded in @xmath24 . the disk of ngc 5102 extends over many arcmin@xmath20 , and this complicates efforts to construct the calibration frames that are needed to remove interference fringes and thermal emission signatures . while the calibration frames could be constructed from observations of a separate background field , time constraints ruled out this approach , and so it was decided to sacrifice the angular coverage of the final processed image to enable the construction of the required calibration frames . images were recorded with ngc 5102 centered on each array of the mosaic in turn ; this corresponds to a 10 arcmin offset between pointings . a four - point square dither pattern with 3 arcsec amplitude was also implemented at each offset position to further facilitate the suppression of stars , cosmic rays , and isolated bad pixels . the raw wircam images were processed with the iiwi pipeline , which performed dark subtraction and the division by flat - field frames . calibration images to remove interference fringes and thermal signatures were constructed by median - combining the unregistered flat - fielded images in each filter , and the results were subtracted from the flat - fielded data . the processed images were then aligned , median - combined , and trimmed to the region having a common exposure time . the region with the full exposure time covers a @xmath25 arcmin area , and this is sufficient to sample much of the ngc 5102 disk . stars in the final images have fwhm @xmath26 arcsec . the photometric measurements were made with the point - spread - function ( psf ) fitting program allstar ( stetson & harris 1988 ) , using source lists , psfs , and preliminary brightnesses obtained from daophot ( stetson 1987 ) tasks . between 50 to 100 stars were used to construct each psf . the psfs were constructed iteratively , with contaminating objects subtracted using progressively improved psfs . allstar computes an uncertainty , @xmath27 , for each object that is based on the quality of the psf fit , but does not account for systematic errors due to crowding or the tendency for the faintest stars to be located on positive noise flucuations . @xmath27 was used to cull objects with large measurement errors from the photometric catalogues . objects with @xmath28 , which tend to be near the faint limit of the data , were rejected . in addition , a plot of @xmath27 versus magnitude reveals a well - defined relation , and outlying objects from the dominant trend were deleted . objects that depart from the dominant trend tend to be either obvious blends in the crowded central regions of ngc 5102 , galaxies , or residual cosmetic defects . artificial star experiments were used to estimate completeness . the artificial stars were assigned @xmath29 and @xmath30 colors that follow the ridgeline of the agb sequence in ngc 5102 . the brightnesses of the artificial stars were measured using the same procedures employed for actual stars , including rejection based on @xmath27 ( see above ) . as with actual stars , an artificial star was only considered to be recovered if it was detected in two filters , and so could be placed on either the @xmath31 or @xmath32 cmds . the completeness fraction varies with position in ngc 5102 . the 50% completeness fraction for stars with r@xmath33 between 3 and 4 kpc occurs near @xmath34 , and is 0.5 magnitudes fainter for the outer regions of ngc 5102 . for comparison , the 50% completeness fraction occurs at @xmath35 throughout most of the wircam image . unlike the @xmath1 data , the completeness fraction in @xmath36 is not as sensitive to location within the galaxy because of the comparatively high contrast between the brightest resolved stars in @xmath36 and the main body of unresolved stars . the artificial star experiments also indicate that many of the stars fainter than the 50% completeness limit in @xmath1 and @xmath36 are probably blends . the @xmath31 cmds of stars in various radial intervals are shown in figure 1 . the distance intervals given in each panel are galactocentric radii , r@xmath33 , that assume a distance modulus of 27.5 and a disk inclination of 64.4 degrees ( davidge 2008a ) . the exponential disk scale length for ngc 5102 measured from 2mass large galaxy catalogue ( jarrett et al . 2003 ) images is 0.9 kpc , and this is in excellent agreement with the scale length measured by freeman ( 1970 ) . while the cmds do not go as deep as those presented by karachentsev et al . ( 2002 ) and davidge ( 2008a ) , they extend out to many disk scale lengths and sample stars over the entire galaxy . stars evolving on the agb form a diffuse cloud of objects with @xmath37 and @xmath29 between 0.2 and 1.2 in the cmds of objects with r@xmath38 kpc . it becomes difficult to identify a concentration of agb stars in the cmds at larger r@xmath33 due to contamination from field stars and background galaxies . still , a modest population of agb stars is present at larger r@xmath33 ( 5 ) . davidge ( 2008a ) detected blue ms stars and rsgs along the western arm of the ngc 5102 minor axis , and objects with @xmath39 and @xmath40 are seen in the r@xmath41 kpc cmds . the blue objects tend to have intrinsic brightnesses that are consistent with them being stars , rather than star clusters . the apparent absence of bright , compact , blue clusters is worth noting , as one might expect these to be present if , as suggested by davidge ( 2008a ) , ngc 5102 experienced a burst of star formation during the recent past . very few objects are detected with @xmath39 and @xmath40 at large r@xmath33 . this is a consequence of the drop in the number of bright ms stars in ngc 5102 at these radii , coupled with the modest number of background galaxies with this color at these magnitudes ( e.g. davidge 2008b ) . there is substantial contamination from background galaxies in the portion of the @xmath31 cmd that contains rsgs , with the result that faint , unresolved galaxies introduce a significant source of uncertainty when probing the stellar content and structure of the low surface brightness outer regions of ngc 5102 . a rsg plume , which is a conspicuous feature in the cmds of spiral galaxy disks ( e.g. davidge 2006 ; 2007 ) , is not seen in the ngc 5102 cmds . still , the spatial distribution of objects with brightnesses and colors that are consistent with rsgs indicate that a modest population of these objects is present throughout the galaxy ( 5 ) . that bright ms stars and rsgs are found over a range of r@xmath33 indicates that there has been star formation throughout the ngc 5102 disk during the past few tens of myr , although the small number of these objects indicates that the overall sfr during this time has been modest . the @xmath32 cmds are shown in figure 2 . the wircam data sample the brightest agb stars in ngc 5102 , which have @xmath42 and @xmath43 . the density of stars on the infrared cmds is lower than on the visible / red cmds because evolution at infrared wavelengths is spread over a larger range of magnitudes , and only the most evolved agb stars are detected in the wircam images ( 3.4 ) . many of the brightest stars in the central 1 kpc are recovered in the wircam data because of the comparatively ( at least with repect to visible wavelengths ) high contrast between these objects and the underlying body of fainter stars . the diminished density of stars on the infrared cmd notwithstanding , an obvious concentration of agb stars can be traced out to r@xmath44 kpc in figure 2 . the cmds of sources with r@xmath45 kpc are dominated by objects that do not belong to ngc 5102 . the plume of objects with @xmath46 and @xmath47 consists of foreground galactic disk stars , while the more diffuse sequence with @xmath48 and @xmath49 is made up of background galaxies . the wircam observations of ngc 5102 are spatially complete out to r@xmath4 kpc , and the decrease in the number density of objects in the cmds with r@xmath50 kpc reflects the progressively smaller areas on the sky that are sampled as one moves to larger r@xmath33 . isochrones from girardi et al ( 2002 ; 2004 ) are compared with the cmds of objects with r@xmath33 between 2 and 3 kpc in figure 3 . davidge ( 2008a ) found that the majority of rgb stars in the disk of ngc 5102 have [ m / h ] between 0.9 and 0.1 , and so only models with z = 0.019 and z = 0.004 are considered here . while kraft et al . ( 2005 ) deduce a super - solar metallicity using spectral synthesis techniques , this is based on observations of the central regions of ngc 5102 , and not the disk ( davidge 2008a ) . the red stars sampled with these data are intrinsically bright and are highly evolved . therefore , some aspects of the physics used to construct the portions of the isochrones that deal with the advanced stages of agb evolution are worth noting . the isochrones used here include evolution during the thermally - pulsing agb ( tp - agb ) phase . this evolution is tracked using the models discussed in the appendix of marigo & girardi ( 2001 ) , which include prescriptions for hot - bottom burning and the third dredge - up . there are uncertainties in key aspects of the model physics , such as the equation of state , the rates of mass loss among highly evolved stars , and the handling of convection and convective boundaries . girardi et al . ( 2000 ) discuss the model physics and the related assumptions for the portions of isochrones used here that cover evolution prior to the tp - agb . the isochrones were transformed onto the observational plane by assuming oxygen - rich photospheres , and so do not match the properties of c stars . finally , an observational complication of using highly evolved agb stars as chronometers is that a large fraction are long period variables ( lpvs ) . with photometric amplitudes that approach a magnitude in @xmath36 , the presence of lpvs blur the boundaries of the region on the cmds that contain the most highly evolved stars . the ms turn - off is consistent with log(t@xmath51 , indicating that star formation occured during the past few tens of myr at intermediate radii in ngc 5102 . as for more evolved stars , the age - metallicity degeneracy confounds efforts to determine ages for these objects from photometric data alone , and the ages estimated for agb stars from the z=0.004 models are 0.2 - 0.3 dex older than those from the z=0.019 models . c stars are important probes of stellar content because they have conspicuous photometric properties , and contribute significantly to the light output from simple stellar systems with ages between roughly 1 gyr ( persson et al . 1983 ) and @xmath52 gyr ( cole & weinberg 2002 ; feast et al . demers & battinelli ( 2005 ) investigated the c star content in the outer regions of m31 , with the majority of the candidate c stars identified on the basis of their @xmath53 spectral - energy distributions . the area of the @xmath31 cmd that demers & battinelli ( 2005 ) found to contain c stars is indicated in the upper panels of figure 3 . the area containing oxygen - rich and carbon - rich agb stars overlaps on the @xmath31 cmd , and so c stars can not be identified from @xmath1 and @xmath0 observations alone . still , it is evident from figure 3 that some c stars may be lurking in the @xmath54 cmds of ngc 5102 , and this is confirmed by the detection of c stars in the wircam observations ( see below ) , where all but the warmest c stars have photometric properties that are very different from those of oxygen - rich giants . line blanketing is much less of an issue at infrared wavelengths than at visible wavelengths , and one consequence is that the infrared cmds sample stars that tend to be older than those in visible / red cmds . comparisons with the isochrones suggest that the majority of stars in the @xmath32 cmds have log(t@xmath55 . still , there are stars near the red edge of the isochrones with @xmath56 that _ may _ have ages approaching 10 gyr , although at least some of these are probably warm c stars ( see below ) . the red envelope of the isochrones in figure 3 occurs near @xmath57 ( z = 0.004 ) and @xmath58 ( z=0.019 ) , due to the assumption of oxygen - dominated photospheres . in fact , there are objects in the @xmath32 cmds with @xmath59 , and some of these are probably c stars . the lmc @xmath32 cmd constructed by nikolaev & weinberg ( 2000 ) from 2mass data contains a prominent c star sequence , and the region of the lmc cmd that contains c stars is indicated in the lower row of figure 3 . of course , some of the red objects in the ngc 5102 cmds are undoubtedly compact background galaxies , the majority of which will have @xmath60 . still , in 5 it is shown that the spatial distribution of the objects with @xmath59 cluster around the main body of ngc 5102 , as expected if a large fraction are c stars . the numbers of c stars and bright m giants in ngc 5102 can be used to gain insight into its evolution during the past few gyr . for a given star - forming history , the ratio of c stars to m giants on the tp - agb can be calculated using the fuel consumption theorum ( e.g. renzini & buzzoni 1986 ) , and these predictions can be compared with the observed ratio . for the current work a model in which the sfr in the disk of ngc 5102 is constant with time , with a disk age of 10 gyr , is considered . there is evidence that the sfr in ngc 5102 has departed significantly from the time - averaged mean sfr during the past @xmath11 gyr ( e.g. davidge 2008a ; kraft et al . 2005 ) , but the goal of the present effort is not to reproduce the observed statistics in ngc 5102 . rather , a constant sfr model simply provides a convenient benchmark for gauging the relative behaviour of the sfr with respect to time when comparisons are made by with observations . the relative contributions that stars evolving on the tp - agb make to the total luminosity of a stellar system were computed using evolutionary fluxes and fuel consumption values from maraston ( 2005 ) . the fuel consumption fractions for c stars with z@xmath13/2 from figure 12 of maraston ( 2005 ) , which was calibrated from c star counts in magellanic cloud star clusters , were also adopted . this has an important implication for c star statistics , since these fuel consumption values imply that all tp - agb stars with ages between log(t@xmath61 ) = 9.0 and 9.3 are c stars ( i.e. 100% of the fuel consumed during tp - agb evolution is done so while the star has a c - rich atmosphere ) . isochrones with ages log(t@xmath61 ) = 10 pass through the @xmath32 cmd ( 3.4 ) , and the numbers of old agb stars in the data can be estimated using the fuel consumption theorum . due to the progressive decline of the evolutionary flux with increasing age , coupled with the decline in the amount of fuel consumed by stars with m @xmath62 m@xmath13 during tp - agb evolution , then tp - agb stars with ages in excess of 2 gyr contribute only modestly ( @xmath63 ) to the total number of oxygen - rich tp - agb stars in the constant sfr model . given the evidence for elevated levels of star formation during intermediate epochs then it is likely that the majority of bright agb stars in the wircam data formed within the past @xmath64 gyr . using the assumptions described above then the total number of c stars and luminous m giants can be estimated if the total stellar mass of the ngc 5102 disk is known . the total brightness of ngc 5102 is @xmath65 ( jarrett et al . 2003 ) , so that m@xmath66 . assuming that the luminosity - weighted age of ngc 5102 falls between 0.5 and 3 gyr , which is the range of dates that davidge ( 2008 ) and kraft et al . ( 2005 ) suggest correspond to significant recent star formation activity , then m / l@xmath67 ( mouhcine & lancon 2003 ) , and so the total stellar mass of ngc 5102 is @xmath68 m@xmath13 . the surface brightness profile of jarrett et al . ( 2003 ) suggests that roughly two - thirds of the @xmath36 light originates from the central portion of the galaxy that deviates from an exponential light profile , and so the disk mass is @xmath69 m@xmath13 . adopting the upper and lower luminosity limits for tp - agb evolution from mouhcine & lancon ( 2002 ) and a salpeter imf for the main sequence component , then the constant sfr model predicts 1800 4400 m giants experiencing tp - agb evolution , and 4300 10800 c stars . given galaxy - to - galaxy variations in star - forming history and metallicity , one might not expect that the number of c stars predicted from the constant sfr model would agree with what is seen in other galaxies . it is thus interesting that the number of c stars predicted from the model , in which 10% of the total stellar mass formed between 1 and 2 gyr in the past , is roughly consistent with c star counts in other galaxies after adjusting for differences in mass . for example , the lmc , which has a total stellar mass @xmath70 m@xmath13 , contains 10@xmath71 c stars ( nikolaev & weinberg 2000 ) , so @xmath72 c stars would be predicted in a galaxy with the same star - forming history as the lmc , but a mass comparable to the disk of ngc 5102 . demers , battinelli , & letarte ( 2003 ) find 413 c stars in ngc 3109 , and the number of c stars predicted for a galaxy with the same m@xmath73 as ngc 5102 from these data is @xmath74 . finally , the number of c stars predicted from the letarte et al . ( 2002 ) survey of ngc 6822 after scaling to the mass of ngc 5102 is @xmath75 . c stars and m giants that are evolving on the tp - agb can be identified based on their location on the @xmath32 cmd . for c stars , the number of objects throughout the disk of ngc 5102 in the region indicated in figure 3 has been counted . the blue - edge of the c star sequence in figure 3 is @xmath76 . however , there is some uncertainty in defining the blue edge of the c star sequence , and so counts were also made for objects with a blue edge at @xmath77 and having the same @xmath36 magnitude boundaries as the c star region in figure 3 . the final c star count is the mean of the counts with the two @xmath30 cut - offs . as for the region of the @xmath32 cmd that contains m giants on the tp - agb , the girardi et al . ( 2002 ) isochrones indicate that the onset of the tp - agb phase of evolution occurs near m@xmath78 for intermediate - age stars , while the agb - tip peaks at m@xmath79 . the isochrones also predict that m giants on the agb have @xmath30 between 0.9 and 1.3 . thus , the number of sources with m@xmath80 between @xmath81 and m@xmath79 and between @xmath82 and 1.3 have been counted ; the mean of the counts with the lower m@xmath80 set at 7.5 , 7.0 , and 6.5 is adopted as the number of tp - agb m giants in ngc 5102 . applying the methodology described above , and correcting for contamination from background galaxies using observations of a control field along the minor axis of ngc 5102 ( the number of contaminating objects is modest when compared with stars in ngc 5102 ) then there are @xmath83 m giants and @xmath84 c stars in the disk of ngc 5102 . the uncertainties are dominated by the dispersion in counts arising from the different photometric selection criteria described in the previous paragraphs . the numbers of m giants agree with the range predicted in 4.1 , and broadening the color boundaries for m giants on the cmd by @xmath85 magnitude increases the number of these stars by 10 20% , further improving the agreement with the constant sfr predictions . in contrast , the number of c stars is significantly lower than predicted by the constant sfr model . highly evolved agb stars may be obscured by circumstellar dust , and this might decrease the numbers of c stars . however , it can be anticipated that this will be a factor only for the stars that are nearing the end of their tp - agb evolution , and it is unlikely that correcting for such stars will boost the numbers of c stars substantially . taken at their face value , the numbers of m giants and c stars in the @xmath32 cmd suggest that @xmath86 of the total stellar mass in the disk formed within the past 1 gyr , but that @xmath87 of the total stellar mass formed within the past 2 gyr . the mean sfr in ngc 5102 during the past 1 gyr is then @xmath11 m@xmath13 year@xmath14 , whereas davidge ( 2008b ) estimated a sfr of at least 1.4 m@xmath13 year@xmath14 based on modelling of the bolometric lf of agb stars along the minor axis of ngc 5102 . given the ( substantial ) uncertainties in computing the mass of ngc 5102 from photometric data , the estimated sfrs are not significantly different . while there are considerable uncertainties in the computation of the numbers of m giants and c stars , such as determining the mass of the ngc 5102 disk , the ratio of c stars to m giants ( c / m ) should yield a more robust means of probing recent star formation . the counts from the cmds indicate that c / m @xmath88 , which is significantly smaller than the ratio predicted from the constant sfr model , which is c / m = 2.4 . the measured c / m ratio is consistent with a recent marked increase in the sfr of ngc 5102 during the past @xmath11 gyr , which produced a high number of bright m giants when compared with the c stars that formed 1 2 gyr in the past . however , the reader is cautioned that a lower than expected c / m ratio could result from other factors . for example , a c / m ratio that is lower than the model prediction could occur if the dominant population in ngc 5102 is either younger than @xmath89 gyr or older than @xmath90 gyr . such ages for the dominant population are in conflict with the large number of intrinsically bright agb stars that are seen throughout the disk of ngc 5102 ( e.g. davidge 2008a ) , and the evidence of elevated sfrs in other parts of the galaxy a few tenths of a gyr in the past ( e.g. kraft et al . 2005 ) . the c / m ratio also depends on metallicity . as discussed by maraston ( 2005 ) , the fraction of fuel that is consumed during tp - agb evolution while a star has a c - rich atmosphere scales with metallicity , as a metal - rich star must dredge up more c in order to bind o in co than a metal - poor star . kraft et al . ( 2005 ) conclude that the central regions of ngc 5102 have a super - solar metallicity , and so a c star survey of this part of the galaxy should find a c / m ratio that is lower than in a comparatively metal - poor environment that had the same recent star - forming history . this being said , it is unlikely that metallicity plays a major role in the low c / m ratio in the disk of ngc 5102 . the metallicity measured by kraft et al . ( 2005 ) is based on the central regions of the galaxy , and the colors of the rgb sequence in the disk measured by davidge ( 2008a ) are indicative of a sub - solar metallicity . in fact , the peak metallicity among rgb stars is [ m / h ] @xmath91 ( davidge 2008a ) , which is lower than the metallicity used to compute the numbers of c stars and m giants in 4.1 . the adoption of a lower metallicity in the calculations will result in a higher predicted c / m ratio , increasing the difference with respect to the observed ratio . stars with ages from @xmath16 myr to at least a few gyr are detected in the megacam and wircam images , and these data can be used to investigate the spatial distribution of star - forming activity with respect to time in ngc 5102 . a caveat is that the ability to resolve stars that formed over a given age range drops as progressively older stars are considered , and isolating sites in a galaxy that formed stars over narrow age ranges becomes problematic . stars of different ages and evolutionary stages were identified based on their locations in cmds , as indicated in figure 4 . the bright ms stars and rsgs that formed during the past few hundred myr are identified from the megacam data , while agb stars that formed within the past few gyr are identified from the wircam data . the region of the megacam cmd that contains rsgs has been divided into three groups , the brightest of which contains stars with ages @xmath92 myr , whereas the faintest samples stars with ages in excess of 100 myr . the various stellar types have distinctly different spatial distributions , and this is demonstrated in figure 5 . while there are only a modest number of objects in each group , it can be seen that ms stars and brsgs occur in largest numbers in the south - western half of the disk , which is the area of the galaxy that also contains hii regions . in contrast , the irsgs and frsg / bagbs , where ` bagb ' refers to ` bright agb ' stars , are more evenly distributed across ngc 5102 . the uniform distribution of these stars is likely a consequence of their ages . with ages that exceed many tens of myr , these stars have probably obtained random velocities through interactions with molecular clouds that have allowed them to diffuse away from their places of birth , thereby blurring any structure in their original distribution . the agb stars with oxygen - rich ( oagb stars ) and carbon - rich ( cagb stars ) atmospheres that were identified from the wircam images define a more compact distribution than the megacam - based samples . given that these stars have ages of a few hundred myr to a few gyr , then this suggests that star formation at this epoch was more centrally concentrated than during the past few hundred myr . it is thus interesting that , in contrast to the spatial distribution of objects identified from the megacam data , the oagb stars are not uniformly distributed . an arm of oagb sources extends to the north east of the galaxy along the major axis , and this same feature is seen among cagb stars . a corresponding feature is not seen to the south west of the galaxy . there is a clear concentration of oagb stars to the west of the galaxy center . although the small numbers of cagb stars makes the identification of an obvious concentration amongst these stars more difficult , the distribution of cagb stars is clearly asymmetric along the east - west axis , in the sense that cagb stars extend @xmath64 kpc to the west of the galaxy center , but only @xmath11 kpc to the east . a corresponding structure is not evident in the frsg / bagb counts as most of the area containing the oagb concentration falls in a gap between ccds . the concentration of agb stars coincides roughly with the center of a loop of [ oiii ] emission discovered by mcmillan et al . ( 1994 ) , for which they estimated an age of @xmath93 10 myr based on its size and expansion velocity . this might suggest that the agb concentration was also a site of very recent star formation . however , while davidge ( 2008 ) found ms stars to the west of the galaxy center , these tend to be fainter , and hence older , than the ms stars in the southern portions of ngc 5102 . in fact , the distribution of stars in figure 5 indicates that the area near the agb cluster does not contain an anomolous number of bright ms stars , with the caveat that the center of the agb cluster falls in a gap between ccds . the absence of hii regions in this part of ngc 5102 , based on the entries in table 4 of mcmillan et al . ( 1994 ) , further argues that it is not a site of star formation at the present day . the stellar content of the agb concentration appears to differ from that of the ngc 5102 disk , and this is demonstrated in figure 6 . the @xmath32 cmds of 1.6 arcmin@xmath20 regions that are centered 50 arcsec immediately to the west ( ` agb cluster ' ) and east ( ` disk east ' ) of the galaxy nucleus are compared in the upper row of figure 6 . there are almost twice as many stars in the agb cluster area , and so the cmd of this feature is more richly populated than that of the eastern disk . the cmd of the disk east field also appears to be broader than that of the agb cluster region . indeed , the disk east field contains 6 stars with @xmath94 , whereas there is only one such star in the agb cluster cmd . the more compact cmd morphology of the agb cluster is consistent with this area being dominated by stars that have a narrower range in age and metallicity than occurs in the main body of the disk . the @xmath30 color distributions of stars with @xmath36 between 19 and 20 ( i.e. with m@xmath73 between 7.5 and 8.5 ) in the disk east and agb cluster fields are compared in the lower row of figure 6 . both distributions peak near j k = 1.1 . however , the color distribution of the agb cluster appears to be skewed by @xmath89 mag to smaller j k colors than that of the disk east field . a kolomogorov - smirnov test indicates that the color distributions differ at the 91% significance level , and so the evidence for a difference in stellar content should be considered to be suggestive , rather than conclusive . deeper photometric studies will provide greater statistical robustness when comparing the stellar content of the agb cluster with that of the rest of the disk . this being said , if the stars in both fields have similar metallicities , then a 0.2 mag difference in j k is consistent with a younger mean age for the agb cluster field , amounting to @xmath95log(t@xmath96 dex , based on the isochrones shown in figure 3 . the distribution of stars in the outermost regions of the ngc 5102 disk may provide clues into its past history . because the disk of ngc 5102 is viewed at a moderately high angle , the projected intrinsic width of the disk complicates efforts to study the distribution of stars , especially at small r@xmath33 along the minor axis . to reduce the possible impact of disk thickness on structural studies , in the following analysis star counts are restricted to a strip along the major axis with a projected width @xmath97 kpc . the numbers of ms , brsg , and irsg sources are too small to allow their radial profiles to be investigated . however , frsg / bagb , oagb , and cagb stars are present in moderately large numbers , allowing limited conclusions to be drawn about their radial distributions . the number of sources in these groups are shown as a function of r@xmath33 in figure 7 . the counts have been corrected statistically for contamination from background galaxies and foreground stars by subtracting number counts measured in control fields . the cmds in figure 1 indicate that incompleteness is an issue in the frsg / bagb star counts for r@xmath98 kpc , and so little weight should be assigned to the frsg / bagb counts near the center of ngc 5102 . it is evident from figure 7 that the oagb and frsg / bagb stars have very different radial distributions , in that the former are more centrally concentrated than the latter . this difference is perhaps not surprising given that there is not a one - to - one correspondence between the stars that fall in the frsg / bagb portion of the @xmath31 cmd and those located in the oagb portion of the @xmath32 cmd , and that stars with different age distributions are found in each of these boxes . as noted in 5.1 , the more compact spatial distribution of oagb stars suggests that star formation @xmath11 gyr ago was more centrally concentrated than during the past @xmath99 gyr . the number density of frsg / bagb stars in figure 7 flattens when r@xmath100 kpc , although ( 1 ) there are substantial errors in the measurements at large r@xmath33 due to the small numbers of objects , ( 2 ) even a modest variation in the number of background galaxies will have a major impact on the behaviour of the frsg / bagb profile at large r@xmath33 , and ( 3 ) the distribution of stars at large radii is asymmetric in both the wircam ( 5.1 ) and megacam ( 5.3 ) datasets . the radial profile of cagb stars , which have systematically different photometric properties than the majority of frsg / bagb objects , also show evidence of flattening near r@xmath9 kpc , although the wircam data are limited in their spatial coverage on the sky and so star counts can not be extended past this radius . still , the number density of oagb stars drops off near r@xmath101 kpc , and the oagb and cagb profiles in figure 7 suggest that the ratio of c stars to m giants increases with r@xmath33 in ngc 5102 , as is seen in other disk galaxies ( e.g. demers & battinelli 2005 ) . the frsg / bagb and cagb profiles in figure 7 suggest that the distribution of stars in the disk of ngc 5102 does not follow a simple exponential profile . erwin , beckman , & pohlen ( 2005 ) investigate the light profiles of barred early - type galaxies , and find that at least 25% of their sample have anti - truncated light profiles . the light profiles of these galaxies at large radii follow an exponential profile that is shallower than at small radii . similar light profiles are seen among unbarred disk galaxies , with the highest frequency of occurence among galaxies with early morphological types ( pohlen & trujillo 2006 ) . two possible mechanisms for generating an anti - truncated light profile are considered here , with the goal of yielding possible insights into the past history of ngc 5102 . anti - truncated light profiles may be the consequence of mass re - distribution in disks during interactions . younger et al . ( 2007 ) argue that the angular momentum of gas that is funneled to smaller radii is transfered to disk stars , causing the latter to move outwards . to the extent that the disks of ` normal ' spiral galaxies likely have formed stars over a wide range of ages prior to any interactions , then most of the stars that move outwards in a spiral galaxy will have ages in excess of a gyr . if the outer disk of ngc 5102 was populated in this way then it should have a luminosity - weighted age that is older than that of the inner disk , which at some point ( or points ) in the past experienced an elevated sfr due to the interaction . the model described by younger et al . ( 2007 ) is not without its problems . the distribution of stars in the outer disk of the interacting galaxy m81 does not support the predicted stellar content trends , in that the outer disk of m81 appears to be _ younger _ than the inner disk . this might indicate that some fraction of the gas in m81 was the beneficiary of angular momentum re - distribution , or that gas was accreted from m82 ( davidge 2009 ) . in addition , secular processes , such as viscosity - induced re - distribution of angular momentum , are expected to act on the order of a few disk crossing times , and these will blur any age signatures that may have been imprinted at the time of formation of the outer disk . an anti - truncated light profile may also result from the accretion of material with an angular momentum distribution that is different from that of the original disk , and this provides a natural explanation for warped outer disks ( van der kruit 2007 ) . if this mechanism is / was at play in ngc 5102 then the oldest stars in the outer disk will be younger than those in the main body of the disk . a difference in age between the inner and outer disk of ngc 5102 might be measureable using integrated light or the brightest resolved stars if the outer disk was assembled from material that was accreted only during the past 1 2 gyr . if the outer disk was accreted earlier then very deep imaging will be required to detect an age difference between the inner and outer disk . secular processes will also blur differences in stellar content with time . we close this section by noting that an extended disk might also result from star burst activity , although it is not clear if an anti - truncated light profile would result . dalla vecchia & schaye ( 2008 ) model galaxy outflows , and find that the gas dispelled by winds initially escapes the disk plane along the minor axis . however , material builds up along the minor axis as gas loss proceeds , with some gas falling back into the disk plane . the resulting increase in density forces gas to escape along the major axis , and the disk boundary moves outwards as ejected gas moves to larger radii . a prediction of this model is that the mass - weighted age of the region near the outer boundary of the disk after the star burst subsides will be younger than at larger radii , with the oldest stars in this part of the galaxy having an age that corresponds to the onset of material being ejected along the major axis . in the case of ngc 5102 this means that the outer disk will be populated by stars with ages @xmath102 gyr . there should then be an extended distribution of oagb stars , which is not seen . regions of a galaxy that have very low stellar density and span large spatial scales can be identified from star counts if the number density of photometrically - selected objects exceeds a threshold defined by the number of foreground stars and background galaxies . davidge ( 2008b ) searched for young stellar groupings in the m81 m82 debris field , and found three objects that have surface brightnesses of 27 28 mag arcsec@xmath103 in @xmath104 . this procedure has been applied to the ms and frsg / bagb objects in the megacam data . source counts were made in @xmath105 pixel bins , which corresponds to @xmath106 arcsec , or @xmath107 kpc at the distance of ngc 5102 . this binning produces a moderately large number of objects per resolution element , while also sampling the spatial scales over which significant changes in galaxy structure might be expected . the results are not critically sensitive to bin size , and the basic conclusions drawn below do not change if @xmath108 pixel or @xmath109 pixel binning factors are employed . aside from the central few arcmin of ngc 5102 , statistically significant over - densities were not found in the numbers of ms objects over the 1 degree@xmath20 megacam field . this is interesting given the low level of contamination from background galaxies with blue colors ( davidge 2008b ) . this null detection suggests that ngc 5102 is not accompanied by kpc - sized companions that contain stars that formed within the past few tens of myr . the situation is very different when the spatial distribution of frsg / bagb objects is considered . a gray scale image illustrating the distribution of frsg / bagbs is shown in the top panel of figure 8 . the star counts in the north west portion of the image are affected by bright stars , which suppress the detection of faint sources over arcmin or larger scales . the brightest of the stars in the megacam image is the @xmath110 star hd 112892 ( @xmath111 cen ) , which is at the center of the large dark spot in figure 8 . still , three large - scale features can be seen . the most conspicuous of these , at the center of the map , is the main body of ngc 5102 , where the density of frsg / bagb objects is highest . the second feature falls along the south west arm of the major axis , where elevated source counts can be traced out to @xmath112 kpc ( @xmath113 scale lengths ) at the distance of ngc 5102 . the third structure is a concentration of frsg / bagbs along the south east arm of the minor axis , that extends out to a projected distance of @xmath114 kpc . what is the statistical significance of the features identified in figure 8 ? to answer this question , the standard deviation about the mean number of counts pixel@xmath14 was measured in the southern quarter of the frsg / bagb map . this region was adopted as the control area since contamination from stars associated with ngc 5102 should be low . the pixels where the number of frsg / bagbs exceeds the mean in the control field at the 2 , 3 , 4 , and 5@xmath115 levels are shown in the bottom row of figure 8 . not unexpectedly , the frsg / bagb counts in the inner regions of ngc 5102 have the highest statistical significance , exceeding those in the control field at more than the @xmath116 level . what is of greater interest for the outer regions of ngc 5102 is that the frsg / bagb number counts in pixels along the south west major axis exceed those in the control area at more than the @xmath117 level this is in direct contrast to the number counts along the north west arm of the major axis , which do not exceed those in the control area at even the @xmath118 level . the distribution of stars in the outer disk of ngc 5102 is thus not homogeneous . rather , there is clumpy sub - structure , such that the stellar densities along the south west arm of the major axis in the outer disk are significantly higher than those along the north east arm . finally , pixels along the south east arm of the minor axis exceed those in the control area between the 2 and @xmath117 levels ; while having lower statistical significance than the structure on the south west major axis , these pixels are clustered together , suggesting that their elevated counts are not statistical flukes . the surface brightnesses of the areas along the major and minor axes of ngc 5102 can be estimated if their stellar content is assumed to be the same as areas of the ngc 5102 disk where the surface brightness has been measured directly . the assumption of similar stellar content is certainly open to debate given the evidence for population gradients in disks ( e.g. bell & de jong 2000 ) . still , this issue can be mitigated to some extent by using fields at moderately large r@xmath33 in ngc 5102 to calibrate surface brightness . a @xmath118 detection in the number counts corresponds to @xmath119 frsg / bagb objects arcsec@xmath103 , whereas a @xmath117 detection corresponds to @xmath120 frsg / bagb stars arcsec@xmath103 . from figure 7 it is seen that at r@xmath121 kpc , the mean density of frsg / bagb stars is @xmath122 arcmin@xmath103 , or @xmath123 arcsec@xmath103 . the surface brightness at this radius is @xmath124 mag arcsec@xmath103 in @xmath125 ( figure 2 of van woerden et al . therefore , features that exceed the frsg / bagb number counts in the control field at the @xmath117 level have a surface brightness @xmath126 mag arcsec@xmath103 , whereas those that exceed the control field counts at the @xmath118 level have a surface brightness of 27.1 mag arcsec@xmath103 . these very low surface brightnesses explain why these objects have eluded detection until now . the bright stellar content of the nearby ( d @xmath127 mpc ) s0 galaxy ngc 5102 has been surveyed with the wircam and megacam imagers on the cfht . with the exception of the crowded central regions of the galaxy , the megacam data cover all of ngc 5102 , while the wircam data are spatially complete out to r@xmath128 kpc . this is the first published study of the stellar content of ngc 5102 at near - infrared wavelengths , and the data have been used to probe the spatial distribution of stars , with particular emphasis on the outer regions of the galaxy . structural information of this nature will provide insights into the past evolution of ngc 5102 , and may identify the trigger of the elevated star - forming activity that occured within the past 1 gyr . there is a large body of literature that deals with the properties of gas - poor disks , and most of the papers focus on the higher mass kindred of ngc 5102 . given that the current paper deals with only a single galaxy , no attempt is made to present a comprehensive review of the literature on s0 galaxies . rather , we simply note here that it is broadly accepted that the progenitors of s0 galaxies are spiral galaxies that have experienced gas depletion . there is an environmental dependence ( e.g. butcher & oemler 1984 ) , and the mechanism causing gas removal has been the subject of much debate , focusing on galaxy - galaxy interactions ( e.g. lavery & henry 1988 ) , or infall from the field into an area with a dense intergalactic medium ( e.g. van dokkum et al . 1998 ) . an active nucleus may also drive gas from a galaxy ( e.g. silk 2005 ; silk & norman 2009 ) , although this is expected to be restricted to disk systems that have massive spheroids . of course , it is possible that not all s0s were subjected to the same gas - removal mechanism . low surface brightness structures such as those found here in the vicinity of ngc 5102 may prove challenging to detect in more distant s0 galaxies , and so ngc 5102 may play an important role for investigating the formation and evolution of gas - depleted disks . this being said , the reader is reminded that some characteristics of ngc 5102 are not typical of s0 galaxies . these include its intrinsic faintness ( m@xmath129 ; karachentsev et al . 2002 ) , the blue central colors and deep balmer absorption lines ( gallagher et al . 1975 ) , the modest total dust mass ( van woerden et al . 1993 ) , the low number of x - ray binaries , and the low globular cluster specific frequency ( kraft et al . 2005 ) . in regards to the modest intrinsic brightness and blue central colors of ngc 5102 , bedregal et al . ( 2008 ) investigate the spectroscopic properties of fornax cluster s0 galaxies that span a range of masses , and find a trend between luminosity - weighted central age and galaxy mass , in the sense that lower mass s0 s have younger central ages . the blue central colors of ngc 5102 may thus be tied to its modest mass . [ [ tracing - star - formation - in - ngc-5102-the - recent - past - and - the - immediate - future ] ] tracing star formation in ngc 5102 : the recent past and the immediate future ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ there is only a modest number of resolved bright ms stars and rsgs throughout the disk of ngc 5102 , demonstrating the low sfr during the past @xmath130 myr . the majority of young stars are in the southern half of the galaxy , which is where the only hii regions are located . this is also where the hi density is highest and there are hi clumps ; for comparison , hi in the northern part of the ngc 5102 disk is more uniformly distributed , with a lower mean density ( van woerden et al . 1993 ) . the hi distribution strongly suggests that barring some external influence the part of the disk that is presently south of the galaxy center will continue to be the dominant area of star formation for time scales of at least a few disk crossing times . rsgs and agb stars span a much wider range of ages than the brightest ms stars , and these objects are more uniformly distributed throughout the disk of ngc 5102 than bright ms stars . this indicates that star formation within the past @xmath131 gyr was not restricted to selected locations in the disk . indeed , the number densities of frsg / bagb stars at intermediate radii along the northern and southern arms of the major axis are similar , indicating that the northern and southern parts of the galaxy had similar time - averaged sfrs @xmath131 gyr in the past . the number counts of objects that are classified as frsg / bagb stars suggest that the light profile of the ngc 5102 disk does not follow a single power law , but breaks to a shallower profile near r@xmath132 kpc ( 5.5 6.7 disk scale lengths ) . the error bars in the frsg / bagb counts are significant at large radii , and background galaxies that at an angular resolution of roughly an arcsec are unresolved introduce a considerable source of uncertainty . still , the majority of early - type disk galaxies have light profiles that become shallower at large radii ( pohlen & trujillo 2006 ) . the spatial distribution of oagb stars , selected from the wircam data , is more compact than that of the frsg / bagb stars . given that the oagb stars tend to be older than those in the frsg / bagb sample , then this suggests that the outer disk of ngc 5102 has a younger mean age than the inner disk . this is similar to what is seen in m81 , where the structural properties of the outer disk have almost certainly been affected by interactions with m82 ( davidge 2009 ) . stars along the major axis of ngc 5102 , which presumably belong to the disk , can be traced out to r@xmath133 kpc , or @xmath134 disk scale lengths , in the southern half of the galaxy . this indicates that the disk may extend to a greater distance than was estimated by davidge ( 2008a ) , who used rgb star counts along the minor axis to trace the disk out to a de - projected radius of @xmath16 kpc . clumpiness in the outer disk of ngc 5102 may explain why the disk radius estimated made by davidge ( 2008a ) along the minor axis differs from that found here . the distribution of frsg / bagb stars in figure 8 indicates that there is structure in the outer disk of ngc 5102 , with a marked concentration of objects along the south west semi - major axis ( 4.3 ) . the disk of ngc 5102 is also markedly warped at large radii in the northern portion of the galaxy . this is evident in the hi map of van woerden et al . ( 1993 ) , where the hi distribution veers to the north west of the northern arm of the major axis as extrapolated from smaller r@xmath33 . oagb stars follow the hi warp . a cluster of background galaxies could masquerade as the southern extension of the ngc 5102 disk , although such a cluster would have to be positioned fortuitously along the major axis of ngc 5102 . deep images with angular resolutions of the order 0.1 arcsec or better will provide the data needed to confirm the nature of the objects that define the low surface brightness structures along the major axis of ngc 5102 . if , as suggested here , the structures are made up of stars that belong to the disk of ngc 5102 , as opposed to background galaxies , then the projected density of extended objects in these structures will be comparable to that in other parts of the area around ngc 5102 . on the other hand , if high angular resolution observations detect an excess population of galaxies then the frsg / bagb counts will need to be re - visited . [ [ the - nature - of - the - agb - concentration - near - the - center - of - ngc-5102-star - cluster - or - satellite ] ] the nature of the agb concentration near the center of ngc 5102 : star cluster or satellite ? ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ the kpc - sized concentration of agb stars immediately to the west of the galaxy center is the most prominent structure found in the vicinity of ngc 5102 , and the nature of this feature is of considerable importance for understanding the recent evolution of ngc 5102 . the relatively blue @xmath30 colors of the majority of agb stars in this area ( 5 ) suggest that this was probably an area of localized star formation in the not - to - distant past . the physical coincidence with a probable superbubble ( mcmillan et al . 1994 ) suggests that star formation may have continued in this stellar concentration up to within the past few tens of myr , and hi is concentrated in this area ( van woerden et al . deeper images should reveal a population of intermediate age ms stars in the agb concentration with a spatial density that is roughly twice that on the eastern side of the disk . the age measured from the main sequence turn off of this stellar concentration will provide a means of timing star burst activity in ngc 5102 . two possible origins for the agb concentration are considered here . one is that the agb concentration is a large cluster in the disk of ngc 5102 . an awkward aspect of this interpretation is that much of the most recent star - forming activity during intermediate epochs in ngc 5102 would have been concentrated in a single region of the disk that is offset from the center of the galaxy . this is not typical for starburst systems , where star formation tends to be centered on the nucleus ( e.g. kewley , geller , & barton 2006 , and references therein ) . still , if the star cluster interpretation is correct then it can not be older than @xmath11 gyr . the crossing time of the ngc 5102 disk is on the order of a few hundred myr , and coherent structures might be expected to dissipate after a few crossing times , due to kinematic heating by interactions with large gas clouds and other star clusters . the difference in colors between the brightest m giants in the disk to the east of the galaxy center and the agb concentration ( figure 6 ) is qualitatively consistent with a relatively ` young ' age for the agb concentration , although it is difficult to assign a reliable absolute age from these data alone . the second interpretation is that the stellar concentration is a satellite galaxy seen in projection against ( or behind ) the disk . kraft et al . ( 2005 ) hypothesize that a gas - rich dwarf galaxy triggered the large episode of star formation in ngc 5102 during intermediate epochs . a long - recognized difficulty with this hypothesis is that ngc 5102 is in an isolated environment . using distances measured by karachentsev et al . ( 2007 ) , the nearest known neighbor of ngc 5102 is the barred spiral galaxy eso383-g087 , which is at a distance of @xmath135 mpc . the satellite interpretation of the agb concentration is thus attractive as it provides a trigger for the burst of star formation in ngc 5102 . comparatively easy tests of the satellite hypothesis are that ( 1 ) a satellite would have left a trail of material as it was disrupted , and this trail might still be visible , and ( 2 ) ngc 5102 should show signs of having experienced a recent interaction . while ambiguous stellar streams have not been detected near ngc 5102 , there are two _ possible _ tidal remnants . the grouping of stars along the minor axis could be the remnant of an interaction involving a satellite with an orbit that was perpendicular to the disk plane . the integrated brightness of the three minor axis pixels in figure 8 that breech the @xmath118 threshold is v @xmath136 , such that m@xmath137 , which is @xmath138 of the integrated brightness of ngc 5102 . mori & rich ( 2007 ) model interactions in which satellites with masses that are no more than a few percent of the total mass of m31 pass through the central regions of m31 . while caution should of course be exercised when extending these results to ngc 5102 , some broad conclusions are worth noting . the satellite is significantly disrupted , and its stars are distributed throughout the extraplanar environment , forming distinct streams and shells during the first @xmath11 gyr . the disk of m31 is not affected significantly as long as the satellite has a mass that is less than @xmath138 of that of the galaxy . a few gyr after the first interaction the extraplanar region is dominated by a diffuse component , with little evidence of streams and shells . the clumpy structures found in the outer disk of ngc 5102 are reminiscent of the extended outer disk of m31 ( ibata et al . 2005 ) . based on its disk - like stellar content , faria et al . ( 2007 ) argue that the m31 ` cluster ' g1 , which falls along the major axis of m31 , might actually be a fragment of the m31 disk that was pulled from the disk during a past interaction . richardson et al . ( 2008 ) find that the stellar content throughout the outer disk of m31 is highly uniform , and argue that the extended disk may be the consequence of the heating and disruption of the thin disk of m31 by a satellite . an extended disk formed in this manner may stay in place for many gyr given the long mixing times in the outer regions of galaxies ( e.g. johnston , hernquist , & bolte 1996 ) . it is further worth noting that the ring - like hi distribution in ngc 5102 is reminiscent of the distribution of gas and dust in m31 , which simulations suggest may be the result of a smaller companion ( presumably m32 ) passing through the disk ( e.g. block et al . the warping of the ngc 5102 disk , which is seen in both the hi and stellar distributions , may also be the consequence of a tidal encounter . there are other tests of the satellite hypothesis , but they will prove challenging to implement . a satellite galaxy would probably contain a substantial old population , with a metallicity that may be distinct from that of ngc 5102 . however , such a population would prove difficult to detect given that the proposed satellite is viewed against / through the crowded inner regions of ngc 5102 . another test is that the radial velocities of stars in the cluster and the disk may differ . an obvious difficulty is that an investigation of the velocities of even the most luminous stars will require larger telescopes than those currently available . this being said , the agb concentration is coincident with a localized peak in the hi distribution , and this feature is not kinematically distinct in hi channel maps ( van woerden et al . 1993 ) . a large velocity difference between ngc 5102 and the supposed satellite may not be present today , as dynamical friction will act to harmonize the kinematic properties of stars in the satellite with those in ngc 5102 , although this may require very long time scales depending on the orbital geometry and relative system masses ( e.g. colpi , mayer , & governato 1999 ) . in 6.3 the collection of stars at a projected galactocentric distance of @xmath114 kpc on the south east minor axis of ngc 5102 was discussed in the context of it being a possible tidal remnant . in this section we consider another possible origin for this structure . galaxies with elevated sfrs may experience outflows , and this can have an impact on their surroundings . the outflow from m82 may trigger star formation in its extraplanar regions ( e.g. davidge 2008c , d ) , as the wind interacts with circumgalactic clouds that presumably were tugged from m82 and/or m81 as they interacted , or condensed from material that was ejected from m82 by winds . the features identified as the ` cap ' ( devine & bally 1999 ) and m82 south ( davidge 2008d ) are perhaps the most obvious signatures of the interplay between the m82 outflow and surrounding material . the ism of ngc 5102 shows only marginal evidence for an outflow at the present day ( shwartz et al . 2006 ) , although kraft et al . ( 2005 ) find diffuse x - ray emission in the central 1 kpc of ngc 5102 , which they attribute to a superbubble that is powered by sne associated with the most recent burst of star - forming activity . however , the sfr in ngc 5102 during intermediate epochs was probably sufficient to power a galaxy wind ( 7.3 of davidge 2008a ) . could the diffuse stellar component along the south east minor axis of ngc 5102 have formed in such an outflow ? an outflow origin for these stars provides a natural explanation for their location along the minor axis of ngc 5102 . however , such structures will probably not be long - lived , and will probably be disrupted over a few orbital timescales around ngc 5102 ( i.e. within @xmath11 gyr ) . there would also have to be a large diffuse gas component surrounding ngc 5102 , and no evidence for this has been found . finally , with m@xmath139 ( 6.3 ) the minor axis structure would be much larger than the largest outflow cluster in m82 ( davidge 2008d ) . thus , while an outflow origin for these stars can not be ruled out , this possibility appears to be unlikely .

we study the dynamical effect of relative velocities between dark matter and baryonic fluids , which remained supersonic after the epoch of recombination . the impact of this supersonic motion on the formation of cosmological structures was first formulated by , in terms of the linear theory of small - scale fluctuations coupled to large - scale , relative velocities in mean - density regions . in their formalism , they limited the large - scale density environment to be those of the global mean density . we improve on their formulation by allowing variation in the density environment as well as the relative velocities . this leads to a new type of coupling between large - scale and small - scale modes . we find that the small - scale fluctuation grows in a biased way : faster in the overdense environment and slower in the underdense environment . we also find that the net effect on the global power spectrum of the density fluctuation is to boost its overall amplitude from the prediction by . correspondingly , the conditional mass function of cosmological halos and the halo bias parameter are both affected in a similar way . the discrepancy between our prediction and that by is significant , and therefore the related cosmology and high - redshift astrophysics should be revisited . the mathematical formalism of this study can be used for generating cosmological initial conditions of small - scale perturbations in generic , overdense ( underdense ) background patches .

1603.09356

the @xmath0-cold dark matter ( @xmath0cdm ) scenario , combined with the theory of cosmic inflation , is the successful , concurrent model describing the past and the present of our universe , consistent with a wide range of observations . in this scenario , cosmological structures grow out of an extremely uniform density field but with tiny fluctuations that are seeded by the cosmic inflation . the growth of the cdm density fluctuations and the growth of baryon density fluctuations are not in perfect synchronization , because baryons were tightly coupled to photons before the epoch of recombination and thus their motion was different from the motion of cdm which only reacts to gravity . only after recombination baryons gradually decoupled from photons , and followed the motion of the cdm under gravity . cosmological observations have verified the @xmath0cdm scenario in scales large enough to make the baryonic physics almost irrelevant ( e.g. @xcite ; @xcite @xcite ) . however , once in the regime where the baryonic physics becomes important , the growth of baryon fluctuations is affected by hydrodynamics and the growth of the cdm is affected by the gravitational feedback from the baryon fluctuations . some of the usual assumptions that are made for treating very large scales , therefore , should be taken carefully or modified when treating relatively small scales . for example , @xcite improved on the previous estimation of the linear density power spectrum in small scales , by replacing the usual assumption made in cosmology that the sound speed of baryons is uniform in space with the fact that the sound speed fluctuates in space in small scales . they showed that more than @xmath1 change occurs in the baryon density power spectrum and even more change in the baryon temperature power spectrum . a sheer inclusion of the sub - dominant , yet non - negligible baryonic component in the analysis changes the prediction on the matter density power spectrum at a few percent level even in large scales , as was shown in the framework of the high - order perturbation theory @xcite . similarly , the relative velocity ( `` streaming velocity '' ) between baryons and the cdm after the recombination should also be considered carefully in cosmology . * th hereafter ) , for the first time , properly calculated the growth of small - scale density fluctuations under the influence of the streaming velocity . small - scale fluctuations are coupled to large - scale streaming velocity fields which are coherent over a few comoving mpc scale . this has its own spatial fluctuation with @xmath2km / s standard deviation at the epoch of recombination , and then decays in proportion to the inverse of the scale factor @xmath3 . its impact on the small - scale structure formation is non - negligible because the streaming velocity remains supersonic ( until the intergalactic medium is strongly heated ) . this then leads to the suppression of small - scale matter density fluctuations . the wave - modes that are the most strongly affected are around @xmath4mpc , and the impact is on the matter power spectrum , the conditional mass function and the halo bias parameter , to name a few ( th ) . subsequent studies have considered the impact of the streaming velocity in the perspective of both cosmology and astrophysics . the boost of amplitude and the shift of the peak of the baryonic acoustic oscillation ( bao ) feature due to the streaming velocity , in the gas intensity mapping or the galaxy survey , were intensively investigated ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . they find that both high - redshift ( e.g. @xcite ; @xcite ) and low - redshift ( e.g. @xcite ) surveys will be affected . the impact of the streaming velocity on bao may be separated out from the impact of the matter density itself ( @xcite ) , which is important because otherwise it will become another nuisance parameter in cosmology . the astrophysical impact of the streaming velocity has been investigated with focus on the formation of the nonlinear structure such as cosmological halos and stellar objects ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . most studies indicate that the formation of minihalos ( roughly in the mass range @xmath5\ , m_{\odot}$ ] ) and the formation of stellar objects in them are suppressed . it may induce baryon - dominated objects such as globular clusters ( @xcite ; @xcite ) , but the actual star formation process leading to globular clusters has yet to be simulated . it may be responsible even for the generation of the primordial magnetic field ( @xcite ) . because minihalos are the most strongly affected among cosmological halos and they are responsible for the early phase of the cosmic reionization process , how they change the high - redshift 21-cm background is also of a prime interest ( @xcite ; @xcite ; @xcite ) . it is noteworthy that some of these numerical simulation results ( @xcite ; @xcite ; @xcite ) , which are based on the initial condition generated by the usual boltzmann solver such as the code for anisotropies in the microwave background ( camb : @xcite ) , need to be re - examined . this is because the impact of the streaming velocity is cumulative and inherent even at @xmath6 ( @xcite ) , at which or later these simulations start with a streaming velocity implemented by hand . the original formalism by th has been re - investigated in terms of the high - order perturbation theory in the wave - number space ( `` @xmath7-space '' henceforth ) , and some `` missing terms '' previously neglected were found important ( @xcite ; @xcite ) . basically , for the large - scale modes responsible for the streaming velocity ( @xmath8/{\rm mpc}$ ] ) , th used a trivial solution for the evolution of the streaming velocity ( @xmath9 ) and the density ( @xmath10 and @xmath11 being overdensities of the cdm and baryons in large scale , respectively ) environment : @xmath12 , @xmath13 . treating this as a new 0th - order solution to the perturbation equations , they then examined how the perturbation in small scales grows . in doing so , they treated @xmath9 as a spatial quantity and perturbation variables in small scales as @xmath7-space quantities . this is basically a high - order perturbation theory , coupling large - scale mode ( @xmath9 ) and the small - scale modes ( @xmath14 and @xmath15 , small scale cdm and baryonic overdensities , respectively ) . however , @xmath9 is tightly linked to the fluctuating @xmath10 and @xmath11 through the density continuity equation , and thus the trivial solution adopted by th can not be used for generically overdense and underdense regions . the continuity equation connects the divergence of @xmath9 to @xmath10 and @xmath11 , and @xcite finds that the divergence of @xmath9 is indeed an important term that one should not ignore . @xcite also works on the generic basis of non - zero @xmath10 and @xmath11 . we improve on the formalism of th by also considering the non - zero overdensities . toward this end , different from @xcite and @xcite , we inherit the original method by th and focus on the impact on small - scale modes : large - scale @xmath9 is treated as the spatial quantity and small - scale overdensities @xmath14 and @xmath15 are treated as the @xmath7-space quantity in the perturbation analysis . most importantly , we explicitly include generically `` non - zero '' @xmath10 and @xmath11 as a new set of spatial quantities , and consequently the divergence of cdm and baryon velocities as well . we find that this leads to a set of mode - mode coupling terms , including the velocity divergence - density coupling . we carefully include all the coupling terms to the leading order in our perturbation analysis . we also include the baryonic physics , namely fluctuations in the sound speed ( @xcite ) , the gas temperature and the photon temperature . our formalism is also suitable for generating initial conditions for n - body+hydro numerical simulations . because the new set of mode - mode couplings is imprinted in the initial condition , the initial condition generator considering the streaming - velocity effect by @xcite , cicsass , should also be improved on if one were to numerically simulate the structure formation inside overdense or underdense regions . this paper is organized as follows . after the introduction , we lay out the basic formalism and describe the statistics of large - scale fluctuations in section [ sec : formalism ] . in section [ sec : result ] , we show results on the matter density power spectrum , the conditional halo abundance and the halo bias as applications of the formalism . we conclude this work in section [ sec : discussion ] with a summary , discussion and future prospects . some details left out in the main body are described in appendices . we start from a set of equations for perturbations of relevant physical variables . all the @xmath16 modes of interest are in sub - horizon scale such that the newtonian perturbation theory holds . let us define the overdensity @xmath17 , where the subscript @xmath18=\{c , b } denotes either the cdm ( c ) or the baryonic ( b ) component , and @xmath19 where @xmath3 is the scale factor and @xmath20 is the proper peculiar velocity of component @xmath18 . when the universe is in the regime where we can ignore fluctuations of photons and neutrinos due to their rapid diffusion after recombination , we have ( e.g. @xcite ; th ) @xmath21 where @xmath3 is the scale factor , @xmath22 is the gradient in the comoving frame , @xmath23 , @xmath24 , @xmath25 , @xmath26 , @xmath27 is the present - day mean density of component @xmath18 in the unit of the critical density , @xmath28 is the sound speed , and @xmath29 is the hubble constant at a given redshift . even though a usual approximation for @xmath28 is a spatially uniform one given by @xmath30 which assumes a mean - density environment undergoing hubble expansion with the mean baryon temperature @xmath31 , for high @xmath16 modes a more accurate treatment is required ( @xcite ) . this requires replacing the pressure term in equation ( [ eq : mater_pressure ] ) in order to find the correct coupling of high-@xmath7 and low-@xmath7 modes in equations ( [ eq : perturbation_spatial ] ) and ( [ eq : perturbation_k ] ) . ] , @xmath32 in equation ( [ eq : master ] ) and considering another rate equation @xmath33 where @xmath34 and @xmath35 are temperature fluctuations of the baryon and the photon , respectively , @xmath36 is the mean photon temperature , @xmath37 is the global electron fraction at time @xmath38 and @xmath39 . for @xmath31 , we use the fitting formula by th . equation ( [ eq : master ] ) allows a trivial solution : @xmath40 , @xmath41 and @xmath42 , where @xmath43 is the initial scale factor . th took this as the zeroth - order solution of a spatial patch and developed a linear perturbation theory of small scale modes inside the patch . the physical process is in principle a coupling of small-@xmath7 and large-@xmath7 modes ( mode - mode coupling ) , which is beyond the linear theory where all modes are assumed to be mutually independent . th used the fact that the relative velocity @xmath44 is coherent over the length scale of a few comoving mpc ( contributed by modes with wave numbers in the range @xmath45 ) , and averaged the `` local '' power spectra of density fluctuations over many such patches with varying @xmath46 . even though a trivial solution exists , there also exists a nontrivial solution to equation ( [ eq : master ] ) , exact to the first order . this nontrivial solution is suited to describe the physics inside patches with non - zero overdensity ( see also @xcite ) . in order to obtain the nontrivial solution , we first linearize equation ( [ eq : master ] ) : @xmath47 where @xmath48 is the matter content with respect to the critical density @xmath49 at @xmath3 , and we ignored second - order terms and also the pressure term @xmath50 . this is indeed a valid approximation in the wave number range ( @xmath45 ) relevant to the coherent @xmath46 ( th ) , where the second - order terms remain much smaller than the first - order terms and the baryonic sound speed keeps decreasing from @xmath51km / s after recombination to make the pressure term negligible . then , equation ( [ eq : background ] ) can be rewritten as @xmath52 where @xmath53 , @xmath54 , @xmath55 , and @xmath56 . in the matter - dominated ( @xmath57 ) flat universe , @xmath58 allows both the growing mode ( @xmath59 , @xmath60 , @xmath61 ) and the decaying mode ( @xmath62 , @xmath63 , @xmath64 ) . @xmath65 allows a slowly decaying ( `` streaming '' ) mode ( @xmath66 , @xmath67 , @xmath68 ) and a compensated mode ( @xmath65=constant , @xmath69 , @xmath70 ) . during @xmath71 , the non - negligible amount of the radiation component ( cmb and neutrinos ) makes @xmath72 , and most of the simple analytical forms above become no longer intact except for \{@xmath73 , @xmath74 } of the streaming mode and \{@xmath65 , @xmath73 , @xmath74 } of the compensated mode . using this mode decomposition , the large - scale perturbations evolve in the following form : @xmath75 where we used upper - case letters to denote the `` background '' fluctuations for each patch of a few comoving mpc , over which we will develop the small - scale perturbation . @xmath76 , @xmath77 , @xmath78 , and @xmath79 are the initial ( at @xmath80 ) values of the growing , decaying , compensated , and streaming modes , respectively . @xmath81 , @xmath82 , and @xmath83 are growth factors of the growing , decaying , and streaming modes , respectively , and they are all normalized as @xmath84 at @xmath80 . we describe the details of these modes in appendix a. it is noteworthy , as is well known already , that both cdm and baryonic components tend to approach the same asymptotes @xmath85 and @xmath86 , indicating that baryons tend to move together with cdms in time . more interestingly , @xmath46 decays as @xmath87 throughout the evolution at any overdensity environment even though @xmath88 and @xmath89 grow roughly as @xmath90 individually ( except in regions with @xmath13 where @xmath88 and @xmath89 decay as @xmath87 ) . this fact may seem to make the analysis by th valid in generic overdensity environments to some extent : th relied on the trivial solution @xmath13 , in which all velocity components decay in time such that @xmath91 , @xmath92 , and most importantly @xmath93 . because the suppression of the matter - density fluctuations ( @xmath94 ) depends not on individual velocity components but on @xmath46 only , different temporal behavior of individual velocity components among the trivial and generic solutions do not matter . nevertheless , quantitative prediction by th will be questioned in section [ sub : pk ] , because we use nontrivial solutions ( equation [ eq : background_sol ] ) which result in the new type of coupling of high-@xmath7 and low-@xmath7 modes in general . we also require the evolution of @xmath95 , which is given by equation ( [ eq : master_temp ] ) : @xmath96 which is evolved in conjunction with equation ( [ eq : master ] ) . here we neglect @xmath97 term due to its smallness ( see also the following discussion ) , even though we do not neglect @xmath98 when initializing @xmath99 ( section [ sub : numerical ] ) . we have found a useful fitting formula for @xmath100 for patches with volume @xmath101 : @xmath102\nonumber \\ & & |\delta_{t,\ , a}|^{4.2591}|\delta_{t,\ , b}|^{-3.2591 } , \label{eq : deltat_fitting}\end{aligned}\ ] ] which provides a good fit to @xmath95 at @xmath103 , and for higher redshift range we simply ignore @xmath95 altogether because of smallness of @xmath95 in general . here @xmath104 , @xmath105 and @xmath106 , and an almost complete coupling of @xmath100 to @xmath107 at @xmath108 yields this simple , empirical relation to @xmath107 . we describe this fitting formula in more details in appendix b. the decoupling of @xmath100 from the cmb is much earlier than the mean value experiences , which occur at @xmath109 , because the larger the @xmath7 is , the earlier the decoupling occurs ( see e.g. figure 1 of @xcite ) . including @xmath97 explicitly may delay this decoupling to some extent , but we leave such an accurate calculation to future work . using the evolution equation for @xmath107 in equation ( [ eq : background_sol ] ) , equation ( [ eq : deltat_fitting ] ) is determined solely by local values of the 4 modes . now we expand equations ( [ eq : master ] ) , ( [ eq : mater_pressure ] ) and ( [ eq : master_temp ] ) to the linear order , taking equation ( [ eq : background_sol ] ) as the zeroth - order solution . we first define the net density , the net velocity ( of the fluid component @xmath110 ) , the net gravitational potential and the net baryon temperature as @xmath111 , @xmath112 , @xmath113 , and @xmath114 , respectively . here @xmath115 and @xmath116 denote the comoving - coordinate position of the center of a background patch and that of a small - scale fluid component , respectively , and we use lower - case letters for small - scale fluctuations . @xmath117 is the gravitational potential sourced only by the background fluctuations such that @xmath118 ( e.g. @xcite ) . similarly , @xmath119 . we then have @xmath120,\nonumber \\ \frac{\partial\delta_{{\rm b}}}{\partial t } & = & -a^{-1}{\bf v}_{{\rm b}}\cdot\nabla\delta_{{\rm b}}-(1+\delta_{{\rm b}})\theta_{{\rm b}}-\theta_{{\rm b}}\delta_{{\rm b}},\nonumber \\ \frac{\partial{\bf v}_{{\rm b}}}{\partial t } & = & -a^{-1}\left({\bf v}_{{\rm b}}\cdot\nabla\right){\bf v}_{{\rm b}}-a^{-1}\nabla\phi - h{\bf v}_{{\rm b}}\nonumber -a^{-1}\frac{k_{b}\bar{t}}{\mu m_{{\rm h}}}\nabla\left\ { \left(1+\delta_{{\rm b}}\right)\delta_{t}+\left(1+\delta_{t}\right)\delta_{{\rm b}}\right\ } -\left[a^{-1}\left({\bf v}_{{\rm b}}\cdot\nabla\right){\bf v}_{{\rm b}}\right],\nonumber \\ \frac{\partial\delta_{t}}{\partial t } & = & \frac{2}{3}\left\ { \frac{\partial\delta_{{\rm b}}}{\partial t}+\frac{\partial\delta_{{\rm b}}}{\partial t}\left(\delta_{t}-\delta_{{\rm b}}\right)+\frac{\partial\delta_{{\rm b}}}{\partial t}\left(\delta_{t}-\delta_{{\rm b}}\right)\right\ } \nonumber + \frac{x_{e}(t)}{t_{\gamma}}a^{-4}\left\ { \left[\left(\frac{5\bar{t}_{\gamma}}{\bar{t}}-4\right)\delta_{t_{\gamma}}\right]-\frac{\bar{t}_{\gamma}}{\bar{t}}\delta_{t}+4\left[\delta_{t_{\gamma}}\delta_{t}+\delta_{t}\delta_{t_{\gamma}}\right]\right\ } \nonumber \\ \nabla^{2}\phi & = & 4\pi ga^{2}\omega_{m}(f_{{\rm c}}\delta_{{\rm c}}+f_{{\rm b}}\delta_{{\rm b}}),\label{eq : perturbation_spatial}\end{aligned}\ ] ] where we marked the terms that can be further ignored in square brackets . first , we can safely ignore any terms containing @xmath35 and @xmath121 , because the former is negligible at @xmath122 compared to @xmath34 ( see figure [ fig : pk_z1000 ] ) and the latter is just too small ( @xmath123 at @xmath80 and decaying in time ) to produce any appreciable impact on the baryon temperature of high @xmath7 modes . secondly , the justification for ignoring @xmath124 is easily seen in viewpoint of the @xmath7-space . with the fourier expansion @xmath125 of a quantity @xmath126 in an actual , real space ( `` @xmath127 space '' henceforth ) , the background velocities have @xmath128 but with the condition @xmath129 . in contrast , the small - scale modes fluctuating against the background patches have intrinsically larger wave number @xmath7 , or @xmath130 . then , at each @xmath131 , @xmath132 . similarly , @xmath133 . we finally note that we do not include the coupling terms between similar wavenumbers in equation ( [ eq : perturbation_spatial ] ) , which will involve quadratic and higher - order polynomicals of @xmath134 and @xmath135 . therefore , the validity of equation ( [ eq : perturbation_spatial ] ) will break down in the non - linear regime . nevertheless , our approach is more suitable for a crude estimattion of the conditional halo mass function ( section [ sub : mass_function ] ) in terms of the extended press - schechter formalism , which is based on the mapping of the linear density growth to the nonlinear growth of the e.g. top - hat density perturbation . -space variance at @xmath80.,scaledwidth=50.0% ] the evolution of small - scale perturbations , therefore , is coupled to large - scale perturbations on which they are sitting . ignoring the terms in square brackets , and shifting the viewpoint to the cdm rest frame in which @xmath136 ( as in @xcite ; th chose the baryon rest frame ) , equation ( [ eq : perturbation_spatial ] ) , in the @xmath7-space , finally becomes @xmath137 where @xmath138 , @xmath139 and @xmath34 now denote fluctuations in the @xmath7-space while @xmath140 , @xmath141 and @xmath142 are fluctuations of a given patch at @xmath143 in the @xmath127 space , given by equation ( [ eq : background_sol ] ) . evolution of small - scale perturbations can be calculated by integrating the rate equation ( equation [ eq : perturbation_k ] ) from some initial redshift , preferentially not too long after the recombination epoch when the relative motion has not yet influenced the evolution . we take @xmath144 as the initial redshift . the initial condition should be generated for both the background quantities and the small - scale modes . for the background , as perturbations in equation ( [ eq : perturbation_k ] ) are @xmath127-space quantities whose distributions are all gaussian , one needs to sample these values in the @xmath127-space accordingly . for the small - scale modes , one just needs to track the evolution of the average value in the @xmath7-space . let us first describe the statistics of background patches that we expect . th calculated the evolution of small - scale ( @xmath145 ) fluctuations under different background patches but only of @xmath13 , and defined `` local power spectrum '' @xmath146 averaged out over all possible opening angles between @xmath46 and @xmath7 . in our case , there are extra dimensions to consider which are @xmath10 , @xmath11 , @xmath147 , and @xmath148 , resulting in a much higher computational demand . fortunately , some of these quantities are in perfect correlation with one another with linear proportionality . in addition , their initial values at @xmath149 completely compose the ensemble at any time through equation ( [ eq : background_sol ] ) . at the minimal level . as the correlation between @xmath150 ( @xmath11 ) and @xmath151 ( @xmath10 ) becomes tighter in time , the initial variation gets gradually diluted , which roughly justifies our restricting the parameter space only to @xmath9 and @xmath10 . ] , it would suffice to just consider variation of @xmath10 in addition to @xmath9 such that the local power spectrum is an explicit function of the two background quantities , or @xmath152 . for the initial condition for background patches , we generate 3d maps of @xmath10 , @xmath11 , @xmath147 , @xmath148 , @xmath88 , @xmath89 , and @xmath95 at @xmath80 on @xmath153 uniform grid cells inside a cubical volume of @xmath154 . we generate fluctuations of discrete modes that are randomized as @xmath155,\nonumber \\ { \rm im}(\delta_{{\bf k } } ) & = & g_{2}n^{3}\left(\frac{p(k)}{2v_{{\rm box}}}\right)^{1/2}{\rm sign}\left[{\tt tf}(\delta_{{\bf k}})\right],\label{eq : random}\end{aligned}\ ] ] for given @xmath7 , where @xmath156 and @xmath157 are random numbers drawn from mutually independent gaussian distributions with mean 0 and standard deviation 1 , @xmath158 stands for any kind of @xmath7-space fluctuations , and ` tf ` is the transfer function of @xmath158 , whose sign should be multiplied because some @xmath158 s oscillate around zero in @xmath7 . the configuration is roughly equivalent to applying a smoothing filter of length @xmath159 mpc . in practice , we use camb ( @xcite ) for @xmath160 , and use the continuity equations ( @xmath161 ) for @xmath162 , with the help of two camb transfer - function outputs at mutually nearby redshifts for time differentiation . @xmath163 is obtained from the relation @xmath164 . @xmath165 is fixed by following the scheme by @xcite : we require @xmath166 at the initial redshift in equation ( [ eq : master_temp ] ) , which results in @xmath167 where all quantities are evaluated at @xmath43 , especially with the help of equation ( [ eq : background_sol ] ) for @xmath168 and two adjacent camb transfer - function outputs for @xmath169 . finally , all these @xmath7-space fluctuations are fourier - transformed to obtain @xmath127-space fluctuations . 3d maps and 2d histograms of several initial quantities are presented in figure [ fig : map ] . fields of @xmath10 and @xmath46 on a part of a slice of the box at @xmath170 are shown in figure [ fig : map](a ) . as expected , the velocity field converges on overdense regions and diverges on underdense regions . we find that in most patches @xmath151 dominates over @xmath150 , and thus the map of @xmath88 looks very similar to figure [ fig : map](a ) . this occurs because baryons lag behind cdms due to their coupling to cmb . @xmath11 ( figure [ fig : map]b ) is coupled to @xmath95 ( figure [ fig : map]c ) more strongly than to @xmath10 . @xmath10 and @xmath147 ( similarly @xmath11 and @xmath148 ) are almost perfectly correlated ( figure [ fig : map]d ) . @xmath10 and @xmath11 are very loosely correlated due to the tight coupling of baryons to photons at the redshift ( figure [ fig : map]e ) , but the correlation becomes tighter in time . @xmath10 and @xmath9 are not correlated ( figure [ fig : map]f ) . because of this fact , the probability distribution function ( pdf ) @xmath171 is simply a multiplication of pdfs @xmath172 and @xmath173 , at the minimal level . due to gaussianity at @xmath174 , we have @xmath175,\nonumber \\ \mathcal{p}(k;\ , v_{{\rm bc } } ) & = & \sqrt{\frac{2}{\pi}}\frac{v_{{\rm bc}}^{2}}{\sigma_{v_{{\rm bc}}}^{3}}\exp\left[-\frac{v_{{\rm bc}}^{2}}{2\sigma_{v_{{\rm bc}}}^{2}}\right],\label{eq : pdf}\end{aligned}\ ] ] where @xmath176 and @xmath177 are the standard deviations of @xmath10 and @xmath46 projected onto one cartesian - coordinate axis , respectively . with our setup , we find that @xmath178 and @xmath179km / s at @xmath174 ( or the root - mean - square of @xmath180 is @xmath181km / s ) . at the minimal level of only allowing the variance in @xmath10 and @xmath9 , the average power spectrum @xmath182 will then be given by the ensemble average @xmath183 where the pdfs and integral arguments are the ones at @xmath174 . of course , a more accurate and straightforward way is to just ensemble - average @xmath184 over the patches from a large - box realization , because for example @xmath10 and @xmath11 are too poorly correlated at @xmath174 . we numerically integrate equation ( [ eq : perturbation_k ] ) to examine the evolution of small - scale ( high-@xmath7 ) fluctuations at any overdense ( underdense ) patch to the linear order , with the help of equations ( [ eq : background_sol ] ) and ( [ eq : deltat_fitting ] ) for the evolution of background quantities . in practice , we used the ode45 modules of matlab(2015b , the mathworks , inc . , natick , massachusetts , united states ) and of gnu octave , which use the 4th - order runge - kutta method , with the relative tolerance @xmath185 and the absolute tolerance @xmath186 . during the evolution , the number of integration steps is the highest for @xmath34 , because its amplitude changes from the initial , very small values around @xmath35 to final , much larger values close to @xmath15 . therefore , taking sub - steps for @xmath35 while coarser steps for other @xmath187 s is expected to boost the computational efficiency , even though we did not yet implement the method in our computation . the end result is then ensemble - averaged over varying @xmath10 and @xmath46 to obtain @xmath182 ( equation [ eq : pk_average ] ) . we first examine how the evolution of @xmath188 depends on the density environment , and compare the result to the prediction by th . figure [ fig : pk_growth ] shows the evolution of @xmath189 of three arbitrarily chosen wave numbers ( @xmath16=\{33 , 150 , 2000 } /mpc ) when @xmath190 in different density environments ( @xmath191=\{-0.01 , -0.005 , 0 , 0.005 , 0.01 } ) . note again that @xmath178 at @xmath80 , and thus these samples correspond to @xmath192 and @xmath193 . first , as expected , the growth of small - scale fluctuations are biased when @xmath194 and anti - biased when @xmath195 , with respect to the mean - density case ( prediction by th ) . secondly , when @xmath191 s are equal in amplitude but opposite in sign , the deviations of @xmath188 from @xmath196 reveal the same trend but only until @xmath197 when @xmath198 and @xmath199 when @xmath200 . afterwards , the bias and the anti - bias are not balanced by more than 1% and such off - balance keeps growing in time . the higher the @xmath201 is , the earlier this unbalance starts . thirdly , the fractional deviation from the mean - density case is almost universal regardless of the value of @xmath16 , and thus the timing of the unbalance is approximately a function only of @xmath202 . finally , in some high-@xmath134 patches , our linear analysis based on equation ( [ eq : perturbation_k ] ) without quadratic and higher - order terms in @xmath135 starts to break down at @xmath203 , because these modes enter the nonlinear regime ( @xmath204 ; see figure [ fig : pk_growth ] ) at this epoch . a similar breakdown of the formalism will occur for perturbations inside very low-@xmath134 patches as well . therefore , a higher - order scheme than our work is required when one is to predict the low - redshift evolution of @xmath135 s . on the other hand , if one were to generate initial conditions for numerical simulations in the linear regime , our formalism would provide the sufficient accuracy . how large - scale overdensity impacts the evolution of small - scale inhomogeneities is reflected in the density continuity equation . in overdense background patches , @xmath140 grows in time and @xmath205 . then , in equation ( [ eq : perturbation_k ] ) , @xmath206 and @xmath207 work as sources terms in addition to @xmath208 to the growth of @xmath138 . this will boost the growth rate of @xmath138 of both overdense ( @xmath209 , @xmath210 ) and underdense ( @xmath211 , @xmath212 ) modes . in contrast , in underdense background patches , these terms suppress the growth rate of @xmath138 of both overdense and underdense modes . the distribution of @xmath213 is gaussian , and therefore for each `` bias '' case with @xmath214 there exists an `` anti - bias '' case with @xmath215 . nevertheless , the unbalance described above is expected to boost the average power spectrum from that by th . the overall effect of including the gaussian distribution of @xmath10 is thus to mitigate the negative impact by the relative velocity , predicted by th , to some extent . in addition , the universality of the unbalance in @xmath7 boosts @xmath182 even in the @xmath7 range ( @xmath216 and @xmath217 ) where the power spectrum is almost unaffected by non - zero @xmath9 ( figure [ fig : pk_compare ] shown in terms of the @xmath7-space matter - density variance @xmath218 ) . note that the result for @xmath219 can not be trusted , because our perturbation theory is based on the condition that large - scale modes ( @xmath220 ) are well separated from small - scale modes in scale . discrepancy of @xmath182 including non - zero @xmath134 s from the prediction by th is negligible at @xmath221 , but later the discrepancy grows in time . of course , individual patches may experience discrepancy in @xmath222 much earlier than this epoch ( figure [ fig : pk_growth ] ) . understanding the abundance and the spatial distribution of cosmological halos is crucial in modern astrophysics and cosmology . in this section , we examine the halo abundance both in the local and the global sense , just as we did for the matter power spectrum . let us first revisit the calculation by th . they adopted the extended press - schechter formalism and calculated the local halo abundance , in terms of the conditional mass function , using the peak - background split scheme . a patch with @xmath134 ( let us use this notation for the matter overdensity , to avoid confusion with the @xmath7-space matter - density variance @xmath223 ) and @xmath9 will have the number of halos per unit eulerian comoving volume per @xmath224 given by @xmath225 , \label{eq : cond_mass_ftn}\end{aligned}\ ] ] where @xmath226 is the critical overdensity of spherical collapse , and @xmath227 is the variance of density field smoothed with the window function @xmath228 corresponding to mass @xmath224 , @xmath229 one should note that @xmath230 should be that of high @xmath16 modes only , or more accurately a reduced value @xmath231 where @xmath232 is the variance of density field smoothed with the window function corresponding to the mass of the patch ( e.g. @xcite ; @xcite ; @xcite ) . one can instead put a lower bound @xmath233 ( @xmath234^{1/3}$ ] ) in the integral of equation ( [ eq : sigma2_th ] ) , which would be identical to the reduced variance if a sharp @xmath16-space window function is used . the global mass function @xmath235 is simply an average of the local mass function , @xmath236 , over the ensemble of patches . when @xmath236 is averaged only over @xmath9 for a given @xmath134 , which is equivalent to visiting only those patches with the same @xmath134 and taking the average , it leads to the conditional mass function @xmath237 . this calculation should be modified , because @xmath227 depends also on @xmath134 through the dependence of @xmath238 on @xmath134 : @xmath239 which will enter equation ( [ eq : cond_mass_ftn ] ) . here we used the sharp @xmath7-space filter , and thus @xmath240^{1/3}$ ] . an overdense patch will then have a boost in @xmath236 from the value by th because @xmath241 and thus @xmath242 , and vice versa ( a decrease from the value by th ) for an underdense patch . obviously , both @xmath243 and @xmath235 will also be affected . it is important to compare our findings to the usual peak - background split scheme and the one by th . in the `` standard '' scheme , if the density field is purely gaussian , all the wave modes are assumed mutually independent in the linear regime . therefore , the local , high-@xmath7 modes have a universal as @xmath244 ) . ] variance @xmath245 whether or not they are placed inside a patch with non - zero @xmath134 . the way how the halo formation is biased in an overdense region is simply through the shift in the density ( @xmath246 ) . th then realized the fact that the variance is not universal but should depend on @xmath9 . because non - zero @xmath9 tends to suppress @xmath245 , the odds to cross @xmath226 decrease relative to the standard picture . we find that there is another dependency of the variance , which is @xmath134 . in other words , we find that there are two biasing effects in an overdense region compared to a mean - density region : getting closer to @xmath226 because of the shift in the density ( @xmath246 , also in the standard scheme ) , and having a larger degree of fluctuation in @xmath187 due to the mode - mode coupling ( e.g. source terms @xmath247 and @xmath248 in @xmath249 in equation [ eq : perturbation_k ] , which is a new finding ) . therefore , by not fully implementing the effect of non - zero @xmath134 , th in effect underestimates and overestimates the halo mass functions in overdense and underdense regions , respectively . we note that this additional bias effect should be present even in the standard picture with @xmath250 , because this is due to the natural coupling between the large - scale and small - scale density perturbations . in this case , however , we are not sure about the quantitative validity of the extended press - schechter formalism on the conditional mass function ( equation [ eq : cond_mass_ftn ] ) , which is based on the linear theory guaranteeing gaussianity at any filtering scales _ without _ the mode - mode couplings . qualitatively , we believe that the boost of local @xmath187 and @xmath245 under @xmath134 should boost the conditional mass function to the level estimated by equations ( [ eq : cond_mass_ftn ] ) and ( [ eq : sigma2_ahn ] ) as described above anyways . we defer a further investigation of this issue , which can be clarified with numerical simulations of the halo formation under different @xmath134 s . discrepancy between the conditional mass functions by this work and by th is significant if we focus on individual patches . figure [ fig : cond_mass_ftn ] illustrates how our prediction differs from that by th . for example , at @xmath251 , under @xmath252 we predict [ 100 - 2000 ] % boost in @xmath243 compared to the values by th ( let us denote them by @xmath253 ) . for @xmath254 , we predict 90 % or more decrease in @xmath243 compared to the values by th . this is the obvious result of the mode - mode coupling of @xmath134 and @xmath135 described above . it is also noteworthy that the discrepancy is the largest for the rarest halos : first , at any redshift , the discrepancy increases as the halo mass increases and secondly , for any given halo mass , the discrepancy decreases in time . the discrepancy among the conditional mass functions by this work , by th and by the standard picture also influences @xmath235 . let us just take the example of @xmath255 ( figure [ fig : global_mass_ftn ] ) . compared to the standard prediction ( @xmath256 ) , conditional mass functions by th stay lower regardless of @xmath134 . this is due to the suppression of structure formation by the relative velocity . in contrast , @xmath243 in this work is either higher or lower than the standard prediction ( @xmath257 ) depending on the halo mass and @xmath134 . at @xmath255 , for @xmath254 @xmath258 and for @xmath252 @xmath259 for any halo mass . the tendency for @xmath243 to overshoot @xmath260 when @xmath261 is not generic , because the influence of the positive @xmath134 s on the structure formation appears only late in its evolution ( see figs . [ fig : pk_growth ] and [ fig : pk_compare ] ) . at any rate , the contrast in @xmath243 among overdense and underdense regions is increased compared to th and the standard picture , and the net effect e.g. at @xmath255 is to boost @xmath235 from @xmath262 . at much higher redshifts , our @xmath235 is almost indistinguishable from that by th , which undershoots @xmath262 . we note that @xmath243 and @xmath235 have the usual problem of not correctly predicting the actual mass function , if one sticks to the original extended press - schechter formalism . minihalos at high redshifts are usually underestimated by the extended press - schechter formalism . the usual peak - background split method suffers from large discrepancies between its prediction and the n - body simulation results for rare halos in general . in this case , a hybrid method to connect the peak - background - split halo bias parameter to the better - fitting mean mass function types ( e.g. @xcite ; @xcite ) is much more appropriate . we will apply this method in the future for a better estimation of the conditional mass function and the @xmath263-space halo bias parameter ( section [ sub : halo - bias ] ) . the conditional mass function we examined in section [ sub : mass_function ] is an indicator of how halo formation is biased toward overdense regions . the halo bias can be viewed also in the the @xmath7-space . this is a crucial parameter in cosmology when trying to probe the fluctuation of the matter density from surveys of galaxies through , for example , the power spectrum analysis . the halo bias parameter in @xmath7-space is defined as @xmath264 where @xmath265 is the power spectrum of halo over - abundance @xmath266 @xmath267 can then be fourier - transformed in order to calculate @xmath265 . we expect @xmath268 to be larger than that predicted by th , because the `` contrast '' in @xmath243 between overdense and underdense regions has increased from that by th ( section [ sub : mass_function ] ; figure [ fig : cond_mass_ftn ] ) . this is clearly seen in figure [ fig : bias ] , where we show @xmath268 s computed in this work and by th . at @xmath255 and for @xmath269 , we find that @xmath268 is about @xmath270 $ ] times as large as the one by th . this is again caused by the mode - mode coupling . at @xmath255 . comparison is made between this work ( black , x ) and th ( red , square ) in the top sub - panel . in the bottom sub - panel , we show the fractional difference @xmath271/b(k)_{{\rm th}}$].,scaledwidth=40.0% ] as was pointed out by th , @xmath268 oscillates in @xmath16 due to the baryonic acoustic oscillations ( bao ) at low @xmath16 s . we will increase the size of the box in future work for a better estimation . ] , which is tied to the modulation of the streaming velocity , the density fluctuation , and the baryon fraction in the existence of the compensated mode @xcite . this makes it difficult to deduce @xmath272 from @xmath265 , not to mention from the galaxy surveys where galaxy formation mechanism , strongly influenced by baryonic physics , is another nuisance parameter ( but see @xcite for how to separate out the streaming velocity effect ) . for the halo mass range treated in this paper , @xmath268 should modulate the distribution of the first stars which grow predominantly inside minihalos . as was also noted by th , the difference in @xmath268 should also influence the formation of much larger - mass halos , which are used for galaxy surveys . it is also possible that the nonlinear effect described by @xcite , or the heating of the intergalactic medium ( igm ) due to the velocity difference between cdm and baryons , is modulated in space depending on the overdensity . then the power spectrum in the 21-cm background , which may be dominated by the velocity fields if the heating is efficient ( @xcite ) , is likely to be boosted . because such a power spectrum shows a very clear bao feature and the 21-cm observation usually suffers from the low sensitivity , the signal boosted even more from the prediction by @xcite will be a very promising target for the high - redshift 21-cm cosmology . let us briefly discuss the halo stochasticity . the halo stochasticity is defined as @xmath273 where @xmath274 is the cross power spectrum between the halo density and the matter density . both in th and in this work , stochasticity is caused by the fluctuation in @xmath243 because patches with the same @xmath134 can have different @xmath9 s which affect @xmath243 . one should note that the fluctuation should be caused also by the sampling variance . the conditional mass functions usually show super - poissonian distributions in @xmath243 even in the standard picture with @xmath250 ( e.g. @xcite ; @xcite ; @xcite ; @xcite ) , and obviously this should cause the stochasticity in addition to that by the varying @xmath9 . because we use the `` mean '' conditional mass function just as th did , in both works @xmath275 does not reflect the sampling variance . including this effect requires the calculation of the sub - cell correlation function ( @xcite ) , which we delay to future work . we investigated the impact of the relative velocity ( streaming velocity ) between cdm and baryons on the small - scale structure formation . th first studied this effect by adopting a trivial solution to the large - scale velocity and density fields . because velocity fields are correlated with density fields , however , such a trivial solution can not accurately describe the physics in regions with non - zero overdensity . we thus improved on the work by th by implementing a non - trivial solution to the large - scale velocity and density fields , and we find that this causes a new type of coupling between large - scale and small - scale modes . this results in boosting the small - scale structure formation in overdense regions and suppressing that in underdense regions , aside from the suppression originating from the streaming velocity . the net effect on the structure formation is to boost the overall fluctuation , in terms of @xmath272 , and thus the `` negative '' effect by th is mitigated to some extent . depending on the wave mode ( @xmath7 ) and the observing redshift , @xmath272 can even be larger than that in the standard picture with @xmath250 . the conditional halo mass function and the halo bias are also affected in similar ways . the results of this work show that the formation and evolution of small - scale structures depend strongly on not only the streaming velocity but also the density environment . the most important aspect of our work is that in contrast to th , who predict that regardless of the underlying density the local matter power spectrum of small - scale structures will be identical as long as @xmath9 is the same , the underlying large - scale ( @xmath276 a few mpc ) overdensity is another key parameter in addition to @xmath9 . this then requires re - examining previous work based on the formalism by th . we already showed that @xmath277 , @xmath272 , @xmath243 , @xmath235 , and @xmath268 are affected . if one were to simulate the nonlinear evolution of density perturbations and the structure formation in @xmath278mpc patches , he should generate initial conditions based on this work . as is seen in figure [ fig : pk_growth ] , we can not neglect the impact of overdensity even when the simulation starts at e.g. @xmath279 , because at @xmath280 the discrepancy between our prediction and that by th is already a few percent at that redshift . both the previous semi - analytical work ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) and the semi - numerical work ( e.g. @xcite ; @xcite ) should be re - examined . attempts to numerically simulate the nonlinear evolution of small - scale structures have been mostly limited to the physics inside mean - density regions ( @xcite ; @xcite ; @xcite ; @xcite ) or special , isolated regions ( @xcite ) . these numerical simulations thus need to be extended to incorporate @xmath134 which varies in space . in doing so , a reasonable method would be to use adaptive mesh refinement ( amr ) codes with nested grids , so that one or a few interesting regions ( @xmath278mpc patches with @xmath281 , for example ) are treated with fine meshes and other regions with coarse meshes for computational efficiency . we also showed that cosmology through galaxy surveys should carefully consider the impact of the mode - mode couping , because the halo bias ( and galaxy bias as well ) would be boosted from not only the standard prediction with @xmath250 but also the prediction by th . cosmology with the intensity mapping may also be affected . the post - reionization intensity mapping targets the large - angle , diffuse 21-cm background from neutral hydrogen atoms inside galaxies ( @xcite ; @xcite ; @xcite ; @xcite ) . because any galaxies , small or large , contribute to this cumulative 21-cm background , such observations will be affected by the streaming velocity through @xmath268 . the pre - reionization intensity mapping targets the large - angle , diffuse 21-cm background from the intergalactic neutral hydrogen atoms ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; ; @xcite ; @xcite ) . even though this is free from the galaxy bias , the streaming velocity may act as a heating mechanism and boost the power spectrum of the velocity field ( @xcite ) , and therefore the new findings of our work should be incorporated . application to the study of the cosmic reionization process is of a prime interest in terms of the high - redshift astrophysics . the complex nature of the process usually requires numerical simulations , and they are performed through either efficient semi - numerical methods ( @xcite ; @xcite ; @xcite ; @xcite ) or fully numerical methods ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . the early phase of cosmic reionization must have been driven by the first stars , possibly forming first in minihalos , as these are the first luminous objects in the universe . a very important factor that modulates the formation of the first stars inside minihalos is the lyman - werner intensity , which have been properly treated in simulations in a box large enough for statistical reliability but implementing subgrid physics for the first star formation inside minihalos ( @xcite ; @xcite ) . to predict reionization scenarios to our best knowledge , especially on its early phase , this work should be properly incorporated because the star formation inside minihalos is strongly modulated by the streaming velocity as well . the results of this paper have rooms for further improvements . this paper is based on the presumption that the fluctuations at a few mpc scale remain linear even at later epochs . however , high density regions will reach the nonlinear regime earlier than the rest , and then our formalism will break down in such regions . the halo bias is more strongly pronounced in the nonlinear patches ( e.g. @xcite ) than in the linear theory , and thus one should use the actual values of the overdensity in such circumstances . one could achieve this goal by adopting the quasi nonlinear calculation ( e.g. 2lpt by @xcite ) , adopting the top - hat collapse model as in @xcite and @xcite , or for the best accuracy running n - body+hydro simulations which resolve the density fluctuation at @xmath276 mpc scale . then , in each patch of a few comoving mpc , equation ( [ eq : perturbation_k ] ) can be integrated with the newly computed values of @xmath134 s . wave modes in the range @xmath282/$]mpc are not accurately treated , because we based our formalism on the separability of the large - scale modes ( @xmath283mpc ) and the small - scale modes ( @xmath284mpc ) . the code we used will be released for the public use , but it requires technical improvements such as allowing parallel computation and porting to more generic computation languages . we will maintain and control its development through the website http://www.chosun.ac.kr/kjahn . we thank p. r. shapiro , f. schmidt and r. barkana for helpful discussions . we also thank the anonymous referee for the clear report which led to a significant , quantitative improvement of the paper . this work was supported by a research grant from chosun university ( 2016 ) and by nrf-2014r1a1a2059811 . when the fluctuation of radiation components are neglected , the growth of large - scale density and velocity fluctuations are well approximated by equations ( [ eq : background ] ) and ( [ eq : background_easy ] ) . their evolution can then be described by the 4 normal modes described in section [ sub : perturbation ] . the growing and decaying modes are the two solutions to the second - order equation @xmath285 where @xmath286 , and can be written also as @xmath287 similarly , the compensated and streaming modes are the two solutions to @xmath288 where @xmath289 , and can be written also as @xmath290 we now take a convention of writing each mode as the product of its initial value at @xmath80 and its growth factor : @xmath291 , @xmath292 , @xmath293 , and @xmath294 , denoting the growing , decaying , compensated , and streaming modes , respectively . these modes comprise @xmath295 and @xmath296 , as @xmath297 and @xmath298 , with the normalization @xmath299 at @xmath80 . the first task in finding these modes is to calculate the growth factors . because the hubble constant @xmath300 ( @xmath301 and @xmath302 have non - negligible radiation components during the period of interest , @xmath303 , growth factors should be calculated numerically . we factor out deviations from the analytical form valid during the matter - dominated ( @xmath57 ) era , as @xmath304 , @xmath305 , and @xmath306 , and then solve for the order - of - unity values of @xmath307 , @xmath308 , and @xmath309 . they are determined by @xmath310 @xmath311 and @xmath312 in practice , the numerical integration of equations ( [ eq : fg])-([eq : fs ] ) is started at @xmath313 ( backward for @xmath314 and forward for @xmath315 ) , with the condition that @xmath316 at @xmath313 because it is a matter - dominated epoch , and @xmath317 at @xmath80 because of our normalization convention . one can instead start the integration from the radiation dominated epoch ( just numerically assuming that equations ( [ eq : fg])-([eq : fs ] ) are all valid at @xmath318 and using the asymptotes for @xmath319 and @xmath320 at that time ) , but we find that equations ( [ eq : fd ] ) and ( [ eq : fs ] ) become stiff if integrated forward in increasing @xmath3 at high @xmath321 . we show the growth factors found this way in figure [ fig : growth ] . ( left panel ) is multiplied to each analytic power - law evolution valid for the matter - dominated era , it results in the actual growth factor @xmath322 ( right panel).,title="fig:",scaledwidth=45.0% ] ( left panel ) is multiplied to each analytic power - law evolution valid for the matter - dominated era , it results in the actual growth factor @xmath322 ( right panel).,title="fig:",scaledwidth=45.0% ] now we can find the initial values of these modes by using the growth factors found above on the transfer function outputs from camb . we algebraically relate two redshift outputs , at @xmath323 and @xmath324 in practice , to find these modes : @xmath325 where @xmath326 are those from camb . the modes found this way are shown in figure [ fig : modes ] . a few things are notable . an unperturbed hubble flow requires @xmath327 , while we find that at @xmath80 @xmath328 and is not constant over @xmath16 . even though one can choose different redshifts to extract these modes and they should not change in principle , the resulting modes at @xmath174 vary depeding on the choice of the redshifts . we believe that this is partly due to our neglect of the fluctuations of radiation components , because they can affect the evolution of density fluctuations . we find that our current choice of @xmath329 and @xmath330 is optimal for @xmath331 : using these modes and evolving them with the growth factors , we find a good match between @xmath332 , @xmath333 `` constructed '' by using these modes and those calculated by camb , at any redshifts with at most a several percent error . we show their comparison in figure [ fig : constructed_vs_camb ] . @xcite follows a similar approach for the mode extraction but uses camb transfer function outputs at @xmath334 . their focus is on the low - redshift galaxy surveys , and thus the accuracy is required mostly at and near the present . in our case , accuracy in @xmath140 and @xmath141 is required mostly at a high redshift range , @xmath335 , because the small - scale modes are continuously affected by large - scale modes from the epoch of recombination and the linear perturbation analysis on small - scales modes breaks down later when they become nonlinear . , plotted in varying line widths . the negative values are plotted by a dashed line for the decaying mode and a dotted line for the streaming mode , after flipping the sign.,scaledwidth=40.0% ] , some error is inherent as seen in the lower panels . the difference is with respect to the values calculated by camb . we allow this degree of error in this paper , but for higher accuracy one is advised to use the camb outputs for the large - scale fluctuations.,title="fig:",scaledwidth=33.3% ] , some error is inherent as seen in the lower panels . the difference is with respect to the values calculated by camb . we allow this degree of error in this paper , but for higher accuracy one is advised to use the camb outputs for the large - scale fluctuations.,title="fig:",scaledwidth=33.0% ] , some error is inherent as seen in the lower panels . the difference is with respect to the values calculated by camb . we allow this degree of error in this paper , but for higher accuracy one is advised to use the camb outputs for the large - scale fluctuations.,title="fig:",scaledwidth=33.0% ] for the evolution of @xmath95 , we integrate equation ( [ eq : deltatemp ] ) on each spatial patch . early on , its value is strongly affected by the initial fluctuation of the cmb temperature . later , it decouples from the cmb fluctuation and is determined mostly by @xmath336 . therefore , one expects a strong correlation between @xmath95 and @xmath336 long after recombination . for example , at @xmath337 and @xmath338 , they follow the linear relation : @xmath339 at @xmath337 , and @xmath340 at @xmath341 . regardless of the variance in @xmath95 , therefore , one can find a fitting formula for @xmath100 after its decoupling from the cmb temperature . the fitting formula is given by equation ( [ eq : deltat_fitting ] ) . we note that equation ( [ eq : deltat_fitting ] ) is valid only when the patch size is 4 comoving mpc . for patches in different size , we believe that a generic form of a two - parameter fit , @xmath342,\nonumber \\ \alpha&=&\frac{\log_{10}\left(\delta_{t,\ , b}/\delta_{t,\ , a}\right)}{(\beta-1)^{\gamma}-(\beta-2)^{\gamma } } , \nonumber \\ y&=&\frac { ( \beta-2)^{\gamma}\log_{10}|\delta_{t,\ , b}| -(\beta-1)^{\gamma}\log_{10}|\delta_{t,\ , a}| } { ( \beta-1)^{\gamma}-(\beta-2)^{\gamma } } , \label{eq : deltat_fit_general}\end{aligned}\ ] ] will serve as a good fit regardless of the patch size . the linear relation between @xmath95 and @xmath336 at @xmath337 and @xmath338 can be found by comparing the two quantities . in case of the 4 comoving mpc patch , we find that @xmath343 and @xmath344 ( equation [ eq : deltat_fitting ] ) provides an excellent fit to @xmath95 when @xmath345 , with the linear relations @xmath346 at @xmath337 and @xmath347 at @xmath341 this is demonstrated in figure [ fig : dtfit ] , where the evolution of different @xmath95 s are shown depending on the initial @xmath348 , together with the corresponding fits . of three arbitrarily chosen spatial patches ( 4 comoving mpc ) with varying @xmath349 ( thick ; solid , dashed , dot - dashed ) . overlaid are the corresponding fitting functions given by equation [ eq : deltat_fitting ] , in thin dotted lines . , scaledwidth=40.0% ]

a scheme for incorporating the creation of radiation and matter into the cosmological evolution is introduced so that it becomes possible to merge the times before and after the creation of radiation and matter in a single scale factor in robertson - walker metric . this scheme is illustrated through a toy model that has the prospect of constituting a basis for a realistic model .

1407.7199

the question of determining the model that best describes the universe is the ultimate goal of cosmology . the energy - momentum content of the present universe seems to be a perfect fluid mainly consisting of a dark sector ( possibly consisting of a dark energy and a dark matter component ) , baryonic matter , and radiation @xcite . in the standard model of cosmology ( namely , @xmath0cdm ) dark matter @xcite and baryonic matter are considered to be dust , dark energy @xcite is taken to be the einstein s cosmological constant , and radiation is described by the usual energy - momentum term for radiation . although the standard model seems to be compatible with observations yet it has some problems . the magnitudes of potential theoretical contributions to cosmological constant ( cc ) are extremely higher than the value of cc deduced from the energy density of the universe @xcite . there are many attempts to solve this problem , namely , the cc problem . nevertheless none is wholly satisfactory . the best option seems to employ a symmetry such as metric reversal symmetry @xcite to cancel cc and then attribute the dark energy to something else e.g. to modified gravity @xcite , or to some scalar field such as quintessence @xcite . cold dark matter ( i.e. dust - like dark matter with no or negligible interaction with itself and with baryonic matter and photons ) scenario of @xmath0cdm as well suffers from some problems such as rotation curves of spiral - like galaxies i.e. cuspy halo problem , missing satellite galaxies problem @xcite . there are many alternatives to cold dark matter ( cdm ) scenario including warm - dark matter @xcite , bose - einstein condensate dark matter @xcite , and scalar field dark matter @xcite . the above considerations essentially hold for the time from the radiation dominated era till the present era . the standard paradigm for the era before the radiation dominated era is an inflationary era ( that serves to solve the problems of the standard cosmology such as horizon , flatness , absence of monopoles problems ) @xcite . usually the inflationary era and the epoch after this era are studied separately . this is not only due to the need to concentrate on each of these and to try to understand each epoch better before a possible unification . in fact the most serious problem in the direction of the unification of the whole cosmic history is the difficulty of merging these two epochs because of the form of the dependence of the energy density of dust and radiation on scale factor ( i.e. on redshift ) . in @xmath0cdm the energy density of radiation dominates over that of inflaton if one goes back to sufficiently large redshifts . this is due to the fact that the energy density of inflaton is essentially constant during inflationary era while the energy density of radiation scales like @xmath1 where @xmath2 is the scale factor . in other words , to have a true unification , the creation of radiation and matter after the inflationary era must be taken into account in the scale factor without destroying the standard cosmology before and after the inflation , and this is not an easy task . the models in literature that unify all eras of cosmological evolution in a single model @xcite are not wholly realistic since they do not include baryonic matter although they are able to produce eras of cosmological evolution with correct equations of state in the corresponding eras , and some have graceful exit from inflationary era . the matter in these models must be identified with dark matter since the energy densities of these models do not contain energy components that scale proportional to @xmath3 for all times ( or at least for a sufficiently long time ) . the models in @xcite use the energy densities expressed in terms of simple functions of hubble parameter and/or scale parameter as the starting point rather than starting from the scale factor . although one may , in principle , determine the scale factor from this information the form of scale factor may be rather complicated in some cases . on the other hand a relatively simple scale factor may result in a rather complicated and unmanageable functional form for the energy density when expressed in terms of the scale factor or the hubble parameter . therefore in some cases it may be more suitable to consider a specific ansatz for scale factor such as in this study and in @xcite . the same approach is adopted in this study . moreover the present study introduces a general prescription to include dust and radiation into unification . in this study , first , in section ii , i introduce a scheme to unify the cosmological evolution before and after the radiation dominated era . then i give a concrete realization of this scheme in section iii . in section iv i discuss the observational compatibility of this scheme in the context of the model introduced in section iii . finally i conclude in section v. the scale factor in this model is a sum of two terms . the first term is a pure dark energy contribution . the second term is responsible for the baryonic matter and radiation terms and additional terms that may be mainly identified with dark matter . there is also an additional term due to coupling between these terms , and this term gives another contribution to the dark energy and dark matter . some of the ideas employed here have been already studied in literature . in this study i do not make a sharp distinction between dark energy and dark matter because the dark energy and dark matter terms are coupled and the equation of state ( eos ) of some terms e.g. eos of the coupling term between dark matter and dark energy terms evolve with time . the superficiality of a distinction between dark energy and dark matter is considered in many studies in literature , either explicitly or implicitly @xcite . this option is quite possible since dark energy and dark matter are not observed directly . what we see observationally is a missing element in the energy - momentum tensor of the einstein equations , other than baryonic matter and radiation , and this missing quantity may be described by two components ; dark energy and dark matter . it is , in principle , equally possible that this quantity is composed of a single component , say , dark fluid . in @xcite i had introduced a universe composed of a dark fluid ( that may be written in terms of two scalar fields ) . in fact the scale factor in that study is essentially @xmath4 in eq.([a2 ] ) of this paper . the present study , somewhat , may be considered as an extension of @xcite where baryonic matter and radiation are included . however there are important differences as well . the main aim of this study is to introduce a scheme to merge the cosmological evolution of the time before and after the production of radiation into a single scale factor with the baryonic matter and the usual radiation terms included . the modified form of @xmath4 in @xcite only serves as a realization of this scheme . furthermore i do not discuss the scalar field identification of the energy density due the part of the scale factor similar to @xmath4 of @xcite ( although it can be easily done ) , and i do not consider the cosmological perturbations of these quantities , and the inflationary era in this study because these points would cause divergence of the main goal of the paper and would increase the volume of this study drastically . i leave these points to future studies . consider the robertson - walker metric @xmath5 i take the 3-dimensional space be flat , i.e. @xmath6 for the sake of simplicity , which is an assumption consistent with cosmological observations @xcite . i let the form of the scale factor be @xmath7 where @xmath8 denotes the present time . we will see that @xmath4 is the part of the scale factor responsible for dark energy and dark matter , and @xmath9 is the one mainly responsible for dust and radiation and additional contribution to dark matter - energy , and we shall see later that a mixing between the sectors due to @xmath10 and @xmath11 act as an additional source of dark energy . we assume that @xmath4 and @xmath9 are chosen in such a way that @xmath12 for all @xmath13 . in general one may identify the dust by a mixture of baryonic matter and dust - like dark matter . the best fit values that i could find by trial and error for the specific toy model considered in this study for implementation of the present scheme seem to prefer the case where the dust term is wholly or almost wholly due to baryonic matter . we first focus on the @xmath9 term and specify it as @xmath14 where @xmath15 is some function that its form will be specified later . ( [ a2 ] ) and ( [ a3 ] ) may be used to relate @xmath16 and @xmath4 , @xmath9 in a more applicable way , and to derive the corresponding hubble parameter . we observe that @xmath17 in a similar way the hubble parameter is found to be @xmath18 where we have used @xmath19 note that @xmath20 by convention . we let @xmath21 where @xmath22 , @xmath23 are some constant coefficients , and @xmath24\,,~ \tilde{h}_2^{(2 ) } \,=\,\alpha_{o2}\left[\frac{\alpha_b}{a^{\frac{3}{2}}}+ \frac{\alpha_r}{a^2}-\frac{\alpha_k}{a}\right]\,,~ \tilde{h}_2^{(3 ) } \,=\,\alpha_c\left(\frac{1}{a^3}-\frac{1}{a}\right ) \nonumber \\ \label{a10a } \end{aligned}\ ] ] where @xmath25 , @xmath26 , @xmath27 , @xmath28 , @xmath29 , @xmath30 are some other constant coefficients . in fact , in ( [ a3a ] ) we could take the simpler form where @xmath31 , @xmath32 , @xmath33 , @xmath34 . this would be enough as long as we are concerned only with merging of the eras before and after the radiation domination , and the resulting model would be compatible with union2 data set at an order of magnitude level . the more involved form in ( [ a3a ] ) is used to make the model phenomenologically more viable . this point will be discussed when we discuss the phenomenological viability of the model in section iv . one may determine @xmath35 in eq.([a4b ] ) by using eq.([a3a ] ) , @xmath36\ ] ] where @xmath37 \label{c2 } \\ k(t)&=&\frac{a_2}{c_1-c_2}x_3\left[c_2\alpha_{o2}\alpha_kx_2- \alpha_c\left(c_1x_1-c_2x_2\right)\right ] \label{c3}\end{aligned}\ ] ] hence one may express ( [ a4b ] ) as @xmath38 where @xmath39 we let @xmath40 where @xmath41 , @xmath42 , @xmath43 , @xmath44 , @xmath45 . because the three dimensional part of metric is taken to be flat the present energy density is equal to the critical energy density the above equations imply that @xmath46 note that , at this point @xmath47 , @xmath48 , @xmath49 , @xmath50 , @xmath51 can not be identified as density parameters since density parameters should satisfy @xmath52+@xmath53+@xmath54+@xmath55+@xmath56=1 . in chapter iv we will see that this condition is not satisfied for the phenomenologically viable sets of parameters , so @xmath47 , @xmath50 , @xmath51 can not be identified as density parameters separately , instead one must define the total density parameter for dark sector by @xmath57=@xmath58 rather than the separate contribution due to @xmath59 and @xmath60 while i identify @xmath48 , @xmath49 as the density parameters corresponding dust and radiation . therefore to retain the physical content of this paper more evident i will not make a distinction between @xmath48 , @xmath49 and the density parameters for baryonic matter , radiation ; @xmath53 , @xmath54 while i keep this distinction for the others i.e. for the ones due to @xmath59 and @xmath60 terms . the @xmath61 and @xmath62 terms result in energy densities that are identified as the energy densities for baryonic matter and radiation . in principle , there may be also contributions due to the @xmath63 and @xmath64 . the sign of the @xmath63 term is negative of the usual stiff matter . it may be identified as stiff matter under pressure so that it has a negative deceleration parameter . the main function of this term is to damp the energy densities of baryonic matter and radiation in the time before the radiation dominated era . the function of the @xmath65 term is similar . it ensures the behavior of the energy density in late times be well - behaved ( i.e. preventing the energy density to grow too fast ( through the @xmath65 term in @xmath66 and @xmath67 ) ) . although the @xmath64 term is similar to that of a negative curvature 3-space it is different from such a term since its origin is the hubble parameter @xmath68 while a usual 3-curvature term arises from the 3-curvature part of metric . note that this term arises even in a flat 3-space in this construction . therefore i identify the @xmath69 and @xmath70 terms in @xmath68 as additional contributions to dark sector . another point worth to mention is ; it is evident that the square of ( [ a4b ] ) ( in conjunction with ( [ a10a ] ) ) results in an @xmath71 term containing @xmath72 and @xmath73 terms that may be identified with the standard baryonic matter and radiation terms , respectively if @xmath74 is taken to be constant while it depends on time in this scheme as it is evident from ( [ a3a ] ) . in fact variation of @xmath74 with time makes it possible to go to zero before the radiation dominated era as desired . therefore , given the considerable success of the standard model at least in the observed relatively small redshifts , the variation in @xmath74 after the matter - radiation decoupling time should be small so that this scheme mimics the standard model at relatively small redshifts where observational data is available . if one takes @xmath75 sufficiently small one may guarantee an almost constant value for @xmath74 for a sufficiently long time ( e.g. from the present time till the beginning of the radiation dominated era ) . we will see in section iv that there exist such values of @xmath74 with reasonable phenomenological viability . another term arising from @xmath76 is the cross term , @xmath77 . this term may be identified as the energy density term due to the transitory time where massive particles that act as radiation at high energies turn into more dust - like at intermediate energies . another term in @xmath78 is @xmath79 . this term will be considered as a pure dark sector term . finally the cross term @xmath80 gives an additional contribution to the dark sector for the phenomenologically viable values of the parameters . it may be easily shown that this term does not necessarily imply strong interaction between the dark fluid and radiation and baryonic matter as its form may suggest if the parameters of the underlying physics at microscopic scale satisfy some restrictions . otherwise one may use screening mechanisms such as @xcite to explain the unobservablity of dark matter - energy . next we derive the general form of the equation of state for this model . we derive the explicit form of the equation of state after ( eos ) after we give the explicit form of @xmath4 in the section . however giving the general form of eos in this scheme provides us a more model independent formula and may be useful for other choices of @xmath4 in future . after using eqs.([a3ac],[a3da],[a3db],[a3dc ] ) one obtains eos , @xmath81 as @xmath82 the terms inside the first parenthesis in the second line correspond to the contribution of the dark sector term @xmath59 . the other terms in the same line correspond to the contributions of dust and radiation and their coupling with dark sector term @xmath59 . the remaining terms are the term corresponding to variation of @xmath74 , the term corresponding to coupling of curvature - like term and the stiff matter under negative pressure with dust and radiation , the term corresponding to coupling of curvature - like term and the stiff matter under negative pressure with @xmath59 , the term corresponding to coupling of curvature - like term and the stiff matter under negative pressure with the other terms , and the contribution of the curvature - like term and the stiff matter under negative pressure , respectively . it is evident from ( [ a11a ] ) that the pressure for baryonic matter is zero as should be , and the pressure for radiation is @xmath83 as expected . a point worth to mention at this point is ; the coupling term between baryonic matter and radiation in eq.([a11a ] ) has an equation of state @xmath84 ( that may be seen by considering the ratio of the @xmath85 in @xmath86 by the corresponding term in @xmath87 i.e. @xmath88 ) . the redshift dependence of this term is between that of baryonic matter and radiation . this time dependence is more natural than the standard picture where there is no such term . massive particles at high energies act as radiation and at lower turns into dust . the coupling term accounts for the transitory time when massive particles pass from radiation to dust state . in order to obtain the evolution of @xmath81 as a function redshift or time explicitly , @xmath59 must be specified . this will be done in the next section . however i give a @xmath81 versus redshift graph in fig.[fig:1 ] for @xmath89 introduced in the next section for a phenomenologically viable set of parameters ( i.e. those with small @xmath90 values and with energy densities for recombination and nucleosynthesis as discussed in section iv ) to have an idea about the evolution of @xmath81 with redshift . to draw this graph i have converted time , t to redshift , z ( for union2.1 data ) through the relation @xmath91 , and then used mathematica to use this relation to make the calculations ( although the original quantities are expressed in terms of time ) . this procedure is applicable for small redshifts . however , in general , it becomes inapplicable due to highly non - linear form of scale factor and hubble parameter since it requires huge ram and cpu for computation , if it can be done at all , and hence requires a separate computational physics project by itself . therefore i have used equation of state versus and energy density versus time graphs ( instead of redshift ) in section iv . in fact , even that option required a long time of order of months to make the necessary computations . now we focus on the @xmath4 term . we take @xmath92^r\,\exp{[-b_1(b_2t)^{-1/s } ] } \label{a13b } \end{aligned}\ ] ] where @xmath93 , @xmath94 , @xmath95 , @xmath96 , @xmath97 are some constants that to be fixed or bounded by consistency arguments or cosmological observations . this scale factor is a generalization of the scale factor in @xcite where @xmath98 , @xmath99 . a similar scale factor is considered in @xcite as well . one of the shortcomings of @xcite is that the present value of the equation of state parameter in that model ( for phenomenologically relevant choices of parameters where the model mimics @xmath0cdm ) is @xmath100 while observations imply that it should be @xmath101 @xmath102 @xcite . in the present study there is an additional contribution due to mixing of the terms due to @xmath10 and @xmath11 and hence there is less need to modify the scale factor in @xcite . however i prefer to adopt the more general form in ( [ a13b ] ) to seek a greater parameter space and to insure the correct equation of state parameter . after determining the @xmath59 we are almost ready to find the explicit values of the energy density and the equation of state . the only missing element for calculation of these quantities is to find @xmath74 , @xmath113 , @xmath114 in ( [ a3ac ] , [ a3da ] , [ a3db ] , [ a3dc ] ) . another point to be addressed is to show that there exist sets of a whose variation with time are small for low redshifts so that the terms that are proportional to @xmath115 and @xmath116 in @xmath117 term may be identified with dust and usual radiation terms , respectively . in order to determine @xmath74 , @xmath113 , @xmath114 ( and to determine the rate of variation of @xmath74 with time ) one should derive an approximation scheme for the evaluation of these quantities because these quantities depend on @xmath118 , @xmath66 , @xmath67 ( that are defined in ( [ a3b ] ) and ( [ a3c ] ) ) , and these quantities , in turn , are defined in a recursive way since @xmath119=@xmath120 ( @xmath121 ) and @xmath122 depend on @xmath16 , and @xmath16 , in turn , depends on @xmath119 through eq.([a4a ] ) . in other words , in order to determine the approximate values of @xmath119 one must identify the zeroth order approximation and a method how to obtain the higher order approximations in an iterative way . one may use the following observations to obtain the zeroth order approximation ; @xmath123 and @xmath124 @xmath125 @xmath126 ( @xmath121 ) i.e. @xmath127 , @xmath128 for small redshifts . this implies that the zeroth order approximation for the scale factor @xmath16 should be taken as @xmath129 hence for phenomenologically viable cases ( where @xmath124 for small redshifts ) one may take the zeroth order approximations as @xmath130 where @xmath131 , @xmath132 , @xmath133 is obtained from @xmath134 , @xmath135 , @xmath136 by replacing @xmath16 by @xmath137 in those expression , for example , @xmath138 then @xmath139 one may get the next order approximation by using @xmath140 the next order quantities @xmath141 , @xmath142 may be obtained from ( [ d1 ] ) and ( [ d2 ] ) by replacing the superindices @xmath143 by @xmath144 where @xmath145 here @xmath146 is obtained from @xmath122 by replacing @xmath16 by @xmath147 . for kth approximation we replace @xmath16 by @xmath148 . in principle this may be done up to arbitrarily higher order approximations but it is quite difficult to calculate even @xmath141 even with the help of computers . in fact i have divided the interval @xmath149 in to coarser subintervals to decrease the cpu time and have used the approximate numerical values in the ith interval ( by assuming @xmath150 to be almost constant in those intervals ) by using the formula @xmath151 to find @xmath141 . i have seen ( by trial and error ) that it is possible to find almost constant @xmath150 and @xmath141 values for many relevant ( i.e. of small @xmath90 values considered in the next section ) choices of parameters , @xmath27 , @xmath152 , @xmath153 , @xmath154 , @xmath154 , @xmath155 , @xmath156 , @xmath22 , @xmath23 , @xmath28 , @xmath157 , @xmath25 , @xmath26 , @xmath29 , @xmath30 . for example the variations of @xmath150 and @xmath141 with time for one of the phenemonologically viable sets in table [ table-9 ] is given in table [ table-1 ] . now we check the phenomenological viability of the model . the observational analysis of the model for all possible values of the parameters , @xmath158 , @xmath152 , @xmath153 , @xmath159 , @xmath154 , etc . is an extremely difficult job ( if not impossible at all ) because expressing the hubble parameter , deceleration parameter etc . in terms of the scale factor is quite difficult since these quantities are highly nonlinear functions of the scale factor in this model . therefore i adopt some guidelines to seek the phenomenologically viable sets of parameters . these guidelines are : + 1- i take the model mimic the standard model i.e. the @xmath0cdm model , at least from the time of decoupling of matter and radiation up to the present time . therefore i take the present time values of the equation of state of the whole universe and the density parameter of the baryonic matter and radiation to be the same as @xmath0cdm . + 2- in searching for the phenomenologically viable parameter space i start from the values of the parameters in @xcite i.e. @xmath98 , @xmath99 , @xmath160 , and @xmath161 since the universe studied in @xcite mimics the true universe roughly . + 3- due to the highly non - linear relation between the hubble parameter and the scale factor i seek the relevant parameter space usually by trial and error rather than a continuous scan of the parameter space . therefore the optimum values obtained here most probably may not correspond to the best possible optimization . rather they hopefully correspond to a good approximation to the best optimal values . in this subsection we use the union2.1 compilation data set to find the optimal values of @xmath158 , @xmath152 , @xmath153 starting from @xmath162 , @xmath98 , @xmath99 . we find the theoretical values of distance moduli , @xmath163 for the redshift values of union2.1 and calculate the corresponding @xmath90 value by using the measured values of @xmath163 and their errors . the expression for distance modulus is @xmath164 where @xmath165 where for small redshifts reduces to @xmath166 where we have used the requirement that @xmath167 at small redshifts as discussed in the preceding section ( see table [ table-7 ] ) , and @xmath168 . in @xmath0cdm @xmath169 is usually expressed in terms of redshift , @xmath170 and hubble parameter @xmath68 , and then the results for different @xmath170 s are compared with the data directly . this is not possible in this model because @xmath68 can not be expressed in terms of @xmath16 in a simple way . therefore in this study first we convert redshift values of union2 to time values by using @xmath171 and then solve it for @xmath172 . the corresponding expression for the theoretical value of the luminosity distance @xmath173 in this case ( i.e. in terms of @xmath172 ) is @xmath174^r } \int_\gamma^1 d\gamma\frac{\beta^{-1+\gamma^{-\frac{1}{s } } } } { [ 1+\frac{\gamma-1}{r}\left(\xi\xi_1-\frac{1}{s}\ln{\beta}\right)]^r } \label{mu2}\ ] ] where @xmath137 is expressed in terms of @xmath158 , @xmath152 , @xmath153 , @xmath175 by using the parameterization given in the preceding section . eq.([mu2 ] ) may be written in more standard form in terms of @xmath176 by using @xmath177 . then we find ( [ mu1b ] ) numerically for each of the @xmath172 corresponding to observational redshifts . finally we find the corresponding @xmath178 values by using the formula @xmath179 where the subscript @xmath180 in @xmath181 and the superscript @xmath143 in @xmath182 stands for the fact that @xmath16 is approximated by its zeroth order approximation i.e. by @xmath89 , the superindices @xmath183 and @xmath184 stand for the theoretical and observational values of @xmath163 , and the subindices @xmath185 denote the values of the corresponding quantity for the @xmath185th data point in union2 data set . one may try a better approximation by replacing @xmath137 in ( [ mu1b ] ) by a better approximation of @xmath16 i.e. by @xmath186 where @xmath187 is defined by eq.([a17a ] ) . in principle , then , one may evaluate the integral ( [ mu1a ] ) after replacing @xmath137 by @xmath186 . however this seems to be inapplicable for standard computers because of the complicated form of the integral . one needs a separate computational physics project for this aim . instead one may try a rough approximation ( hopefully better than @xmath89 ) ; we take the @xmath188 term in the integral to outside of the integral with its @xmath172 value being the bound of the integral . this approximation is a good approximation provided that @xmath189 does not vary much in the time interval between @xmath8 and the time corresponding to the given redshift value . otherwise the higher order approximation may worsen the approximation rather than improving . the corresponding formulas ( in the first order approximation ) become @xmath190^r } \int_\gamma^1 d\gamma\frac{\beta^{-1+\gamma^{-\frac{1}{s}}}}{[1+\frac{\gamma-1}{r}\left(\xi_1 -\frac{1}{s}\ln{\beta}\right)]^r } \label{mu4}\ ] ] @xmath191 after trial and error i have found many sets of parameters with relatively small @xmath178 , @xmath90 values . for example the @xmath178 , @xmath90 values for two phenemonologically viable sets of parameters are given in table [ table-9 ] where the reduced @xmath178 , @xmath192 , and the reduced @xmath90 values @xmath193 are in the order of 1 ( where 580 is the number of data points , and 5 , 12 are the number of free parameters @xmath152 , @xmath153 , @xmath158 etc . to be adjusted ) . the sets of parameters ( that i could by trial and error ) with relatively small @xmath90 values satisfy @xmath194 , @xmath195 . by using this information one may check the validity of ( [ a10ba ] ) and determine if one may identify @xmath47 , @xmath196 , @xmath51 by the corresponding density parameters ; @xmath52 , @xmath55 , @xmath56 for the phenomenologically relevant parameters by using eq.([a10b ] ) and eq.([a3db ] ) and eq.([a3dc ] ) . we observe that @xmath197=@xmath156 , @xmath198=@xmath199=@xmath200=1 , @xmath33 and for relevant values of the parameters hence , after using ( [ a10b ] ) , we obtain @xmath201 we observe that for phenemonologically viable sets of parameters , for example , for those in table [ table-9 ] we have @xmath202 and ( [ a10ba ] ) may be satisfied since @xmath203= @xmath204@xmath205 . we notice that @xmath206 . however one may define a total density parameter for the dark sector by @xmath207 then the density parameters satisfies the necessary condition , @xmath208 . in other words , @xmath59 and @xmath60 terms can not be identified as separate contributions to dark sector , rather they must be considered as just a single object in order not to introduce an ambiguity in their identification . in this subsection we investigate if this model is compatible with the cosmological depiction of the recombination and nucleosynthesis , at least , at the order of magnitude level . in a similar vein as the preceding subsection we require this model mimic the standard model , @xmath0cdm , as much as possible . we assume that the radiation and the baryonic matter are in thermal equilibrium in the eras of recombination and nucleosynthesis since we adopt the same equations of thermal equilibrium as @xmath0cdm . therefore , in the following , first we drive the condition for thermal equilibrium for this model . then we find the sets of parameters with least @xmath90 values that may produce successful recombination and nucleosynthesis eras . the correct choices should have sufficient radiation energy densities in these eras . in other words the redshift at the recombination time , @xmath209 should be in the order of @xmath210 where * denotes time of last scattering surface ; and in the nucleosynthesis era the energy density of neutrinos should reach energy densities of the order of @xmath211 . we seek an approximate , rough agreement with @xmath0cdm since the search of the parameter space is done by trial and error rather than a systematic search of the whole parameter space . therefore a detailed , thorough analysis and compatibility survey would be too ambitious especially considering this is a toy model . before checking if there exist a set of parameters compatible with recombination and nucleosynthesis we should check if the thermal equilibrium is maintained in these eras in for the given set of parameters because we adopt the standard analysis in @xmath0cdm , and that analysis assumes existence of thermal equilibrium . as is well known , if there is thermal equilibrium then we should have @xmath212 where @xmath213 is the rate of the interaction between radiation and the matter and @xmath68 is the hubble parameter . however the implementation of this condition in this model is not exactly the same as in @xmath0cdm . in the case of recombination era the implementation of this condition does not give exactly the same result as @xmath0cdm since , in @xmath0cdm the recombination takes place in radiation dominated era and the total energy density is almost wholly due to radiation while , in this model , the total energy density of the universe at this era is not almost wholly due to radiation although the equation of state parameter for phenomenologically relevant cases is similar that of radiation dominated universe at the time of recombination and we require the radiation energy density to be the same or almost the same as @xmath0cdm . in the case of nucleosynthesis , even the equation state parameter in this model does not mimic that of a radiation dominated universe . therefore we should derive the corresponding conditions for thermal equilibrium for this model . the condition for thermal equilibrium in the recombination era is @xmath214 here we have used the identities , @xmath215 where @xmath216 is the @xmath0cdm value and @xmath217 at the time of recombination and is not constant in this model , and i have used the pdg values , @xmath218 , @xmath219 . note that @xmath213 in ( [ rn5 ] ) is the same as the @xmath0cdm value while @xmath68 is different from the @xmath0cdm value . next consider the condition on thermal equilibrium at and before the time of nucleosynthesis . in thermal equilibrium we have @xmath220 where we have used the identity similar to ( [ rn5a ] ) , where @xmath221 and the subindex @xmath222 are replaced by @xmath223 and @xmath152 , respectively and the ratio is evaluated at the time of nucleosynthesis . in this case , as well , @xmath224 is the same as its @xmath0cdm value while the expression for @xmath68 in terms of temperature is different since @xmath225 and is not a constant ( i.e. it gives a different value when evaluated at different time during nucleosynthesis ) in this model while @xmath226 in @xmath0cdm . during thermal equilibrium the ratio of neutrinos to all nucleons , @xmath227 is given by @xmath228 where @xmath229 is the rest mass energy difference between a neutron and a proton , @xmath230 . after the thermal equilibrium between the neutrinos and the nucleons are lost i.e. after decoupling the value of @xmath227 further decreases due to decay of free neutrons as @xmath231 } \label{rn9}\ ] ] where @xmath232 is the @xmath227 of eq.([rn8 ] ) at the time of decoupling , and @xmath233 seconds is the lifetime of a free neutron . therefore the effect of this model is to change the value of @xmath232 ( that depends on @xmath223 ) and probably the value of @xmath227 as well . now we are ready to check the viability of this model . i could give only four graphs and three tables that partially summarize the results of my calculations related to this and the next paragraphs in order not to expand the size of the paper too much . otherwise the size of the manuscript would be almost doubled . first we check the viability of the model for recombination and nucleosynthesis eras . to this end i have used the equations ( [ a11a],[a16c],[a3db],[a3dc ] ) in the zeroth order approximation where @xmath234 ( as discussed before eq.([a3ba ] ) ) to draw @xmath81 , @xmath235 , versus time graphs by using a mathematica code that i have prepared for this aim for the sets of the parameters , @xmath152,@xmath153,@xmath158,@xmath236 , @xmath155 , @xmath156 , @xmath22 , @xmath23 , @xmath237,@xmath238 , @xmath239 , @xmath157,@xmath25 , @xmath26 , @xmath8 @xmath240 , that correspond to some relatively small @xmath90 values obtained in preceding subsection . then i have tried to find at least one set of parameters with phenomenologically viable @xmath241 , @xmath235 , @xmath242 values i.e. @xmath241 , in the range @xmath243 @xmath244 ; @xmath245 ( in the range of redshifts @xmath246 ) at the time of recombination , and @xmath247 at the time of nucleosynthesis where i have approximated @xmath118 , @xmath66 , @xmath67 by @xmath248 , @xmath249 , @xmath250 ( that are defined in eq.([a3ba ] ) ) , and @xmath16 by @xmath137 ( that is defined in ( [ a13b ] ) ) as discussed in the preceding section . i have found two sets of parameters given in table [ table-9 ] that satisfy these conditions . a comment is in order at this point . the zeroth order approximation is reliable only for small redshifts . however this approximation is reliable at any redshift if one is only interested in the energy density - redshift relation . this may be seen as follows : assume that the energy density @xmath87 is related to redshift @xmath170 by @xmath251 in the zeroth order approximation ( where @xmath252 is an arbitrary function ) , and in an approximation better than the zeroth order we have @xmath253 i.e. @xmath254 . then the energy density after the correction is @xmath255 . if one rescales @xmath256 as @xmath257 then one obtains the same redshift and energy density values . in other words the redshift - energy density relation is invariant under such corrections . however this is not true for the redshift - time relation . if the approximation is not a good approximation to the true value then the redshift - time relation will be distorted . this , in turn , may cause the distortion of the value of the equation of state and the distortion of the variation of the energy densities with time in an amount depending on the reliability of zeroth order approximation . keeping these observations in mind i content to use zeroth order approximation for the times of recombination and nucleosynthesis because even employing zeroth order approximation needs a lot of computer cpu and ram , and in many cases the use of first order approximation as well does not improve the situation . we will come back to these points when discuss the times of recombination and nucleosynthesis . next i have checked if thermal equilibrium is maintained at the times of recombination and nucleosynthesis and if recombination and nucleosynthesis are realized in this model . one may have an idea on thermal equilibrium at the time of recombination by using the values of table [ table-9 ] at @xmath258 and eq.([rn5 ] ) . however a more rigorous way is to draw @xmath259 ( that may be obtained from eq.([rn5 ] ) ) versus time graphs to determine the time intervals ( and then the corresponding redshift intervals ) where @xmath260 for each of the sets a and b. in fact i have used @xmath261 for the relevant redshifts . the resulting intervals are the intervals where thermal equilibrium is maintained as shown in figure [ fig:15 ] for the set b in table [ table-9 ] . the smallest redshifts where the thermal equilibrium is lost are @xmath262 ( @xmath263 with @xmath264 ) and @xmath265 ( @xmath266 @xmath267 ) for the sets a and b , respectively . this implies that the photon electron decoupling takes place before the time of last scattering at an energy of @xmath268 ev@xmath269 ev and @xmath270 ev@xmath271 ev for the sets a and b , respectively ( assuming the transition being directly to the ground state of hydrogen atom ) to be compared to the value of photon energy of about @xmath272 ev@xmath273 ev for @xmath0cdm at the time of last scattering . this , in turn , implies that photon electron decoupling in this model for the sets of parameters a and b is at a smaller redshift than @xmath0cdm where thermal equilibrium is maintained until decoupling . ( thermal equilibrium would be maintained till @xmath274 in @xmath0cdm if recombination of electrons and protons to form neutral atoms had not taken place as may be seen from ( [ rn5 ] ) by setting @xmath275 ) . in fact the corresponding times for decoupling are already smaller than that of @xmath0cdm by five orders of magnitude . a detailed comprehensive separate study is need to see if these imply some interesting phenomenologically viable alternatives or just an artifact of the toy model and/or the sets of parameters considered . this may also be due to the limitation of the applicability of zeroth order approximation that we have discussed above . @xmath234 is not violated badly at the time of recombination for the most of the relevant sets of parameters . for example for the sets of parameters given in table [ table-9 ] the first order approximation results in @xmath276 i.e. @xmath277 and does not vary much at the time of recombination . therefore it seems that the effect of the limitation of the applicability of zeroth order approximation to the time of recombination must be limited . however this shift does not introduce a major problem since the redshift values , hence the photon energy density at recombination remains almost the same and thermal equilibrium is maintained . next i have checked if thermal equilibrium is maintained at the peaks in table [ table-9 ] where the energy densities are sufficient for nucleosynthesis . i have used eq.([rn7 ] ) to find the range of temperatures where thermal equilibrium is maintained . i have found that this condition is satisfied for @xmath278 ( provided that @xmath279 ) for the second peaks . this value gives us @xmath232 in ( [ rn9 ] ) by using ( [ rn8 ] ) as @xmath280 which is quite large compared to the @xmath0cdm value of @xmath281 . the time that takes @xmath282 drop to 0.07 mev ( that is when @xmath283 ) in this model is something like @xmath284 seconds . therefore @xmath232 does not drop significantly through eq.([rn9 ] ) . in other words the final result @xmath285 is much larger than the @xmath0cdm value @xmath286 ( which agrees well with observations ) . probably the main source of this discrepancy is inapplicability of zeroth order approximations to redhifts and energy densities to this era to obtain correct energy density - time relations . the variations of @xmath287 and @xmath74 are quite large and their values are quite different than those at @xmath288 at the time of nucleosynthesis that makes the applicability of the zeroth order approximation extremely difficult to obtain correct energy density - time relation . in other words the main source of the discrepancy may be due to the fact that the real time of free decay may be of the order of @xmath289 seconds in this model instead of 90 seconds . the use of first order approximation does not improve the situation because in the calculation of first order approximations to @xmath287 and @xmath290 one uses the zeroth order approximation @xmath291 in the integrals for @xmath292 , @xmath121 . this , in turn , results in over contribution of large redshifts and hence larger and more varying @xmath292 with respect to their true values since @xmath293 and they get smaller i.e. @xmath287 gets larger at larger redshifts . therefore the energy density versus time graphs in the figures [ fig:9 ] , [ fig:10 ] and [ fig:15 ] must be considered with some care : the time values in those graphs should be taken with utmost care especially in the case of nucleosynthesis while the magnitudes of energy densities and the corresponding redshifts are expected to be the same as the exact values . all these points must be studied in more detail in future studies . however i have been able to show that this scheme can produce a model that mimics the standard model : there is a current accelerated epoch whose present equation of state ( for the whole universe ) is -0.7 ( that is , at least , roughly in agreement with observations e.g. see the value in table [ table-9 ] for a phenomenologically relevant set of parameters ) . before this epoch @xmath81 changes sign and the time near this sign change may be considered as the matter dominated era . although the sign change of @xmath81 occurs at a later time in this model compared to @xmath0cdm the time and the redshift of onset of the accelerated era ( i.e. @xmath294 ) are comparable with those of @xmath0cdm . there is an epoch before the matter dominated era where @xmath81 is on average close to @xmath83 , and may be identified by radiation dominated era , and the time of the maximum value of @xmath81 may be considered as the time when the universe was like stiff matter or denser ( as in the cores of stars ) . then @xmath81 changes sign again reaches to two minima peaks as mentioned before and eventually approaches to -1 as time goes to zero ( due to the @xmath59 , in particular the first part of it ) and this epoch probably may be considered as the inflationary era . moreover the model is able to give relatively small reduced @xmath178 and @xmath90 values for union2.1 data set , and it can , at least roughly , account for recombination and nucleosynthesis times . i think this is a sufficiently well starting point for a toy model whose main aim is to embody the creation of matter and radiation in the scale factor of robertson - walker metric . however there is a great deal of points to be clarified and addressed in future studies such as checking the whole parameter space of this model by using a more elaborate software and to use more powerful computers that may give scan the whole parameter space in a better approximation than the one given here , and considering a more detailed analysis of the recombination and nucleosynthesis epochs , studying the evolution of cosmological perturbations in this model , and considering possible extensions of this model towards a more realistic model . in this study a scheme for obtaining a scale factor ( in robertson - walker metric ) that may account for the times before , during , and after the radiation dominated eras is introduced . the prescription to obtain the scale factor in this model is quite simple ; first one introduces a scale factor for the pure dark sector , and then the full scale factor is obtained by a relation between these two scale factors . the result is a scheme to produce the scale factor for the whole universe , including baryonic matter , radiation , and dark energy - matter ( i.e. dark sector ) in such a way that the times before , during , and after radiation dominated era are expressed by a single scale factor in robertson - walker metric . different choices of the pure dark sector scale factor ( denoted by @xmath10 in this paper ) and different choices of the relation between @xmath4 and the scale factor of the full universe , @xmath16 give different models . as an illustration of this scheme a model with a specific scale factor for the pure dark sector and a specific relation between @xmath4 and @xmath16 is considered . the phenomenological viability of this model is checked through its compatibility with union2.1 data set , and with recombination and nucleosynthesis by using trial error and mathematica software for almost randomly chosen sets of parameters . two sets of parameters with relatively small @xmath90 values for union2.1 data set , and that are compatible with successful recombination and nucleosynthesis at an order of magnitude level are found . these results are encouraging in view of the fact that only a tiny portion of the whole parameter space could be considered in this way . a separate , detailed , and comprehensive computational project with more advanced software codes and/or powerful computing facilities that may scan the full parameter space and may employ better approximation schemes is needed to reach a definite view on the observational viability of this scheme and/or this model . moreover the effect of this model on cosmological perturbations should be considered and possible implications and extensions of this scheme to inflationary era should be studied in future . furthermore different pure dark sector scale factors and different options to relate pure dark sector and the full universe scale factors may be considered in future to see the full range of possibilities that this scheme may offer . i would like to thank professor joan sol@xmath295 for reading the manuscript and for his valuable comments . i would also like to thank professor gregory gabadadze for reading the manuscript , and to dr . a. aviles for his help in the initial phase of writing the mathematica code used in this study . 99 e.w . kolb and m.s . turner , _ the early universe _ , ( westview press , usa , 1994 ) ; + s. dodelson , _ modern cosmology _ , ( academic press , usa , 2003 ) ; + s. weinberg , _ cosmology _ , ( oxford univ . press , new york , 2008 ) g. bertone , d. hooper , j. silk , _ particle dark matter : evidence , candidates and constraints _ , _ phys . rept . _ * 405 * , 279 ( 2005 ) , hep - ph/0404175 e.j . copeland , m. sami , s. tsujikawa , _ dynamics of dark energy _ , _ int d _ * 15 * , 1753 ( 2006 ) , hep - th/0603057 ; + m. sami , _ models of dark energy _ , _ lect . notes phys . _ * 720 * , 219 ( 2007 ) ; + j. frieman , m. turner , d. huterer , _ dark energy and the accelerating universe _ , _ ann . rev astrophys._*46 * , 385 ( 2008 ) , arxiv:0803.0982 s. weinberg , _ the cosmological constant problem _ , _ rev . phys . _ * 61 * , 1 ( 1989 ) ; + s. nobbenhuuis , _ categorizing different approaches to the cosmological constant problem _ , _ found . _ ; * 36 * , 613 ( 2006 ) ; + f. bauer , j. sol@xmath295 , and h. @xmath314tefan@xmath315i@xmath316 , _ the relaxed universe : towards solving the cosmological constant problem dynamically from an effective action functional of gravity _ , _ phys . b _ * 688 * , 269 ( 2010 ) ; + s. basilakos , f. bauer , j. sol@xmath295 , _ confronting the relaxation mechanism for a large cosmological constant with observations _ , _ jcap _ * 1201 * , 050 ( 2012 ) ; + j. sol@xmath295 , _ cosmological constant and vacuum energy : old and new ideas _ , _ j. phys ser . _ * 453 * , 012015 ( 2013 ) r. erdem , _ a symmetry for vanishing cosmological constant in an extra dimensional toy model _ , _ phys . b _ * 621 * , 11 ( 2005 ) + g. t hooft , s. nobbenhuuis , _ invariance under complex transformations , and its relevance to the cosmological constant problem _ , _ class . * 23 * , 3819 ( 2006 ) + r. erdem , _ a symmetry for vanishing cosmological constant : another realization _ , _ phys . b _ * 639 * , 348 ( 2006 ) + m.j . duff , j. kalkkinen , _ signature reversal invariance _ , _ nucl . b _ * 758 * , 161 ( 2006 ) + r. erdem , _ a symmetry for vanishing cosmological constant _ , _ j. phys . a _ * 40 * , 6945 ( 2007 ) + r. erdem , _ a way to get rid of cosmological constant and zero point energy problems of quantum fields through metric reversal symmetry _ , _ j. phys . a _ * 41 * , 235401 ( 2008 ) s. nojiri , s.d . odintsov , _ unified cosmic history in modified gravity : from f(r ) theory to lorentz non - invariant models _ , _ phys . rept . _ * 505 * , 59 ( 2010 ) , arxiv:1011.0544 b. ratra and p.j.e . peebles , _ the cosmological consequences of a rolling homogenous scalar field _ , _ phys . * 37 * , 3406 ( 1988 ) + r.r . caldwell , r. dave , and p.j . steinhardt , _ the cosmological imprint of an energy component with general equation of state _ , _ phys . * 80 * , 1582 ( 1998 ) , astro - ph/9708069 + p. binetruy , _ the cosmological constant versus quintessence _ , _ int . phys . _ * 39 * , 1845 ( 2000 ) + v. sahni , _ the cosmological constant problem and quintessence _ , _ class . grav . _ * 19 * , 3435 ( 2002 ) g. gentile and p. salucci , _ the cored distribution of dark matter in spiral galaxies _ , _ mon . not . soc . _ * 351 * , 903 ( 2004 ) a. klypin , a.v . kravtsov , o. valenzuela , and f. prada , _ where are the missing galactic satellites _ , _ astrophys . j _ * 522 * , 82 ( 1999 ) j. sommer - larsen and a. dolgov , _ formation of disk galaxies : warm dark matter and the angular momentum problem _ , _ astrophys . j. _ * 551 * , 608 ( 2001 ) + m. viel et . _ constraining warm dark matter candidates including sterile neutrinos and light gravitinos with wmap and the lyman-@xmath317 forest _ , _ phys . d _ * 71 * , 063534 ( 2005 ) s - j . sin , _ late time cosmological phase transition and galactic halo as bose liquid _ * 50 * , 3650 ( 1994 ) , hep - ph/9205208 ji and s - j . sin , _ late time phase transition and the galactic halo as a bose liquid : 2 . the effect of visible matter _ , _ phys . d _ * 50 * , 3655 ( 1994 ) , hep - ph/9409267 . bhmer and t. harko , _ can dark matter be a bose - einstein condensate ? _ , _ jcap _ * 06 * , 025 ( 2007 ) + l.a . urena - lopez , _ bose - einstein condensation of relativistic scalar field dark matter _ , _ jcap _ * 01 * , 014 ( 2009 ) . + i.r . rodriguez - montoya , a. perez - lorenzana , e.d.l . cruz - burelo , y. giraud - heraud , and t. matos , _ cosmic bose dark matter _ , _ phys . d _ * 87 * , 025009 ( 2013 ) r. dick , _ the dilaton as a candidate for dark matter _ , in _ heidelberg 1996 , dark matter in astro and particle physics _ , 395 ( 1996 ) , hep - th/9609190 + y.m . cho and y.y . keum , _ dilatonic dark matter and unified cosmology : a new paradigm _ , _ class . _ * 15 * , 907 ( 1998 ) + t. matos and f.s . guzman , _ scalar fields as dark matter in spiral galaxies _ , _ class . * 17 * , l9 ( 2000 ) + t. matos and l.a . urena - lopez , _ quintessence and scalar dark matter in the universe _ , _ class . * 17 * , l75 ( 2000 ) d. lyth , a. riotto , _ particle physics models of inflation and the cosmological density perturbation _ , _ phys . rep . _ * 314 * , 1 ( 1999 ) , hep - ph/9807278 + a. mazumdar , j. rocher , _ particle physics models of inflation and curvaton scenarios _ , _ phys . rep . _ * 497 * , 85 ( 2011 ) , arxiv:1001.0993 j.a.s . lima , s. basilakos , j. sol@xmath295 , _ expansion history with decaying vacuum : a complete cosmological scenario _ , _ mon . not . r. astron . soc . _ * 431 * , 923 ( 2013 ) , arxiv:1209.2802 ; + e.l.d . perico , j.a.s . lima , s. basilakos , j. sol@xmath295 , _ complete cosmic history with a dynamical @xmath318 term _ , _ phys d _ * 88 * , 063531 ( 2013 ) , arxiv:1306.0591 ; + s. basilakos , j.a.s . lima , j. sol@xmath295 , _ from inflation to dark energy through a dynamical @xmath0 : an attempt at alleviating fundamental cosmic puzzles _ , _ int . d _ ( 2013 ) ( work awarded honorable mention by the gravity research foundation ) , arxiv:1307.6251 r. erdem , _ a simple toy model for a unified picture of dark energy , dark matter , and inflation _ , preprint arxiv:1105.0345 f.s . guzman , t. matos , _ quintessence - like dark matter in spiral galaxies _ , _ rev . fis . _ * 49 * , 203 ( 2003 ) , arxiv : astro - ph/0003105 . t. matos , d. nunez , f.s . guzman , e. ramirez _ geometric conditions on the type of matter determining the flat behavior of the rotational curves in galaxies _ , _ gen . _ * 34 * , 283 ( 2002),arxiv : astro - ph/0005528 . a. arbey , _ dark fluid : a complex scalar field to unify dark energy and dark matter _ , _ phys . d _ * 74 * , 043516 ( 2006 ) , astro - ph/0601274 . l. chimento , m.i . forte , _ unified model of baryonic matter and dark components _ , _ phys . lett . b _ * 666 * , 205 ( 2008),arxiv:0706.4142 a. aviles , j.l . cervantes - cota _ dark matter from dark energy - baryonic matter couplings _ , _ phys . d _ * 83 * , 023510 ( 2011),arxiv:1012.3203 k. nakamura et al . ( particle data group ) , _ j. phys . g _ * 37 * , 075021 ( 2010 ) p.a.r . abe et al . ( planck collaboration ) , _ planck 2013 results . cosmological parameters _ , ( 2013 ) arxiv:1303.5076 j. khoury , a. weltman , _ chameleon fields : awaiting surprises for tests of gravity in space _ , _ phys . lett . _ * 93 * , 171104 ( 2004 ) , astro - ph/0309300 + j. khoury , a. weltman , _ chameleon cosmology _ , _ phys . d _ * 69 * , 044026 ( 2004 ) , astro - ph/0309411 + p. brax , c. van de bruck , a - c . davis , j. khoury , a. weltman , _ detecting the dark energy in orbit - the cosmological chameleon _ , _ phys . d _ * 70 * , 123518 ( 2004 ) , astro - ph/0408415 k. hinterbichler , j. khoury , _ symmetron fields : screening long - range forces through local symmetry restoration _ , _ phys . rev . lett . _ * 104 * , 231301 ( 2010 ) , arxiv:1001.4525 + k. hinterbichler , j. khoury , a. levy , and a. matas , _ symmetron cosmology _ , _ phys . d _ * 84 * , 103521 ( 2011 ) , arxiv:1107.2112 k.a . olive and m. pospelov , _ environmental dependence of masses and coupling constants _ , _ phys . * 77 * , 043524 ( 2008 ) , arxiv:0097.3825 o. akarsu , s. kumar , r. myrzakulov , m. sami , and l. xu , _ cosmology with hybrid expansion law : scalar field construction of cosmic history and observational constraints _ , _ jcap _ * 01 * , 022 ( 2014 ) , arxiv:1307.4911 . versus redshift z graphs for @xmath0cdm ( for pdg and planck values ) and for this model ( for two sets of parameters with small @xmath90 values ) , namely , for @xmath0cdm with @xmath319 , @xmath320 , @xmath321 ( solid green ) and with @xmath322 , @xmath323 , @xmath321 ( dashed black ) ; and for this model for the set of parameters in table 9 ( solid blue ) and for the set of parameters ( @xmath152=1.58 , @xmath153=5.3 , @xmath158=3.1 , @xmath240=0.21 , @xmath324=0.975 , @xmath325=1 , @xmath156=@xmath326 , @xmath23=0.999 , @xmath237=0.1 , @xmath327=0.05 , @xmath25=0.9 , @xmath238= @xmath239=0.7 ) ( dotted red ) . here the sub - index @xmath328 refers to dust . ] ( dotted blue ) , @xmath242 ( dot - dashed red ) , @xmath329 ( solid black ) , @xmath235 ( dashed green ) , @xmath330 ( solid yellow ) versus @xmath175 graphs for the set b in table 9 for the first energy density peak in the interval @xmath331 . in this graph @xmath81 , @xmath242 , @xmath235 , @xmath330 are given as multiples of @xmath332 , @xmath333 , @xmath334 , @xmath335 , @xmath336 , respectively . ] ( dotted blue ) , @xmath242 ( dot - dashed red ) , @xmath329 ( solid black ) , @xmath235 ( dashed green ) , @xmath330 ( solid yellow ) versus @xmath175 graphs for the set b in table 9 for the second energy density peak in the interval @xmath337 . in this graph @xmath81 , @xmath242 , @xmath235 , @xmath330 are given as multiples of 10 , @xmath333 , @xmath334 , @xmath338 , @xmath339 , respectively . ] .the zeroth and the first order approximations to @xmath290 ; @xmath341 and @xmath342 as function @xmath175 for the set of parameters , @xmath152=.138 , @xmath153=12 , @xmath158=3.3 , @xmath343=0.975 , @xmath155=1 , @xmath156=0.002745 , @xmath22=1 , @xmath23=0.9986 , @xmath237=0.03 , @xmath157=@xmath344 , @xmath25=0.5 , @xmath26=1 , @xmath238=1 , @xmath239=0.8 , @xmath345 . note that first order value @xmath141 for @xmath346 and smaller values of @xmath172 are not evaluated since the iteration procedure is not applicable for such small times because of the time intervals , @xmath347 that we have used in eq.([ax ] ) is much coarser than @xmath348 . [ cols="^,^,^,^,^,^,^,^,^,^,^,^,^,^,^",options="header " , ] set&@xmath90&@xmath349 & @xmath350&@xmath351 & @xmath352 & @xmath353&@xmath354 + a&623.203&-0.69&@xmath355&1148 & @xmath356&@xmath357&0.245 + a&@xmath358&@xmath359 & @xmath360 & @xmath361 & @xmath362 & @xmath363&@xmath364 + a&@xmath365 & 12839&@xmath366&@xmath367 & -1.67&@xmath368(@xmath369 ) & 6.22 + & @xmath370&@xmath371 & @xmath372 & @xmath373&-8.5 & @xmath374 ( @xmath375 ) & ( @xmath376 ) + set&@xmath90&@xmath241 & @xmath350&@xmath351 & @xmath352 & @xmath353 & @xmath354 + b&576.69 & -0.6948 & @xmath355&1148 & @xmath377 & @xmath378 & 0.257 + b & @xmath358&@xmath359 & @xmath360 & @xmath361 & @xmath362 & @xmath363&@xmath364 + b&@xmath379 & 7827&@xmath380&@xmath381&-1.67 & @xmath382 ( @xmath383 ) & @xmath384 + & @xmath370 & @xmath385 & @xmath386 & @xmath387 & -9.6 & @xmath388 ( @xmath389 & ( @xmath390 ) +

we report systematic hcn j=1 - 0 ( and co ) observations of a sample of 53 infrared ( ir ) and/or co - bright and/or luminous galaxies , including seven ultraluminous infrared galaxies , nearly 20 luminous infrared galaxies , and more than a dozen of the nearest normal spiral galaxies . this is the largest and most sensitive hcn survey of galaxies to date . all galaxies observed so far follow the tight correlation between the ir luminosity and the hcn luminosity initially proposed by solomon , downes , & radford , which is detailed in a companion paper . we also address here the issue of hcn excitation . there is no particularly strong correlation between and the 12 m luminosity ; in fact , of all the four bands , the 12 m luminosity has the weakest correlation with the hcn luminosity . there is also no evidence of stronger hcn emission or a higher ratio of hcn and co luminosities for galaxies with excess 12 m emission . this result implies that mid - ir radiative pumping , or populating , of the j=1 level of hcn by a mid - ir vibrational transition is not important compared with the collisional excitation by dense molecular hydrogen . furthermore , large velocity gradient calculations justify the use of hcn j=1 - 0 emission as a tracer of high - density molecular gas ( ) and give an estimate of the mass of dense molecular gas from hcn observations . therefore , may be used as a measure of the total mass of dense molecular gas , and the luminosity ratio may indicate the fraction of molecular gas that is dense . psfig.sty 2 2

astro-ph0310341

though co traces most of the molecular gas mass in galaxies , it does not necessarily trace active star - forming regions where the gas density is more than 10 times higher than the average . these regions are better traced by high dipole moment molecules like hcn , cs and hco@xmath6 . consequently , these molecules have often been used as probes of physical conditions in giant molecular cloud ( gmc ) cores the regions of active star formation in the milky way . yet there have been few systematic studies of dense gas in gmcs in the disk of the milky way . lee , snell , & dickman ( 1990 ) have found that for milky way disk gmcs , the ratio of co / cs intensity is 156 , much larger than for galactic center clouds and an order of magnitude larger than for the archetypal ulig arp 220 ( solomon , radford , & downes 1990 ) . some dense clumps in some nearby gmcs have also been surveyed in cs ( lada , bally , & stark 1991 ; plume 1997 ) and in hcn ( helfer & blitz 1997b ) . although there are some recent attempts in galactic plane hcn , cs , and co observations ( helfer & blitz 1997b ; mcquinn 2002 and references therein ) , only the central region of the galaxy has been extensively mapped with dense molecular gas tracers ( , cs , bally 1987 , 1988 ; hcn , jackson 1996 ; lee 1996 ) . systematic and unbiased large - scale observations of the dense molecular gas in galactic disk gmcs need continued study since the large - scale ( kpc ) distribution of dense gas in the milky way disk ( cf . mcquinn 2002 ) is a basis for comparison with dense molecular gas observations in external galaxies ( , paglione 1998 ) . hcn and cs are probably the most frequently observed interstellar molecules after co. because of their higher dipole moments ( @xmath7 2.0 3.0 d ) , they require about 2 orders of magnitude higher densities for collisional excitation than co ( @xmath8 0.1 d ) . hcn is one of the most abundant high dipole moment molecules and traces molecular gas at densities @xmath9(h@xmath10 @xmath5 ( compared to densities at @xmath11 @xmath5 traced by co ) . hcn j=1 - 0 emission has been so far detected toward the nuclei of @xmath12 10 nearby galaxies ( nguyen - q - rieu , nakai , & jackson 1989 ; nguyen - q - rieu 1992 ; henkel 1990 ) , 10 relatively distant luminous and ultraluminous infrared galaxies ( ligs and uligs with ir luminositiesm to 1000 @xmath2 m ) is calculated according to the prescription in sanders & mirabel ( 1996 ) using measurements from all 4 flux bands . ] @xmath13 and @xmath14 , respectively ) including three less luminous , `` normal '' spiral galaxies ( solomon , downes , & radford 1992 ) , 12 nearby starburst and/or normal galaxies ( helfer & blitz 1993 ) , 10 interacting galaxies and infrared galaxy mergers ( aalto 1995 ) , 13 seyfert galaxies ( curran , aalto , & booth 2000 ) , and a few nearby co - bright galaxies ( israel 1992 ; sorai 2002 ; kuno 2002 ) . there are significant overlaps in the galaxy samples among these various observations , and the total number of galaxies with hcn detections is still @xmath15 , particularly the total hcn emission measured globally from the galaxies . a recent summary of hcn observations in centers of nearby galaxies can be found in sorai ( 2002 ) and shibatsuka ( 2003 ) . in a few of the nearest galaxies , interferometer hcn maps have also been obtained in the nuclear regions ( carlstrom 1988 ; 1990 ; radford 1991 ; downes 1992 ; brouillet & schilke 1993 ; jackson 1993 ; helfer & blitz 1995 , 1997a ; tacconi 1994 ; shen & lo 1995 ; paglione , tosaki , & jackson 1995b ; kohno 1996 ; kohno , kawabe , & vila - vilaro 1999 ) . cs emission , another dense molecular gas tracer , has been observed in the central regions of @xmath16 galaxies as well including interferometry imaging ( mauersberger & henkel 1989 ; mauersberger 1989 ; sage , shore , & solomon 1990 ; helfer & blitz 1993 ; xie , young , & schloerb 1994 ; paglione 1995a ; peng 1996 ; reynaud & downes 1997 ; wild & eckart 2000 ) . the presence of large amounts of dense molecular gas in the central @xmath12 1 kpc of galaxies has been well established ( for an earlier review , see mauersberger & henkel 1993 ) . the main goal of this paper is to present a systematic hcn survey of a large sample of normal spiral galaxies and relatively distant ligs and uligs . this includes an attempt to measure the total hcn emission in nearby large spiral galaxies by mapping the hcn distribution along the major axes ( gao 1996 , 1997 ) . we examine correlations between hcn emission and various bands and discuss the excitation mechanisms of hcn . we here argue that hcn is primarily excited by collision with molecular hydrogen at high density and thus can be used as a tracer of dense molecular gas . we concentrate here on presenting observations of hcn j=1 - 0 only since it is almost always the strongest line among these high dipole moment molecules , and thus the easiest and most often observed dense gas tracer in external galaxies ( gao 1996 ) . we have not only doubled the total number of galaxies with hcn j=1 - 0 detections in the literature for the last decade , but we have also extensively mapped the inner disk hcn emission in a few nearby galaxies in order to estimate the total hcn emission . we will explore in detail the extent and radial distribution of dense molecular gas in nearby galaxies in a future paper and present the detailed hcn mapping observations there ( y. gao & p.m. solomon in preparation ) . in a companion paper ( gao & solomon 2004 , paper ii ) , we fully discuss the interpretation and implications of our hcn survey . we describe the sample selection and characteristics of our hcn survey in the next section . in 3 , we discuss our various observations and the observational strategies conducted at different telescopes . the survey results are presented in 4 , where we show all the hcn j=1 - 0 and co spectra , tabulate the observational data , and summarize the global quantities for all galaxies . we have confirmed the tight correlation between the ir and hcn emission , initially proposed by solomon ( 1992 ) based on the hcn observations of a sample of 10 representative galaxies , in a large and statistically significant sample of more than 60 galaxies when both samples are combined . we have also checked for any significant dependence of the hcn emission upon the excess mid - ir emission as indicated by an flux ratio of the 12 and 100 @xmath2 m emission or the ratio of the 12 @xmath2 m and total ir emission to address the possibly enhanced excitation of hcn j=1 - 0 emission by mid - ir radiative pumping through a vibrational transition . moreover , the use of hcn as a tracer of high - density molecular gas is discussed using large velocity gradient ( lvg ) calculations . finally , we summarize the main points of this hcn survey in 5 . our goal is to survey a large sample of galaxies ( @xmath17 ) with a wide range of ir luminosities where the total hcn j=1 - 0 emission from the whole inner disk or entire galaxy can be measured . essentially , all galaxies with strong co and ir emission were chosen for our hcn observations . our hcn survey sample is mainly drawn from the co observations of solomon ( 1997 ) , sanders , scoville , & soifer ( 1991 ) , tinney ( 1990 ) , solomon & sage ( 1988 ) , and young ( 1989 , 1995 ) . almost all galaxies with an antenna temperature measured by various telescopes to be greater than 100 mk ( or 50 mk for ligs ) were included as candidate galaxies . these cutoffs were specifically used in order to select all truly co - bright galaxies in the hope of detecting their hcn emission ( which is typically 1020 times weaker than co ) with a rms sensitivity at the level of @xmath18mk . most ir - bright galaxies with @xmath19 are included as well in our sample ( rice 1988 ; strauss 1992 ) . we thus selected @xmath16 nearest co and/or ir bright large galaxies with @xmath20 , and a few tens of relatively distant yet still nearby `` normal '' spiral galaxies and ligs . moreover , most northern uligs in the local universe ( cz @xmath21 20,000 ) were also added to enlarge our sample at this extremely high ir luminosity . we detected hcn j=1 - 0 emission from 52 galaxies , and only one galaxy was not detected . table 1 lists all 53 sources , along with some basic properties of the galaxies useful for our observations and data analysis . the nearest large galaxies that require mapping to estimate the total hcn emission are highlighted in boldface in table 1 , and the detailed observational results of these galaxies will be discussed separately ( y. gao & p.m. solomon 2004 , in preparation ) . seven uligs are marked with a star symbol in table 1 . our hcn survey sample also contains essentially a statistically complete sample for northern ( @xmath22 ) galaxies with @xmath23 jy , which includes about 30 of the brightest galaxies . for the purpose of the detailed statistical analysis to be presented in paper ii , we included almost all hcn data available in the literature where the total hcn emission from galaxies can be estimated to augment our sample . there are about a dozen mostly ir - luminous galaxies with total hcn emission measured , almost entirely from solomon ( 1992 ) , but m51 and ngc 4945 are from nguyen - q - rieu ( 1992 ) and henkel ( 1990 ) , respectively . even with these supplements and those intentionally added uligs , the survey sample is not severely biased toward the high ir luminosity end . as shown in figure 1a , although most of our sample galaxies are ir bright or the brightest in the local universe , more than half have ir luminosities less than @xmath24 , and among them most have @xmath25 . there seems to be a slight bias , however , toward high co luminosity galaxies ( fig . 1b ) , , more co - luminous or molecular gas - rich galaxies have been selected . this bias mainly arises because most normal spiral galaxies selected are co bright and quite rich in molecular gas , and essentially all ligs and uligs have high co luminosities . nevertheless , the galaxy number distribution in hcn luminosity appears to be rather unbiased with comparable numbers of galaxies with both low and high hcn luminosities , and the entire hcn luminosity distribution covers a large range of almost 3 orders of magnitude ( fig . we used the former nrao 12 m telescope at kitt peak and the iram 30 m telescope at pico veleta near granada , spain , for most of our hcn observations . the fcrao 14 m telescope with the 15 element focal - plane array receiver , quarry ( a configuration of three rows of five beams with a row spacing of @xmath26 and a beam spacing of @xmath27 ) , was initially used for mapping some of the nearest strong sources . we conducted several observing runs at each telescope from 1993 april to 1995 december . additional observations were conducted remotely at the nrao 12 m telescope in 1997 may and june to confirm some weak detections and to further improve the signal - to - noise ratios for some hcn spectra . the system temperatures at the redshifted hcn line frequencies were typically 220 , 250 , and 500 k on average on the antenna temperature scales of @xmath28 , @xmath29 , and @xmath29 for the nrao , iram and fcrao telescopes , respectively . we converted the observed antenna temperature to the main - beam temperature scale ( @xmath30 ) for all our observations at different telescopes . table 2 lists various telescope beam efficiencies measured at different telescopes at both the 2.6 and 3.4 mm wavebands . at @xmath31 mm , we adopted the continuum sensitivity ( the temperature - to - flux conversion for a point source ) on the @xmath30 scale as @xmath32 , 29.0 , and 23.5 jy k@xmath33 for the iram 30 m , nrao 12 m , and fcrao 14 m telescopes , respectively . at the iram 30 m , we used two 512 @xmath34 1 mhz filter banks and the autocorrelator in its widest bandwidth configuration as back ends . these back ends were connected to three sis receivers ( 3 mm , 2 mm , 1 mm ) that were usually used to simultaneously observe hcn j=1 - 0 , cs ( 3 - 2 ) , and co j=2 - 1 lines respectively . the telescope beam diameters ( fwhm ) for co j=1 - 0 and hcn j=1 - 0 were @xmath35 and @xmath36 , respectively . the nrao 12 m was equipped with dual - polarization 3 mm sis receivers . we connected the dual receivers to a pair of @xmath37 mhz filter banks as well as the two 600 mhz wide hybrid spectrometers . the telescope beam diameters ( fwhm ) for co j=1 - 0 and hcn j=1 - 0 were 55@xmath38 and 72@xmath38 , respectively . at the fcrao 14 m , the corresponding back ends for the 15 cooled schottky diode mixers of the quarry array were 15 spectrometers of @xmath39 mhz bandwidth . the telescope beam spacing of @xmath27 within each row of quarry is about the beam size ( fwhm ) of the 14 m telescope at 3 mm . the iram 30 m telescope used a wobbling secondary with a beam throw of @xmath40 to achieve flat baselines . similarly , a nutating subreflector with a chop rate of @xmath41 hz was used at the nrao 12 m together with a position switch ( the so - called beam - switch plus position - switch , or `` bsp mode '' ) . the beam throws used were in the range of @xmath42 to @xmath43 ( the larger beam throw was used for nearest large galaxies with @xmath44 ) . thus , the separations between the off reference positions and the on source position are in the range of @xmath45 to @xmath46 . for a couple of the largest galaxies ( @xmath47 ) observed at the 12 m , we employed position switching in azimuth with offsets comparable to the expected co source sizes to ensure that the reference positions are completely outside of the sources observed . at the fcrao 14 m , we used double - position switching in azimuth with offsets comparable to the diameters of galaxies . the quarry receivers of the fcrao 14 m telescope were efficient in mapping the co emission in the inner disks of nearby galaxies , but the sensitivity achievable limited our extensive mapping of the weak hcn emission outside the nuclear regions . no additional beams except for the central beam and sometimes the adjacent two inner beams were detected in hcn with the 14 m . therefore , most of the hcn mapping was eventually conducted with the 12 m and 30 m telescopes after the initial trial observing runs with the quarry at the fcrao 14 m . we mainly used the iram 30 m to observe most ( rather distant ) uligs in the sample and to map some nearby starburst galaxies of mostly smaller optical diameters , given the smaller telescope beam compared with the source size and extent for achieving effective mapping . essentially all other observations were conducted with the nrao 12 m so that one beam measurement covered all emission from relatively distant galaxies including some interacting / merging galaxy pairs . almost all nearby spiral galaxies were mapped along the major axes with the nrao 12 m . these 12 m observations included both the hcn observations initially attempted with the fcrao 14 m quarry array , and in several cases , the major axis mapping observations from the iram 30 m in order to properly compare the total hcn emission estimated with different telescopes ( see table 3 ) . this is essential to better characterize the radial hcn distribution in galaxies ( y. gao & p.m. solomon 2004 , in preparation ) . all observed positions were chosen by checking and comparing the positions of the peak radio continuum emission as determined from the vla snapshot survey ( condon 1990 ) or the best positions ( with uncertainties @xmath48 ) then listed in the ned . the observed spectra from all telescopes ( except for the fcrao 14 m ) were smoothed to @xmath128 mhz resolution , or @xmath49 and @xmath50 at 2.6 and 3.4 mm , respectively . occasionally , in several of the weakest hcn sources , the spectra were smoothed to 16 mhz to boost the signal - to - noise ratio . linear baselines were removed from the spectra after emission line windows were chosen to estimate the integrated line intensities . we converted the line temperatures to the main - beam brightness temperature scale @xmath30 ( tables 2 and 3 ) . in many cases , co lines were also observed from the same telescope for comparison and for determination of the hcn line emission windows . all hcn observations of galaxies so far have shown that the significantly detected hcn line profiles have the same line widths as the co lines ( , solomon 1992 ; helfer & blitz 1993 ; gao 1996 ) . with the co spectra obtained from the same positions or available in the literature from the same telescope , the integrated line intensities of the weakly detected hcn lines can be more accurately estimated . calibrations at all three telescopes were done using the standard chopper wheel method ( kutner & ulich 1981 ) . the absolute antenna temperature scale was checked by either observing the standard calibration sources or by the measured co line strengths of the targets ( mostly for the nearest strong sources ) as compared with the standard spectra at the telescopes and the known co spectra available in the literature . pointing was checked by observing quasars or planets about every 23 hours , with typical errors of @xmath51 for the iram 30 m , and @xmath52 10@xmath38 for the nrao 12 m and the fcrao 14 m . focus was also monitored a few times per day mostly using planets . more detailed discussions of the observations and the antenna temperature scales at different telescopes can be found in gao ( 1996 , 1997 ) . we present all hcn spectra in figs . 2 and 3 for the iram 30 m and the nrao 12 m telescopes , respectively , including a few weak ( @xmath53 ) detections . for most galaxies , we also show the co spectra taken at the same position from the same telescope for comparison . a few co spectra in figure 2 , however , are co j=2 - 1 lines , which have much smaller beam sizes compared with that of the hcn j=1 - 0 spectra . in some cases , the velocity range of co emission , obtained from the published co spectra in the literature with the same telescopes or comparable telescope beam sizes , is indicated instead . ngc 253 , ngc 2903 , m82 , and ngc 6946 were mapped by both the iram 30 m and the nrao 12 m and therefore appear in both figs . 2 and 3 . only the central hcn spectra of those nearby galaxies highlighted in table 1 are displayed here ; the detailed hcn mapping will be shown in a future paper ( y. gao & p.m. solomon 2004 , in preparation ) . we detected hcn emission in essentially all of the sources ( except ngc 1055 ) listed in table 1 , though a few galaxies were only marginally detected . the observed line intensities ( of both hcn and co if applicable ) are listed in table 3 together with other measured quantities useful for analysis . the derived global properties of the line luminosities and their ratios using methods discussed below ( 4.2 ) are listed in table 4 . fully sampled mapping is essential in order to estimate the total hcn emission in large spiral galaxies ( gao 1996 , 1997 ; y. gao & p.m. solomon 2004 , in preparation ) because a substantial fraction of the total hcn emission is distributed outside the @xmath54 kpc nuclear regions ( which are covered by the central beams in these observations ) . therefore , we did not add more nearby galaxies with hcn detections from their centers from the literature into our hcn sample . those data only measure lower limits to the hcn luminosity since only the central beam detections were made for these nearby large spiral galaxies , and the hcn emission outside the central @xmath54 kpc nuclear regions is unknown . notice that about 15% of the galaxies in our hcn survey sample also have either upper or lower limits in the total hcn emission or luminosity ( tables 3 & 4 ) . the upper limits in the hcn measurements correspond to weak detections ( @xmath55 ) or non - detections ( @xmath56 , only for ngc 1055 ) or to nearby galaxies with good detections in the center but with only weak detections or non - detections at the off - center positions . the lower limits in the hcn luminosity are for a few nearby galaxies with good hcn detections from the galaxy centers , but no systematic mapping or detections at off - center positions ( ngc 2903 is exceptional since it was mapped with the iram 30 m , yet more sensitive and extensive mapping is still required to estimate the total hcn emission ) . this is similar to the case of more than a dozen nearby galaxies with only central beam hcn observations in the literature ( thus lower limits to the total hcn emission ) . we include our own limits , but not the hcn limits in the literature in the final hcn sample . for sources much smaller than the telescope beam , the hcn luminosity is calculated using @xmath57 ( see solomon 1997 for details , @xmath58 with a unit of k pc@xmath59 is introduced to distinguish it from normal @xmath60 , which has proper power units ) , where @xmath61 is the solid angle subtended by the source area , @xmath62\ ] ] is the observed integrated line intensity ( as converted to the beam - diluted main - beam temperature scale ) , @xmath63 is the intrinsic source brightness temperature , @xmath64 is the solid angle of the source convolved with the diameter ( fwhm ) of the telescope gaussian beam @xmath65 , and @xmath66 and @xmath67 are the luminosity distance=75 mpc@xmath33 and @xmath68 were used in this paper ] and the angular size distance to the source , respectively , related by @xmath69 $ ] mpc . for large galaxies where hcn was observed at several positions along the major axis , the total integrated hcn flux over the source , @xmath70 is estimated following the empirical formula adopted in solomon & sage ( 1988 ; also sage 1987 ) to calculate the total co emission . the total hcn emission is obtained by summing all the measured line fluxes along the major axes with weights determined from considerations of @xmath65 , the ( source ) disk size , @xmath71 , and the disk inclination , and assumption of an axisymmetric source distribution . thus , for nearby galaxies , @xmath72 the co luminosity is calculated similarly with the same formula listed above ( switching hcn with co ) . for brevity , we hereafter use the symbol @xmath60 rather than @xmath58 for the line luminosity in unit of k pc@xmath59 for both co and hcn emission . for those nearby galaxies that we observed with different telescopes , the agreement in the estimated total hcn fluxes is rather good ( table 3 ) , given the weak hcn lines detected outside the nuclear regions in most cases . for all positions observed in hcn , the permitted velocity ranges are available either from our own , or published , co observations . as discussed in 3 , the weak detections or upper limits for the hcn line intensities can be better estimated when the hcn line emission windows are known / assumed from the co lines , and thus the hcn results are rather robust . the uncertainties in the total hcn ( and occasionally co ) luminosities are up to @xmath73 including the calibration errors of different telescopes , but in most cases only @xmath74 or better . the largest uncertainties are only present in a few galaxies with the marginal detections ( @xmath55 ) and in some nearby large galaxies with very limited points observed where extensive mapping along their major axes is still required for more accurate estimates of the total hcn emission . figure 4 shows a tight linear correlation between @xmath1 and @xmath0 over 3 orders of magnitude in ir luminosity . the originally proposed correlation , based on 10 representative galaxies ( solomon 1992 ) , is also noted for comparison . we recalculated , ( see $ 4.2 ) based on the measured line intensities @xmath75 and @xmath76 listed in table 1 of solomon ( 1992 ) . obviously , the new hcn data from a sample more than 5 times larger closely follow the trend established by solomon ( 1992 ) and have about the same scatters . a least squares fit including published data ( almost entirely from solomon 1992 ) , yet excluding our hcn limits , yields a slope of 1.00 @xmath77 0.05 , with a correlation coefficient r=0.94 ( r@xmath78 ) . the correlation implies @xmath79 . when we include all galaxies in the fit and take the limits given in table 4 as the hcn luminosities , the linear relation remains almost the same with @xmath80 and the same correlation coefficient . we have therefore confirmed the tight correlation between ir and hcn luminosities in a large , statistically significant sample . ligs / uligs have the highest @xmath1 and highest ratio of / , and many of them have @xmath1 even larger than @xmath81 of the milky way , which implies a relatively high fraction of dense molecular gas as well as an extremely large reservoir of dense molecular gas ( discussed further in 4.5 ) . the detailed discussion of the tight correlation between @xmath0 and @xmath1 and other relationships , as well as their implications , will be fully explored in paper ii . in this paper , we would like to show that hcn emission is indeed a good tracer of dense molecular gas ( in the next two sections ) . therefore , the tight correlation between far - ir and hcn suggests that active star formation is strongly related to the mass of dense molecular gas and the main source of the strong ir emission in ligs / uligs is mostly from active star formation . in figure 5 , we have divided the sample into two subsamples : ligs and uligs with @xmath82 are plotted as filled circles , and less luminous galaxies are shown as open circles . apparently , all ligs / uligs have larger than 10@xmath83 , whereas the less luminous normal spiral galaxies have @xmath84 . we also note that several uligs have / as high as @xmath85 , whereas most normal galaxies have / nearly 10 times smaller . other systematic differences between these two subsamples can also be easily seen in various other correlations presented in paper ii . we here try to empirically demonstrate that the radiative pumping , or populating , of the j=1 level of hcn through a vibrational transition at 14 @xmath2 m wavelength , does not enhance significantly the excitation of the hcn j=1 - 0 emission . since the mid - ir vibrational transition of hcn falls within the 12 @xmath2 m bandpass and there are no other direct measurements of the 14 @xmath2 m emission available , we use the 12 @xmath2 m flux as an approximation . we show in figure 5 two correlation plots that involve the 12 @xmath2 m emission to illustrate that the enhancement in the excitation of hcn through a vibrational transition , by this potential mid - ir radiative pumping , as postulated by aalto ( 1995 ) , is not important . previous hcn modeling also found that the radiative pumping from the dust continuum emission has little effect on the excitation of the rotational transitions in the vibrational ground state of hcn ( e.g. , stutzki 1988 ; paglione , jackson , & ishizuki 1997 ) . although there is still a good correlation between the 12 @xmath2 m and hcn emission ( a squared correlation coefficient of r@xmath86 ) , the relationship is not particularly strong as compared with the much tighter correlation between ir and hcn emission ( fig . 4 , r@xmath78 ) . in fact , the correlation between the 12@xmath2 m and hcn emission appears to be the weakest among the correlations between the four waveband and hcn emission . both the 60 and 100@xmath2 m luminosities correlate much better with the hcn luminosity . for example , the correlation between 100@xmath2 m and hcn emission ( r@xmath87 ) is significantly better than that between 12@xmath2 m and hcn emission ( r@xmath86 ) . a three - parameter fit of hcn emission as a function of both 12@xmath2 m and 100@xmath2 m emission clearly shows that most of the contribution to the fit is from the 100@xmath2 m emission alone , whereas there is little from the 12@xmath2 m emission . should the mid - ir pumping be an important effect in the excitation of hcn in the vibrational ground state of hcn , a tighter correlation between the 12@xmath2 m and hcn luminosities than any correlations between the hcn and 60@xmath2 m , 100@xmath2 m , and total ir luminosities would be expected . we observe the contrary . we can also use the luminosity ratio , rather than the single luminosity alone , to normalize the correlations in order to eliminate the distance and galaxy size dependencies inherent in the luminosity correlations ( more examples are in paper ii ) . there is only a weak correlation between @xmath88 and / ( r@xmath89 ) , whereas the correlation between @xmath90 and / is still fairly strong ( r@xmath91 ) . clearly , these results suggest that the mid - ir emission is only loosely correlated with the hcn emission , and the correlation between the far - ir emission and the hcn emission is significantly tighter in comparison . this essentially rules out the possibility that the mid - ir pumping in the vibrational ground state of hcn could be one of the significant contributors to the hcn j=1 - 0 emissivity . to strengthen these results , we compare the hcn luminosity and the ratio of hcn and co luminosities ( / ) with the excess 12 @xmath2 m emission estimated by either the ratio of 12@xmath2 m and total ir or the ratio of 12@xmath2 m and 100@xmath2 m ( or 12@xmath2 m and 60@xmath2 m ) luminosities . we plot the luminosity ratio of @xmath92 versus in figure 5a and the luminosity ratios of / versus @xmath93 in figure 5b . obviously , no correlations are observed in these plots and there is no sign of increased hcn emission or an increased / ratio for galaxies with a 12 @xmath2 m emission excess . similar plots for relationships between the luminosity ratios of / and @xmath92 or between / and @xmath94 show no correlations either . therefore , there is little evidence for a direct connection between the 12 @xmath2 m and hcn emission , even for galaxies with excessive 12 @xmath2 m emission ( fig . 5 ) . the contribution of mid - ir radiative pumping to the hcn excitation must be negligible when it is compared to the collisional excitation by dense molecular hydrogen ( see 4.5 ) . gao & solomon ( paper ii ) show that can be quite accurately predicted from the ir ( or far - ir ) and co luminosities . in this three - parameter model fit , appears to be slightly more important , but is also a significant predictor for . here we can add @xmath95 as an additional parameter into the model fit for to check whether @xmath95 is a non - negligible predictor for as well . this could test the significance of the mid - ir emission contribution to hcn as compared with that of the far - ir and co emission . we find that the model fit changes only marginally with no improvement in the correlation coefficient or the rms scatter of the fit . in fact , no difference is noticeable in the model fit whether @xmath95 is added or not . therefore , even though there exists a correlation between 12@xmath2 m and hcn emission , the correlation is significantly worse than the correlations between the hcn and ir emission of three other bands , and the contribution to hcn from 12 @xmath2 m emission is negligible . we conclude that there is no clear evidence to claim that the mid - ir pumping can significantly enhance the hcn emission . for the archetypal agn / starburst hybrid galaxy ngc 1068 , which has the strongest mid - ir emission in our hcn sample , the / ratio is also one of the highest . in particular , a very high hcn / co ratio ( @xmath96 ) within @xmath97 pc nuclear region is observed ( jackson 1993 ; tacconi 1994 ; helfer & blitz 1995 ) in the interferometric maps , which appears to be the mid - ir compact core around agn ( le floch 2001 ) . is this an example that the mid - ir pumping has significantly enhanced the global hcn emission ? apparently not , since the innermost @xmath98100 pc nuclear region around the agn contributes little to the average hcn / co ratios in the circumnuclear ( @xmath990.51 kpc ) starburst ring where most of molecular gas is located . also in ngc 1068 , the ratio @xmath100 is much higher than in any other galaxy in the sample . yet , it is still not certain whether the extreme excess of the mid - ir emission contributes significantly to the excitation of hcn . this is because galaxies with the lowest @xmath92 ratios can have even larger hcn luminosities than ngc 1068 ( fig . 5a ) and galaxies with the lowest @xmath101 ratios appear to have similar or even higher / ratios than ngc 1068 ( fig . thus , globally the high / ratio and high in galaxies may have nothing to do with the mid - ir emission excess . again , there is no indication that the strong mid - ir emission is linked to the high ratio of hcn / co or large hcn luminosity even for galaxies with the most extreme mid - ir emission excesses . the mass of molecular gas in galaxies is typically derived from the co luminosity using the `` standard '' co - to - h@xmath102 conversion factor @xmath103 ( , solomon & barrett 1991 ; young & scoville 1991 and references therein ) , estimated from the galactic disk gmcs in the milky way . although this conversion corresponds to @xmath104 @xmath105 ( k ) @xmath33 , rather than the recent calibration of @xmath106 @xmath105 ( k ) @xmath33 ( dame 2001 ; hunter 1997 ) , we use it simply for easy comparison with previous co studies of galaxies . as discussed in downes , solomon , & radford ( 1993 ) , solomon ( 1997 ) , and downes & solomon ( 1998 ) , however , we caution that it might overestimate the molecular gas mass in uligs by factors of 35 . unlike normal spiral galaxies , where a large fraction of co emission comes mainly from gmcs distributed in the inner disk , the co distribution in uligs is extremely concentrated in the nuclear regions and may fill the entire volume since even the intercloud medium may be molecular ( solomon 1997 ; downes & solomon 1998 ) . high - resolution co imaging of uligs has indeed shown extremely high concentration of co emission in the central few hundred parsecs ( , downes & solomon 1998 ) . for uligs , the conversion factor is thus lower than for normal spiral galaxies . nevertheless , we still adopt the standard co - to - h@xmath102 gas mass conversion ( eq . [ 5 ] ) for convenience since 90% of the galaxies in our sample are not uligs . the ratio of /m(h@xmath102 ) , often referred to as the star formation efficiency , is also listed in table 4 . observations of h@xmath107cn ( nguyen - q - rieu 1992 ; wild 1992 ) indicate that extragalactic hcn emission is also optically thick , just like co emission . the conversion factor between hcn and the mass of the dense molecular gas , however , is poorly constrained as there is no direct calibration from gmcs or gmc core regions in the disk of the milky way . from observations of higher transition cs emission , plume ( 1997 ) find a typical gmc core has a diameter of 1 pc and mass of 3800 . then the rms velocity of a virialized cloud with these parameters is @xmath108 , consistent with the observed fwhm line width . we have carried out large velocity gradient ( lvg ) calculations that include radiative trapping for hcn j=1 - 0 to determine the mass of dense molecular gas from the measured hcn brightness temperatures ( , kwan & scoville 1975 ) , using the collision rates given by green & thaddeus ( 1974 ) . for an hcn abundance ( relative to h@xmath102 ) of @xmath109 ( , irvine , goldsmith , & hjalmarson 1987 ; lahuis & van dishoeck 2000 ) and a velocity gradient of @xmath108 pc@xmath33 , this gives [ hcn]/@xmath110 pc@xmath111 . for a kinetic temperature of 2050 k and the intrinsic hcn line brightness temperature @xmath112 5 k , this gives @xmath113 @xmath105 ( k ) @xmath33 or @xmath114 for h@xmath102 at a density @xmath115 @xmath5 , where @xmath116(h@xmath102 ) means the dense molecular gas as traced by hcn emission , and @xmath117 is the optical depth of the hcn j=1 - 0 line . for higher intrinsic hcn line brightness temperature @xmath118 k , higher gas density n(h@xmath10 @xmath5 , and the same abundance and velocity gradient , however , lvg results give @xmath119 @xmath105 ( k ) @xmath33 , or @xmath120 alternatively , for a virialized cloud core of an average density @xmath121 @xmath5 and @xmath122 35 k ( , radford , solomon , & downes 1991 ) , @xmath123 almost same as that obtained from lvg approximation ( eq . [ 7 ] ) . the lvg calculations above give perhaps an upper limit for the conversion factor between hcn and the mass of dense molecular gas ( cf . mauersberger & henkel 1993 ) since the much lower intrinsic hcn line brightness temperature @xmath124 or 15 k was used besides other approximations in the lvg estimates . higher @xmath125 indeed tends to result in a smaller conversion factor allowed by lvg ( eq . [ 7 ] ) . hcn hot cores in regions of massive star formation usually have extremely high temperature up to hundreds of degrees ( e.g. , boonman 2001 ) . thus , it is likely that a much smaller conversion factor holds under these extreme conditions . moreover , there are some other mechanisms possible to help excite hcn molecule , , excess hcn abundance ( bergin , snell , & goldsmith 1996 ; lahuis & van dishoeck 2000 ) resulting from the active star - forming environment and high supernovae events , and collisions with electrons . although these are most likely insignificant ( , mid - ir radiative pumping discussed in previous section ) and the dominant mechanism should still be collisions with molecular hydrogen , they could work collectively to lower the conversion factor given in equations ( 6 ) and ( 7 ) . other constraints such as the total molecular gas mass estimated from the co observations , the higher frequency co line observations , and the dynamical mass estimations ( , solomon 1997 ; downes & solomon 1998 ) , all seem to point to a smaller conversion factor , particularly for ligs / uligs . thus , a lower conversion factor given in equation ( 8) is more likely . in fact , since the luminosity ratio / in uligs can be as large as @xmath126 ( , mrk 231 , mrk 273 and 17208 - 0014 ) , the true conversion factor between hcn and the mass of dense molecular gas should obviously be much smaller than the upper limit allowed by the lvg calculation ( eq . [ 7 ] ) in order to reconcile with the total molecular gas mass estimated from the standard co - to - h@xmath102 conversion ( eq . [ 5 ] ) . particularly for uligs , even equation ( 8) might still overestimate the dense molecular gas mass since @xmath125 could be indeed very high ( @xmath127 k ) and the standard co - to - h@xmath102 conversion of equation ( 5 ) should be reduced by a factor of 5 in uligs ( downes & solomon 1998 ) . the main difference between normal spiral galaxies and uligs in the hcn - to - h@xmath102 conversion , however , is the drastically different molecular line brightness temperatures . while @xmath128 k is valid for most normal spiral galaxies , a factor of 5 higher @xmath129 k was deduced for uligs undergoing extreme starbursts ( downes & solomon 1998 ) , which argues for a much lower hcn - to - h@xmath102 conversion factor than those given in the above equations . nevertheless , this also indicates that uligs with the highest / ratio are likely to have most ( @xmath130 ) of their molecular gas at high density ( @xmath4 @xmath5 ) and that / ratio can indeed be used as an indicator of the fraction of dense molecular gas in galaxies . on the other hand , / ratio also indicates the average molecular gas density in galaxies . given the large uncertainty of the hcn - to - h@xmath102 conversion , the same as is true for co - to - h@xmath102 conversion , we suggest that a conversion factor of @xmath131 ( eq . [ 8 ] ) is the more reasonable one to use , and we adopt this relation to estimate the total dense molecular gas mass in galaxies as tabulated for /@xmath132 in table 4 . accurate determination of the conversion factor between hcn and the mass of dense molecular gas needs further extensive studies including higher frequency hcn observations ( , jackson 1995 ; paglione 1997 ) and more detailed modeling . we present systematic hcn j=1 - 0 observations ( complemented with co observations ) of a sample of 53 ir / co - bright and/or luminous galaxies , which is the largest hcn sample and most sensitive hcn survey of external galaxies so far . ultraluminous infrared galaxies have the highest hcn luminosities , usually several times larger than the co luminosity of our own galaxy . many luminous infrared galaxies have hcn luminosities comparable to the co luminosity of the galaxy . the ratio of hcn and co luminosities can be as large as @xmath126 in ultraluminous infrared galaxies , nearly an order of magnitude higher than most normal spiral galaxies . we compared the hcn line emission with the far - ir emission ( an indicator of the rate of high - mass ob star formation ) . all galaxies surveyed follow the tight correlation between ir and hcn luminosities initially proposed by solomon ( 1992 ) . since stars form only in dense clouds and the hcn - emitting gas ( i.e. , the dense molecular gas ) and ir emission appear closely and physically related , then active star formation is the dominant source of ir emission in luminous and ultraluminous galaxies . this is fully discussed in paper ii . there is no particularly strong correlation between the 12 @xmath2 m and hcn luminosities . the correlations between 100 @xmath2 m ( and 60 @xmath2 m ) and hcn emission are much better correlated in comparison . galaxies with excess 12 @xmath2 m emission do not show stronger hcn emission or higher hcn / co luminosity ratios . thus , mid - ir radiative pumping of hcn excitation is of little significance when compared with the collisional excitation by dense molecular gas . we also discussed the use of hcn j=1 - 0 emission as a tracer of higher density molecular gas ( @xmath4 @xmath5 ) and adopted a hcn - to - h@xmath102 conversion factor @xmath133 ( k pc@xmath134 . luminous and ultraluminous infrared galaxies usually have much higher hcn brightness temperature , which might result in a lower conversion factor . analyses including lvg calculations indicate that the hcn luminosity measures the mass of dense molecular gas and the hcn / co luminosity ratio indicates the fraction of molecular gas in a dense phase . we thank simon radford , dennis downes , and mark heyer for help with some of the observations . we appreciate the generous support and allocation of observing time from the nrao 12 m , the iram 30 m , and the fcrao 14 m . we would also like to acknowledge edwin bergin , paul goldsmith , and ron snell for providing their lvg code . the careful review and helpful comments of the anonymous referee are greatly acknowledged . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , caltech , under contract with the national aeronautics and space administration . aalto , s. , booth , r.s . , black , j.h . , & johansson , l.e.b . 1995 , a&a , 300 , 369 bally , j. , stark , a.a . , wilson , r.w . , & henkel , c. 1987 , apjs , 65 , 13 bally , j. , stark , a.a . , wilson , r.w . , & henkel , c. 1988 , apj , 324 , 223 bergin , e.a . , snell , r.l . , & goldsmith , p.f . 1996 , apj , 460 , 343 boonman , a.m.s . , stark , r. , van der tak , e.f.s . 2001 , apj , 553 , l63 brouillet , n. , & schilke , p. 1993 , a&a , 277 , 381 carlstrom , j.e . 1988 , in _ galactic and extragalactic star formation , _ ed . rudritz , & m. fich ( dordrecht : reidel ) , 571 carlstrom , j.e . , jackson , j. , ho , p.t.p . , & turner , j.l . 1990 , in _ the interstellar medium in external galaxies _ , eds . hollenbach & h.a . thornson ( washington , dc : nasa cp 3084 ) , 337 condon , j.j . , helou , g. , sanders , d.b . , & soifer , b.t . 1990 , apjs , 73 , 359 curran , s.j . , aalto , s. , & booth , r.s . 2000 , , 141 , 193 dame , t.m . , hartmann , d. , & thaddeus , p. 2001 , apj , 555 , 487 downes , d. , radford , s.j.e . , guilloteau , s. , 1992 , a&a , 262 , 424 downes , d. , solomon , p.m. , & radford , s.j.e . 1993 , apj , 414 , l13 downes , d. , & solomon , p.m. 1998 , apj , 507 , 615 gao , y. 1996 , ph.d . thesis , suny at stony brook gao , y. 1997 , pasp , 109 , 1189 gao , y. , & solomon , p.m. 2004 , , 606 , may 1 ( paper ii , astro - ph/0310339 ) green , s. , & thaddeus , p. 1974 , apj , 191 , 653 helfer , t.t . , & blitz , l. 1993 , apj , 419 , 86 helfer , t.t . , & blitz , l. 1995 , apj , 450 , 90 helfer , t.t . , & blitz , l. 1997a , apj , 478 , 162 helfer , t.t . , & blitz , l. 1997b , apj , 478 , 233 henkel , c. , whiteoak , j.b . , nyman , l .- . , & harju , j. 1990 , a&a , 230 , l5 horellou , c. , casoli , f. , combes , f. , & dupraz , c. 1995 , , 298 , 743 hunter , s.d . , 1997 , apj , 481 , 205 irvine , w.m . , goldsmith , p.f . , & hjalmarson , . 1987 , in interstellar processes , eds . hollenbach & h.a . thronson , jr . ( dordrecht : reidel ) , 561 israel , f.p . 1992 , a&a , 265 , 487 jackson , j.m . , eckart , a. , cameron , m. , wild , w. , ho , p.t.p . , pogge , r.w . , & harris , a.i . 1991 , apj , 375 , 105 jackson , j.m . , paglione , t.a.d . , ishizuki , s. , & nguyen - q - rieu . 1993 , apj , 418 , l13 jackson , j.m . , paglione , t.a.d . , carlstrom , j.e . , & nguyen - q - rieu . 1995 , apj , 438 , 695 jackson , j.m . , heyer , m.h . , paglione , t.a.d . , & bolatto , a.d . 1996 , apj , 456 , l91 kohno , k. , kawabe , r. , & vila - vilaro , b. 1999 , apj , 511 , 157 kohno , k. , kawabe , r. , tosaki , t. , & okumura , s.k . 1996 , apj , 461 , l29 kuno , n. , nakai , n. , sorai , k. 2002 , pasj , 54 , 555 kutner , m. , & ulich , b. 1981 , , 250 , 341 kwan , j. , & scoville , n.z . 1975 , apj , 195 , l85 lada , e.a . , bally , j. , & stark , a.a . 1991 , apj , 368 , 432 lahuis , f. , & van dishoeck , e.f . 2000 , , 355 , 699 le floch , e. , mirabel , i.f . , laurent , o. , charmandaris , v. , gallais , p. , sauvage , m. , vigroux , l. , & cesarsky , c. 2001 , , 367 , 487 lee , y. , snell , r.l . , & dickman , r.l . 1990 , apj , 355 , 536 lee , c.w . 1996 , apjs , 105 , 129 mauersberger , r. , & henkel , c. 1989 , a&a , 223 , 79 mauersberger , r. , & henkel , c. 1993 , rev . modern astron . , 6 , 69 mauersberger , r. , henkel , c. , wilson , t.l . , & harju , j. 1989 , a&a , 226 , l5 mcquinn , k.b.w . , simon , r. , law , c.j . 2002 , apj , 576 , 274 nguyen - q - rieu , nakai , n. & jackson , j.m . 1989 a&a , 220 , 57 nguyen - q - rieu , jackson , j.m . , henkel , c. , truon - bach & mauersberger , r. 1992 , apj , 399 , 521 paglione , t.a.d . , jackson , j.m . , ishizuki , s. , & nguyen - q - rieu 1995a , aj , 109 , 1716 paglione , t.a.d . , tosaki , t. , & jackson , j.m . 1995b , apj , 454 , l117 paglione , t.a.d . , jackson , j.m . , & ishizuki , s. 1997 , apj , 484 , 656 paglione , t.a.d . , jackson , j.m . , bolatto , a.d . , & heyer , m.h . 1998 , apj , 493 , 680 peng , r. , zhou , s. , whiteoak , j.b . , lo , k.y . , & sutton , e.c . 1996 , apj , 470 , 821 plume , r. , jaffe , d.t . , evans , n.j . , ii , 1997 , apj , 476 , 730 radford , s.j.e . , delannoy , j. , downes , d. , 1991 , in iau symp 146 , _ dynamics of galaxies and their molecular cloud distributions _ , ed . f. combes , & f. casoli ( dordrecht : kluwer ) , 303 radford , s.j.e . , solomon , p.m. , & downes , d. 1991 , apj , 368 , l15 reynaud , d. & downes , d. 1997 , , 319 , 737 rice , w. , lonsdale , c.j . , soifer , b.t . , 1988 , apjs , 68 , 91 sage , l.j . 1987 , ph.d . thesis , suny stony brook sage , l.j . , shore , n. , & solomon , p.m. 1990 , apj , 351 , 422 sanders , d.b . , mazzarella , j.m . , kim , d .- c . , surace , j.a . , & soifer , b.t . 2003 , aj , 126 , 1607 sanders , d.b . , scoville , n.z . , & soifer , b.t . 1991 , apj , 370 , 158 sanders , d.b . , & mirabel , i.f . 1996 , ara&a , 34 , 749 sandqvist , a. , jrster , s. , & lindblad , p.o . 1995 , , 295 , 585 shen , j. & lo , k.y . 1995 , apj , 445 , l99 shibatsuka , t. , matsushita , s. , kohno , k. , & kawabe , r. 2003 , pasj , 55 , 87 solomon , p.m. , & barrett , j.w . 1991 , in _ dynamics of galaxies and their molecular cloud _ _ distributions _ , ed . combes , f. , & casoli , f. ( dordrecht : kluwer ) p. 235 solomon , p.m. , downes , d. & radford , s.j.e . 1992 , apj , 387 , l55 solomon , p.m. , downes , d. , radford , s.j.e . , & barrett , j.w . 1997 , apj , 478 , 144 solomon , p.m. , radford , s.j.e . , & downes , d. 1990 , apj , 348 , l53 solomon , p.m. & sage , l.j . 1988 , apj , 334 , 613 sorai , k. , nakai , n. , kuno , n. , & nishiyama , k. 2002 , pasj , 54 , 179 strauss , m.a . , huchra , j.p . , davis , m. , yahil , a. , fisher , k.b . & tonry , j. 1992 , apjs , 83 , 29 stutzki , j. , genzel , r. , harris , a.i . , herman , j. , & jaffe , d.t . 1988 , 330 , l125 tacconi , l.j . , genzel , r. , blietz , m. , cameron , m. , harris , a. & madden , s. 1994 , apj , 426 , l77 tinney , c.g . , scoville , n.z . , sanders , d.b . & soifer , b.t . 1990 , apj , 362 , 473 wild , w. , & eckart , a. 2000 , a&a , 359 , 483 wild , w. , harris , a.i . , eckart , a. , 1992 , a&a , 265 , 447 xie , s. , young , j.s . , & schloerb , f.p . 1994 , apj , 421 , 434 young , j.s . , kenney , j.d . , tacconi , l. 1986 , apj , 311 , l17 young , j.s . , xie , s. , kenney , j.d.p . , & rice , w.l . 1989 , apjs , 70 , 699 young , j.s . , & scoville , n.z . 1991 , ara&a , 29 , 581 young , j.s . , xie , s. , tacconi , l. , 1995 , apjs , 98 , 219 ngc 253 & 2.1 & 4.6 & 0.27 & 0.059 & 77.8 & 9.6 ic 1623 & 46.7 & 130.5 & 8.5 & 0.065 & 55.0 & 7.5 ngc 660 & 3.7 & 7.3 & @xmath1370.26 & @xmath1370.036 & @xmath21142.3 & 10.6 ngc 695 & 46.6 & 92.9 & 4.3 & 0.046 & 108.4 & 10.5 mrk 1027 & 25.7 & 41.7 & 1.89 & 0.045 & 136.0 & 12.9 ngc 891 & 2.6 & 11.0 & 0.25 & 0.024 & 104.0 & 5.4 ngc 1022 & 2.6 & 4.2 & 0.20 & 0.047 & 131.5 & 13.1 ngc 1055 & 2.1 & 13.3 & @xmath210.37 & @xmath210.028 & @xmath13756.9 & 3.3 ngc 1068 & 28.3 & 20.7 & 3.61 & 0.174 & 78.5 & 28.6 ngc 1144 & 25.1 & 108.9 & 2.67 & 0.025 & 94.0 & 4.8 ngc 1365 & 12.9 & 58.7 & 3.10 & 0.053 & 41.6 & 4.6 ic 342 & 1.4 & 9.4 & 0.47 & 0.050 & 30.2 & 3.4 ngc 1614 & 38.6 & 24.5 & 1.25 & 0.051 & 306.8 & 33.2 * viizw31 & 87.1 & 125.0 & 9.8 & 0.078 & 88.9 & 14.6 * 05189@xmath1382524 & 118.1 & 67.0 & 6.2 & 0.093 & 190.4 & 36.9 ngc 2146 & 10.0 & 12.5 & 0.96 & 0.071 & 122.4 & 16.8 ngc 2276 & 6.2 & 10.2 & 0.40 & 0.039 & 156.0 & 12.8 arp 55 & 45.7 & 125.0 & 3.8 & 0.030 & 120.2 & 7.6 ngc 2903 & 0.83 & 2.3 & @xmath1370.09 & @xmath1370.036 & @xmath21104.0 & 7.6 * ugc 05101 & 89.2 & 50.8 & 10.0 & 0.197 & 89.2 & 36.7 m82 & 4.6 & 5.7 & 0.30 & 0.053 & 153.3 & 16.9 ngc 3079 & 4.3 & 24.0 & @xmath121.0&@xmath120.042&@xmath1243.5 & 3.7 * 10566 + 2448 & 93.8 & 61.5 & 10.2 & 0.166 & 92.0 & 32.0 arp 148 & 36.5 & @xmath13747.0 & 4.0 & @xmath210.085 & 91.2 & @xmath2116.2 ngc 3556 & 1.35 & @xmath1374.5 & @xmath1370.09 & @xmath120.020 & @xmath21155.0 & @xmath216.3 ngc 3627 & 1.26 & 4.4 & @xmath1370.08 & @xmath1370.017 & @xmath21160.0 & 6.0 ngc 3628 & 1.01 & 7.1 & 0.24 & 0.034 & 43.5 & 3.3 ngc 3893 & 1.15 & 4.1 & 0.23 & 0.056 & 50.4 & 5.9 ngc 4030 & 2.14 & 15.2 & 0.54 & 0.036 & 40.0 & 3.0 ngc 4041 & 1.70 & 3.9 & 0.18 & 0.046 & 94.8 & 9.1 ngc 4414 & 0.81 & 4.6 & 0.16 & 0.033 & 54.8 & 3.7 ngc 4631 & 2.0 & 2.3&@xmath120.09&@xmath120.039&@xmath12222.2 & 18.5 ngc 4826 & 0.26 & 1.3 & @xmath1370.04 & @xmath1370.030 & @xmath2165.5 & 4.1 ngc 5005 & 1.4 & 8.2 & 0.41 & 0.049 & 21.5 & 2.2 ngc 5055 & 1.1 & 8.6 & @xmath1370.10 & @xmath1370.012 & @xmath21111.0 & 2.7 ngc 5135 & 13.8 & 31.3 & 2.73 & 0.087 & 50.6 & 9.1 m83 & 1.4 & 8.1 & 0.35 & 0.043 & 40.5 & 3.7 * mrk 273 & 129.9 & 65.0 & 15.2 & 0.234 & 85.5 & 41.8 ngc 5678 & 3.0 & 17.2 & 0.75 & 0.044 & 40.0 & 3.6 ngc 5713 & 4.2 & 8.1 & 0.22 & 0.027 & 188.9 & 10.7 ngc 5775 & 3.8 & 10.9 & 0.57 & 0.052 & 66.7 & 7.2 * 17208@xmath1380014 & 234.5 & 146.9 & 37.6 & 0.256 & 62.4 & 33.4 18293@xmath1383413 & 53.7 & 85.5 & 4.03 & 0.047 & 66.6 & 13.2 ngc 6701 & 11.2 & 34.0 & 1.38 & 0.041 & 40.6 & 6.9 ngc 6921 & 11.4 & 17.5 & @xmath122.81&@xmath120.160&@xmath1241.0 & 13.6 ngc 6946 & 1.6 & 9.2 & 0.49 & 0.053 & 33.5 & 3.6 ngc 7130 & 21.4 & 44.9 & 3.27 & 0.071 & 65.4 & 10.0 ic 5179 & 14.1 & @xmath1226.4 & 3.42 & @xmath120.129 & 41.2&@xmath1211.2 ngc 7331 & 3.5 & @xmath13710.7 & @xmath1370.44 & @xmath120.041 & @xmath2179.8 & @xmath216.8 ngc 7469 & 40.7 & 37.1 & 2.19 & 0.059 & 185.4 & 23.0 ngc 7479 & 7.4 & 26.7 & 1.12 & 0.042 & 66.1 & 5.8 * 23365 + 3604 & 142.0 & 85.0 & 15.0 & 0.176 & 94.6 & 35.0 mrk 331 & 26.9 & 52.1 & 3.35 & 0.064 & 80.3 & 10.7

we perform an objective classification of 170,000 galaxy spectra in the sloan digital sky survey ( sdss ) using the karhunen - love ( kl ) transform . with about one - sixth of the total set of galaxy spectra which will be obtained by the survey , we are able to carry out the most extensive analysis of its kind to date . the formalism proposed by connolly and szalay ( 1999a ) is adopted to correct for gappy regions in the spectra , and to derive eigenspectra and eigencoefficients . from this analysis , we show that this gap - correction formalism leads to a converging set of eigenspectra and kl - repaired spectra . furthermore , kl eigenspectra of galaxies are found to be convergent not only as a function of iteration , but also as a function of the number of randomly selected galaxy spectra used in the analysis . from these data a set of ten eigenspectra of galaxy spectra are constructed , with rest - wavelength coverage . the eigencoefficients describing these galaxies naturally place the spectra into several classes defined by the plane formed by the first three eigencoefficients of each spectrum . spectral types , corresponding to different hubble - types and galaxies with extreme emission lines , are identified for the 170,000 spectra and are shown to be complementary to existing spectral classifications . from a non - parametric classification technique , we find that the population of galaxies can be divided into three classes which correspond to early late- through to intermediate late - types galaxies . this finding is believed to be related to the color separation of sdss galaxies discussed in earlier works . bias in the spectral classifications due to the aperture spectroscopy in the sdss is small and within the signal - to - noise limit for majority of galaxies except for the reddest nearby galaxies and large galaxies ( kpc ) with prominent emissions . the mean spectra and eigenspectra derived from this work can be downloaded from http://www.sdss.org .

astro-ph0407061

the most successful scheme used to date to classify galaxies is the morphological classification of hubble @xcite . the utility of this simple classification scheme ( a compression of the available morphological types to approximately seven classes ) has become apparent through its application to numerous extragalactic studies . spectral classifications have a number of natural advantages over the morphological classifications of hubble in that they are more easily related to the physical processes that are ongoing within a galaxy ( e.g. , star formation ) and that they do not require us to obtain high resolution imaging of a large number of galaxies . as such , they are well suited to studying the cores of galaxies in the distant universe . as was found for the classification of the spectra of stars , classifying the spectra of more complicated systems such as galaxies or quasars ( qsos ) can provide a better understanding of the physical processes that determine the formation and evolution of these sources . moreover , if there exist mechanisms by which galaxies can be classified using a handful of representative parameters , this classification can be thought of as a compression of the information contained within the spectra . from such an approach , one might be able to derive simple mechanisms for exploring the physics of the spectral properties of galaxies using large data sets . recent massive spectroscopic surveys , e.g. , the anglo - australian observatory 2-degree - field ( 2df ) galaxy survey ( colless et al . 2001 ) and sloan digital sky survey ( sdss ; york et al . 2000 ) provide us with the opportunity to address the classification of galaxy spectra using hundreds of thousands of galaxy spectral energy distributions ( seds ) . one technique that has gained popularity for studying the distribution of seds is the karhunen - love ( kl ) transform . the power of this approach is that it enables a large amount of data to be decomposed and compressed into independent components in an objective way . applications of this technique can be found in the classifications of galaxies ( connolly et al . 1995 ; folkes et al . 1996 ; sodre & cuevas 1997 ; bromley et al . 1998 ; galaz & de lapparent 1998 ; ronen et al . 1998 ; folkes et al . 1999 ) , qsos ( francis et al . 1992 ; boroson & green 1992 ; yip et al . 2004 ) and stars @xcite . this paper is organized as follows . in section [ section : data ] , we describe the spectral data used in our analysis . in section [ section : klintro ] , we discuss the details of the karhunen - love transform and the gap - correction formalism . in section [ section : conv ] , we pose the problems to be addressed with this paper , and show the results of a convergence analysis on the kl gap - correction formalism . in section [ section : kl ] , we derive the eigenspectra and eigencoefficients for the full sdss data set , and implement a classification scheme . in section [ section : repeat ] , we discuss the reliability of this classification . a simple model is used to describe the population of galaxies in section [ section : model ] . in section [ section : app ] , we discuss the applications of the kl eigenspectra obtained in this work . in section [ section : aperturebias ] , we discuss the aperture bias effect on the current classification scheme . finally , in section [ section : conclusion ] we conclude our results and discuss some possible future directions based on this work . as part of the sloan digital sky survey ( york et al . 2000 ) spectra are taken with fibers of 3 arcsec diameter ( corresponding to 0.18 mm at the focal plane for the 2.5 m , f/5 telescope ) . all sources are selected from an initial imaging survey using the sdss camera described in gunn et al . ( 1998 ) with the filter response curves as described in fukugita et al . ( 1996 ) , and using the imaging processing pipeline of lupton et al.(2000 ) . the astrometric calibration is described in pier et al.(2002 ) . the photometric system and monitoring are described in detail in smith et al . ( 2002 ) and hogg et al . ( 2001 ) respectively . to date , there are three complete samples of sdss spectra : the main galaxy sample ( strauss et al . 2002 ) , the luminous red galaxy sample ( lrgs ; eisenstein et al . 2002 ) , and the qso sample ( richards et al . 2002 ) . from these data we select the main galaxy sample for our analysis and use only those galaxies defined as being of survey quality : a signal - to - noise lower - limit of approximately 16 . the galaxies in this sample have r - band petrosian magnitudes @xmath2 and petrosian half - light surface brightnesses @xmath3 mag arcsec@xmath4 , defined to be the mean surface brightness within a circular aperture containing half of the petrosian flux ( called the petrosian half - light radius ) . the spectral reductions used are the standard sdss 2d analysis pipeline ( idlspec2d v4.9.8 , as of 18th of april , 2002 ) and the 1d specbs pipeline ( schlegel et al . the resultant spectra are flux- and wavelength - calibrated , and sky - subtracted . from these data approximately two hundred galaxies are removed , as they have zero flux in all pixels . this results in a final sample of 176,956 galaxy spectra . the median redshift is about 0.1 , and we find that about 6.5% of the sample have redshifts @xmath5 kms@xmath6 , so that their petrosian half - light radii can be substantially larger than the 3 arcsec aperture of the fiber ( strauss et al . 1,854 spectra of the final sample are found to be duplicated observations ; identified as being within a search radius of 2 arcsec and a redshift tolerance of 0.01 . all spectra are shifted to a common rest frame , and rebinned to a vacuum wavelength coverage of @xmath7 . the binning of the spectra is logarithmic , with a velocity dispersion of 69 kms@xmath6 . this procedure emphasizes the blue end of the optical spectrum , enabling our analysis to focus on the ca h and ca k lines , and the balmer break . the resultant spectra cover rest - wavelength range @xmath7 over 3839 pixels . the karhunen - love transform ( or principal component analysis , pca ) is a powerful technique used in classification and dimensional reduction of data . in astronomy , its applications in studies of multivariate distributions have been discussed in detail ( efstathiou & fall 1984 ; murtagh & heck 1987 ) . in this paper , we limit ourselves to its applications to spectral energy distributions . the basic idea is to derive a lower dimensional set of _ eigenspectra _ @xcite from a very large set of input seds . each sed can be thought of as an axis in a multidimensional hyperspace , @xmath8 , where @xmath9 denotes the @xmath10-th wavelength in the @xmath11-th galaxy spectrum . for the moment , we assume that there are no gaps in each spectrum ; we will discuss the ways we deal with gappy regions later . from the set of spectra we construct the correlation matrix @xmath12 where the summation is from @xmath13 to the total number of spectra , and @xmath14 is the normalized @xmath11-th spectrum , defined for a given @xmath11 as , @xmath15 the eigenspectra are obtained by finding a matrix , @xmath16 , such that @xmath17 where @xmath18 is the diagonal matrix containing the eigenvalues of the correlation matrix . @xmath16 is thus a matrix whose @xmath11-th column consists of the @xmath11-th eigenspectrum @xmath19 . we solve this eigenvalue problem by using singular value decomposition . the observed spectra are projected on to the eigenspectra to obtain the eigencoefficients . in these projections , every pixel in each spectrum is weighted by the error associated with that particular pixel , @xmath20 , such that the weights are given by @xmath21 . the observed spectra can be decomposed , with no error , as follows @xmath22 where @xmath23 is the total number of eigenspectra . it is straightforward to see that @xmath23 equals the total number of wavelength bins in the spectrum . previously , we assumed that the spectra are without any gaps . in reality , however , there are several reasons for gaps to exist : for example , different rest - wavelength coverages , the removal of sky lines , bad pixels on the ccd chips etc . leave gaps at different rest frame wavelengths for each spectrum . all can contribute to incomplete spectra . the principle behind the gap - correction process is to reconstruct the gappy spectrum using its principal components . the first application of the method to analyze galaxy spectra is due to connolly and szalay ( 1999a ) , which expands on a formalism developed by everson and sirovich for dealing with two - dimensional images @xcite . initially , we fix the gap regions by some means , for example , linear - interpolation . a set of eigenspectra are then constructed from the gap - repaired galaxy spectra . afterward , the gaps in the original spectra are corrected with the linear combination of the kl eigenspectra . the whole process is iterated until the eigenspectra converge , which we define in the next section . according to everson and sirovich s work on artificially masked two - dimensional images , they claimed that the iteration process gives convergent images . however , the question about whether the principal components resulting from the correction procedure on realistic gappy galaxy spectra would converge is unknown and is to be addressed with this work . there are some questions to be solved in our analysis before the kl eigenspectra and hence the classification itself become robust and meaningful . these questions are : do the resultant eigenspectra converge and , if so , how many iteration steps are required , what is the dependency of the quality of the kl - repaired spectra on how the gaps are initially corrected , how much information is contained in the eigenspectra and most importantly , how many galaxy spectra are needed in order to derive a convergent set of eigenspectra ? several authors have tried to assess the performance of a kl analysis in a number of different quantitative ways . an example of this is the @xmath24 assessment @xcite in which the authors calculated the difference between the observed spectrum and the spectrum reconstructed with the principal components in order to determine the number of components needed for reconstructing a quasar spectrum . with the implementation of gap - corrections in our analysis , this comparison of only _ one _ spectrum to another may not suffice . we are more interested in how well the _ set _ of eigenspectra describes the distribution of spectra rather than a one - to - one comparison . for example , how does the set of eigenspectra differ as the gap - correction procedure progresses ? given two subspaces , each formed by a set of eigenspectra obtained with different conditions ( e.g. , at different points in the iterative gap correction or computed with different numbers of observed spectra ) , we require a method that will quantitatively compare one set of eigenspectra with another . in other words , instead of just comparing two spectra , a mechanism is desired to compare two subspaces , which are spanned by a finite number of spectra respectively . mathematically it can be stated , as in @xcite , that two spaces , @xmath25 and @xmath26 , are in common if @xmath27 where @xmath28 and @xmath29 are the sum of projection operators of space @xmath25 and @xmath26 respectively , and @xmath30 is equal to the dimensionality of each space . we assume that these eigenbases have the same dimensions for a meaningful comparison . the sum of the projection operators , @xmath28 , of a space is given by the sum of the outer products @xmath31 where @xmath32 are the basis vectors which span the space @xmath25 ( e.g. , merzbacher 1970 ) . a basis vector is an eigenspectrum if @xmath25 is considered to be a set of eigenspectra . the two spaces are disjoint if the trace quantity is zero and are identical if the quantity is equal to the dimension of the subspace . this provides a quantitative way of measuring the commonality of two subspaces ( i.e. , how similar the two subspaces are ) . in investigating the convergence behavior of eigenspectra as a function of the number of iterations , we define one of the spaces to be that formed by a finite number of eigenspectra obtained after initially interpolating over the gap regions , but _ without _ gap correction . the other spaces are defined to be those formed by the same finite number of eigenspectra but at different iterations in the gap correction . the sum of projection operators in the first case is named @xmath33 and in the latter , @xmath34 . the subspaces are named @xmath35 and @xmath36 respectively , where @xmath11 denotes the @xmath11-th iteration . the dimension of the space is @xmath37 , which is the number of eigenspectra forming the subspace . the trace quantity @xmath38 as a function of iteration is plotted in figure [ fig : eigspectracede]a , in which the kl transform is applied to @xmath39 randomly - chosen sdss galaxy spectra , where @xmath37 is set to be @xmath40 and @xmath41 , meaning that the subspaces are spanned by the the first five and the first ten eigenspectra respectively . repairing of the galaxy spectra in the iteration procedure is performed with @xmath42 eigenspectra . it should be noted that @xmath43 is independent from @xmath37 and that @xmath37 is always smaller or equal to @xmath43 . the traces are normalized by the corresponding @xmath37 in each curve to simplify comparison . initially , let us concentrate on the curves with open symbols in figure [ fig : eigspectracede ] . these curves denote that the initial gappy regions are approximated by linear - interpolation . in this linear - interpolation method the flux of a pixel , @xmath44 , in the gap is simply approximated by the average of its neighbors , so that @xmath45 the trace quantity decreases gradually as the iteration step increases , indicating that the space @xmath36 is less and less in common with @xmath35 . this implies that the kl gap - correction and eigenspectra construction are changing the spectrum of a galaxy within the gappy regions . as such the eigenspectra from the kl - repaired spectra differ progressively more from those formed from the original spectra . the above is true for both @xmath46 and @xmath41 . the above is generally valid for @xmath37 from @xmath47 . as the iteration increases , the slope of the curve decreases , which implies that a we have converging set of eigenspectra . the choice of linear - interpolation in the initial correction for the gappy regions is arbitrary . in fact , if the gap formalism is robust , the quality of the kl - repaired spectrum and the eigenspectra should be independent of the way the observed spectra are initially repaired . we test an alternative method of correcting for gaps , where the flux at each wavelength bin in the gap region is approximated by the mean of all other spectra within that region , i.e. , the flux of a pixel in a gap is approximated by , @xmath48 where @xmath49 is the normalized flux at @xmath50 . we call this method mean - interpolation . with this alternate method the trace quantity also converges as we can see from figure [ fig : eigspectracede]a , but at a higher value than those in the case of linear - interpolation . this behavior shows that , in the case of mean - interpolation , the eigenspectra constructed after the gap correction deviate _ less _ from the initial interpolated spectra than in the case of linear - interpolation . the rate of convergence is faster when using the mean - interpolation method . this suggests that the mean - interpolation provides a better initial estimate of the true spectra within the gap regions . we do not , therefore , require as many iterations as in the case of linear - interpolation . this is important as each step in repairing the spectra and constructing the eigenspectra is computationally expensive when large amounts of data are under consideration . figure [ fig : eigspectracede]b shows the convergence behavior of the sets of eigenspectra given in figure [ fig : eigspectracede]a except that the trace quantity now compares the subspace from one iteration to the subspace from the next iteration , i.e. , @xmath51 . as expected , the convergence with the number of iteration steps can be seen in both methods , but with this more sensitive measurement the convergence is now no longer found to be monotonic . this implies that we may need more iterations , than it first appeared from our previous example in order to obtain a convergent set of eigenspectra . again , the mean - interpolation method is shown to converge to a consistent set of eigenspectra faster than for linear - interpolation . consequently , in the following all gaps in the spectra will be fixed using the mean - interpolation method , unless otherwise specified . an example of the actual performance in the mean - interpolated and repaired spectra is shown in figure [ fig : mirepairedspec ] . the data set is the same @xmath52 galaxies as before , except that one randomly chosen spectrum is artificially masked in the region [ 4500 , 5000] . the upper panel of figure [ fig : mirepairedspec]a shows that the mean - interpolated region is already close to that of the original spectrum before masking . the lower panel shows the kl - repaired spectrum at @xmath53 overlaid on the original spectrum , using all @xmath41 eigenspectra . the spectra are offset by an arbitrary amount for illustration . there is a substantial improvement in retrieving the original spectrum as the iterations proceed . to compare the kl - repaired and unmasked spectrum quantitatively , we apply a similar convergence measure as described previously . the convergence measure in this case is defined to be @xmath54 where @xmath55 is the projector of the kl - repaired spectrum with @xmath43 eigenspectra in the gaps , and @xmath56 is that of the unmasked spectrum . in figure [ fig : mirepairedspec]b , the trace quantity versus the number of iterations is plotted for the case corresponding to figure [ fig : mirepairedspec]a . @xmath57 means that the repaired spectrum is identical with the original unmasked one . we see that after the initial few iterations , the two become more similar . after 30 iterations , the kl - repaired spectrum converges to that of the original spectrum with a high degree of accuracy , the difference in the trace quantities of @xmath58 % . combining the results of the convergence measures in both kl - repaired spectra and the previously discussed eigenspectra , it can be concluded that the convergence of the eigenspectra implies the convergence of the repaired spectrum , and vise versa . furthermore , the quality of a repaired spectrum should not depend on the initial gap - interpolation technique . figure [ fig : milirepairedspec ] shows the kl - repaired spectra , using linear- and mean - interpolations for the initial gap approximation , @xmath42 and at @xmath53 . the two are shown to be very similar to each other . the insert shows the corresponding @xmath59 as a function of iteration . the convergence behavior seems different in both cases , nevertheless they are approaching each other with a difference in the actual value about 0.3% which is small as can be seen in the plots of the spectra in the main graph . this is a desired result , because if the whole formalism is robust , the repaired spectrum should not be different due to different procedures used in the initial gap - fixing . another important aspect of the kl - eigenspectra construction is the number of observed spectra necessary as the input . in principal , as we increase the number of galaxy spectra , a more representative and general set of eigenspectra should result . the question remains , however , how much more generality would be gained by including more observed spectra in the analysis ? fundamentally , does there exist a minimum number of input galaxy spectra such that the eigenspectra set start to converge ? this is important because we can thus use a minimum number of randomly - chosen observed spectra in the survey to derive a set of eigenspectra which nevertheless contain all the necessary information within the full data set . figure [ fig : eigentrace_2310_newn ] shows an attempt to answer this question . in these figures , the commonality percentages of two subspaces spanned by ( a ) @xmath60 ( b ) @xmath61 and ( c ) @xmath41 eigenspectra are plotted versus the number of galaxy spectra n@xmath62 used in the sample . the commonality is similar to that previously discussed for the trace quantities except here we compare the set of eigenspectra derived from n@xmath62 input galaxy spectra with that from a smaller number of spectra n@xmath63 . this is defined as follows @xmath64 where @xmath65 is the sum of projectors of the subspace spanned by @xmath66 eigenspectra , derived from @xmath67 galaxy spectra , using @xmath68 eigenspectra for gap - repairing . the number of iterations for the gap - correction is 20 . the smallest number of spectra we consider is @xmath69 ( = n@xmath70 ) and the largest number @xmath71 . the galaxies in each case are randomly selected from the full sdss sample . for all cases , convergent trends are present as more spectra are included . for the case of two eigenspectra ( figure [ fig : eigentrace_2310_newn]a ) we can see that only about @xmath72 galaxy spectra are needed in order to construct the first two eigenspectra to an 0.5% accuracy when compared to the final converged set . the inclusion of the third mode ( figure [ fig : eigentrace_2310_newn]b ) requires more spectra to obtain a similar accuracy though the convergence behavior is very similar to ( a ) ( with slightly larger fluctuations ) . nevertheless , only about @xmath73 galaxy spectra are needed for 90% commonality . these results are consistent with the fact that most types of galaxy spectra can be described with 2 to 3 colors @xcite and therefore a random sampling of a few thousand galaxies can be expected to cover the full color distribution for these galaxies . in figure ( c ) , it is interesting that the convergent behavior is different from that in ( a ) and ( b ) . with a sample size smaller than about @xmath74 to @xmath75 , the improvement in the set of 10 eigenspectra is small . however , once the number of spectra used exceeds that threshold , the convergent rate dramatically increases . this finding suggests that there exists a minimum number of galaxy spectra that we need to include in our kl analysis in order to fully sample the true distribution of galaxy spectra . combining this with the fact that the higher - order modes in the eigenspectra tend to correspond to spectral features in galaxies with prominent emission lines ( this will be discussed in detail in section [ section : kl ] ) and the fact that those galaxies only comprise about 0.1% of the whole sample , the behavior in ( c ) can be understood as the effect of including galaxies with relatively extreme spectral properties . when we randomly pick about @xmath73 galaxies from the sample there are a few emission line galaxies . as more spectra are included we eventually reach a threshold where we begin to sample the extreme emission line galaxies . once we have included a small number of these emission line galaxies ( with a sample size of @xmath76 galaxies we would expect three to four emission line galaxies ) the information they contain is now incorporated within the kl eigenbases . the dramatic increase in the convergence rate come from the fact that , while rare , these extreme emission line galaxies can still be described by a handful of spectral components ( i.e. , once we have a small number of them in the sample we can map out their full distribution ) . to conclude , there exist a minimum number of galaxy spectra we need to observe in order to derive a convergent set of eigenspectra . approximately @xmath77 spectra are sufficient for a 90% convergence level with ten eigen - components ( figure [ fig : eigentrace_2310_newn]c ) . this is sufficient to characterize the spectral types of 99.9% of galaxies within the local universe . these results are , however , purely empirical , based on randomly selecting spectra from the current data set . thus , there is no concrete evidence to support the present result that @xmath77 galaxy spectra are all we need in deriving the most complete set of eigenspectra . there may exist populations of galaxies that comprise much less that 0.1% of the full galaxy sample that our current analysis is not sensitive to . in general , for a larger data set ( e.g. , at the completion of the survey ) , new galaxy types , if any , may call for more spectra to be included when constructing the eigenspectra . the first @xmath41 kl eigenspectra of the 170,000 sdss galaxies are shown in figure [ fig:10eigspec ] , derived from @xmath79 iterations and using 10 eigenspectra for gap repairing . the eigenspectra are publicly available ( from the website http://www.sdss.org ) . the first eigenspectrum is the mean of all galaxy spectra in our sample . the continuum is similar to a sb - type . as we would expect from the mean of all spectra , nebular lines and other emission lines , as well as absorption lines such as ca h and ca k , are present within this spectrum . the second eigenspectrum has one zero crossing , positive toward longer wavelengths , at around 5200 , which marks the wavelength at which the modulation in the continuum level relative to the 1st eigenspectrum occurs . in the third component , there is a zero crossing in the continuum , negative toward longer wavelengths , at around 6000 . the higher the order of the eigenspectrum , the larger the number of zero crossings which in turn adds high - frequency features to the final spectrum as these higher order modes are added or subtracted . in the higher - order modes , the eigenspectra are dominated by emission and absorption lines because each of these eigenspectra comprises emission and absorption lines plus a small fluctuation of the continuum level around zero . we illustrate this point later in the paper . statistically , the amount of information contained in each eigenspectrum is given by the eigenvalue of the correlation matrix of that particular mode . table [ tab : weight ] lists the weights of the first @xmath43-modes of eigenspectra , where the weights are normalized to unity . we find that the first three eigenspectra contain more than @xmath80 of the total variance of the data set . it is known that there is a one - parameter description of galaxy spectra which correlates with the spectral type of a galaxy @xcite . this parameter , @xmath81 , is the mixing angle of the first and second eigencoefficients . explicitly , @xmath82 where @xmath83 and @xmath84 are the eigencoefficients of the first and second modes of a galaxy respectively . furthermore , the inclusion of the third component discriminates the post - starburst activity @xcite . to follow this classification scheme , we define here @xmath85 where @xmath86 is the third eigencoefficient . here we adopt the normalization @xmath87 the first three eigencoefficients of the whole sample are plotted in figures [ fig : a2vsa1 ] and [ fig : a3vsa2 ] in the forms of @xmath84 versus @xmath83 and @xmath86 versus @xmath84 . more than 99% of the total galaxy population is located on the locus in figures [ fig : a2vsa1 ] , in which the second eigencoefficients have values from @xmath88 0.25 to -0.75 . the appearance of this locus is very similar to previous works @xcite . red galaxies have positive , and relatively large second eigencoefficients , while blue galaxies have smaller , or in some cases negative values . from figure [ fig : a3vsa2 ] we clearly see that by introducing the third eigen - component , there is a group of galaxies being separated out from the main group . these galaxies , with negative @xmath86 and negative @xmath84 values , exhibit post - starburst activity in their spectra . a much smaller group ( about 0.1% ) with positive @xmath86 and negative @xmath84 values is also seen . these are outliers and will be discussed later . the resulting @xmath81 versus @xmath89 is plotted in figure [ fig : phitheta ] for all galaxies in the sample excluding those galaxies with @xmath90 ( 118 objects ) . these @xmath90 sources tend to be either artifacts within the data ( m. subbarao , private communication ) or spectra that are not visually confirmed as a galaxy spectrum . in figure [ fig : phitheta]a , the sequence from red to blue to extreme emission line galaxies is illustrated . the boxes drawn show the regions from which a set of spectral types are identified . they range from the early - type at the top of the plot to emission line galaxies at the bottom . the spectra for these subsamples are shown in figure [ fig : meanspec](a - f ) , ranging from red to emission line galaxies . the spectral energy distributions shown are the mean of all the _ observed _ spectra classified to be in the range ( @xmath91 , @xmath92 , @xmath93 , @xmath94 ) , where the superscripts @xmath95 and @xmath96 denote the starting and ending values bounding the range . the actual values are chosen such that the resultant mean spectra agree with the galaxy spectra of each type in kennicutt s atlas of nearby galaxy spectra @xcite . the flux levels are in good agreement with those in the atlas , which leads us to believe that the classification is physically sound as well as having statistical rigor . spectra with similar spectral features are therefore seen to be clustered by the kl procedure . due to the smoothing of spectral inhomogeneities with a large number of galaxies , the resultant mean spectra have very high signal - to - noise levels . this result demonstrates the power of the kl transform for calculating mean ( or composite ) spectra ( e.g. , eisenstein et al . 2003 for the mean spectrum of the sdss massive galaxies ) . table [ tab : number ] shows the number of observed galaxy spectra in each of the regions described previously . we stress that the sum of all galaxies listed in the table is not equal to the total number of galaxies in the data set because the ranges chosen comprise a subset of the full ( @xmath97)-plane . the early late- to intermediate late - types galaxies ( with @xmath98 , @xmath99 ) dominate within the whole data set which agrees with the well - known fact that late - type galaxies dominate the field populations in terms of number counts . apart from the main locus in figure [ fig : phitheta ] the region marked `` ( b ) '' identifies a group of outliers , forming approximately 0.1% of the full sample ( @xmath100 sources ) . these are unusual sources that arise due to artifacts within the reduction pipelines , errors within the spectra themselves or possibly due to new classes of astrophysical sources . in later processing runs ( idlspec2d v4.9.8 , as of 13th of august , 2002 ) only 68 of these sources remain in the main galaxy sample . approximately half of them have the zwarning flag set to 4 , which indicates that there are large errors in the redshift estimations . this results in less than 0.02% of the spectroscopic sample having spectra that can be considered unphysical ( a testament to the remarkable accuracy and performance of the current spectroscopic reduction pipelines ) . considering all of these sources as a whole 90% have signal - to - noise ratios ( s / n ) higher than the mean survey quality ( the @xmath101 is 15.9 in the data set ) . of the remaining 30 galaxies within this outlier class , most have relatively high redshifts ( @xmath102 ) as assigned by the pipelines . some of these sources classed as galaxies by the pipelines do not appear to be galaxy spectra when inspected visually . on the other hand , for those that _ are _ galaxies as inspected by us , we found that the pipelines have assigned incorrect redshifts to some of these spectra . as expected , the gap - repairing procedure fails in those objects and the resulting expansion coefficients have unphysical values . an example is shown in figure [ fig : weirdspec_176455.eps ] , according to the assigned redshift , this object has a redshift @xmath103 , which is obviously incorrect from the locations of + h@xmath104 + lines , as shown in the insert ( this galaxy should have a redshift of @xmath105 ) . the outcome is that the magnitudes of the 2nd and the 3rd eigencoefficients obtained by a kl of all the objects in this group are roughly the same but with different _ signs _ , meaning that no lines that are representative of typical spectral types are found . this result suggests that kl technique is a powerful tool for identifying artifacts within any spectral reduction procedure . the above results show that the classification is successful in allowing the galaxy types to be identified using the first three eigencoefficients and that it may serve as a way for error checking . how do the eigenspectra actually perform in reconstructing the spectra ? figure [ fig:3modes](a - f ) shows , for the same range of ( @xmath106 ) as above , the means of all kl - reconstructed spectra . a kl - reconstructed spectrum , using @xmath43-eigenspectra , is given by @xmath107 it should be noted that the kl - reconstructed spectrum is different from the kl - repaired spectrum we mentioned previously ( in that case the repairing is in the gap regions only ) . for convenience , the _ mean _ of all kl - reconstructed spectra in a given range is abbreviated as `` kl - reconstructed spectrum '' in the following sections unless otherwise specified . comparing the mean spectra in figure [ fig : meanspec](a - f ) with the reconstructed ones in figure [ fig:3modes](a - f ) ( 3 modes are used ) , the continuum levels and most emission lines are in excellent agreement with the mean spectra ( except for the galaxies with extreme emissions , which we will discuss later ) . these results are consistent with previous claims that two eigenspectra are enough to describe most of the spectral types in galaxies @xcite . our present classification scheme of using two mixing angles of the first three eigencoefficients is also justified . the 3-mode kl - reconstructed spectra shown in figure [ fig:3modes](a - f ) also suggest that to reconstruct some of the lines and line ratios , more eigenspectra are necessary . figure [ fig : recon_lines_details ] shows in detail the emission lines that require more than 3 modes for reconstruction . these figures show early - type to late - type galaxies ( from top to bottom ) . with the first eight eigenspectra , the amplitude of the line for galaxies with classification angles in the range ( 7.5 , 20 , 86 , 92 ) can be correctly recovered . similarly , the first four modes are sufficient for the and h@xmath104 reconstruction for galaxies with classification angles in the range ( 5 , 6 , 80 , 100 ) . progressing to bluer galaxies with classification angles in the range ( 0 , 2 , 80 , 100 ) , the first eight modes are enough for the reconstruction . similarly , figure [ fig : recon_lines_details_2 ] shows the cases for galaxies with prominent emission features . we find that the first four modes are enough to reconstruct the amplitudes and line - ratios [ 5008.240]/[4960.295 ] and h@xmath108[4862.68]/[4960.295 ] for those with classification angles in the range ( -12 , -8 , 80 , 100 ) . for galaxies in the range ( -60 , -40 , 120 , 135 ) , the line - ratio [ 6585.27]/[6549.86 ] is correct with three eigenspectra , while the first eight modes are enough to further retrieve the amplitudes of the two lines . the maximum differences between the amplitudes of the mean and the reconstructed lines ( which we define as the error of the reconstruction of a particular line ) in the above - mentioned cases are about 10% . therefore , for all but the most extreme emission line galaxies , eight spectral components , or modes , are sufficient to reconstruct the spectral line ratios to an accuracy of about 10% ( a factor of 500 in compression of information within the galaxy spectra ) . for the reconstruction of galaxies with extreme emissions , however , the performance is not satisfactory when using a small number of eigenspectra . nevertheless , ten eigenspectra are sufficient to recover the continuum level ( see the mean and kl - reconstructed spectra in the enlarged continuum region ) . the residuals of the mean spectra and the kl - reconstructed spectra are shown in figure [ fig : sbmresidue ] , where ( a - d ) correspond to the reconstructions with 3 , 4 , 5 and 10 eigenspectra respectively . there are substantial improvements in using ten eigenspectra , especially in nebular lines and lines , and various line ratios . the typical errors in the fluxes of lines remains around @xmath109% . this is not a surprising result . on one hand , the result follows because of the increasingly dominant role of lines in the higher - order modes compared with the continuum . on the other hand , statistics also play a factor . the early and intermediate - type galaxies dominate the population of galaxies while emission and extreme emission line galaxies comprise just a few percent of the total population . thus , galaxies with significant emission call for more eigenspectra and higher - order modes in their reconstructions . besides the statistical reasons , the inevitable variations in line - widths of emission lines make it comparatively difficult to reconstruct them accurately using linear combinations of eigenspectra . due to the fact that the spectral features in extreme emission line galaxies are distinct from other types of galaxies , they still reveal themselves in the plane ( @xmath97 ) . thus , for the main purpose of this work , which is obtaining a robust and objective classification of galaxies , the less satisfactory performance of reconstructing some emission lines in galaxies with extreme emission lines does not have a significant effect . however , if detailed diagnosis of lines ( for example , the flux - ratio of two lines ) in those galaxies are of interest , then more modes are needed . better yet , a separate analysis using kl with those emission line galaxies is suggested . the inclusion of all the modes in the kl - reconstruction of a given spectrum should , in principle , reproduce all spectral features ( including the noise ) . as higher - order modes contain higher frequency signals and smaller variances of the sample , the inclusion of only a few lower - order modes would thus suppress the noise present in the spectrum . examples of the comparison between the observed spectra and the kl - reconstructed ones are shown in figure [ fig : inspec_reconspe_comprandom](a - c ) . in each case , the spectrum is reconstructed with the first 10 eigenspectra , and normalized at 5500 . from these noise - free reconstructed spectra , it becomes a simple task to identify and classify the spectral lines . any classification scheme has to be repeatable in order to be useful . if we measure the spectrum of a galaxy on different nights in different conditions , would the classification still be the same ? to answer this question , we are fortunate in that many galaxy spectra in the sdss data set are taken on multiple nights @xcite . a total of 1,854 galaxies were found in our sample to be not unique ( i.e. , they have been observed and reduced independently ) . a further thirty thousand galaxies were found in the sdss spectroscopic data to have been observed on multiple nights with different observing conditions ( often these individual observations do not meet the signal - to - noise requirements of the sdss spectral observations ) . the quantitative interpretation of the repeatability of classifications based on these plates may be difficult due to the variation in signal - to - noise ratios . nevertheless all repeat observations are selected for this part of our work . of the thirty - thousand sources , only those with flags primtarget , objtype and class equal galaxy are selected ( together with the requirement that all sources are present in the most up to date reductions ) . this selection results in thirteen - thousand objects in the final sample . figure [ fig : repeat]a and figure [ fig : repeat]b show a comparison between the @xmath81 and @xmath89 values assigned by our classification scheme to those galaxies with the highest signal - to - noise and the classification derived from the repeat observations . the solid line corresponds to the location of the one - to - one correspondence between the two measures . the dispersions in @xmath81 and @xmath89 are @xmath110 and @xmath111 respectively . this agreement is excellent , as these angle dispersions correspond to small changes in the resulting repaired spectra . the agreement also spans a large range in both classification angles . the implication of this finding is that a truly reliable and repeatable classification scheme is obtained which validates the repeatability of the spectrophotometry of the sdss . in order to determine a representative signal - to - noise ratio for each spectrum the median signal - to - noise ratio is adopted ( the flag sn_median in spzbest-_plate_-_mjd_.fits ) . the dependence of the rms error in the measured angles on the signal - to - noise of the observations ( where the signal - to - noise is selected to be the lower one of any pair of observations ) is shown in figure [ fig : sigsnphitheta ] . from the current data set we observe a weak trend with larger discrepancies in the classification for those observations with lower signal - to - noise ratios . the mean ( absolute ) discrepancies in the classification angles ( @xmath112 and @xmath113 ) and mean signal - to - noise ratios are calculated in the ranges of signal - to - noise ratio ( @xmath114 ) , ( @xmath115 ) , ( @xmath116 ) , ( @xmath117 ) and larger than 30.0 . the dependence is very similar in both the @xmath81 and @xmath89 angles . the error bars are set by the root - mean - square fluctuations in both quantities . the vertical line marks the calculated mean signal - to - noise ratio of all the galaxies defined as meeting the survey quality ( a signal - to - noise of 15.9 ) . for those sources meeting the survey quality signal - to - noise criteria , the maximum errors in the two mixing angles are , approximately , three degrees in @xmath81 and two degrees in @xmath89 . this result shows that the classifications based on the sdss spectra are repeatable and robust to the variable signal - to - noise within the spectroscopic data . the fact that the signal - to - noise dependence is weak suggests that the noise within the spectra are essentially poisson such that the projection of a noisy spectrum does not add substantial artifacts into the expansion coefficients . despite this weak dependence in the distribution of expansion coefficients with signal - to - noise we do find instances where the spectral properties of the galaxies change between pairs of observations . for example , in one case we find that the strength of the o ii lines change by about 20% between two separate observations . it is not clear whether this difference is due to a calibration error or due to variability in the source . from studies of the luminosity function of galaxies it has been shown that the distribution of galaxies comprise a number of populations or classes . it is , therefore , natural to ask how many classes are present within the sdss spectroscopic data and how many galaxies occupy each class . we plot in figure [ fig : phihistogram ] the frequency distribution of @xmath81 , ( for the moment we neglect @xmath89 because the extreme emission line galaxies contribute less than a percent to the full galaxy distribution ) . the bin width is @xmath118 and the histogram is normalized to unity . visually , there appear to be two to three dominant `` classes '' or subtypes within the @xmath81-distribution . to further investigate the number of subpopulations , we adopt the akaike information criterion ( aic ) . aic is widely used in model selection in a number of different disciplines . the details of aic and its application in astronomy can be found in connolly et al . basically , in aic , a score is assigned to the model distribution , allowing a quantitative comparison with the true distribution of the data . naturally , more parameters within a model yield a better fit to the data . to counter this the aic penalizes the score based on the number of parameters within a model . the aic score is given by the following , @xmath119 for a given model . in this definition , @xmath120 is the log - likelihood function , and @xmath121 is the number of parameters in the model . as a result , the higher is the score the better the model . the first step in the analysis is to choose a functional form for the model ; a gaussian model is adopted here . we fit models with increasing numbers of gaussian components to the @xmath81-distribution and show the resultant aic score in figure [ fig : aicscore ] as a function of the number of gaussians @xmath122 in the linear mixture model . the insert shows an enlargement of the region around @xmath123 . we find that for @xmath124 , the scores start to flatten off ( 0.01% difference in the aic score ) , whereas the major improvement occurs at @xmath125 . in a statistical sense , @xmath124 give the best score and therefore it would appear that at most six subgroups might contribute to the distribution of the @xmath81 values . we do note , however , that there is no underlying physical motivation for assuming a gaussian mixture model and that as the number of gaussians in the model exceeds four the individual gaussian contain no direct physical meaning . that is to say , the individual populations for @xmath126 actually overlap , forming redundant descriptions . thus , we estimate that a linear model of a mixture of three gaussians is sufficient for modeling the populations of galaxies in our data set . the form of the model is as follows @xmath127 where @xmath128 is the number density ( normalized , @xmath129 ) as a function of @xmath81 , and @xmath130 is a gaussian function of @xmath81 @xmath131}^2 } \ .\ ] ] the three gaussians are illustrated by the dotted lines in figure [ fig : phihistogram ] . comparing the mean values ( in @xmath81 ) of the gaussian distributions with the ranges of @xmath81 over which the mean spectra of different galaxy types are derived ( see figure [ fig : meanspec ] , also see table [ tab : number ] ) , these distributions correspond approximately to early through to intermediate types . because the first two eigenspectra , i.e. , @xmath81 , roughly describe the color of a galaxy , the different sub - populations we obtain should relate to the color separation found in the sdss edr galaxies @xcite in which the bimodal @xmath132 color distribution corresponds to early- ( e , s0 and sa ) and late- ( sb , sc and irr ) types @xcite . besides , optical colors of all galaxies in the sdss were found to be correlated very strongly with @xmath133 color ( i.e. , the @xmath134 color for galaxies at redshift @xmath135 ) , which was also found to be double - peaked @xcite . as we have shown , the eigencoefficients that describe a galaxy spectrum correlate strongly with its intrinsic spectral type . we will leave for a later paper a detailed investigation of the correlations inherent within the eigenbases and their relations to physical spectral energy distribution models such as bruzual and charlot ( 1993 ) . in the following section we will just note a number of the interesting correlations present within the galaxy spectra and eigenspectra . each galaxy spectrum can be constructed through a linear combination of the eigenspectra . while the relative weights of these combinations ( i.e. , the expansion coefficients ) have been shown to provide a basis for the classification of the galaxy spectra , the details of the individual eigenspectra provide insight into the relative correlations between the emission and absorption lines within a spectrum together with its continuum shape . spectral lines that are typically anti - correlated will appear anti - correlated in the the second eigenspectrum ( e.g. , one with positive emission and the other as an absorption feature ) . figure [ fig : pclines ] plots the first three eigenspectra ( with the first eigenspectrum the lower spectrum on the plot ) with the typical emission and absorption lines identified by the sdss spectroscopic pipelines @xcite overlaid . the 2nd and 3rd eigenspectra are flipped and offset by an arbitrary amount to improve the clarity of this figure . what is immediately apparent from this figure is that the majority of the nebula lines are highly correlated . an increase in the star formation rate within a galaxy will result in a general increase in the luminosity of all emission lines . while we expect this correlation in the hydrogen lines , it does not necessarily have to be the case for other lines such as [ ] : the physical processes that give rise to these lines are different ( i.e. , radiative vs collisional excitation ) . the most obvious anti - correlation arises for the na d line at 5800 . the first eigen - component shows the sodium absorption ( commonly associated with neutral gas at a temperature of a few thousand degrees ) to be present in the mean spectrum of galaxies . from the second eigenspectrum , we see that as the star formation rate of the galaxy increases ( i.e. , we add the second component to the mean spectrum increasing the emission line strengths ) the intensity of the na d absorption line decreases . if the majority of the na line comes from stellar lines ( averaged over the 157,000 galaxies in this sample ) then this relation is to be expected due to the increase in the population of o stars with increasing star formation rate . a similar relation is seen for the mg triplet . considering the eigen - components individually we see that the mean galaxy spectrum for the main galaxy sample has a significant component in the h@xmath104 and [ ] emission lines with weaker emission from [ ] and no real evidence for balmer emission below h@xmath108 . adding in the second eigenspectrum has the result of increasing the overall star formation within a galaxy ( i.e. , both the blue continuum increases together with the nebular emission lines ) . the second eigenspectrum has very strong balmer absorption indicative of post starburst activity within a galaxy . the third component is dominated by line emission . there is very little of the stellar absorption of the balmer emission lines as it is seen in the second eigenspectrum . a combination , therefore , of the first and third component will enable the reconstruction of pure absorption or emission line spectra . within the third component the ca k and ca h absorption lines are strongly anti - correlated with the emission lines . increasing the contribution of the third eigenspectrum has the net effect of increasing the line emission together with decreasing the strong absorption line features . this result is understood by the fact that absorption features in a galaxy are mainly due to older stellar populations , and many emission lines , especially nebular lines , are due to the ionization of the interstellar medium within the galaxy by hot stars . it is , therefore , clear that the correlations present within the eigenspectra provide a reasonable description of the physical processes that occur within typical galaxy spectra . a more detailed description of these correlations will be the subject of a followup paper . perhaps the most striking feature within these spectra is that the second eigenspectrum shows the hydrogen emission lines in h@xmath136 , h@xmath137 , h@xmath138 , and h@xmath108 exhibiting stellar absorption . the clarity of this effect comes from the high resolution of the sdss spectroscopic data ( relative to other large spectroscopic samples such as the 2df ) together with the accurate control we have on the spectrophotometric calibration of the individual spectra . figure [ fig : eigspec_selfabs_all ] shows an enlarged region of interest for the first four eigenspectra . comparatively , h@xmath108 is the weakest in terms of this effect , while h@xmath104 shows no apparent effect . the absorption features are also observed in higher - order modes , but they are not shown here since the first few modes dominate . we find that the majority of the signal for the stellar absorption comes from the second eigen - component . there is a smaller contribution from the fourth component but the contribution from this component describes the variation in the widths of the hydrogen lines rather than their amplitudes . as we have shown in figure [ fig:3modes ] using just three modes we can recover galaxy spectra with and without strong stellar absorption . the consequences of this are two - fold . the fourth eigen - component appears not to contribute significantly to the stellar absorption signal , as noted above . secondly , the fact that just two modes can recover the shape of the stellar absorption suggests that the mechanism that describes the magnitude of this process is , on average , relatively simple ( as would be expected given the correlation between the spectral properties and stellar composition of the galaxies ) . this would imply that modeling the stellar absorption and correcting for its effect on the emission line properties of galaxies should be a straightforward process in a statistical sense , even in the presence of low signal - to - noise data . the sdss uses a fixed aperture of 3 diameter for its spectroscopic observations . this can , in principal , lead to biases in the current spectral classification scheme if , for example , a fiber samples only the central bulge of a nearby intermediate or late - type galaxy ( resulting in the assignment of an early - type spectral class ) . as this effect depends on the apparent size of a galaxy when compared to the fiber diameter it has the potential to induce redshift and luminosity dependent biases in any analysis using the kl classifications @xcite . studies of the effect of aperture bias on observed parameters ( e.g. , star - formation rate ) can be found in , e.g. , baldry et al . ( 2002 ) , prez - gonzlez et al . ( 2003 ) and brinchmann et al . the questions we address here are : @xmath139 is there an aperture bias using the kl approach @xmath140 how can we quantify this bias and @xmath141 can we correct for the aperture effects to obtain bias - free galaxy types ? we estimate the effect of aperture bias by calculating , for a given galaxy , the difference in the classification ( in this case the @xmath81 angle ) derived from the total galaxy flux compared to that derived from the central 3 arcseconds . the dependence of this classification error on the redshift and the physical size of the galaxy serves to quantify the bias in our sample . we assume that the apparent diameter of each galaxy can be approximated as twice the petrosian half - light radius ( petro50 ) in the r - band . the physical sizes of the galaxies are then calculated by assuming @xmath142 , @xmath143 and @xmath144 . the aperture magnitudes of all galaxies are initially k - corrected to redshift @xmath135 using the code by blanton et al . ( 2003b ) version 1.16 prior to estimating the spectral types . type assignment for the total flux and fiber flux is performed using the photometric redshift code of connolly et al . ( 1999b ) . the input spectral templates are constructed as linear - combinations of the first 3 eigenspectra from this work , with the resolution in both @xmath81 and @xmath89 set to @xmath145 . in the following discussion we will express the distance dependence of the relation as function of @xmath146 , where @xmath147 is the highest redshift at which a galaxy of a given absolute magnitude would pass the sample selection criteria . this provides a pseudo volume independent analysis . figure [ fig : dphiallgalcontour ] shows the difference in the classifications of galaxies , @xmath81(total)@xmath148(3 ) ( @xmath149 ) , as a function of @xmath146 and galaxy type . the bin sizes of smoothing are 0.02 in @xmath146 and @xmath145 in @xmath150 . galaxies of sizes from @xmath151 kpc are included , whereas galaxies of @xmath81(total)@xmath152 are excluded for there are less then 1% of them . lighter components in the greyscale image correspond to the fraction of galaxies that would be classified as an earlier type ( i.e. , redder ) if the total flux was used rather than the 3 arcsec flux . darker components correspond to galaxies that are of later type ( bluer ) when using the total flux . the percentages of galaxies residing within these contours are listed in table [ tab : aperturebiasnum ] . from our repeatability test , the mean signal - to - noise limited classification is @xmath153 . with the assumption that the typical signal - to - noise limit in the @xmath81 angle estimation for the whole galaxy is the same as that for the inner 3 , the derived signal - to - noise limit in @xmath150 is @xmath154 . there are about half of the galaxies ( @xmath88 40% ) in our sample in which the type - differences are within the estimated signal - to - noise limit . aperture effects on the spectral classification clearly exist . for blue galaxies ( i.e. , @xmath155 ) , @xmath150 increases for nearby galaxies . this is to be expected as the flux from the inner 3 arcsec is more likely to be dominated by the presence of a bulge component . similarly , for galaxies classified as red based on their total flux ( i.e. , @xmath156 ) , errors in the classification angles @xmath150 increase rapidly with decreasing distance ( i.e. , @xmath157 ) . this implies that the cores of red galaxies are redder than the color estimated from the total flux . the dependence of the aperture bias on the physical size of a galaxy is illustrated in figure [ fig : dphivszoverzmax6types ] . we divide the above sample into 6 ranges of galaxy type , from the reddest in figure ( a ) to the bluest in figure ( f ) . the differences in the classification angles @xmath150 are plotted in each figure as a function of @xmath146 for physical sizes ranging from @xmath151 kpc ( black line ) , @xmath158 kpc ( dotted line ) and @xmath159 kpc ( dashed line ) . the two horizontal lines mark the uncertainty on the classification due to the survey signal - to - noise limits . for the red galaxies in figure ( a ) to ( c ) the bias is constant or decreases with effective distance and is , essentially , negligible when compared to the uncertainties on the classification . for distances @xmath157 , the bias is above the signal - to - noise limit so that the type deduced from the total flux is redder than that from the central 3 . as we would expect figures ( b ) and ( c ) show a dependence on galaxy size for the classifications with larger galaxies exhibiting a redder classification when only considering the 3 arcsec flux . this size and redshift dependence extends to the blue galaxies ( figure ( d ) through ( f ) ) . for these galaxies , however , a more pronounced dependency is shown on the classification bias with galaxy size . overall there is a general aperture bias for all physical sizes of galaxy ( @xmath151 kpc ) that approaches the intrinsic error on the classification as the effective redshift @xmath146 approaches unity . the exception to this arises when we consider galaxies with prominent emission lines ( figure ( f ) ) which , counter - intuitively , exhibit _ larger _ bias the more distant they are . one of the possible reasons for this is our use of three eigenspectra in constructing the spectral templates , whereas 10 modes and more are typically required to accurately reconstruct these observed spectra . while we observe an aperture bias in the sdss sample it is relatively small when compared to the intrinsic classification errors and is essentially negligible for most galaxies . moreover its dependence on size and redshift is relatively mild and straightforward to correct . for those galaxies with non - negligible bias a simple correction can be made using the lookup table shown in figure [ fig : dphiallgalcontour3arcsec ] , which is identical to figure [ fig : dphiallgalcontour ] with the exception that the ordinate axes is the spectral type inside 3 , @xmath160 . given the @xmath160 and @xmath146 the correction to our spectral classification can be determined directly from figure [ fig : dphiallgalcontour3arcsec ] with the size - averaged bias . from the application of karhunen - love transform , an objective classification of @xmath88170,000 galaxy spectra in the sdss is performed . with a quantitative convergence criteria defined , gappy galaxy spectra can be repaired and kl eigenspectra and eigencoefficients derived . for most of the galaxy types , three eigenspectra are sufficient for describing the continua and emission lines to a high degree of accuracy with a maximum error in line - reconstructions of approximately 10% . typically ten modes are needed in the reconstruction of galaxies with extreme emission lines with errors of @xmath109% in the line fluxes . we find that a two - parameter @xmath161-classification scheme can discriminate between spectra corresponding to all spectral types used in the current classification scheme ( including galaxies with extreme emission lines ) . this classification is robust to repeat observations ( at a level of a few degrees in the classification angles ) due to the accurate spectrophotometric calibration of the sdss data set . we find a weak dependence in the classification on the signal - to - noise of the spectra . this effect is , however , smaller than the typical dispersion between repeat observations and is negligible at signal - to - noise levels at which the sdss spectra are defined as being of survey quality . we find that there exists a minimum number of randomly selected spectra that are necessary to statistically represent the information within the full sample ( i.e. , to be representative of the true distribution of galaxies ) . for a set of ten eigenspectra ( i.e. , ten eigenspectra enable the reproduction of both quiescent and active galaxies ) the number of spectra required is around @xmath74 to @xmath75 . this is due to the need to sample a minimum number of randomly selected galaxies in order to include galaxies with extreme emission line properties in our data set ( as they comprise only 0.1% of the full galaxy sample ) . we find that the bias on the spectral classification due to the fixed aperture spectroscopy is , on average , small and is negligible for all galaxies except for the reddest galaxies that are very close by ( @xmath162 ) and for those galaxies that are large physically ( @xmath163 kpc ) with prominent emission lines . a look - up table is constructed for the correction of this bias . there are several future directions related to this work . with the present continuous classification scheme , which simplifies the distribution of galaxies into a handful of parameters , studies of the statistics of the physical properties of galaxies become more tractable . the clustering and spectral properties of these classifications will be addressed in a future paper . the generality of these techniques are applicable to any set of spectra and has been recently applied to the sdss qso catalog ( yip et al . 2003 , 2004 ) . we thank mariangela bernardi , ravi sheth and michael blanton for comments and discussions . cwy is partially supported by zaccheus daniel fellowship . ajc and cwy acknowledge partial support from an nsf career award ast99 84924 , a nasa ltsa grant nag5 8546 and a nsf itr award 0121671 . funding for the creation and distribution of the sdss archive has been provided by the alfred p. sloan foundation , the participating institutions , the national aeronautics and space administration , the national science foundation , the u.s . department of energy , the japanese monbukagakusho , and the max planck society . the sdss web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium ( arc ) for the participating institutions . the participating institutions are the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , university of pittsburgh , princeton university , the united states naval observatory , and the university of washington . connolly a. j. , budavri t. , szalay a. s. , csabai i. & brunner r. j. 1999b , in asp conf . 191 , _ photometric redshifts and high redshift galaxies _ , ed . weymann r. j. , storrie - lombardi l. j. , sawicki m. , & brunner r. j. ( san francisco : asp ) , 13 . lupton r. , gunn j. e. , ivezi z. , knapp g. r. & kent s. 2001 , in asp conf . 238 , _ astronomical data analysis software and systems x _ , ed . harnden f. r. jr . , primini f. a. , & payne h. e. ( san francisco : astr . spc . pac . ) , p. 269 . .the number of galaxies in the range ( @xmath164 ) . these data are a subset of the full sample . the galaxy types listed are the possible morphological types , estimated by comparing the spectral features of the mean spectrum constructed in each range with spectra in @xcite and therefore they are for reference only . [ tab : number ] [ cols="^,^,^,^,^",options="header " , ]

in an all - pay auction , only one bidder wins but all bidders must pay the auctioneer . all - pay bidding games arise from attaching a similar bidding structure to traditional combinatorial games to determine which player moves next . in contrast to the established theory of single - pay bidding games , optimal play involves choosing bids from some probability distribution that will guarantee a minimum probability of winning . in this manner , all - pay bidding games we d the underlying concepts of economic and combinatorial games . we present several results on the structures of optimal strategies in these games . we then give a fast algorithm for computing such strategies for a large class of all - pay bidding games . the methods presented provide a framework for further development of the theory of all - pay bidding games . + michael menz^1^ , justin wang^2^ , jiyang xie^3^ + * 3>p.3 ^1^yale university & ^2^yale university & ^3^yale university [email protected] & [email protected] & [email protected]

1504.02799

at the conclusion of an all - pay auction , all bidders must pay the bids they submitted , with only the highest bidder receiving the item . with this idea in mind , one can play a variant of a two - player game using an all - pay auction to decide who moves next instead of simply alternating between players . for example , one could play all - pay tic - tac - toe with @xmath0 chips . each round both players privately record their bids and then simultaneously reveal them . if player a bids @xmath1 and his opponent bids @xmath2 , player a would get to choose a square to mark and the next round of bidding would begin wih player @xmath3 having @xmath4 chips , player @xmath5 having @xmath6 chips . note that the chips have no value outside the game and only serve to determine who moves - the ultimate goal is still just to get three - in - a - row . another variant of the game could have only the player who wins the move pay his / her bid , i.e. deciding who moves next using a first - price auction . these games were first studied formally in the 1980s by richman , whose work has since then been greatly expanded upon . intuitively , there is less risk in these `` richman games '' for the player losing the bid . if your opponent bids @xmath0 for a certain move , it makes no difference whether your bid was @xmath7 or @xmath8 . all that matters is that your opponent s bid was higher . a surprising consequence of this single - pay structure is that for every state of a game , there exists a `` richman value '' @xmath9 for each player that represents the proportion of the total chips that player would need to hold to have a deterministic winning strategy . in this situation , the player with the winning strategy can tell her opponent what bid she will be making next without affecting her ability to ultimately win . for zero - sum games , this means that unless a player s chip ratio is exactly @xmath9 , then one of the players must have such a winning strategy @xcite , @xcite . our objective is to begin the formal study of all - pay bidding games . returning to the above example where your opponent bids @xmath0 chips and you are indifferent between bidding @xmath7 and @xmath8 , it is clear this is no longer true for an all - pay bidding mechanism . you would be very disappointed had you bid @xmath7 , as your opponent would be paying just @xmath10 chip on net to make a move . had you bid @xmath8 , though , you might feel pretty good about not moving this current turn , as the @xmath0 extra chips may make a bigger difference for the rest of the game . thus , there are at least two bidding scenarios which intuitively seem like very good positions to be in : winning the bid by a relatively small number of chips or losing the bid by a relatively large number of chips . this behavior suggests that , unlike in richman games , in all - pay bidding games one of the players will not necessarily have a deterministic winning strategy . instead , players must randomize their bidding in some way . thus , we must appeal to the concept of mixed bidding strategies in nash equilibria . before presenting formal definitions and results , we provide a sample all - pay bidding game to illustrate some of the main features of playing these games . alice and bob , each with @xmath0 chips , are playing all - pay bidding tic - tac - toe . each turn alice and bob secretly write down a bid , a whole number less than or equal to their total number of chips . they then reveal their bids and whoever bid more gets to decide who makes the next move . we say a player has * advantage * if , when players bid the same amount , that player gets to decide who makes the next move . the question of deciding how to assign advantage is one we encountered early on . for our games , we give advantage to the player with more chips , then arbitrarily let alice have advantage when alice and bob have the same number of chips . a number of other mechanisms would also suffice , such as alternating advantage or having a special `` tie - breaking '' chip that grants advantage and is passed each time it is used . our choice was made in the interest of computational simplicity and to eventually allow extension to real - valued bidding . _ first move . _ both players have @xmath0 chips . alice bids @xmath2 , bob bids @xmath1 . bob wins the right to move and plays in the center of the board . _ second move . _ alice has @xmath6 chips , bob has @xmath4 chips . alice wants to win this move to keep pace with bob , but also does not see why it should be worth more than the first , so she only slightly increases her bid to @xmath11 . bob , thinking that alice may want to win this move more , is content to let alice win and collect chips by bidding @xmath8 . alice wins the right to move and plays in the top - left corner of the board . _ third move . _ alice has @xmath4 chips , bob has @xmath6 chips . alice bids @xmath12 , bob bids @xmath1 , so alice wins the right to move and plays in the top - center of the board . _ fourth move . _ alice has @xmath13 chips , bob has @xmath14 chips . alice is one move away from winning and decides to risk it and bid all of her @xmath13 chips . unfortunately for her , bob has guessed her move and has himself bid @xmath13 as well . because bob has more chips overall , he uses his advantage to win the tie and plays in the top - right corner of the board , blocking alice s victory and setting himself up for one . _ fifth move . _ alice has @xmath13 chips , bob has @xmath14 chips . bob has more chips and is just a move away from winning , so he can bid everything , play in the bottom - left corner and win the game . in normal tic - tac - toe , both players can guarantee a draw by playing well , but as we see from this example , the result of a game of all - pay tic - tac - toe involves far more chance . for example , at the fourth move in the above game , alice could have guessed bob might bid @xmath13 and chosen to `` duck '' by bidding @xmath8 . in this case bob would win the move and play as before , but now the chip counts would be @xmath15 to @xmath1 in alice s favor , and alice can bid @xmath1 and then @xmath13 to win the next two moves and win in the left column . it is easy to see that if a player knows what his opponent will bid at each move , he can win the game easily . thus , in the vast majority of all - pay bidding games , optimal play can not be deterministic . though we do not return to tic - tac - toe in this paper , it served as a test game for much of our research . using our results , we built a computer program to play all - pay bidding tic - tac - toe optimally . the program can be played against at http://biddingttt.herokuapp.com . the theory behind this program , which is not specific to tic - tac - toe , will be the focus of the rest of the paper . our ultimate goal is to characterize the optimal strategies for a general class of all - pay bidding games . the game consists of iterations of both players bidding for the right to move followed by one of the players making a move . in turn , an optimal strategy will also have two parts : the bid strategy and the move strategy . for a given position in the game ( e.g. a configuration of the tic - tac - toe board ) and chip counts for each of the players ( e.g. alice has @xmath6 chips , bob has @xmath4 chips ) , the bid strategy must tell players how to best randomize their bets ( e.g. alice bids @xmath8 chips half the time , @xmath13 chips half the time ) while the move strategy must tell whoever wins the bid the best move to make ( e.g. where to play on the tic - tac - toe board ) . the problem of determining move strategy is largely combinatorial in nature and remains similar to its analog in richman games . we can still represent the space of game states as a directed graph , and there is a not always a single best move that each player can make upon winning the bid . that is , the best move could also depend on each player s chip counts moving forward . the focus of our work , then , will be on determining the optimal bidding strategy for any game position and chip counts . naturally , this should depend on a player s chances of winning in any of the possible subsequent game situations ( i.e. after a single move and updated chip counts ) . for purposes of initial analysis , we will assume that these future winning probabilities are known , and see how the bidding strategy can be determined from this information . then , by using the recursive nature of the directed graph , we will be able to start from the `` win '' and `` loss '' nodes ( where the probability is just @xmath10 or @xmath8 ) to find the optimal bidding strategies and winning probabilities for any game situation . for the rest of this paper , we will often refer to a bidding strategy as just a `` strategy '' when it is clear that the focus is just on the bidding side of the game . here , a strategy will be a probability vector where the @xmath16th coordinate corresponds to the probability a player will bid @xmath16 chips . further , a nash equilibrium for a game situation will just be a pair of strategies so that neither player has an incentive to deviate . this means that each player s strategy will maximize his / her minimum probability of ultimately winning from the next turn of the game . it quickly becomes apparent that a naive recursive algorithm using linear programming is feasible only for games with very few moves . thus , in the interest of being able to practically calculate the optimal bidding strategies for general games , we prove some structural results on the nash equilibria . in particular , useful structure arises when we study a particular class of games that we dubbed `` precise '' , which roughly speaking are games where having one more chip is strictly better than not . the key result is a surprising relationship between opposing optimal strategies that allows one to immediately write a nash equilibrium strategy for the player without advantage if given a nash equilibrium strategy for the player with advantage . this relationship , ( [ reverse_thm ] ) , which we call the reverse theorem , is a critical step toward the calculation of optimal strategies for precise games . further , by assigning an arbitrarily small value in the game to each chip , we get a precise game that is very similar to the original game . we show that the optimal strategies we can calculate for these new precise games will indeed converge to optimal strategies for our possibly imprecise games . our theoretical results ultimately culminate in a fast algorithm for computing optimal probabilistic bidding strategies . together with a move strategy for the combinatorial side of the game , this gives a complete characterization of optimal play for all - pay bidding games . let @xmath17 denote a single turn of a two - player all - pay bidding game @xmath18 where player @xmath3 is endowed with @xmath19 chips and player @xmath5 is endowed with @xmath20 chips . the underlying combinatorial game @xmath18 is a two - player zero - sum game , represented by an acyclic , colored , directed graph with two marked vertices , @xmath21 and @xmath22 . the game begins by placing a token at some starting vertex . at each turn , a player moves the token to an adjacent vertex . player @xmath3 wins if the token reaches @xmath21 and player @xmath5 wins if the token reaches @xmath22 . by saying the graph is colored , this means that edges are one of two colors , say red and blue , such that @xmath3 can only move the token along red edges and @xmath5 can only move the token along blue edges . to ensure consistency in the bidding strategy from turn to turn , we seek to avoid situations where the winner of a bid can be put in zugzwang - i.e. where it would be better to not move at all . thus , the bid winning player , rather than simply being able to move next , gets to determine who moves next . with this condition , @xmath18 can an asymmetric game where zugzwang is possible , like chess and many other popular two player games . the * payoff * , or value of the game , for player a at @xmath17 is denoted by @xmath23 $ ] and is equal to the probability that player a wins the game under optimal play . that is , we set @xmath24 and @xmath25 and calculate payoffs recursively . similarly , let @xmath26 denote the probability that player @xmath5 wins the game . often , when the chip counts or specific combinatorial game are not relevant to the discussion , the payoffs will be shortened to @xmath27 and @xmath28 . note that @xmath29 as we only study games that can not end in ties ( for the game of tic - tac - toe above , we can arbitrarily let one of the players win all draws ) . thus a * payoff matrix * for player a in @xmath17 is denoted by @xmath30 and is given by @xmath31 where @xmath32 and @xmath33 are the set of game positions that can be moved to from @xmath34 by @xmath3 and @xmath5 respectively . the @xmath35 entry corresponds to player @xmath3 s probability of winning the game after @xmath3 bids @xmath36 and @xmath5 bids @xmath16 . note this is well - defined because the game is zero - sum : by moving to the game state that minimizes player @xmath3 s payoff , player @xmath5 is maximizing his own payoff at the same time ( and vice - versa ) . similarly , let @xmath37 denote the payoff matrix for player @xmath5 . we notice that if player @xmath3 bids @xmath38 and player @xmath5 bids @xmath39 , this is equivalent to player @xmath3 bidding @xmath40 and player @xmath5 bidding @xmath41 for any @xmath42 because the players are paying each other . thus , we have that payoff matrices are toeplitz , or diagonal - constant . we will write player @xmath3 s and player @xmath5 s payoff matrices for @xmath17 as @xmath43 respectively . we pause to consider a simple example . let the underlying game be one where player @xmath3 needs to make two moves to win , while player @xmath5 needs to make only one more move to win . suppose player @xmath3 has @xmath44 chips while player @xmath5 has @xmath45 chips . then we would get the following payoff matrices for player @xmath3 and player @xmath5 @xmath46 respectively a * strategy * for player a in @xmath17 is given by an @xmath47-dimensional column vector with all non - negative entries that sum to 1 . the @xmath16-th entry of this vector ( where we start indexing at @xmath8 ) gives the probability that player @xmath3 will bid @xmath16 chips . similarly , a * strategy * for player @xmath5 in @xmath48 and is given by a @xmath49-dimensional column vector satisfying the same conditions . we denote a nash equilibrium strategy in the game @xmath48 as @xmath50 for player @xmath3 and as @xmath51 for player @xmath5 . often times we will not be too explicit with the size of these vectors . it should be clear from context . note that the @xmath16th row of @xmath52 corresponds to the payoffs of each of @xmath3 s pure strategies if her opponent @xmath5 bids @xmath16 . letting @xmath53 be the @xmath16th row of @xmath52 , we have @xmath54 , a weighted average of @xmath3 s pure payoffs when @xmath5 bids @xmath16 . thus , @xmath55 is player @xmath3 s probability of winning if her strategy is @xmath56 and her opponent purely bids @xmath16 . for example , if we have @xmath57 this means by playing @xmath56 , player @xmath3 wins @xmath58 of the time if player @xmath5 only bids @xmath8 and wins @xmath59 of the time if player @xmath5 only bids @xmath10 . now , if player @xmath5 s strategy is @xmath60 , @xmath61 , another weighted average of @xmath3 s payoffs for each of @xmath5 s pure strategies . thus , @xmath62 is exactly @xmath3 s payoff if she plays @xmath56 and her opponent plays @xmath60 . @xmath63 is @xmath5 s payoff in the same situation . continuing with the above example , if we now let @xmath64 then @xmath65 . so given strategies @xmath56 and @xmath60 for players @xmath3 and @xmath5 , player @xmath3 has a @xmath66 probability of winning . we compile these results in the lemma below . [ basics ] let @xmath67 and @xmath68 be payoff matrices for players @xmath69 and @xmath70 , respectively , in @xmath71 . then the following statements are true . * the diagonals of @xmath67 and @xmath68 are constant , i.e. the payoff matrices are toeplitz . * let @xmath72 be the appropriately sized matrix whose entries are all 1 . then @xmath73 . * suppose @xmath74 is a nash equilibrium . then @xmath75 if @xmath76 and @xmath77 if @xmath78 . this lemma provides the basic structure from which many of our main proofs will follow from later . it is clear that @xmath79 , since player @xmath3 can always bid as if he did not have the extra chip . we now define a class of games pivotal to our analysis in which this inequality is strict . formally , a game @xmath34 is called * precise * if in every successor state to @xmath34 , it is strictly better to have one more chip . [ precise_matrices ] we note that in particular , this guarantees a certain strict monotonicity among the entries of the payoff matrices . in particular , winning the bid by one less chip is always strictly preferable , as is losing by one more chip . thus we have that for the player with advantage , @xmath80 for @xmath81 and @xmath80 for @xmath82 . a similar relationship holds for the player without advantage , except @xmath83 and @xmath84 . a strategy @xmath85 has * length * @xmath86 if @xmath87 and @xmath88 @xmath89 . a strategy @xmath85 is * gap - free * if @xmath90 if and only if @xmath91 @xmath92 . the definition of length encapsulates the observation that unless the game is close to completion , players will never bid a large proportion of their chips . the second definition seems more arbitrary at the moment , but it plays a pivotal role in the following proposition and will serve to greatly simplify the language throughout the paper . [ grounded_gapfree ] let @xmath71 be precise . any equilibrium strategy for the player with advantage is gap - free and bids @xmath8 with nonzero probability , while any equilibrium strategy for the other player is gap - free and bids 1 with nonzero probability . if the player with advantage has an equilibrium strategy of length @xmath93 , any equilibrium strategy for the other player has length @xmath93 or @xmath94 . suppose without loss of generality that player @xmath69 has advantage , and let @xmath95 and @xmath96 be equilibrium strategies for players @xmath69 and @xmath70 respectively . we claim that if @xmath97 , * @xmath98 implies @xmath99 , and * @xmath99 implies @xmath100 . if @xmath98 and @xmath101 , player @xmath70 should alter his strategy so that he bids @xmath102 with probability @xmath103 and @xmath104 with probability 0 . this saves player @xmath70 a chip whenever he would have bid @xmath104 without changing any possible outcome of these bids , and all other possibilities are unchanged . by precision , this new strategy is strictly better than @xmath105 for player @xmath70 , a contradiction . this proves ( i ) . if @xmath106 and @xmath107 , player @xmath69 should alter her strategy so that she bids @xmath102 with probability @xmath108 and @xmath104 with probability 0 . as in the previous case this new strategy is strictly better for player @xmath69 , a contradiction , proving ( ii ) . together , ( i ) and ( ii ) complete the proof except in the case when @xmath109 . however , in this case an optimal strategy for player @xmath69 is to also bid 0 with probability 1 , and it follows that @xmath110 is not precise . this characterization of equilibrium strategies is what motivated our restriction to precise games . in the presence of precision , an easily observable , yet highly unexpected relationship between opposing optimal strategies appears . this relationship forms the foundation for the rest of our results . the * reverse * of a length @xmath111 strategy @xmath85 is given by @xmath112 where the number of trailing zeroes will be clear from context . [ reverse_thm ] suppose that @xmath71 is precise , and that @xmath113 is an equilibrium strategy for the player with advantage . then @xmath114 is an equilibrium strategy for the player without advantage . suppose without loss of generality that player @xmath69 has advantage , and @xmath115 has length @xmath116 . by lemma [ basics ] and proposition [ grounded_gapfree ] , we have @xmath117 \left [ \begin{array}{c } s_0 \\ s_1\\ \vdots \\ s_{\ell-1 } \\ 0 \\ \vdots \\ 0\\ \end{array } \right ] = \left [ \begin{array}{c } w_0 \\ v_a \\ \vdots \\ v_a \\ w_{\ell } \\ \vdots \\ w_{b } \end{array } \right],\ ] ] where @xmath118 . we claim further that @xmath119 . suppose for a contradiction that @xmath120 . then if @xmath105 is an equilibrium strategy for player @xmath70 , by lemma [ basics ] and proposition [ grounded_gapfree ] it is of the form @xmath121 where @xmath122 , but possibly @xmath123 . when played against @xmath124 , @xmath105 gives a payoff of @xmath125 . let @xmath126 be player @xmath70 s payoff against @xmath124 when he plays the shifted strategy @xmath127 . since @xmath74 is a nash equilibrium , @xmath128 . on the other hand , player @xmath69 can guarantee a payoff of @xmath129 against @xmath105 by using the strategy @xmath130 since the probability of any given difference in bids occurring is the same in @xmath131 as in @xmath132 . therefore @xmath133 , so @xmath134 , whence @xmath135 . expanding this , we find @xmath136 suppose @xmath137 . then , we must have @xmath138 , which solves to get @xmath139 . if @xmath140 , the equation solves the same way to get @xmath139 . thus , either way we have a contradiction of @xmath141 . thus , @xmath139 . together with , this gives @xmath142 by lemma [ basics ] we have @xmath143 , so @xmath144 \left [ \begin{array}{c } s_{\ell-1 } \\ s_0 \\ 0 \\ \vdots \\ 0 \end{array } \right].\ ] ] for @xmath145 we have @xmath146 by equation . in other words , @xmath147 guarantees player @xmath70 his highest possible payoff against @xmath124 , so he has no incentive to deviate from @xmath148 if player @xmath69 uses @xmath124 . we now show player @xmath3 has no incentive to deviate from @xmath56 against @xmath149 . if @xmath150 , the payoff for player @xmath5 if player @xmath69 bids @xmath16 will be @xmath151 . by the formulation of precision in terms of payoff matrices in remark [ precise_matrices ] , we have strict inequalities @xmath152 through @xmath153 , so @xmath154 . thus player @xmath3 loses utility if she bids any amount greater than @xmath155 with positive probability . one also readily sees that if she alters her distribution of bids @xmath156 this will not change her payoff against @xmath149 . it follows that @xmath157 is a nash equilibrium as claimed . the reverse theorem reveals a strong relationship between opposing player s strategies . using it , we can now fully characterize the set of optimal strategies for both players in precise games . [ advantageunique ] if @xmath71 is precise , the player with advantage has a unique equilibrium strategy . let player @xmath3 have advantage . suppose that @xmath56 and @xmath158 are distinct equilibrium strategies for player @xmath3 . let @xmath56 and @xmath158 have lengths @xmath111 and @xmath159 respectively . by the reverse theorem , player @xmath5 has strategies @xmath149 and @xmath160 which have lengths @xmath111 and @xmath159 respectively . suppose @xmath161 . assume , without loss of generality , that @xmath162 . then @xmath160 is a nash equilibrium strategy for @xmath5 with length greater than @xmath56 which contradicts propostion [ grounded_gapfree ] . thus , @xmath163 . assume , without loss of generality , that @xmath164 . that is , we assume , that if @xmath5 bids @xmath111 against @xmath56 he will do no better than if he were bidding @xmath111 against @xmath158 . it is possible he will do strictly worse as bidding @xmath111 is not necessarily a part of player @xmath5 s optimal strategy . consider the following function : @xmath165 we claim that for any @xmath38 for which @xmath166 is a valid strategy , @xmath166 is an optimal strategy . note that @xmath166 has entrywise sum of @xmath10 so @xmath166 is at least valid for @xmath167 $ ] . consider : @xmath168 for @xmath169 , @xmath170 so @xmath171 . for @xmath172 , @xmath173 so @xmath174 . if player @xmath5 bids anything greater than @xmath111 then he will do strictly worse than if he bid @xmath111 , because he will win by more than he would by bidding @xmath111 . therefore , @xmath166 guarantees player @xmath3 a payoff of at least @xmath27 . choose the maximal @xmath175 for which @xmath176 is valid . because @xmath166 has entrywise sum of @xmath10 , it is only invalid if @xmath166 has a negative entry . thus , at this maximal @xmath176 has at least one zero entry . either @xmath176 has length less than @xmath111 , a @xmath8 in its first entry , or is not gap - free . each of these is impossible ( above , prop [ grounded_gapfree ] ) . therefore distinct optimal strategies @xmath56 and @xmath158 can not exist . in most precise games , both players have unique optimal strategies . it is possible , however , to construct a game in which the player without advantage has multiple optimal strategies . we give a characterization of these as well . if @xmath85 is a strategy let @xmath177 represent a new strategy where anytime one would bid @xmath16 in @xmath85 he will bid @xmath178 in @xmath177 . [ noadvantagestrategies ] let @xmath71 be precise and let player @xmath3 have advantage . the following statements hold : 1 . player @xmath5 has a unique strategy of minimal length . this strategy is @xmath149 . 2 . if player @xmath5 has more than one optimal strategy , then another optimal strategy is of the form @xmath179 . all other optimal strategies for player @xmath5 are of the form @xmath180.\ ] ] throughout this proof we will use a method from the proof of theorem [ advantageunique ] . suppose we have two strategies @xmath181 and @xmath182 such that wherever @xmath181 is non - zero so is @xmath182 . then we define @xmath183 we showed above that @xmath184 gives an optimal strategy as long as it is valid . if we choose @xmath185 to be maximal so that @xmath186 is valid , then @xmath186 gives an optimal strategy with a @xmath8 in some spot where @xmath85 was nonzero . let us call the strategy produced by this method @xmath187 . we begin with ( 1 ) . by the reverse theorem , player @xmath5 has a strategy @xmath149 which is of the same length as @xmath56 . by proposition [ grounded_gapfree ] , player @xmath5 can not have a strategy shorter than @xmath56 . therefore , @xmath149 is a strategy of minimal length for player @xmath5 . suppose @xmath85 is another strategy of minimal length for player @xmath5 . then @xmath188 is either of lesser length , is not gap - free , or has a @xmath8 in the first entry . the first two possibilites are impossible by proposition [ grounded_gapfree ] . in the third case , we can apply the same method again to get @xmath189 which is either of lesser length , not gap free , or has @xmath8 s in the first two entries . each of these is impossible by proposition [ grounded_gapfree ] . we now proceed to ( 2 ) . suppose player @xmath5 has more than one optimal strategy . then by ( 1 ) it must be of length greater than @xmath149 . let @xmath111 be the length of @xmath149 . by proposition [ grounded_gapfree ] , any other optimal strategy of player @xmath5 must be of length @xmath190 . let @xmath85 be such a strategy . . then we can take @xmath192 which must have a @xmath8 in the first coordinate lest we contradict proposition [ grounded_gapfree ] . we must show that @xmath193 . because @xmath194 is toeplitz , @xmath195 therefore @xmath179 guarantees player @xmath5 at least his optimal payoff unless player @xmath3 plays @xmath8 . suppose that if player @xmath3 bids @xmath8 then @xmath179 gives player @xmath5 a payoff of @xmath9 less than his optimal payoff of @xmath28 . then define a strategy , @xmath196 for @xmath197 sufficiently small so that @xmath198 has all positive entries . then @xmath199 is a valid strategy that guarantees player @xmath5 his optimal payoff if player @xmath3 bids anything from @xmath10 to @xmath200 . it guarantees player @xmath5 more than his optimal payoff if player @xmath3 bids @xmath8 as : @xmath201 @xmath199 is a strictly better strategy than @xmath202 as player @xmath3 always bids @xmath8 with nonzero probability . @xmath202 is optimal so this is impossible . thus , @xmath179 is an optimal strategy . that it is equal to @xmath202 will follow from ( 3 ) . finally we prove ( 3 ) . @xmath149 and @xmath179 are optimal strategies so any convex combination of the two is optimal . let @xmath203 be an optimal strategy for player @xmath5 that is not a convex combination of the two . then , @xmath203 must be of length @xmath200 . therefore we can take @xmath204 . this gives a strategy which is either of length @xmath111 , is not gap - free , or has multiple @xmath8 s at the begining . the latter two possibilities are impossible by proposition [ grounded_gapfree ] . @xmath149 is the unique optimal strategy of length @xmath111 so : @xmath205 @xmath206 note that @xmath207 and both coefficients must be postive or else the first or last entry of @xmath203 will be negative . thus , @xmath203 is a convex combination of @xmath149 and @xmath179 . in most of the above proofs we assume that @xmath48 is a precise game . in many games with small associated graphs , this is not the case . the simplest example is a game where in the associated graph the only directed edge goes to @xmath21 . then player @xmath3 always wins , so the chip counts do no matter whatsoever . thus , we apply a small adjustment to the payoff matrices for players @xmath3 and @xmath5 . pick a small @xmath208 . we now define @xmath209 as @xmath210 where @xmath211 is given by the @xmath212 toeplitz matrix @xmath213.\ ] ] intuitively , we can think of @xmath214 as a payoff matrix that gives payoff @xmath38 for each chip a player has at the end of a turn . @xmath215 is then given by the strategy that maximizes player @xmath3 s minimum payoff under @xmath209 . @xmath216 is this payoff . while the payoff no longer corresponds exactly to winning probability , the game @xmath217 is still zero - sum , with total utility @xmath218 split between the two players . we generalize our lemma [ basics ] to this new game : the game represented by @xmath209 is precise . each entry of @xmath209 represents a successor state of the game where each player has some number of chips . from the way we have defined @xmath211 , for any successor state in which having one more chip provided an equal payoff in @xmath48 , having one more chip will now provide a payoff exactly @xmath38 greater . a natural question arising from this adjustment is whether or not it gives a good approximation of the actual payoff for @xmath48 and the actual nash equilibria . the following theorem shows that by choosing a small enough @xmath38 , @xmath219 , and @xmath220 can be made arbitarily close to @xmath221 and some nash equilbrium strategy @xmath56 . [ convergence ] with @xmath56 as described above , @xmath222 @xmath223 @xmath224 we notice that ( 1 ) follows directly from the definition of @xmath225 : @xmath226 we now consider ( 2 ) . we can define three functions : @xmath227 notice that for all @xmath228 , @xmath229 . we also see that @xmath230 @xmath231 therefore , @xmath232 this leaves ( 3 ) , the proof of which is more nuanced . we must first develop some more theory of all - pay bidding games . in many bidding games , the random distribution governing optimal play does not involve bidding above some threshold . in a game of bidding tic - tac - toe where each player begins with 100 chips , a player should not bid 100 on the first turn . by the reverse theorem , the two players , have optimal strategies of equal length . suppose in some bidding game @xmath48 , both players have strategies of length @xmath111 . then we can consider the * restricted * game , @xmath233 , where both players can bid at most @xmath234 on the first turn and play returns to normal thereafter . in such a restricted game players are still able to play the length @xmath111 optimal strategy they would have employed in the original game . is this strategy still optimal ? [ restrictedgame ] if @xmath235 are optimal length @xmath111 strategies in @xmath48 that provide the payoffs @xmath27 and @xmath236 respectively , then they are optimal in @xmath233 and provide the same payoffs . @xmath237 is the @xmath238 top - left minor of @xmath239 as the games are identical after the first move . thus , both players bidding less than @xmath111 in @xmath48 is equivalent to the players making the same bids in @xmath233 . thus , @xmath240 gives the first @xmath111 entries of @xmath241 . the minimum entry of @xmath241 is @xmath27 so the minimum entry of @xmath240 is at least @xmath27 . thus , @xmath56 guarantees at least the payoff @xmath27 . using the same logic for @xmath242 , we obtain the @xmath60 guarantees the payoff at least @xmath236 . the total payoff is exactly @xmath10 so player @xmath3 gets payoff @xmath27 and can not do better and player @xmath5 gets the payoff @xmath236 and can not do better . furthermore , recall that precision is a characteristic of the successor states in a game . the possible successors of a restricted game are a subset of the successors of the normal game . thus , if a game is precise then its restricted game is also precise . we are now able to state a powerful result for the restricted game that will allow us to prove some important results for general bidding games . [ preciseinvertible ] in a precise game @xmath48 , if @xmath56 , an optimal strategy of minimal length , has length @xmath111 , then @xmath237 is invertible . suppose by way of contradiction that there exists @xmath243 such that @xmath244 . define @xmath245 by @xmath246 for @xmath247 and @xmath248 for @xmath249 . then @xmath250 is a vector with @xmath8 in it first @xmath111 entries . in particular , @xmath56 has all positive entries so there exists @xmath252 such that @xmath253 and @xmath254 have all positive entries . we note that : @xmath255 @xmath256 for @xmath247 . suppose the sum of the entries of @xmath257 is less than @xmath10 . then there exists @xmath258 such that the sum of the entries of @xmath259 is equal to @xmath10 . then @xmath259 is a valid strategy for player @xmath3 . that gives payoff @xmath260 against player @xmath5 s first @xmath111 pure strategies . thus , against @xmath149 , @xmath259 is better than @xmath56 so @xmath261 is not a nash equilibrium . contradiction . then suppose the sum of the entries of @xmath257 is greater than @xmath10 . then the sum of the entries of @xmath262 is less than @xmath10 so the same argument holds . then suppose the the sum of the entries of @xmath257 equals @xmath10 . then @xmath257 and @xmath56 are optimal in @xmath233 . @xmath233 is precise , however , so there exists only one optimal strategy of minimal length for either player in @xmath233 . therefore @xmath39 must equal @xmath8 . a method for computing optimal strategies for the player with advantage , say player @xmath3 , now becomes apparent . given the length of the player s unique optimal strategy we can consider the payoff matrix of the restricted game . by the reverse theorem , player @xmath5 has a gap - free strategy of the same length . then the restricted payoff matrix multiplied by player @xmath3 s optimal strategy must give a constant vector . the inverse of our restricted payoff matrix multiplied by some non - zero constant vector will therefore give a scalar multiple of player @xmath3 s optimal strategy . [ strategyformula ] let player @xmath3 have advantage . in a precise game @xmath48 if @xmath56 has length @xmath111 then @xmath263 as discussed above @xmath264 is a scalar multiple of @xmath56 . the sum of the entries of @xmath56 is @xmath10 so we need only divide by the sum of the entries of @xmath264 . this is given by @xmath265 . this theorem gives an explicit and rapid method for computing optimal strategies for a player with advantage . combined with the reverse theorem , we will be able to develop a method for computing optimal strategies for both players in any simple bidding game . first , we will return to ( 3 ) of theorem [ convergence ] . recall our conjecture that as @xmath266 , @xmath267 . the above theorem gives even more weight to this claim as together they give a method for approximating optimal strategies for imprecise games via a convergent sequence of strategies for precise games . we begin by partially extending the invertibility of the restricted payoff matrix to imprecise games . the importance of this result is not immediately obvious , but it will be integral to the proof of part ( 3 ) of theorem [ convergence ] . for simplicity , we will sometimes write @xmath237 as @xmath268 and @xmath242 as @xmath269 . [ wow ] if player @xmath3 has a length @xmath111 optimal strategy for @xmath270 then at least one of @xmath237 and @xmath271 is invertible . for simplicity , let @xmath272 and @xmath273 . if @xmath52 is invertible , we are done , so suppose @xmath52 is not invertible . let @xmath274 be in the nullspace of @xmath52 . because @xmath220 is gap - free , there exists @xmath275 sufficiently small such that @xmath276 are valid strategies for player @xmath3 . then @xmath277 each successive row in @xmath5 is 1 greater in each entry than the previous row . suppose that the sum of the entries of @xmath278 is equal to @xmath8 . then , @xmath279 thus , @xmath280 is a constant vector . if @xmath281 then @xmath282 by lemma @xmath283 , @xmath225 is invertible so @xmath280 can not equal @xmath8 . therefore , either @xmath284 or @xmath285 results in a better payoff for player @xmath3 than @xmath220 for @xmath225 contradicting the optimality of @xmath220 . therefore the sum of the entries of @xmath278 is not @xmath8 . we can then let the sum of the entries of @xmath278 be equal to @xmath10 . then @xmath286 we will return to @xmath278 momentarily . we can compute that @xmath287 . since @xmath220 is a nash equilibrium , @xmath288 @xmath289 @xmath290 where @xmath291 . then we can substitute @xmath278 into the equation : @xmath292 let @xmath293 and @xmath294 so that : @xmath295 let @xmath296 be a @xmath238 matrix with rows @xmath297 . then @xmath298 therefore through this seemingly arbitrary construction we obtain that @xmath299 which is invertible by lemma [ preciseinvertible ] . thus , @xmath300 is invertible . the last several results have dealt with payoff matrices of restricted games . the payoff matrix of a restricted game is , by definition , dependent on the length of a player s optimal strategy . the following lemma further demonstrates the relevance of the lengths of the players optimal strategies . [ weakconvergence ] if there exists @xmath301 and @xmath302 such that for all @xmath303 , @xmath304 then @xmath305 exists and is an optimal strategy . let @xmath306 . we will treat @xmath307 as imprecise so the proof holds for both precise and imprecise games . by proposition [ wow ] , at least one of @xmath308 and @xmath309 is invertible . suppose @xmath52 is invertible . then the limit @xmath310 exists . as @xmath38 goes to @xmath8 , @xmath311 is nonzero and has entry wise sum of @xmath10 . thus , @xmath85 is all nonnegative and also has entry wise sum of @xmath10 . finally , @xmath312 thus @xmath85 is optimal . if @xmath52 is not invertible , then @xmath194 is invertible . by the reverse theorem , for all @xmath303 , @xmath313 . therefore , we can apply the same argument as above to @xmath314 . while the above lemma s potential power is clear , we have not yet demonstrated that the conditions it requires are met by any games . we need some restrictions on the length of optimal strategies as we adjust chip value in order to effectively use the above results . the next lemma and its corollary give us the necessary structure . [ openneighborhoods ] let player @xmath3 have advantage . let @xmath315 . the set of @xmath316 such that @xmath317 is open in @xmath318 . the length of @xmath220 is an integer and is bounded above by @xmath19 . hence @xmath301 exists . pick some @xmath316 so that @xmath319 . suppose there exists no @xmath320 such that for all @xmath321 , we have @xmath322 . then we can define a sequence @xmath323 by @xmath324 so that @xmath325 . there exist only a finite number of possible values for @xmath326 so there must be at least one @xmath327 such that@xmath328 has a convergent subsequence @xmath329 with @xmath330 for all @xmath331 . by theorem [ convergence ] , @xmath332 @xmath333 then @xmath334 for which we have that @xmath335 @xmath85 is then an optimal strategy in @xmath336 . @xmath85 is the limit of length @xmath337 strategies so it has length at most @xmath337 . therefore @xmath338 . the player with advantage has exactly one optimal strategy so an appropriate open neighborhood must exist . let @xmath301 be as above and let @xmath52 be invertible . also assume @xmath220 is of constant - length on some interval @xmath339 . then there exists vectors @xmath340 such that for all @xmath341 , @xmath342 let @xmath343 note that @xmath85 is not necessarily optimal or even a valid strategy . it satisfies two notable properties . the sum of the entries of @xmath85 is @xmath10 and @xmath344 is a constant vector . consider , @xmath345 @xmath346 is a constant vector as each row in @xmath5 differs by a vector of all @xmath10 s from the row above it . a vector of all @xmath10 s multiplied by @xmath347 is @xmath8 as both @xmath220 and @xmath85 have entrywise sum of @xmath10 . let this constant vector be denoted @xmath348 . then @xmath349 @xmath350 @xmath351 note the @xmath352 terms cancel , and that @xmath353 is a constant vector . thus , because @xmath348 is also a constant vector , we know that @xmath354 is a constant vector , which we call @xmath355 . then @xmath356 @xmath357 note that @xmath358 is a scalar multiple of @xmath85 . let this scalar be @xmath197 . we have the relation : @xmath359 we see that @xmath197 is a function of @xmath38 , and must be the unique scalar that causes @xmath360 to have entrywise sum of @xmath10 . thus @xmath197 is given by : @xmath361 @xmath362 @xmath363 @xmath364 is a constant because @xmath365 , and @xmath85 are . let it be denoted @xmath366 . @xmath367 @xmath368 is a vector independent of @xmath38 . let it be denoted by @xmath182 . thus , @xmath369 thus , on an @xmath38-interval on which @xmath220 is of constant length , @xmath220 is given by @xmath370 . further , each entry of @xmath220 is given by a linear equation @xmath371 . note that the above lemma does not make use of anything specific to player @xmath3 or @xmath5 . thus , it also applies to @xmath372 if the necessary conditions hold . [ lengthstable ] let player @xmath3 have advantage . let @xmath301 be as above . then there exists @xmath373 such that for all @xmath374 , @xmath375 . let @xmath376 be chosen such that @xmath375 . by proposition [ wow ] , at least one of @xmath52 and @xmath194 is invertible . suppose first that @xmath52 is invertible . by lemma [ openneighborhoods ] , there exists an open interval @xmath339 containing @xmath38 on which @xmath220 is constant - length . let @xmath339 be the largest such open interval . we are able apply the above lemma . there exists vectors @xmath340 such that for all @xmath377 @xmath342 for @xmath378 , the value of a chip is greater than the value of winning the game so neither player will ever bid more than @xmath8 . thus , @xmath379 . suppose that @xmath380 . on this interval @xmath220 is given by @xmath370 for some @xmath340 . therefore , the @xmath381-th entry of @xmath220 is either strictly increasing , strictly decreasing , or constant . by lemma [ weakconvergence ] and the uniqueness of nash equilibrium strategies for the player with advantage . @xmath382 @xmath383 if @xmath384 or @xmath385 have length @xmath301 then by lemma [ openneighborhoods ] there is an open interval about @xmath19 or @xmath20 respectively on which optimal strategies have length @xmath301 so @xmath339 is not maximal . thus , both @xmath384 and @xmath386 must have length less than @xmath301 . this implies that : @xmath387 @xmath388 a linear equation has at most one zero unless @xmath389 and @xmath390 are both @xmath8 . @xmath391 however so this can not be the case . therefore @xmath19 must be equal to @xmath8 . then @xmath220 is constant - length on some interval which has @xmath8 as an endpoint . now suppose that @xmath194 is invertible . we can perform the same operations on the optimal strategy of minimal length @xmath372 for player @xmath5 and then apply the reverse theorem to achieve the same result for player @xmath3 . given this structure , we can complete our discussion of convergence . let player @xmath3 have advantage . then @xmath392 exists and is optimal . by corollary [ lengthstable ] , there exists @xmath393 such that for all @xmath374 , @xmath375 . these are the necessary conditions to apply lemma [ weakconvergence ] which gives the result . although we have developed results on the structure of optimal bidding in all - pay bidding games , we have yet to fully describe how these optimal strategies can be found . in this section , we put together our results for precise games with our convergence results for imprecise games to give an algorithm to calculate the optimal bidding strategy for any state in an all - pay bidding game . in this section , we will discuss the algorithm we developed to quickly calculate an optimal strategy . our algorithm first assigns to each chip an arbitrarily small but positive value @xmath38 . this adjusted game is precise , so we will be able to take advantage of the structure we have shown for precise games . in particular , we will be able to use theorem [ strategyformula ] , which gives a formula for the unique bidding strategy belonging to the player with advantage , in terms of the payoff matrix and optimal length : @xmath394 from the convergence results in the previous section , the resulting strategy will be able to approximate an optimal strategy for player @xmath3 in an imprecise game to any desired degree of accuracy . note that this strategy is not guaranteed to be a unique optimal strategy in the unadjusted game if the unadjusted game is not precise . once @xmath56 is known , we know by convergence that @xmath149 will have to be an optimal strategy for player @xmath5 . @xmath261 is then within any desired degree of accuracy of a nash equilibrium for the unadjusted game . for now we will assume the payoff matrix is known . then , to implement theorem [ strategyformula ] we just need to invert the appropriate minor of that matrix , multiply by a vector of @xmath10 s , and rescale so that the entries of the resulting vector sum to @xmath10 . the problem now is to find this optimal length in a precise game where the payoff matrix is given . the next two lemmas will allow us to use binary search to find the optimal length quickly . [ lengthsearch1 ] let the game be precise . let @xmath395 be a vector of all @xmath10 s . then for all @xmath396 , @xmath397 will have all nonnegative entries . by similar reasoning as in lemma [ preciseinvertible ] , we know that @xmath398 will be invertible for all @xmath399 . we naturally consider the game @xmath400 . let @xmath401 and @xmath402 be @xmath3 s and @xmath5 s optimal strategies in this game . note that if @xmath403 , then by definition of @xmath111 , we have that @xmath404 gives a constant ( nonnegative ) vector , so @xmath405 will be @xmath56 scaled by @xmath406 . this will have all nonnegative entries because @xmath56 is a strategy . we can extend this reasoning to when @xmath407 if we know that @xmath401 still has length @xmath331 , as it must also give some constant payoff , @xmath408 , in @xmath400 . suppose @xmath401 does not have length @xmath331 . then @xmath401 has length @xmath409 . let @xmath408 be the value of @xmath400 for player @xmath3 . suppose that @xmath410 . then , we can make strategy @xmath158 for player @xmath3 in @xmath48 , by extending @xmath401 to the full game , where @xmath411 if @xmath412 , and is @xmath8 otherwise . then , note that @xmath413 if @xmath414 . because @xmath415 , @xmath416 where @xmath417 is the maximal number of chips useable in the @xmath400 game . thus , if @xmath418 , @xmath419 by definition of nash equilibrium for @xmath400 . this is @xmath3 s payoff against @xmath158 if @xmath5 purely bids @xmath420 . but since @xmath3 s maximal bid in @xmath202 is @xmath421 , that means if @xmath5 uses a pure strategy where she bids @xmath422 chips , she will just be winning the same bids by more chips , which can not be better in any way . thus , @xmath423 . thus , for all @xmath424 , @xmath425 , so @xmath158 is a nash equilibrium for @xmath17 as well . but @xmath158 has length @xmath426 , so it would have to be distinct from @xmath56 because it has a different length . this can not be the case as @xmath3 s optimal strategy is unique . thus , we have a contradiction and @xmath401 can not have length less than @xmath331 . thus , @xmath401 has length @xmath331 , so by the same argument as the @xmath403 case , all the entries of @xmath427 are nonnegative . note that because none of the above reasoning depended upon player @xmath3 having advantage , if @xmath428 , we can apply the above argument from player @xmath5 s perspective . [ lengthsearch2 ] let the game be precise . let @xmath395 be a vector of all @xmath10 s . then for all @xmath429 , either @xmath430 is not invertible or @xmath397 will have all nonnegative entries . assume @xmath430 is invertible . we begin by showing there is no valid length @xmath431-strategy for player @xmath3 that produces the same payoff for player @xmath5 s first @xmath200 pure strategies . suppose there does exist such a strategy @xmath85 . let @xmath9 be the payoff that @xmath85 produces against player @xmath5 s first @xmath200 pure strategies ( pure bids from @xmath8 up to @xmath111 ) . note that because @xmath56 has length @xmath111 there is a nash equilibrium strategy @xmath149 of length @xmath111 for player @xmath5 . we consider three cases : ( 1 ) @xmath432 : : + since @xmath5 bids at most @xmath433 , we only need to consider the first @xmath111 coordinates of @xmath434 and @xmath435 . by our assumption , @xmath436 so @xmath85 is strictly better than @xmath56 against @xmath149 . thus @xmath56 can not be a nash equilibrium strategy , which is a contradiction . ( 2 ) @xmath437 : : + if @xmath438 then let player @xmath5 use the strategy @xmath439 . it is easy to verify that @xmath439 produces the payoff @xmath440 against player @xmath3 s first @xmath200 pure strategies . thus , by similar reasoning as in the previous case , @xmath439 is strictly better than @xmath149 against @xmath56 , so @xmath149 can not be a nash equilibrium strategy , which is a contradiction . ( 3 ) @xmath441 : : + if @xmath442 , then expanding the first @xmath200 coordinates of @xmath435 results in the equations @xmath443 for @xmath444 . considering the game from player @xmath5 s perspective , note that @xmath439 gives @xmath5 a payoff of @xmath236 against @xmath3 s first @xmath200 strategies . in particular , @xmath5 s payoff against @xmath3 bidding @xmath111 will be @xmath445 @xmath5 s payoff @xmath38 against against @xmath3 bidding @xmath200 will be @xmath446 note that because player @xmath3 is winning ties , @xmath447 , as in each case @xmath3 is winning by one more chip . thus , @xmath448 which means @xmath449 similarly , @xmath5 s payoff if @xmath3 bids anything greater than @xmath331 will be greater than @xmath236 . thus , @xmath439 is a nash equilibrium strategy of length @xmath331 for player @xmath5 . note that if @xmath331 is at least @xmath450 , then @xmath439 will have length at least @xmath450 , which will be a contradiction if @xmath56 has length @xmath111 . since @xmath451 , this means we must have @xmath452 . then , @xmath439 is a nash equilibrium strategy of length @xmath200 , so by theorem [ noadvantagestrategies ] it must be of the form @xmath453 for @xmath454 . in turn , @xmath85 must be of the form @xmath455 . we can now write @xmath456 as @xmath457 , which can be expanded to the equation @xmath458 + by considering the first coordinate , we get the equation @xmath459 so we must have @xmath460 as well . therefore , @xmath461 but since the game is precise , there must be an inequality for all the coefficients : @xmath462 so @xmath463 because not all the @xmath464 s are @xmath8 . thus , we have a contradiction , and @xmath331 can not be @xmath200 either . + thus , no such strategy @xmath85 can exist , so if @xmath430 is invertible , @xmath397 must have some negative terms . we can implement the binary search algorithm as follows . let the lower bound , @xmath465 , start as @xmath10 . let the upper bound , @xmath466 , start as @xmath467 . * function * lsearch(@xmath52 , @xmath465 , @xmath466 ) by lemmas [ lengthsearch1 ] and [ lengthsearch2 ] , this algorithm will return the length of the optimal strategy for player @xmath3 . we can then apply our formula to directly compute player @xmath3 s unique optimal strategy . the reverse of this strategy is an optimal strategy for player @xmath5 . this completes the algorithm . from our results on the convergence of strategies , this algorithm is also able to approximate , with any desired degree of accuracy , optimal strategies for imprecise games . so far , our results apply to the strategy for bidding on a single turn in an all - pay bidding game . this assumes some prior knowledge of successor game states that allows the payoff matrix to be already known . thus , to use our algorithm to compute nash equilibria for any all - pay bidding game state , we need some way of first finding the payoff matrix . by noting that the payoff matrices for end states ( where one player has already won ) can be set as @xmath8 and @xmath10 for win and loss , we use recursion from the end states of the game to find the payoff matrix for an arbitrary turn . consider a combinatorial game @xmath34 represented as a directed graph @xmath468 with two vertices marked as @xmath21 and @xmath22 and a token placed at some vertex of the graph . we can think of each vertex as the starting position of a subgame of @xmath34 . thus for player @xmath3 with @xmath19 chips and player @xmath5 with @xmath20 chips , the token on vertex @xmath278 , we write the game as @xmath469 . let @xmath470 give all the vertices that can be moved to from @xmath278 . we can compute @xmath27 as follows : @xmath471 then if @xmath3 bids @xmath16 and @xmath5 bids @xmath36 and @xmath3 makes a move then @xmath3 s payoff will be @xmath472 because @xmath3 will seek to maximize his probability of winning over all of his possible sucessor states . if @xmath3 bids @xmath16 and @xmath5 bids @xmath36 and @xmath5 makes a move then @xmath3 s payoff will be @xmath473 because @xmath5 will seek to minimize @xmath3 s probability of winning over all of her possible sucessor states . therefore @xmath474 as each player will consider the best possible scenario if he moves and the worst possible scenario if their opponent moves . then @xmath56 , @xmath60 can computed from this payoff matrix @xmath475 , using our algorithm from before . note this allows us to recurse up the directed graph from states @xmath21 and @xmath22 , first with values for those states , then values for the states one move away ( i.e. @xmath9 such that either @xmath21 or @xmath476 ) , then states two moves away , and so on . an arbitary @xmath477 matrix can be inverted in @xmath478 time using the gauss - jordan method . there exist , however , many more efficient algorithms specific to toeplitz matrices . in particular , the levinson - trench - zohar algorithm can solve a toeplitz system in @xmath479 time @xcite . for an @xmath477 matrix , the binary search algorithm requires @xmath480 iterations . each iteration requires solving one toeplitz system and scanning one vector for negative values . thus , the algorithm runs in time on the order of @xmath481 . thus finding an optimal strategy and corresponding payoff for a given payoff matrix requires time on the order of @xmath482 . a simple implementation of our recursive algorithm would take time growing exponentially with the depth of @xmath483 . we can greatly speed up this process by storing each @xmath484 that is computed . then when @xmath484 must be computed again the value can be looked up rather than recomputed . in the worst case , the program must compute @xmath27 for every possible combination of chips at every vertex . because the sum of the chips is constant , this requires at most @xmath485 computations . thus , the entire algorithm runs in time on the order of @xmath486 where @xmath487 . for comparison , a linear programming algorithm to achieve the same results would require time on the order of @xmath488 @xcite . we would like to thank michael landry and sam payne for exposing to us the research potential surrounding all - pay bidding games and giving guidance in the research process . we would also like to thank aviezri fraenkel for suggesting the idea of integrating economic and combinatorial games and ilan adler for his helpful comments on an earlier draft of this paper . this paper was also supported in part by nsf grant career dms-1149054 .

the shape of the light curve peak of radioactive powered core collapse stripped envelope " supernovae constrains the ejecta mass , nickel mass , and kinetic energy by the brightness and diffusion time for a given opacity and observed expansion velocity . late time light curves give constraints on the ejecta mass and energy , given the gamma ray opacity . previous work has shown that the principal light curve peaks for sn iib with small amounts of hydrogen and for hydrogen / helium deficient sn ib / c are often rather similar near maximum light , suggesting similar ejecta masses and kinetic energies , but that late time light curves show a wide dispersion , suggesting a dispersion in ejecta masses and kinetic energies . it was also shown that sn iib and sn ib / c can have very similar late time light curves , but different ejecta velocities demanding significantly different ejecta masses and kinetic energies . we revisit these topics by collecting and analyzing well sampled single band and quasi bolometric light curves from the literature . we find that the late time light curves of stripped envelope core collapse supernovae are heterogeneous . we also show that the observed properties , the photospheric velocity at peak , the rise time , and the late decay time , can be used to determine the mean opacity appropriate to the peak . the opacity determined in this way is considerably smaller than common estimates . we discuss how the small effective opacity may result from recombination and asymmetries in the ejecta .

1411.5975

clocchiatti & wheeler ( 1997 ) noted several apparent properties of stripped envelope " supernovae : ( 1 ) some events completely trap gamma rays for hundreds of days . the only known examples of these slow light curves were sn ib . ( 2 ) some events display a similar photometric evolution with an intermediate later time rate of decline , similar to sn 1993j , despite their spectral variety , sn iib , sn ib , sn ic . ( 3 ) some events decline especially rapidly from maximum , show an especially large peak to tail contrast , but show a slope comparable to those of the intermediate slope class ( point 2 ) at very late times , over 100 d after maximum . these events all seemed to be sn ic . drout et al . ( 2011 ) found that the behavior of the light curve peaks of stripped envelope events were rather similar , with sn ib and sn ic having statistically indistinguishable decline rates up to @xmath0 40 d after maximum , implying similar ejecta mass and energy . taddia et al . ( 2014 ) find that sn ic and sn ic bl have shorter rise times than sn iib and sn ib , but that the immediate post peak declines are similar for all categories . similar decline rates shortly after peak , with some dispersion , were also noted by cano ( 2013 ) and lyman et al . this evidence for similar post peak decline rates belies the long standing evidence for late time light curve dispersion . here , we compile light curve data for late time light curves of stripped envelope supernovae and re - investigate the uniformity of the late time behavior . we concentrate here on light curve properties . see modjaz et al . ( 2014 ) for a recent compilation of spectral properties and @xcite for a compilation of light curves . section 2 defines the analytical basis of our analysis . section 3 presents the data . section 4 gives a discussion and section 5 summarizes our conclusions . we analyze data from the literature on the light curves of stripped envelope supernovae . we did our own compilation , but see cano ( 2013 ) for an independent compilation , including some long term data of the same events we present here . we estimate the ejecta mass and energy from the observed rise time and photospheric velocity near peak light , and an effective opacity , assumed to be constant . following arnett ( 1982 ) we write @xmath1 where @xmath2 is an integration constant , c the speed of light , @xmath3 is the effective uvoir opacity , @xmath4 is the velocity at the photosphere with @xmath5 in units of @xmath6 , and @xmath7 the rise time to maximum light . this equation is based on assumptions of homologous expansion and self similar diffusion with a power source at the center of spherical ejecta . two further assumptions are that @xmath4 is a reasonable proxy for the scaling velocity in the model and that the rise time to maximum light is a reasonable proxy for the effective timescale in the model , @xmath8 , where @xmath9 is the model diffusion time and @xmath10 the model hydrodynamical time . in practice , the photospheric velocity will be affected by the distribution of density and opacity and hence may not precisely represent a fixed scaling velocity , the rise time is not exactly equal to @xmath11 even in the context of the basic model ( chatzopoulos et al . 2013 ) , and the opacity will be constant neither in space nor time . the corresponding kinetic energy is then @xmath12 which implicitly defines the mean squared expansion velocity . this velocity can not be measured directly , so a relation must be adopted between this velocity and the velocity at the photosphere . we adopt the relation for a constant density sphere , @xmath13 . this error was corrected in arnett ( 1996 ) where the correct relation is given , @xmath14 , but the error has propagated in the literature in , among other works , valenti et al . ( 2008a ) , chatzopoulos , wheeler & vinko ( 2012 ) and chatzopoulos et al . ( 2013 ) who wrote @xmath15 without carefully specifying the prescription for the velocity . ] . with this relation we can then write @xmath16 so @xmath17 is proportional to @xmath18 and hence sensitive to the choice of @xmath4 and its dispersion among events . for the late - time tail , we adopt the formalism of clocchiatti & wheeler ( 1997 ) and define a characteristic timescale @xmath19 where c is a dimensionless structure constant dependent on the slope of the density profile and is typically c @xmath0 0.05 , and @xmath20 is the opacity to gamma rays . using equation ( [ mass ] ) for @xmath21 , this can also be written @xmath22^{1/2}t_r = 32 { \rm ~d } \left(\frac{\kappa_{\gamma}/0.03}{\kappa/0.1}\right)^{1/2 } v_{ph,9}^{-1/2 } \left(\frac{t_r}{10 ~d}\right),\ ] ] where we take as a fiducial optical opacity @xmath23 @xmath24 g@xmath25 and as a fiducial gamma ray opacity @xmath26 @xmath24 g@xmath25 . we take the radiated power to be @xmath27 where l@xmath28 is the total input from gamma rays and positrons from the radioactive decay of @xmath29ni and @xmath29co according to @xmath30,\ ] ] where a is a scaling constant that depends on the initial mass of @xmath29ni , @xmath31 d is the decay time of @xmath29ni , and @xmath32 d is the decay time of @xmath29co . for this work , we neglect the difference in the deposition functions of gamma - rays and positrons @xcite , but comment on the separate effect of positrons below . figure [ models ] gives simple model bolometric light curves with various degrees of gamma ray leakage as determined by the timescale @xmath33 . to estimate @xmath21 , @xmath17 , and @xmath33 , we use data from the literature . some of the supernovae that we discuss here have estimated ubvri quasi bolometric light curves , but not all . we have used the r band as a proxy that allows us to treat most of the events in a somewhat homogeneous way , but also compare to bolometric data where it is available . for events with only r band data , we neglect bolometric corrections and assume that the r band light curves are proxies of the bolometric luminosity to within a constant scaling factor ( see lyman , bersier & james 2014 for a discussion of estimating bolometric corrections for stripped envelope supernovae ) . figure [ rbc ] gives a comparison of the r band and uvoir light curves of three events , type iib sn 1993j , and two broad lined sn ic , sn 1998bw , and sn 2002ap , normalized at peak light . this figure shows that , for these events , the r band is a reasonable representation of the quasi bolometric light curve to times of hundreds of days . we have ignored reddening corrections for our r band light curves , but such corrections have been made by the original authors who present the quasi bolometric data . it seems unlikely that the results we present here are simply the result of variance in reddening , but this must be borne in mind . in some cases of sparse or absent r band data , we employ v band or other available data . the sample for which there is good rise time data as well as long time tail data is still sparse . the peak is used to estimate the rise time , @xmath7 . for many of the more recent events and for some earlier nearby events , there is pre maximum photometry . in the best cases , there are reasonably accurate estimates of the time of explosion ( sn 1993j , richmond et al . 1994 ; sn 2008d , malesani et al . 2009 , modjaz et al . in other cases , the earliest data are up to two weeks prior to peak and thus are a good representation of the rise time ( sn 1998bw , patat et al . 2001 ; sn 1994i , richman et al . 1996 ; sn 1999dn , benetti et al . 2011 ; sn 2007y , stritzinger et al . 2009 ; sn 2007gr , valenti et al . 2008a ; sn 2009jf , sahu et al . 2011 ; sn 2011bm , valenti et al . 2012 ; sn 2011dh , ergon et al . 2014 , marion et al . note that even in these cases , the r band peak may be delayed from the bolometric peak by several days . we have used the time to r band peak in most cases , but recognize that some ambiguity is introduced in this way . in yet other cases there is some pre maximum data , but not as extensive as the previous cases . for cases where there is pre maximum data more than a magnitude dimmer than maximum , we use the time from first observation to the peak in r band as an estimate for the rise time . this is an underestimate by perhaps several days , or typically 10 20% . examples are sn 1983n ( fes data ; clocchiatti et al . 1997 ) , sn 1996cb ( qui et al . 1999 ) , sn 2002ap ( tomita et al . 2006 ) , sn 2004aw ( taubenberger et al . 2006 ) , and sn 2009bb ( pignata et al . 2011 ) . in other cases , there is little or even no pre maximum data . in these cases , we make estimates based on analogies to other events ( sn 1983v , clocchiatti et al . 1997 ; sn 1984l , swartz & wheeler 1991 ) or on the decline from maximum and estimate the time to decline by a magnitude from peak ( sn 1990b , clocchiatti et al . the light curves of the stripped envelope events are typically asymmetric around the peak with more rapid rise than decline , so this post maximum decline timescale almost surely overestimates the rise time . in these various cases , we have put our estimates in square brackets in table [ tab : timescale ] to highlight the uncertainty . the rise time comes in squared in the estimates of the mass and energy , so these estimates should be considered with caution . taddia et al . ( 2014 ) give a table containing observed r band rise times . most of these agree with our choices , but there are some discrepancies . we chose 17 d for the rise for sn 1993j versus 22.53 d for taddia et al . our estimate is measured from the early minimum in the light curve , ignoring the initial fireball decline phase . we assume the subsequent rise is more representative of other events that were not caught so early and of the diffusion time we are attempting to constrain . we chose 15 d for sn 2007y from stritzinger et al . ( 2009 ) , versus the 21 d assigned by taddia et al . the latter seems to involve some extrapolation even beyond the early v band data , but our value is surely a lower limit . estimates of velocity are taken from photospheric velocities given in the literature . for uniformity , we have tried to select whenever possible the velocity of the p cygni absorption minimum of fe ii @xmath345169 measured at r band maximum . while this line may be relatively free of blending , as may affect estimates based on si ii , there are still issues . in particular , the velocity measured in this way often declines rather rapidly from pre maximum to post maximum epochs . velocities at bolometric maximum several days earlier would usually have been somewhat higher and there is the general concern as to whether any velocity measured in this way properly represents the scale velocity of the underlying model . such considerations comes in especially sensitively in estimates of the energy . cano ( 2013 ) and lyman et al . ( 2014 ) also give compilations of photospheric velocities . in general , our estimates agree with theirs to within 20% . exceptions are sn 2004aw where we determined 16,000 , cano gave 11,800 and lyman et al . give 11,000 , and sn 2009bb , for which we determined 18,000 , cano gave 15,000 , and lyman et al . gave 17,000 . we use the late time light curves to directly determine an estimate of @xmath33 and one standard deviation error bars by minimizing the @xmath35 of the fit of equation [ radiated ] to the data . to minimize the effect of scatter in the peak data , we only use the data after 50 d from maximum to determine the observed value of @xmath33 . table [ tab : timescale ] also gives the measured values of @xmath33 . the error given for sn 2007y is especially , and probably artificially , low because there is sparse late time data that is easy to fit . for some objects in the sample here , clocchiatti et al . ( 2008 ) also estimated values of @xmath33 from ubvri quasi bolometric light curves . comparing the values derived here ( first numbers , r band ; the numbers in parentheses are our estimate of the value based on quasi bolometric data ) with those given by clocchiatti et al . ( second numbers ) we find : sn 1994i , 115 d vs 65 d ; sn 1998bw , 192 ( 283 ) d vs 120 d ; sn 2002ap , 186 ( 202 ) d vs 142 d ; sn 1990b , 113 d vs 88 d. there are differences of 10 to 30% with our estimates somewhat on the high side . the difference is most likely due to our simple treatment that implicitly has the trapping of positrons decreasing in the same manner as for the gamma rays . if positrons are trapped , as assumed by clocchiatti et al . , then for values of @xmath36 100 to 200d , the positron contribution will be comparable to that of the gamma rays by 200 300 d. with this extra contribution to the luminosity , a light curve of given late time slope can be accommodated with a smaller value of the parameter @xmath33 . in the first portion of our discussion comparing the observed decay time versus that predicted from the peak properties , as expressed in equation [ t02 ] , differences of @xmath37% are not significant since we are interested in discrepancies of factors of several . in [ discuss ] where we address the issue of estimating ejecta masses and energies from the tail , the decay time enters as the square and hence more sensitively . we address the issue of positron trapping again there . figures [ sniib ] , [ snib ] , and [ snic ] give the light curves we have compiled . for comparison purposes , the light curves have been shifted in time to a common peak and have been normalized to an arbitrary magnitude of 15 . figure [ sniib ] gives a sample of 4 sn iib . they are all rather similar near peak , with similar rise times and early declines , but substantial departures from homogeneity set in beginning about 30 d after maximum . the long term light curves of these events are distinctly steeper than that given by full trapping of gamma - rays from decay . there is a clear dispersion of the late time decay times despite the similarity of the light curves near peak and the common spectral class . sn 1993j and sn 2011dh show very similar late time light curves for nearly 350 d. sn 2003jd falls more steeply around 30 d after peak , but has a similar rate of decline at later times . sn 1996cb is slower at late time than the others . its decay after 100 d is similar to that of total trapping , but a contribution from circumstellar interaction can not be ruled out . figure [ snib ] gives a sample of 7 sn ib . we have plotted v band data for sn 1983n , rband data for sn 2007y , and bolometric data for sn 2008d and sn 2008ax . sn 2008ax is often classified as a sn iib because some hydrogen was seen early on . by maximum , it looked like a classic sn ib . we have classified it as a sn ib here , but that particular choice does not affect our analysis in any substantial way . these seven events again are rather similar on the rise . sn 1984l flattens to be comparable to the radioactive decay slope , but sn 1984l was a late time radio emitter , so there might be some contribution to the light curve from collision with a circumstellar medium . sn 1999dn and sn 2009jf are similar on the tail for the first 100 d. sn 2009jf subsequently falls at a rate roughly comparable to cobalt decay until at least 250 d , but sn 1999dn falls more steeply at 300 to 400 d. sn 1983n , sn 2007y , sn 2008d and sn 2008ax fall at very similar rates at late times , with sn 2008d being the steepest of this group . figure [ snic ] gives a sample of 9 sn ic . we have plotted b band data for sn 1983v . there appears to be some true dispersion in the early peak widths . sn 2011bm has a distinctly wider peak . sn 1994i falls especially rapidly after maximum . sn 2009bb nearly rivals sn 1994i in terms of steepness of immediate post maximum decline . there is clearly a large dispersion in the rate of decay of the late time tails of the sn ic . sn 2011bm continues to decline slowly , with a late time decay very comparable to that expected for decay . time light curves of the broad line type ic sn 1998bw and sn 2002ap are very similar ; sn 2004aw declines somewhat more slowly than the two broad line events , and sn 1990b somewhat more rapidly . the light curve of sn 1994i flattens at very late times ( 300 d ) , which may have to do with partial trapping of positrons ( clocchiatti et al . 2008 ) . figure [ similar ] gives a selected sample of light curves that seem , by eye , to fall into two categories in terms of late time tails . one group has a decline roughly comparable to decay , and the other is significantly steeper . among the first category with rather shallow late time declines , type ib sn 1984l , type ic sn 2011bm , and perhaps type ib sn 2009jf are the only examples in our sample of late - time light curves that track decay ( corresponding to very large values of @xmath33 ) . as remarked above , time light curves of the type ib sn 1999dn and sn 2009jf are very similar up to 100 d , with sn 1999dn perhaps tracking decay . later data on sn 1999dn fall below this early common trend . either sn 1999dn was not powered by decay earlier , but perhaps by circumstellar interaction , or some other effect , perhaps dust formation , arose after 200 d to suppress the optical light curve . in any case , most stripped envelope supernovae decay considerably faster than expected for complete trapping of . the assortment of light curves of other sn iib , sn ib , and sn ic , including some broad line sn ic , that fall in the second category in figure [ similar ] are also rather similar to one another , but with a distinctly different , steeper slope than those that might follow decay . interestingly , this group comprises 11 of our sample of 20 events . this group includes sn 1983v , but we note that the data are b band , so the decline rate may be exaggerated . the light curves of sn 1983v in both v band and b band were very similar to those of sn 1993j in the respective bands ( clocchiatti et al . 1997 ) , and the light curve we plot here is very similar to that of sn 1993j in r band at late times , but it is appropriate to regard this comparison with some caution . there appears to be some dispersion in peak width for this particular subsample . figure [ fast ] presents data for two events that decay more rapidly at late times than most . both are sn ic . sn 1994i seems to fall more rapidly from peak than sn 1990b . figures [ similar ] and [ fast ] illustrate that neither spectral type nor similar peak light curves determine the slope of the late time tail , and that events of different spectral type can have remarkably similar late time tails , as emphasized by clocchiatti et al . ( 1996 , 1997 ) . the spectra and light curves of stripped envelope supernovae near maximum light are not sufficient to fully classify them . lcccccc 1983n & ib & @xmath38 & @xmath39$]@xmath40 & @xmath41 & @xmath42 & v + 1983v & ic & @xmath43 & @xmath44$]@xmath40 & @xmath45 & @xmath46 & b + 1984l & ib & @xmath47 & @xmath44$]@xmath40 & @xmath48 & @xmath49 & v + 1990b & ic & @xmath50 & @xmath44$]@xmath40 & @xmath51 & @xmath52 & r + 1993j & iib & @xmath53 & @xmath54 & @xmath55 & @xmath56 & bol + 1994i & ic & @xmath57 & @xmath58 & @xmath59 & @xmath60 & r + 1996cb & iib & @xmath61 & @xmath62 & @xmath63 & @xmath64 & r + 1998bw & ic & @xmath43 & @xmath65 & @xmath66 & @xmath67 & r + 1999dn & ib & @xmath47 & @xmath68$]@xmath40 & @xmath69 & @xmath70 & r + 2002ap & ic & @xmath71 & @xmath72$]@xmath40 & @xmath73 & @xmath74 & bol + 2003jd & iib & @xmath43 & @xmath44$]@xmath40 & @xmath45 & @xmath75 & r + 2004aw & ic & @xmath76 & @xmath77$]@xmath40 & @xmath78 & @xmath79 & r + 2007y & ib & @xmath80 & @xmath81 & @xmath82 & @xmath83 & r + 2007gr & ic & @xmath84 & @xmath81 & @xmath85 & @xmath86 & r + 2008d & ib & @xmath47 & @xmath87 & @xmath88 & @xmath89 & r + 2008ax & ib & @xmath90 & @xmath91 & @xmath92 & @xmath93 & bol + 2009bb & ic & @xmath94 & @xmath95$]@xmath40 & @xmath96 & @xmath97 & r + 2009jf & ib & @xmath47 & @xmath62 & @xmath98 & @xmath99 & r + 2011bm & ic & @xmath53 & @xmath100 & @xmath101 & @xmath102 & r + 2011dh & iib & @xmath103 & @xmath91 & @xmath63 & @xmath104 & r + @xmath40 [ tab : timescale ] table [ tab : timescale ] gives for our sample of events , the observed values of @xmath4 , @xmath7 , and @xmath33 and the value of @xmath33 estimated from equation [ t02 ] . we have also attempted to assign representative uncertainties to our estimates of @xmath4 and @xmath7 . for @xmath4 we have assigned an uncertainty of 10% , or about @xmath105 for the typical observed velocities . while not exact , this is a representative uncertainty . for @xmath7 , we have also assigned an uncertainty of 10% for many of the events , 2 or 3 d for typical events , but an uncertainty of 20% for the events where , for various reasons , the rise time is especially ill determined . the values of @xmath7 for these events are given in square brackets in table [ tab : timescale ] to draw attention to these events . we have then propagated these uncertainties through the estimate of @xmath33 from equation [ t02 ] and for other quantities to be discussed in [ discuss ] . we have assumed for this exercise that the observed values of @xmath4 , @xmath7 , and @xmath33 are uncorrelated , but this may not be completely true . the predicted value of @xmath33 in equation [ t02 ] only depends linearly on the rise time and on the square root of the velocity . uncertainty in these parameters are relatively unimportant to that particular quantity . the effects in which we are interested are typically factors of several in @xmath33 , as illustrated in table [ tab : timescale ] . for illustration , we have also made an analytic fit to the tail for the group of 11 events that form a nearly common locus in figure [ similar ] . restricting the analysis to more than 50 d from maximum , we can neglect the contribution of the @xmath106 term in equation ( [ decay ] ) . anticipating that @xmath33 is fairly large , then at 50 d the term @xmath107 in equation ( [ radiated ] ) is essentially zero and can also be neglected . making those two assumptions , one can write the magnitude @xmath108 at the two epochs , 50 d and 400 d , where the similar track in figure [ similar ] passes through about magnitude 17 and 22.5 , respectively . using these values , the resulting parameter is @xmath109 d , corresponding to @xmath110 , a reasonable estimate to the range of values for these 11 events presented more formally in table [ tab : timescale ] . figure [ compare ] shows the observed value of @xmath33 versus that estimated from the peak properties using equation [ t02 ] . this exercise shows clearly that the values of @xmath4 and @xmath111 , and hence the values of @xmath21 and @xmath17 , derived from the peak behavior are poor predictors of the late time behavior . the discrepancies are not subtle . the directly measured values of @xmath33 occasionally agree within a factor of two of those estimated from the peak properties , but the discrepancies are routinely factors of 2 to 4 , and can be as much as a factor of 10(sn 1984l ) . in no case is the estimate of @xmath33 from the peak properties greater than that determined from direct observations of the late time tail . this analysis defines four dilemmas associated with the late time light curves of stripped envelope supernovae . ( 1 ) the relatively similar peak properties often belie a considerable spread in the properties of the late time tail . ( 2 ) even for events with similar peaks and tails , the properties derived from the peak seem to be incommensurate with the behavior of the tail . ( 3 ) events with different spectral types can , nevertheless , have rather similar late time decay . ( 4 ) both typical sn ic and broad lined sn ic can have similar late time decay despite the nominal significant differences in energy associated with the broad lines of the latter . to elucidate these various , related , dilemmas , we can write expressions for the ejected mass and energy in terms of the observed properties as follows . from the peak , we have from equations [ mass ] and [ ke2 ] @xmath112 and @xmath113 and from the tail we have from equation [ t01 ] , invoking @xmath114 , @xmath115 and @xmath116 the complex relationships among the early and late photometric behaviors of different spectral types of stripped envelope supernovae can be illustrated by comparing light curves of several well observed events that have data in table [ tab : timescale ] . the two type ib sn 1994l and sn 2008d have very similar rise times and photospheric velocities , yet very different late time tails , illustrating point ( 1 ) . the typical type iib event sn 2011dh and the hydrogen contaminated type ib event sn 2008ax had similar rise times and photometric velocities of @xmath117 d and @xmath118 , respectively . for similar opacities , one might expect similar ejecta masses and energies and thus similar tails . the tails were similar , but also similar to that of type ic sn 2007gr , a congruence that defies easy explanation , point ( 3 ) above . furthermore , not all events with similar tails have similar peak properties . sn 2007gr had a somewhat more rapid rise , @xmath119 d , and lower photospheric velocity near peak , @xmath120 . both of these differences suggest that , for similar opacities , the ejecta mass ( @xmath121 ; equation [ masspeak ] ) and the kinetic energy ( @xmath122 ; equation [ kepeak ] ) should be less than for sn 2011dh and sn 2008ax . the fact that the late time tail of sn 2007gr is similar to those of sn 2011dh and sn 2008ax and others of different spectral class would then seem to require an unexpected correlation of the ejecta mass and energy . the broad line type ic sn 2002ap had a rise time of @xmath123 d and a photospheric velocity of @xmath124 . based on the peak properties , and again assuming similar opacities , sn 2002ap should have about the same ejecta mass as sn 2007gr , but a kinetic energy that is larger by about a factor of 4 . it is difficult in this simple framework to understand how their late time tails are similar , illustrating point ( 4 ) . equations [ masstail ] and [ ketail ] show that the constraint of similar late time light curves , those with similar values of @xmath33 , implies that @xmath125 and @xmath126 . with these relations , the similar tails imply that the ejecta mass of sn 2002ap must be larger than that of sn 2011dh and sn 2008ax by a factor of @xmath127 and the energy larger by a factor of @xmath128 while maintaining similar late time decay . even for individual events with similar peak and tail properties , it is difficult to reconcile the two phases , point ( 2 ) above . the constraints from the width of the peak , @xmath129 and @xmath130 for a constant opacity and given light curve rise time , are different from those associated with the slope of the tail , @xmath131 and @xmath132 for a given @xmath133-ray opacity and late time decay time . a particular supernova can only have one @xmath21 and one @xmath134 , so both of these sets of constraints need to be at least approximately satisfied . since the peak relations predict @xmath135 and the tail relations predict @xmath136 , there is a conflict . poznanski ( 2013 ) makes an empirical case that the energy scales as the cube of the progenitor mass for sn iip , but no current core collapse simulations predict @xmath136 or @xmath135 . the resolution of this peak / tail conflict may involve consideration of the mean uvoir opacity near the peak , @xmath3 . we note from equations [ masspeak ] and [ kepeak ] that the observed parameters do not constrain the ejected mass and energy but only the degenerate combinations @xmath137 and @xmath138 . by invoking the constraint that the ejected mass determined from peak parameters be equal to the ejected mass constrained from tail parameters , the mean opacity can be formally expressed in terms of constants ( @xmath139 and @xmath140 depend slightly on density structure ) , the reasonably well known value of @xmath20 , and of the observed rise time , photospheric velocity near peak , and the decay time as @xmath141 for typical parameters this can be expressed as @xmath142 where @xmath143 is the rise time in units of 10d and @xmath144 is the tail decay time in units of 100d . we emphasize that this expression for the opacity is basically determined by the observed parameters , @xmath4 , @xmath111 , and @xmath33 . equations [ masspeak ] , [ kepeak ] , [ masstail ] , and [ ketail ] suggest other interesting relations . for a given ejected mass and energy , there is a degeneracy between @xmath3 and the rise time , @xmath111 , such that one can write @xmath145 for events with similar rise times and decay times , this implies that @xmath146 . using equations [ masstail ] and [ ketail ] , one can also write @xmath147 and @xmath148 giving the photospheric velocity ( mean ejecta velocity ) as functions of the ejected mass and energy . we note that equations [ vm ] and [ ve ] are degenerate in the quantities @xmath149 and @xmath150 . we have assumed that @xmath20 is less dependent on explosion parameters than @xmath3 , but have kept the @xmath20 dependence explicit to remind of that dependence . * note added in proof : * _ a small , but significant variation in the arguments we present here would be to consider the mass given by equation [ masspeak ] to not represent @xmath151 , but the quantity @xmath152 , the mass involved in diffusion for which the opacity @xmath153 2 g is relevant . from this perspective , the lhs of equations [ opacity1 ] and [ opacity2 ] would become @xmath154 and hence @xmath155 or @xmath156 for @xmath153 @xmath24 g@xmath25 , equation [ mratio2 ] would then give @xmath157 for fiducial parameters , strongly suggesting that only a rather small portion of the ejecta is ionized . with this redefinition of the mass from equation [ masspeak ] , corresponding changes should be made in equations [ t02 ] , [ opacityme ] , [ vm ] , and [ ve ] . _ the mean opacity derived in this way tends to be @xmath158 2 g for events with typical peak properties and a late time decline of @xmath159 d or @xmath110 that characterizes the various stripped envelope events considered in the examples above ; this is considerably smaller than the estimates normally applied in this context , @xmath153 2 g . we note that the two component model fit of @xcite to sn 2002ap required a mean opacity of order 0.01 @xmath160 g@xmath25 to fit the observed evolution of the photospheric velocity . if this analysis , constrained by the decay properties of the tail , is reasonably correct , then the effective opacity that determines the peak properties may be quite low . many estimates of the ejected mass and energy in the literature that assume or compute a higher opacity could be too low by as much as a factor of 10 . this analysis shows that the mean opacity that dictates the peak properties of stripped - envelope supernovae may have been over estimated in many cases . if so , this would help to resolve dilemma ( 2 ) outlined above and suggest that results derived from the peak alone may be suspect . this analysis has not resolved the other dilemmas , but suggests some new perspectives to bring to the study of stripped envelope and perhaps other core collapse supernovae , suggesting correlations between the energy and ejected mass required to account for the tail behavior and how the opacity and expansion velocity might depend on the ejected mass and energy . lcccccc 1983n & ib & @xmath161 & @xmath162 & @xmath163$]@xmath40 & @xmath164$]@xmath40 & @xmath165$]@xmath40 + 1983v & ic & @xmath166 & @xmath167 & @xmath168$]@xmath40 & @xmath169$]@xmath40 & @xmath170$]@xmath40 + 1984l & ib & @xmath171 & @xmath172 & @xmath173$]@xmath40 & @xmath174$]@xmath40 & @xmath175$]@xmath40 + 1990b & ic & @xmath176 & @xmath177 & @xmath178 & 5.1 & 0.047 + 1993j & iib & @xmath179 & @xmath180 & @xmath181 & @xmath182 & @xmath183 + 1994i & ic & @xmath184 & @xmath180 & 4.2 & 3.3 & 0.017 + 1996cb & iib & @xmath185 & @xmath186 & @xmath187 & @xmath188 & @xmath189 + 1998bw & ic & @xmath190 & @xmath191 & 21 & 25 & 0.010 + 1999dn & ib & @xmath192 & @xmath193 & @xmath194 & @xmath195 & @xmath196 + 2002ap & ic & @xmath197 & @xmath198 & 34 & 47 & 0.0027 + 2003jd & iib & @xmath199 & @xmath200 & @xmath201$]@xmath40 & @xmath202$]@xmath40 & @xmath203$]@xmath40 + 2004aw & ic & @xmath204 & @xmath205 & @xmath206$]@xmath40 & @xmath207$]@xmath40 & @xmath208$]@xmath40 + 2007y & ib & @xmath209 & @xmath210 & @xmath211 & @xmath212 & @xmath213 + 2007gr & ic & @xmath214 & @xmath215 & @xmath216 & @xmath217 & @xmath218 + 2008d & ib & @xmath219 & @xmath220 & @xmath221 & @xmath222 & @xmath223 + 2008ax & ib & @xmath224 & @xmath225 & @xmath226 & @xmath227 & @xmath228 + 2009bb & ic & @xmath225 & @xmath229 & @xmath230$]@xmath40 & @xmath231$]@xmath40 & @xmath232$]@xmath40 + 2009jf & ib & @xmath233 & @xmath234 & @xmath235 & @xmath236 & @xmath237 + 2011bm & ic & @xmath238 & @xmath239 & @xmath240 & @xmath241 & @xmath242 + 2011dh & iib & @xmath243 & @xmath244 & @xmath245 & @xmath246 & @xmath247 + @xmath40 @xmath248 ; @xmath249 ; @xmath250 ; @xmath251 ; @xmath252 [ tab : mekappa ] we summarize the implications of these considerations for the supernova sample we analyze here in table [ tab : mekappa ] . columns 3 and 4 give the estimates of the ejected mass and energy derived from peak properties , scaled to a value of the mean opacity of 0.1 2 g , from equations [ masspeak ] and [ kepeak ] . columns 5 , 6 , and 7 give the ejected mass and energy according to equations [ masstail ] and [ ketail ] and the opacity according to equation [ opacity1 ] that would be required to yield the latter values of mass and energy . we have propagated the uncertainties in the observed parameters @xmath4 , @xmath111 , and @xmath33 to express the uncertainties in the various derived quantities . as noted in [ anal ] , we formally assume that the uncertainties in these observed quantities are uncorrelated . equations [ opacityme ] , [ vm ] , and [ ve ] , for instance , suggest that this may not be true . we reserve a deeper study of this issue for future work . as also noted in [ anal ] , our previous determinations of the late time decay parameter , @xmath33 , are probably overestimates because we treated the leakage of positrons as similar to that of gamma rays . while positrons may not be completely trapped , there is evidence that they are at least partially so in sn 1994i ( clocchiatti et al . the uncertainty in the positron deposition function therefore leads to an intrinsic uncertainty in the quantitative value of @xmath33 that comes in squared in equations [ masstail ] and [ ketail ] . assuming that positrons are completely trapped would lead to an underestimate of @xmath33 if they are only partially trapped . we have accounted for the uncertainty in positron trapping and other uncertainties in the following way in table [ tab : mekappa ] . for events for which our uncertainties in the observed value of @xmath33 were large ( sn 1984l , sn 2003jd , sn 2004aw , sn 2009bb ) we take the results for ejected mass , energy , and mean opacity based on the slope of the tail to be essentially undetermined , but present the values formally derived in square brackets . we also regard events without decent r band or quasi bolometric light curves ( sn 1983n , sn 1983v , sn 1984l ) to be ill undetermined in this context and also present the associated data in square brackets . for events for which we have determined a late time decay here with an exaggerated treatment of positron leakage , the majority of the events , we take the results to be upper limits on ejecta mass and energy and a lower limit on the mean opacity . finally , for the few events for which clocchiatti et al . ( 2008 ) estimated @xmath33 assuming complete trapping ( sn 1994i , sn 1998bw , sn 2002ap , sn 1990b ) , we give estimates based on those values of @xmath33 ( 65d , 120d , 142d , and 88d , respectively ) , recognizing that these could be lower limits on the ejecta mass and energy and upper limits on the mean opacity . clocchiatti et al . do not give uncertainties for their values of @xmath33 so we have not assigned formal uncertainties for these four events . from the uncertainties in @xmath4 and @xmath111 alone , there would be uncertainties in the derived quantities of the order of 10 to 20% . the limits we derive for sn 1984l are quite extreme , as follows from the large value of @xmath253d we find to fit the flat tail . as the ejecta mass increases , it should become more difficult to determine a precise value of @xmath33 . the standard deviation we find , 109 d , is substantial but still rather small compared to the best fitting value . at the 3@xmath254 lower limit on @xmath33 , the estimates would be less by about a factor of 4 , but still large . we may have underestimated the uncertainty here . in any case , the propagated uncertainties from all the observed values make the estimates for sn 1984l from the tail properties indeterminate . as noted above , the large uncertainties on sn 2003jd , sn 2004aw , and sn 2009bb mean that the formal results for them are also of little import . for the other events with upper limits to the ejecta mass and energy in table [ tab : mekappa ] , there is a suggestion of considerable spread in derived values . of special interest are the results for the events where the observed values of @xmath33 were determined in the context of a model that included positron trapping . for sn 1990b , the results for the ejecta mass and energy were of order a factor of 2 larger using the tail properties rather than the peak properties with the chosen mean opacity . for sn 1994i , the discrepancy was even larger , about a factor of 5 . the cases of greatest interest were the broad lined events , sn 1998bw and sn 2002ap , where the discrepancy in prediction is an order of magnitude or more . this analysis has enunciated four dilemmas associated with the late time light curves of stripped envelope supernovae . ( 1 ) there is more heterogeneity in the tail properties than suggested by analyses of the peak alone . ( 2 ) the properties derived from the peak are often inconsistent with the properties of the tail . ( 3 ) events with different spectral types can have rather similar tails . ( 4 ) both typical sn ic and broad lined sn ic can have similar late time decay despite the implied differences in energy . the values for ejecta mass and energy derived from the peak and the parameters that determine them frequently predict too steep a late time decline , especially in simple models that assume a constant opacity over the peak of @xmath255 2 g . the physics of the late time light curves is fairly simple , so the discrepancies may lie in the physics of the peak , despite the similarity of the immediate post peak decline determined by drout et al . ( 2011 ) , cano ( 2103 ) , lyman et al . ( 2014 ) and taddia et al . care must be taken with values of the ejecta mass and kinetic energy derived with simple estimates based only on the properties of the peak . our primary goal was not to derive ejecta masses and kinetic energies , rather to caution that these quantities can be sensitive to the way the data are selected and employed . nevertheless , we can compare these quantities with related work . we note that drout et al . ( 2011 ) employed a measure of the rate of decline to determine estimates of masses and energies , but that this time was in some instances shorter than the observed portion of the rise . the rise times we present in table [ tab : timescale ] are typically a factor of 2 3 larger than those given for the same events by drout et al . drout et al . also assigned @xmath4 in only two bins , 10,000 for sn ib and normal sn ic and 20,000 for sn ic - bl . this will tend to mute the true dispersion of the physical properties of the supernovae that are related to differences in observed velocity . also in contrast , cano ( 2013 ) gives mean values for @xmath4 for sn ib of 8027 , for sn ic of 8470 , and for sn ic - bl of 15,114 . cano does not discriminate between @xmath4 and @xmath256 as we have here . for our constant opacity peak models with @xmath23 2 g , we tend to assign lower values of @xmath21 and @xmath17 than does cano ( for sn 1983n , sn 1999dn , sn 2002ap , sn 2007gr , sn 2008d , sn 2009jf , and sn 2011bm ) , but agree rather closely in several cases ( sn 1994i , sn 2003jd , sn 2007y , and sn 2009bb ) . we assign higher values than cano for sn 2004aw . we can not compare directly to taddia et al . ( 2015 ) because their sample was based on sdss objects not discussed in the general literature . the fits of taddia et al . were mostly for light curves of less than 60 d after maximum . their fitting procedure also involved a construction that employed more parameters than do ours . in a sense , the ratio of our values , @xmath257 , is another parameter determined by our process . lyman et al . ( 2014 ) also use a measure of the decline rate , rather than the rise , to determine the characteristic light curve width " and employ a two parameter fitting procedure similar to that of taddia et al . to fit light curve properties up to about 80 d after maximum . the analysis of the light curves of stripped envelope supernovae has a long history . @xcite invoked clumping as a possible way to address the difficulty of reproducing the peak / tail contrast . @xcite discussed the possibility that the dynamics were in some fashion non homologous " with part of the ejecta remaining in a dense core . @xcite constructed a two component model that could fit the data better by dint of having extra parameters . they ascribed the second component to a dense inner region , analogous to the non homologous " component of clocchiatti & wheeler , to asymmetries in the explosion . subsequent work has also invoked simple models with more parameters and hence more elaborate fits ( cano et al . 2014 ; lyman et al . 2014 ; taddia et al . 2015 ) . the enduring problem has been to get the physics of the peak to agree with the physics of the tail . most suggested solutions have focused on the processes of gamma ray deposition . here we suggest that the treatment of the opacity may be critical . in truth , the two are intimately related and must be treated together and self - consistently . a possible factor that could yield a narrower peak for a given ejecta mass and kinetic energy and account for the low mean opacity suggested by equation [ opacity2 ] is recombination that would give a time dependent discontinuity in the opacity , akin to that in hydrogen in sn iip . @xcite made the first detailed quantitative models of sn ib light curves based on helium cores from massive stellar models . they also presented one of the few careful exegeses of the nature of recombination fronts and their effect on opacity in this context . whereas many simple models invoke a constant mean opacity over the peak of stripped envelope models , the opacity is very sensitive to temperature , density , and expansion effects ( hflich , mller & khokhlov 1993 ; pinto & eastman 2000 ) . interestingly , the models of ensman & woosley produced less peak to tail contrast than the observations , rather than giving too steep a tail decline , the problem we have focused on here . in hindsight , this is because ensman & woosely employed rather massive helium star models that retained a substantial fraction of the ejecta mass as helium . in their models , all the helium and much of the oxygen recombined before maximum , yielding a significant mass that had low opacity and contributed little to the optical diffusion time . most of the opacity over the peak was contributed by the relatively low mass of the inner iron and silicon layers . the low effective opacity ( ensman & woosley mention a typical value of 0.003 2 g ) allowed them to fit the peak of sn 1983n with a rather large ejecta mass , but the whole ejecta mass contributed to the subsequent gamma - ray trapping , hence yielding excessively flat late time light curves . subsequently , models of stripped envelope light curves have invoked smaller ejecta masses . detailed hydrodynamic models that have included complex prescriptions for the opacity have yielded rather large mean opacities , @xmath255 2 g . this class of models nearly always found the late time light curve to decline too steeply . attempts to correct this problem have often addressed adjustments to the structure that would affect the gamma ray trapping . the issue then arises as to whether these models that faded too quickly on the tail had , by assumption or computation , too large an effective opacity over the peak and hence too small an ejecta mass and too small an effective gamma ray optical depth . the uvoir opacity that governs the shape of the peak is intrinsically complex with dependence on density , temperature , and composition and on nonlte , nonthermal and expansion effects . recombination depends on adiabatic expansion and cooling ; subsequent reionization and increase in the opacity depends on the deposition of energy from radioactive decay . recombination can lead to a situation where the opacity is a sensitive function of both space and time with the net effect being to decrease the effective opacity and thus yield a narrower peak light curve for a given ejecta mass . the models of @xcite amply illustrate that there can be significant recombination effects on the mean effective opacity even in spherically symmetric models . in the context of sn 1994i , @xcite also noted that the recombination of oxygen could lead yield a narrower peak ( see also piro & morozova ( 2014 ) . some numerical models in the literature set a floor on the opacity . it is possible that this floor is set too high . this raises the issue of how much cold , neutral material may exist in the ejecta of various types of stripped envelope supernovae near peak . there must be some excited he i in both sn iib and sn ib . what is less clear is whether or not there may be a substantial amount of helium that is not excited by gamma - ray deposition . @xcite argue that even with an appreciable layer of helium , a supernova would not appear as a sn ib without outward mixing of . in principle , there could be a considerable amount of unexcited helium even in sn ic . in each spectral type , and especially for sn ic , the presence of a significant mass of cold , neutral helium will also affect the spectrum . a distinct identifying feature of sn ic is the strong absorption corresponding to . this is one of the strongest features in the spectra of sn ic , being distinctly stronger than in sn ib @xcite . models that include a significant mass of unexcited helium decelerate the oxygen and hence tend to produce oxygen lines that are too shallow , red , and narrow to correspond to the observed feature in typical sn ic @xcite . asymmetries in the explosions of stripped envelope supernovae almost surely play a significant role . from spectropolarimetry , we know that stripped envelope supernovae are asymmetric and that different elements are ejected at different angles @xcite . it is instructive to contemplate the image of the sn iib cas a , rent by jets and riddled with irregularities due to nickel bubbles @xcite . this structure is probably a good representation of any realistic stripped envelope supernova . gamma ray deposition on the tail is relatively simple , being independent , for instance , of composition and relatively insensitive to density irregularities . the physics of the @xmath133-ray deposition around the peak depends on the distribution of and . any outward mixing " of would tend to release @xmath133-rays more quickly and hence yield a more rapid rise , but then also lead to a more rapid late time decay , which is already a problem with many models . there may be some turbulent microscopic " mixing , but even rayleigh taylor instabilities tend to form distinct macroscopic plumes in the non linear limit @xcite . the early deposition of gamma rays will not be azimuthally symmetric and hence neither will be the opacity , in contrast to the conditions imposed in any spherically symmetric model . the current analysis suggests that the mean opacity around peak is less than frequently assumed or computed . the opacity is likely to be dependent on asymmetries that have not yet been adequately explored . the small effective opacity we derive suggests that the deposition and heating are irregular in real 3d explosions and leave substantial cold , low opacity , mass near peak that , nevertheless , traps @xmath133-rays on the tail . consideration of the nature of the opacity near peak might point the way to resolving the second dilemma that we have outlined , the self - consistency of the peak and tail properties . this perspective may also give some insight into the first dilemma , that there may be more variety in tail behavior than is clear at first blush from the peak properties . perhaps the peak opacity and diffusion time is controlled by the fraction of the ejecta comprised of iron and silicon peak elements , and the mass of those elements is more uniform than the total ejecta mass . it is striking that a substantial subset exists with similar peaks and similar tails comprising events of all stripped envelope spectral types , sn iib , sn ib , and sn ic , and with a range of photospheric velocities . amid the diversity we emphasize , there is a uniformity that demands deeper understanding . the third dilemma , that stripped envelope events of different spectral type , and hence different compositions , ejecta masses and perhaps energies , can have similar late time light curves remains a challenge . we have shown that solutions can be found with appropriate choices of the mean opacity , but that choice implies a scaling with the ejecta mass and energy ( equation [ opacityme ] ) that is far from obvious . similar concerns remain for the fourth dilemma , how both typical and broad line sn ic can have similar late time light curves . the implication is that the energy is somehow correlated with the ejected mass in an unexpected way . our results invoking the variation in the opacity in accord with equation [ opacity2 ] imply a rather large range in ejecta masses and energies as given in table [ tab : mekappa ] and generally larger estimates of ejecta mass and energy than other studies . taken at face value , these results would require a substantial reconsideration of the progenitor evolution . small ejecta masses are not compatible with the observed class of wolf rayet stars . larger ejecta masses might be . consideration of larger mass progenitors must also confront issues of predicted rates of explosions compared to observations and issues of the sites of the explosions of the various variety of stripped envelope events @xcite . models of the late time spectra of sn 1985f were more compatible with the explosion of a helium core of 8 than a core of 4 ( fransson & chevalier 1989 ) . some of the issues concerning the light curves of stripped envelope supernovae might be clarified with more investigations of this type to determine the ejecta mass by independent means . even within the simple framework we employ here there are refinements that would make a quantitative difference . the constants " @xmath139 , c , and the 3/5 " that relates the mean velocity to the photospheric velocity are functions of the density distribution and could be explored for a variety of assumed power law exponents @xcite . the nature of positron trapping could be explored more thoroughly . in the future , better sampled light curves from peak to tail would be valuable , and the most accurate estimates of the observed uvoir quasi bolometric light curve should be employed . we do not think these aspects would make a qualitative difference compared to the effects of asymmetry , ionization , and recombination we discuss here . clocchiatti et al . ( 1997 ) emphasized the depths of the conundrum presented by events with similar late time tails but different spectral types and different photospheric velocities . from equation ( [ t01 ] ) , two events with the same value of @xmath258 must have a ( nearly ) constant value of @xmath259 . the similar late time light curves would then require some very unexpected correlation between the kinetic energy and ejecta mass and with the the mean , and hence presumably photospheric , velocity . now , as then , the problem posed by the similar light curves and different velocities remains a challenge . we thank stefano benetti , nando patat , stefan taubenberger and stefano valenti for sharing their bolometric light curve data , dave arnett and jozsef vink for a discussion of arnett s light curve model , maryam modjaz , federica bianco and mattias ergon for comments , david branch for asking a key question , and sean couch for a valuable perspective . this research is supported in part by nsf ast-1109801 and by the millennium center for supernova science through grant p06 - 045-f funded by programa iniciativa cientfica milenio de mideplan . " arnett , w. d. 1982 , , 253 , 785 arnett , d. 1996 , supernovae and nucleosynthesis : an investigation of the history of matter , from the big bang to the present , princeton : princeton university press , 1996 . , barbon , r. , benetti , s. , cappellaro , e. , et al . 1995 , , 110 , 513 benetti , s. , turatto , m. , valenti , s. , et al . 2011 , , 411 , 2726 bianco , f. b. , modjaz , m. , hicken , m. , et al . 2014 , , 213 , 19 cano , z. 2013 , , 434 , 1098 cano , z. , maeda , k. , & schulze , s. 2014 , , 438 , 2924 cappellaro , e. , mazzali , p. a. , benetti , s. , et al . 1997 , , 328 , 203 chatzopoulos , e. , wheeler , j. c. , & vinko , j. 2012 , , 746 , 121 chatzopoulos , e. , wheeler , j. c. , vinko , j. , horvath , z. l. , & nagy , a. 2013 , , 773 , 76 clocchiatti , a. , wheeler , j. c. , benetti , s. , & frueh , m. 1996 , , 459 , 547 clocchiatti , a. , wheeler , j. c. , phillips , m. m. , et al . 1997 , , 483 , 675 clocchiatti , a. , & wheeler , j. c. 1997 , , 491 , 375 clocchiatti , a. , suntzeff , n. b. , phillips , m. m. , et al . 2001 , , 553 , 886 clocchiatti , a. , wheeler , j. c. , kirshner , r. p. , et al . 2008 , , 120 , 290 clocchiatti , a. , suntzeff , n. b. , covarrubias , r. , & candia , p. 2011 , , 141 , 163 dessart , l. , hillier , d. j. , li , c. , & woosley , s. 2012 , , 424 , 2139 drout , m. r. , soderberg , a. m. , gal - yam , a. , et al . 2011 , , 741 , 97 ensman , l. m. , & woosley , s. e. 1988 , , 333 , 754 ergon , m. , sollerman , j. , fraser , m. , et al . 2014 , , 562 , a17 ergon , m. , jerkstrand , a. , sollerman , j. , et al . 2015 , arxiv:1408.0731 fransson , c. , & chevalier , r. a. 1989 , , 343 , 323 hammer , n. j. , janka , h .- t . , mller , e. 2010 , , 714 , 1371 hflich , p. , mller , e. , & khokhlov , a. 1993 , , 268 , 570 kelly , p. l. , filippenko , a. v. , modjaz , m. , & kocevski , d. 2014 , , 789 , 23 kleiser , i. k. w. , & kasen , d. 2014 , , 438 , 318 lyman , j. d. , bersier , d. , & james , p. a. 2014 , , 437 , 3848 maeda , k. , mazzali , p. a. , deng , j. , et al . 2003 , , 593 , 931 malesani , d. , fynbo , j. p. u. , hjorth , j. , et al . 2009 , , 692 , l84 marion , g. h. , vinko , j. , kirshner , r. p. , et al . 2014 , , 781 , 69 matheson , t. 2001 , , 113 , 1155 maund , j. r. , wheeler , j. c. , baade , d. , et al . 2009 , , 705 , 1139 milisavljevic , d. , & fesen , r. a. 2013 , , 772 , 134 modjaz , m. , li , w. , butler , n. , et al . 2009 , , 702 , 226 modjaz , m. , blondin , s. , kirshner , r. p. , et al . 2014 , , 147 , 99 pastorello , a. , kasliwal , m. m. , crockett , r. m. , et al . 2008 , , 389 , 955 patat , f. , cappellaro , e. , danziger , j. , et al . 2001 , , 555 , 900 pignata , g. , stritzinger , m. , soderberg , a. , et al . 2011 , , 728 , 14 pinto , p. a. , & eastman , r. g. 2000 , , 530 , 757 piro , a. l. , & morozova , v. s. 2014 , , 792 , ll11 poznanski , d. 2013 , , 436 , 3224 qiu , y. , li , w. , qiao , q. , & hu , j. 1999 , , 117 , 736 richmond , m. w. , van dyk , s. d. , ho , w. , et al . 1996 , , 111 , 327 richmond , m. w. , treffers , r. r. , filippenko , a. v. , et al . 1994 , , 107 , 1022 sahu , d. k. , gurugubelli , u. k. , anupama , g. c. , & nomoto , k. 2011 , , 413 , 2583 stritzinger , m. , mazzali , p. , phillips , m. m. , et al . 2009 , , 696 , 713 swartz , d. a. , & wheeler , j. c. 1991 , , 379 , l13 taddia , f. , sollerman , j. , leloudas , g. , et al . 2014 , arxiv:1408.4084 taubenberger , s. , pastorello , a. , mazzali , p. a. , et al . 2006 , , 371 , 1459 tomita , h. , deng , j. , maeda , k. , et al . 2006 , , 644 , 400 tsvetkov , d. y. , volkov , i. m. , sorokina , e. , et al . 2012 , peremennye zvezdy , 32 , 6 valenti , s. , elias - rosa , n. , taubenberger , s. , et al . 2008 , , 673 , l155 valenti , s. , benetti , s. , cappellaro , e. , et al . 2008 , , 383 , 1485 valenti , s. , taubenberger , s. , pastorello , a. , et al . 2012 , , 749 , l28 vink , j. , blake , r. m. , srneczky , k. , et al . 2004 , , 427 , 453 zhang , t. , wang , x. , zhou , x. , et al . 2004 , , 128 , 1857

transiting planets around rapidly rotating stars are not amenable to precise radial velocity observations , such as are used for planet candidate validation , as they have wide , rotationally broadened stellar lines . such planets can , however , be observed using doppler tomography , wherein the stellar absorption line profile distortions during transit are spectroscopically resolved . this allows the validation of transiting planet candidates and the measurement of the stellar spin - planetary orbit ( mis)alignment , an important statistical probe of planetary migration processes . we present doppler tomographic observations which provide a direct confirmation of the hot jupiter kepler-13 ab , and also show that the planet has a prograde , misaligned orbit , with . our measured value of the spin - orbit misalignment is in significant disagreement with the value of previously measured by from the gravity - darkened _ kepler _ lightcurve . we also place an upper limit of ( 95% confidence ) on the mass of kepler-13 c , the spectroscopic companion to kepler-13 b , the proper motion companion of the planet host star kepler-13 a.

1406.0512

observations over the past few years have shown that many transiting exoplanets ( principally hot jupiters ) have significantly non - zero orbital inclinations . this is , in most cases , measured via the rossiter - mclaughlin effect , which is a probe of the sky - projected orbital inclination ( @xmath3 ) , also known as the spin - orbit misalignment ( e.g. , * ? ? ? * ; * ? ? ? @xcite noted two different regimes in the distribution of @xmath3 versus stellar @xmath4 . planets orbiting cooler stars ( @xmath5 k ) tend to have aligned orbits ( with a few notable exceptions ) , while those orbiting hotter stars ( @xmath6 k ) have a much wider distribution of spin - orbit misalignments that is consistent with isotropic @xcite . several hypotheses have been put forward to explain these two regimes . @xcite proposed that most hot jupiters are emplaced on highly inclined orbits by processes such as planet - planet scattering ( e.g. , * ? ? ? * ) or kozai cycles ( e.g. , * ? ? ? @xmath7 k marks the location on the main sequence where cooler stars have deep , massive convective zones , whereas hotter stars do not . @xcite hypothesized that cooler stars convective zones are able to efficiently tidally couple to the planet and damp out the planetary orbital inclination within the main sequence lifetime , whereas those of hotter stars are not . @xcite recently presented simulations confirming the plausibility of this idea . @xcite instead proposed that hot jupiters are emplaced by disk migration within an inclined disk , coupled with the same tidal dissipation hypothesis as @xcite . the disk is torqued out of alignment with the stellar spin axis by gravitational interactions with a transitory binary companion on an inclined orbit in the birth cluster . further simulations along these same lines by @xcite and @xcite included magnetic and gravitational interactions between the host star and the disk ; both found that this remains a viable misalignment mechanism . @xcite had earlier found that magnetic interactions between the star and disk alone could torque the star out of alignment with the disk . @xcite argued that time variability in the bulk angular momentum of material being accreted by a protoplanetary system could result in a spin - orbit misalignment between the star and the planet - forming disk . another mechanism was proposed by @xcite , who modeled angular momentum transport via internal gravity waves within hot stars , and suggested that such angular momentum transport could drastically change the rotational properties of the stellar atmosphere on short time scales . the rotation of the stellar atmosphere , which is what is probed by all spin - orbit misalignment measurement techniques , would not reflect the bulk rotation of the star . an apparent spin - orbit misalignment could thus be generated even when the bulk angular momentum vectors of the star and planet are in fact well aligned . furthermore , @xcite called into question whether tidal damping could affect inclinations as proposed by @xcite and @xcite . @xcite found that in order for tidal damping not to result in significant semi - major axis changes , inclinations must be driven to @xmath8 , which is not observed . @xcite , however , showed that the latter two of these states would eventually decay to the zero inclination state . in general , these hypotheses fall into two categories : either the planets have changed their orbital plane after their formation , or the planetary orbit and stellar rotation axes are misaligned for reasons unrelated to planet evolution , and are related to star formation or stellar physics . measurement of the spin - orbit misalignments of a statistically significant sample of long - period planets ( which should not have undergone significant tidal damping ) and multi - planet systems ( which should not have undergone violent migratory processes ) around both hot and cool stars will help to discriminate between these hypotheses . the vast majority of the measurements of the spin - orbit misalignments of transiting exoplanets have come via radial velocity observations of the rossiter - mclaughlin effect , where distortions in the stellar line profile during the transit are interpreted as an anomalous radial velocity shift . an alternative method , which we utilize , is doppler tomography , which has been used to probe spin - orbit misalignments for both planets ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and stars ( e.g. , * ? ? ? * ) . here , the spectral line profile distortions are spectroscopically resolved and tracked over the course of the transit . the motion of the line profile perturbation during the transit is a probe of the spin - orbit misalignment @xmath3 . while for the most rapidly rotating planet - host stars @xmath3 can be measured purely from photometry due to the effects of gravity darkening on the surface brightness profile of the star ( e.g. , * ? ? ? * ; * ? ? ? * ) , this method results in a four - fold degeneracy between @xmath9 and @xmath10 . doppler tomography can break this degeneracy . in addition to measurements of @xmath3 , doppler tomography can be used to validate transiting planet candidates around rapidly rotating stars . these stars are not amenable to follow - up using high precision radial velocity observations due to their significantly rotationally broadened stellar lines . detection of the doppler tomographic transit signature allows us to verify that the transiting object is indeed orbiting the expected star , i.e. , that the system is not a background eclipsing binary blended with a brighter foreground star . by examining the line shape we can also rule out scenarios where the transiting object is another star , as we will be able to see an additional set of absorption lines superposed upon those of the primary . the limitation , however , is that doppler tomography can not measure the mass of the transiting object , and thus we can not distinguish between a hot jupiter , a brown dwarf , and a small m dwarf . all of these have similar radii and the latter of these would , in many cases , have an insufficient flux ratio to make a detectable imprint upon the visible light spectrum of the primary . to date the only transiting planet candidate validated using doppler tomography is wasp-33b @xcite . there are , however , a number of planet candidates discovered by the _ kepler _ mission around rapidly rotating stars which can be validated using doppler tomography . we have begun a program using the telescopes at mcdonald observatory to validate suitable candidates , with a particular focus on longer - period candidates . these will provide a test of the hypotheses described above , as these planets should not have undergone significant tidal damping and so should retain their primordial orbital alignments . in this paper we describe our doppler tomography code and present our observations of the hot jupiter kepler-13 ab . although kepler-13 ab has been validated as a planet using doppler beaming and ellipsoidal variations ( e.g. , * ? ? ? * ) , it is one of the most favorable _ kepler _ targets for doppler tomography and thus presents a good test of our code . the kepler-13 ( aka koi-13 , bd+46 2629 ) system has long been known to be a proper motion binary @xcite . @xcite determined that it consists of two a - type stars with similar properties ( see table [ starknowledge ] ) , which are separated by 1.12 @xcite . @xcite also determined that the transiting planet kepler-13 ab ( detected by * ? ? ? * ) orbits the brighter of the two binary components , kepler-13 a. despite the resulting blend , as the separation between kepler-13 a and b is much smaller than the size of one of _ kepler _ s pixels , the inferred radius for kepler-13 ab remains in the planetary range , albeit at the highly inflated end of that range . this is unsurprising , considering the luminous host star and close orbital proximity of the planet to the star , and consequently high planetary temperature . @xcite detected a third stellar component in the system in an eccentric binary orbit about kepler-13 b via the reflex motion of star b. they determined that this companion , kepler-13 c ( denoted kepler-13 bb by * ? ? ? * ) , has a mass of @xmath11 and an orbital period of 65.8 days . kepler-13 ab thus orbits one member of a stellar triple system ; alternatively , due to the massive nature of the planet kepler-13 ab , the system could be considered to be a hierarchical quadruple . kepler-13 a is distinguished as one of the hottest stars to host a confirmed planet ( @xmath12 k ) . stellar parameters for the three stars in the kepler-13 system are given in table [ starknowledge ] , while planetary and transit parameters are summarized in table [ oldknowledge ] . as kepler-13 ab is a hot jupiter , it is one of the hottest known planets ; @xcite estimated @xmath13 k using the secondary eclipse depth in the _ kepler _ passband . kepler-13 ab was first validated by @xcite through detection of a gravity - darkening signature in the transit lightcurve from _ kepler_. this also enabled them to measure the spin - orbit misalignment , albeit with degeneracies , to be @xmath14 or @xmath15 . @xcite , @xcite , @xcite , @xcite , and @xcite detected doppler beaming and ellipsoidal variations due to the planetary orbit , and used these to measure the mass of kepler-13 ab to be @xmath16 , putting it firmly below the deuterium burning limit . many of these different authors , however , found conflicting values for some of the transit and system parameters , especially the impact parameter @xmath17 , ranging from 0.25 to 0.75 ( see table [ oldknowledge ] for the planetary parameters ) . while the orbital plane of kepler-13 ab has been shown to be precessing , resulting in changes in the transit duration and impact parameter @xcite , the rate of change of the impact parameter found by @xcite , @xmath18 yr@xmath19 , is much too small to account for these discrepancies . while @xcite found no evidence for orbital eccentricity , recently @xcite measured a secondary eclipse time offset by @xmath20 seconds from that expected assuming a circular orbit . this could be caused by either a very small eccentricity ( @xmath21 ) , or a bright spot on the planetary surface offset to the west of the substellar point . kepler-13 a is rapidly rotating ( @xmath22 km s@xmath19 ; * ? ? ? * ) and bright for a _ kepler _ target ( @xmath23 ) , making it an excellent target for doppler tomography . while there is a previous measurement of @xmath3 via gravity darkening @xcite , as noted above this method can not distinguish between prograde and retrograde orbits . we can break this degeneracy with doppler tomography . with this work kepler-13 ab becomes the first planet with measurements of @xmath3 from both photometric and spectroscopic techniques , an important consistency check . additionally , @xcite showed that , in addition to the stellar @xmath4 , the planetary scaled semi - major axis @xmath24 and mass ratio @xmath25 are correlated with the degree of alignment . a measurement of the spin - orbit misalignment for kepler-13 ab helps to expand the parameter space , as it is a particularly massive planet orbiting close to a massive star . lcc & system parameters & + @xmath26 ( pc ) & & @xmath27 + @xmath28 ( gyr ) & & @xmath29 + @xmath30 ( mag ) & & @xmath31 + & kepler-13 a & + @xmath32 ( mag ) & & 9.9 + @xmath4 ( k ) & & @xmath33 + @xmath34 ( cgs ) & & @xmath35 + @xmath36fe / h@xmath37 $ ] & & 0.2 + @xmath38 ( km s@xmath19 ) & @xmath39 & @xmath40 + @xmath41 & & 2.05 + @xmath42 & & 2.55 + & kepler-13 b & + @xmath32 ( mag ) & & 10.2 + @xmath4 ( k ) & & @xmath43 + @xmath34 ( cgs ) & & @xmath44 + @xmath36fe / h@xmath37 $ ] & & 0.2 + @xmath38 ( km s@xmath19 ) & @xmath45 & @xmath46 + @xmath41 & & 1.95 + @xmath42 & & 2.38 + & kepler-13 c & + @xmath47 ( days ) & @xmath48 & + @xmath49 & @xmath50 & + @xmath51 ( km s@xmath19 ) & @xmath52 & + @xmath41 & @xmath53 & + observations of kepler-13 ab were taken with two telescopes located at mcdonald observatory , the 9.2 m hobby - eberly telescope ( het ) and the 2.7 m harlan j. smith telescope ( hjst ) . the het utilizes a fiber - fed cross - dispersed echelle spectrograph , the high - resolution spectrograph ( hrs ; * ? ? ? the fibers have a diameter of 2 , and so our observations include blended light from both kepler-13 a and b ( the mutual separation is @xmath54 ; * ? ? ? this complication is discussed in more detail later in the text . the robert g. tull spectrograph ( ts23 ; * ? ? ? * ) on the hjst , on the other hand , is a more traditional slit coud spectrograph . there is no facility to correct for image rotation , and so the relative contributions to the spectrum from kepler-13 a and b vary throughout the course of an observation . while this can , in principle , be corrected for , guiding errors will also cause similar but unpredictable variations . we therefore do not attempt such a correction . our hrs observations were taken with a resolving power @xmath55 , while the ts23 observations have @xmath56 . the spectral range of hrs is @xmath57 to @xmath58 , while that of ts23 is @xmath59 to @xmath60 ; however , none of the orders redward of @xmath61 were used due to telluric contamination and lack of stellar lines . the exposure time was 300 seconds for all het observations and 900 seconds for all hjst observations . the mean per pixel signal - to - noise ratio of the continuum is 159 for the het data and 51 for the hjst data ; the mean snrs for the individual datasets are listed in table [ observations ] . we observed parts of nine transits of kepler-13 ab , seven with the het and two with the hjst ; see table [ observations ] . the transit of 2011 november 5 ut was simultaneously observed with both the het and the hjst . an additional out - of - transit spectral line template observation was obtained with the het on 2013 june 28 ut , in order to better determine the out - of - transit line profile . lcccc 2011 jun 8 & het / hrs & @xmath62 & 150 & 11 + 2011 jun 15 & hjst / ts23 & @xmath63 & 53 & 16 + 2011 jul 6 & het / hrs & @xmath64 & 198 & 16 + 2011 jul 8 & het / hrs & @xmath65 & 183 & 15 + 2011 aug 21 & het / hrs & @xmath66 & 162 & 16 + 2011 sep 13 & het / hrs & @xmath67 & 172 & 15 + 2011 nov 5 & het / hrs & @xmath68 & 135 & 15 + 2011 nov 5 & hjst / ts23 & @xmath69 & 48 & 11 + 2012 jun 7 & het / hrs & @xmath70 & 138 & 17 + 2013 jun 28 & het / hrs & template & 120 & 12 we perform data reduction using the same iraf pipelines utilized by the mcdonald observatory radial velocity planet search program for het / hrs ( e.g. , * ? ? ? * ) and hjst / ts23 ( e.g. , * ? ? ? the extracted spectra are then divided by the blaze - profile function , and any residual curvature is removed by fitting a second - order polynomial using a @xmath71-clipping routine and normalizing . the first step in the analysis of the time series line profiles is to extract these line profiles from our spectra . essentially , we wish to compute the average line profile for each spectrum . we note that in computing an average line profile across a spectrum we ignore variations in the limb darkening parameter as a function of wavelength . as we are interested in the variations in the line profile as a function of time , rather than the detailed line shape , this should not have a significant effect upon our results . the extraction of the average line profiles from the spectra proceeds in several steps . all steps involve fitting a model spectrum to the data . in all cases this model is produced using the least squares deconvolution method of @xcite . in this method , a model spectrum is produced by convolving a model line profile with a series of appropriately weighted delta functions at the wavelengths of the spectral lines . we fit this model spectrum to the data using the least squares methods of @xcite , as implemented in the idl function mpfit and derivatives . we first select several orders of the spectrum with many telluric lines and few or no stellar lines . we produce a model telluric spectrum using least squares deconvolution using a telluric line list ( obtained from the geisa database ) , and assuming a gaussian line profile . this model spectrum is fit to the data , leaving only the velocity offset between the extracted spectrum and the telluric rest frame as a free parameter . we assume that the telluric rest frame is identical to the spectrograph rest frame ( @xmath72 the wind speed , which is much smaller than the velocity scales of interest to us ) , and so we shift the spectra into this frame . telluric lines have been shown to be a stable velocity standard ( e.g. , * ? ? ? * ; * ? ? ? the individual spectra display a rms scatter in the telluric velocities of @xmath73 m s@xmath19 , again much smaller than both the velocity scales of interest and the instrumental resolution , although there is a zero - point offset of @xmath74 km s@xmath19 between the spectrograph s intrinsic wavelength calibration and the telluric velocity frame . now that we have a velocity frame fixed to the earth , we correct for the earth s orbital and rotational motion and shift the spectra into the solar barycentric rest frame . next we co - add each set of spectra taken on each night , creating several nightly master spectra . for each nightly master spectrum we create a model stellar spectrum . this is produced by obtaining a line list from vienna atomic line database ( vald ; * ? ? ? the line list includes the wavelength of each line , as well as a line depth calculated by vald using stellar model atmosphere parameters appropriate to our target . we produce an analytic rotationally broadened line profile using eqn . 18.14 of @xcite . this profile includes only the effects of rotation ; at this stage in the process , we only require an approximately correct line shape . we then fit the model spectrum to each nightly master spectrum , leaving only the velocity offset between the stellar and solar barycentric frames as a free parameter . now that we have obtained these nightly velocity offsets , we shift all of the spectra into the stellar barycentric rest frame . we note that this assumes that there is no significant acceleration of the star over the course of one night s observations ( typically one to a few hours ) . as kepler-13 is a small separation visual binary where one component is itself a single - lined spectroscopic binary , we undertook a small modification to this step for this system . due to the motion of kepler-13 b in velocity space , fitting a single line profile results in a bias in the velocity offsets of the spectra that is correlated with the orbital phase of kepler-13 b. in order to correct for this , we instead fit a model spectrum produced using two analytic rotationally broadend line profiles , with a time - dependent velocity separation given by the orbital elements of @xcite . we determined the contrast between the two profiles by fitting two model line profiles to final extracted line profiles using the unmodified code . now that all of the spectra are fixed to the same velocity frame , we co - add all of the out - of - transit spectra to create a template spectrum . we create a model spectrum using the same methodology as described above . here , however , we fix the velocity offset between model and data at zero and leave the depth of each line as a free parameter . we thus obtain best - fit line depths from our high signal - to - noise template spectrum . the final step is to extract the time series line profiles themselves . for each spectrum we again produce a model spectrum . the line depths are fixed at the best - fit values found earlier . here the free parameters are the depth of the line profile in each pixel . an example of one of these fits is shown in fig . [ spectrum ] . for each spectrum we compute the average line profile by computing the weighted mean of the line profiles extracted from each order . each order s line profile is weighted by the product of the signal - to - noise at the center of that order and the total equivalent width of all lines in that order , after @xcite . any orders with noisy line profiles ( i.e. , the scatter in the continuum is greater than an empirically determined value ) are excluded from the computation of the weighted mean . the line profiles from the different orders are also regridded to a common velocity scale . we then perform the same process on the template spectrum to obtain an out - of - transit template line profile . we subtract this template line profile from each of the time series line profiles , resulting in the time series line profile residuals , which display the transit signature . for kepler-13 , we must again modify this step due to the complicated nature of the system . as will be discussed later , simply subtracting the line profile from our out - of - transit template results in significant systematics because the overall line profile varies as a function of time due to the orbit of kepler-13 b. in order to correct for this we subtract from each line profile the average line profile from that night of observations . while this subtracts off some of the transit signal , it eliminates almost all of the systematics in the time series line profile residuals . now that the time series line profile residuals have been computed , we must extract the transit parameters from these data . we compute a model for the time series line profile residuals and fit this to the data . the model is constructed by numerically integrating over the stellar disk , summing the contributions from each surface element to the overall line profile . we divide the stellar disk into approximately 8,000 surface elements . we utilize cartesian coordinates for the integration and subsequent computations . we assume a gaussian line profile with standard deviation 5 km s@xmath19 for each surface element ; these are then appropriately doppler shifted , assuming solid body rotation , and scaled by a quadratic limb darkening law . we also neglect macroturbulence ; see [ results ] for further discussion of our assumptions on the lack of differential rotation and macroturbulence . in order to improve computational efficiency , we do not perform the full integration for each exposure . instead , we first compute the out - of - transit line profile . then , we compute the location of the planet at the beginning and the end of each exposure ( assuming a circular orbit ) , and for each surface element compute the fraction of the exposure for which that surface element is obscured by the planet . for each surface element , we diminish the out - of - transit line profile by the line profile contribution from that surface element , multiplied by the fraction of the exposure for which that surface element is covered by the planet . finally , we convolve each line profile with a model instrumental point spread function . the steps outlined above are applicable for computing a model for an arbitrary transiting planet . however , for kepler-13 ab we need to take some extra care because of the presence of the binary companion kepler-13 b and its orbit about kepler-13 c ; we must include kepler-13 b s moving line profile in our model . we use the orbital elements for kepler-13 b s orbit about kepler-13 c presented by @xcite to calculate the velocity of kepler-13 b at each exposure . we then compute a rotationally broadened line profile for kepler-13 b using the model described above , doppler shift it and scale it relative to the kepler-13 a profile , and add it to the line profile for kepler-13 a. including this profile and the resulting dilution of the spectroscopic transit signature is necessary to accurately model the data . ideally , we would simply fit for all relevant parameters ( @xmath3 , @xmath17 , @xmath38 ) simultaneously . as our time series line profiles are derived from the average of many lines across a wide region of the spectrum , and the limb darkening and therefore the detailed line shape change as a function of wavelength , our model line profiles do not fit the average line profile to better than a few percent in the wings of the profile . this poses difficulties for extracting @xmath38 , as well as the transit parameters . we therefore adopted a two - stage fitting process , first extracting @xmath38 from a single line and then @xmath3 and @xmath17 from the time series line profile residuals . for each sequential parameter extraction we used a markov chain monte carlo ( mcmc ) to sample the likelihood function of the model fits to the data . in all cases we used four chains each of 150,000 steps , cutting off the first 20,000 steps of burn - in . in addition to our free parameters for each fit , we also wished to incorporate prior knowledge from the literature , e.g. on the transit duration for kepler-13 ab . we thus set gaussian priors upon these parameters ; that is , assuming that the errors are gaussian , we can define an `` effective '' @xmath75 statistic @xmath76 where @xmath77 denotes the data , @xmath78 the model , @xmath71 the calculated error on each data point , @xmath79 the value of parameter @xmath80 at the given iteration of the markov chain , @xmath81 the value of parameter @xmath80 from the literature , and @xmath82 the uncertainty on parameter @xmath80 from the literature , and we are summing over @xmath83 data points and @xmath80 model parameters where we have prior information . first , we model a single line , the ba ii line at @xmath36141.7 , chosen because it is deep but unsaturated and isolated . we fit the nightly master spectra with models of the line profiles of kepler-13 a and b , neglecting any contribution from the transiting planet . we leave the @xmath38 of each star , the contrast between the two stars , and eight nightly velocity offsets as free parameters . we set gaussian priors upon two quadratic limb darkening parameters for each star , each with a width 0.1 , and upon the five parameters determining the radial velocity variation of kepler-13 b ( @xmath47 , epoch , @xmath49 , @xmath84 , @xmath51 ) . for the limb darkening coefficients we use coefficients in the sloan _ r _ band ( as this is the closest standard photometric band to the ba ii @xmath36141.7 line ) , taken from the tables of @xcite for an atlas model atmosphere and interpolated to the stellar parameters of kepler-13 a and b as presented by @xcite using the jktld code . we use the methods of @xcite to obtain even sampling in limb darkening space . for the orbital parameters , we set the initial value and prior width to the best - fit value and 1-@xmath71 uncertainty , respectively presented by @xcite ; see table [ starknowledge ] . second , we fit the time series line profile residuals with an appropriate model using another mcmc . here we leave @xmath3 and @xmath17 as free parameters , and set priors on the @xmath38 of kepler-13 a and contrast between kepler-13 a and b ( with the prior value and width set to the median values and 1-@xmath71 uncertainty , respectively , on these parameters from the first mcmc ) , and the limb darkening coefficients of kepler-13 a , transit depth @xmath85 , transit duration , planetary orbital period , and planetary orbital epoch , with all values and uncertainties / prior widths taken from @xcite . we fix the @xmath38 and orbital parameters of kepler-13 b at values from our first mcmc and @xcite , respectively , in the interests of computational efficiency and as uncertainties in these parameters should have a minimal effect on the line profile residuals . we note that in principle it is possible to measure the time of mid - transit and the transit duration directly from the spectroscopic data . additionally , @xmath85 and @xmath86 may be measured independently ( the width of the transit signature depends on @xmath85 , while the area under the transit signature is proportional to @xmath86 ) . if a system is affected by dilution , the measured value of @xmath86 will be smaller than that inferred from the measurement of @xmath85 from the transit signature width , which is unaffected by dilution . in practice , however , given finite spectral resolution , limited time resolution , and relatively low signal - to - noise , these values are best determined from _ kepler _ photometry . we thus incorporate these parameters via priors in our mcmcs . in our second set of mcmcs , we fit the model directly to the time series line profile residuals . alternatively , we also use a method of binning the spectra to increase the signal - to - noise ratio . this method rests upon the following observation . neglecting differential rotation of the star and assuming a circular orbit for the planet , the rate of motion of the planetary transit signature across the line profile ( @xmath87 ) will be constant . given the transit duration , each value of @xmath87 corresponds to a single value of the velocity difference between the locations of the transit signature at ingress and egress , @xmath88 . in geometrical terms , the path of the planetary transit signature in the time series line profile residual plots will be a straight line . for a given value of @xmath88 , the transit signature will occur at some velocity @xmath89 in the @xmath90 spectrum . we shift each of the @xmath83 line profile residuals by @xmath91 , such that the transit signature will occur at the same velocity for each shifted line profile residual , and then bin together all of the shifted line profile residuals . if we have the correct value of @xmath88 , the transit signatures in each line profile residual will tend to add constructively , and we will obtain a single high signal - to - noise transit signature . if we have an incorrect value of @xmath88 , the transit signatures will not add coherently , and the diluted transit signature will be below the noise floor . we define the velocity scale of the shifted line profile residuals such that it is @xmath92 , the velocity of the transit signature at the transit midpoint . for a grid of possible values of @xmath88 ( @xmath93 ) , we perform this shifting and binning operation , and visualize this as a two - dimensional map of the deviation from the out - of - transit line profile as a function of @xmath92 , @xmath88 . we model these shifted and binned data by producing model time series line profile residuals in the same manner as above , and then shifting and binning these in the same manner as we have treated the data . we then extract transit parameters from the shifted and binned data using an mcmc similar to the one for the unbinned data described above . while mathematically a complicated , usually double - valued relationship exists between ( @xmath3 , @xmath17 ) and ( @xmath92 , @xmath88 ) , qualitatively there exists a simple relationship between ( @xmath92 , @xmath88 ) and the path of the transit signature across the stellar disk . for solid - body rotation , and defining a coordinate @xmath94 on the visible disk of the star perpendicular to the projected stellar rotation axis , each velocity on the line profile maps to a single value of @xmath94 , i.e. , @xmath95 @xcite . @xmath92 and @xmath88 together fix the @xmath94 coordinates of ingress and egress , @xmath96 and @xmath97 , respectively . for each pair of @xmath96 , @xmath97 there are two possible paths across the stellar disk : one with low @xmath3 , high @xmath17 and one with high @xmath3 , low @xmath17 , resulting in the double - valued function that maps ( @xmath3 , @xmath17 ) to ( @xmath92 , @xmath88 ) . in general , positive values of @xmath88 correspond to @xmath98 , and @xmath99 corresponds to @xmath100 , while @xmath101 corresponds to @xmath102 and @xmath101 corresponds to @xmath103 . in order to verify that our code is working correctly , we analyzed one of the doppler tomographic datasets on wasp-33 b presented by @xcite . these observations were taken using the hjst on 2008 november 12 ut . we are able to reproduce their results ( fig . [ wasp33 ] ) , an important test of our code . we measure the quality of the data by the root - mean - squared ( rms ) scatter of the continuum ; for our wasp-33 data , this amounts to 0.010 of the depth of the line profile . @xcite did not provide a quantitative measure of the noise level in their data , but qualitatively our noise floor appears to be somewhat lower than that of the previous work . we furthermore find a best - fitting model using our mcmc . for wasp-33 there are variations of the line shape of a few percent due to non - radial pulsations of the host star , and so , unlike for kepler-13 a , we are able to model the line shape to within the uncertainties from the pulsations . we thus conduct only a single mcmc , fitting for @xmath38 , @xmath3 , and @xmath17 simultaneously . we use limb darkening coefficients interpolated to the stellar parameters from @xcite using jktld , but here use the @xcite values for the @xmath32 band . we obtain values of @xmath104 km s@xmath19 , @xmath105 , and @xmath106 . note that these uncertainties take into account only statistical errors and do not include systematic errors , which will be discussed later for the case of kepler-13 ab . working from the mcdonald data , @xcite obtained @xmath107 km s@xmath19 , @xmath108 , and @xmath109 . we attribute the differences between our measured parameters and those of @xcite to the complication of the stellar non - radial pulsations . we also shift and bin our wasp-33 data , as described above . the resulting map is shown in fig . [ wasp33shift ] . there are two strong peaks in the map , one due to the planetary transit and the other due to non - radial pulsations . we attempted to extract transit parameters from these data using our mcmc , but due to the non - radial pulsations we could not obtain a satisfactory fit . for kepler-13 , using our first mcmc we measure projected rotational velocities for the two stars of @xmath110 km s@xmath19 and @xmath111 km s@xmath19 , which agree to within @xmath112 with the @xmath38 values presented by @xcite . in fig . [ koi13_avg ] we show the time series line profiles extracted from the het data , produced by subtracting the out - of - transit template line profile from each of the time series line profiles . significant systematics are visible , of amplitude @xmath113 of the depth of the line profile . most of these systematics result from differences between the time series line profiles and the out - of - transit template line profile due to the motion of kepler-13 b in velocity space . this is illustrated in fig . [ koi13_nights ] , where we have subtracted the average line profile _ from each night _ from each of the time series line profiles . [ koi13joined ] is identical to fig . [ koi13_nights ] , except using all of our het data . due to these systematics , for the remainder of the analysis we subtract the nightly average line profile from the time series line profiles , and we do not use the out - of - transit template data . the rms scatter of the continuum is 0.022 times the line depth . the transit signature is immediately apparent visually . that the planetary orbit is prograde can be determined by inspection , as the transit signature is over the blueshifted hemisphere of the star at ingress and moves across to the redshifted hemisphere by egress . we also shift and bin the het data ( see fig . [ koi13binned ] , top ) . again , the transit signature is clearly detected . our best - fit values and 1-@xmath71 uncertainties from the mcmcs are shown in table [ values ] . we present values from both directly fitting the data and fitting the shifted and binned data ; these two methods give consistent results . the binned data have smaller uncertainties , but in order to be conservative and as the direct fits have a reduced chi - squared closer to 1 ( @xmath114 for the direct fit , @xmath115 for the shifted and binned fit ) , we quote these values . we find a best - fit spin - orbit misalignment of @xmath116 , in disagreement with the value of @xmath117 found by @xcite . we also find @xmath118 . we note that the quoted uncertainties on these parameters are the formal statistical uncertainties , given the assumptions made in our models . they do not include systematic uncertainties , which we discuss in detail later in this section . in fig . [ koi13resids ] we show the time series line profile residuals with the best - fit model , using these parameters , subtracted off . we note that our data also permit a second solution , with @xmath119 and @xmath120 . this solution , however , has a slightly worse value of reduced chi - squared ( @xmath121 ) and moreover implies a physically unrealistically low value for the stellar mean density , @xmath122 g @xmath123 . we calculated the stellar mean density using eqn . 9 of @xcite , which is , using the nomenclature used in this article , @xmath124}{\sin^2(\tau_{14}\pi / p)}\bigg)^{3/2}\ ] ] where @xmath47 is the planetary orbital period and @xmath125 is the transit duration , both measured from _ kepler _ note that the inferred stellar mean density depends only upon our measurement of @xmath17 and does not directly depend upon @xmath3 . given this stellar mean density and the stellar surface gravity measured by @xcite ( @xmath126 ) , we have two independently - measured parameters which physically depend only on the stellar mass and radius ; thus , we can estimate the stellar mass and radius implied by @xmath127 and see whether it is compatible with the other system parameters . a value of @xmath122 g @xmath123 implies a stellar radius of @xmath128 and mass of @xmath129 , parameters which are incompatible with the @xcite value of @xmath130 k , as well as the other measured parameters of the system . performing the same exercise for @xmath131 results in a stellar mass and radius consistent with those found by @xcite and @xcite . the full @xmath75 space for our data is shown in fig . [ chi2space ] . we also observed two transits of kepler-13 ab using the hjst . these data are shown in fig . [ koi13hjst ] . like for the het , in order to produce the time series line profile residuals , we subtract off the average line profile from each night rather than an out - of - transit line profile from both nights . the data are at a much lower signal - to - noise level than our het data ( the rms scatter of the normalized continuum is 0.037 times the line depth ) , and the transit is not readily apparent to the eye in the time series line profile residual map . we apply the bin - and - shift method to the hjst data ( see fig . [ koi13binned ] , bottom ) . here , we recover the same transit signature seen in the het data , albeit at lower signal - to - noise . here we measure values of @xmath132 and @xmath133 . the spin - orbit misalignment is in mild disagreement with the value from the direct fit to the het data , at a level of @xmath134 for @xmath3 , while there is a strong @xmath135 disagreement between the impact parameter found from the het and hjst data . one possible cause is the varying degree of contamination from kepler-13 b during the observations due to field rotation ( as noted above , the ts23 is a slit spectrograph ) . another possible cause is the poorer time resolution of the hjst data as compared to the het ( exposure times were 900 s for the hjst and 300 s for the het ) . in the spectroscopic data the impact parameter is constrained , in part , by how quickly the transit signature increases ( decreases ) between first and second ( third and fourth ) contacts . thus , the lower time resolution of the hjst could introduce larger systematic uncertainties in these data . additionally , the values above include only statistical uncertainties , which overstate the true degree of discrepancy between the het and hjst values . we have , however , been unable to positively identify the source of this discrepancy . the formal uncertainties on our values for @xmath3 and @xmath17 quoted earlier are the statistical uncertainties given the assumptions that we have made in our models ( no differential rotation or microturbulence , etc . ) and do not contain information on systematic sources of uncertainty , which we will now discuss . one possible source of systematic errors is the presence of differential rotation , which we have neglected in our models . @xcite analyzed the line profiles of a and f dwarfs for evidence of differential rotation . they found no stars with @xmath136 k that exhibited differential rotation . @xcite , however , used fourier analysis of the _ kepler _ lightcurves of a stars to infer that these stars exhibit a similar degree of differential rotation to the sun . we constructed a modified version of our models that include differential rotation , and conducted a version of our first mcmc , fitting to the line profile shape , in order to constrain the differential rotation . for a differential rotation law @xmath137 , where @xmath138 is the latitude on the stellar surface , the differential rotation parameter @xmath139 can be defined as @xmath140 ( for the sun , @xmath141 ; * ? ? ? we note that we also need to include the stellar inclination @xmath83 with respect to the line of sight in this model ; however , we find @xmath83 to be totally unconstrained . the results of this exercise indicate the presence of a small amount of differential rotation . overall , we find @xmath142 ; however , there does exist a degeneracy such that higher values of @xmath143 result in larger preferred values for @xmath139 : we find @xmath144 for @xmath145 and @xmath146 for @xmath147 , the value found by @xcite . this is consistent with the results of @xcite , who found splitting of the frequency spectrum peak associated with rotation , likely due to differential rotation . in order to test the effects of this level of differential rotation on our measurement of the transit parameters , we modified our second mcmc to include differential rotation . we added two parameters , @xmath139 and the stellar inclination @xmath83 . @xmath83 was allowed to float , while , due to the dependence of the best - fit @xmath139 on @xmath83 , we included a variable prior on @xmath139 depending on the value of @xmath83 . marginalizing over @xmath83 in 5@xmath148 bin sizes , we found the mean and standard deviation of @xmath139 for each bin and used these as the prior center and width for the new mcmc . from this mcmc , we obtain @xmath149 and @xmath150 . we note that the presence of even strong differential rotation can not bring our value of @xmath3 into agreement with that found by @xcite . we also neglected macroturbulence in our models , which could potentially induce systematic uncertainties in our measured values of @xmath3 , @xmath17 . measurements of macroturbulence in a dwarfs in the literature are somewhat lacking . @xcite found varying degrees of macroturbulent broadening for b dwarfs , ranging from none to several tens of km s@xmath19 ( they note that this `` macroturbulence '' is not necessarily physical turbulence ) . @xcite measured macroturbulent broadening of order @xmath151 km s@xmath19 for two late a dwarfs . @xcite argued that `` macroturbulence '' in early - type stars is actually due to the collective action of many low - amplitude pulsational modes ; early - type stars which do not pulsate should not show this type of macroturbulence . even with _ s photometric precision , there is little evidence for any pulsation of kepler-13 a which could result in this type of macroturbulence . @xcite conducted simulations of convection in the outer layers of massive stars due to an opacity peak produced by fe ionization . they found that such zones can cause surface granulation and consequent small - scale velocity fields in stellar photospheres . they find , however , that this effect does not occur for stars with @xmath152 for galactic metallicities , and is furthermore more prominent at low surface gravities . as kepler-13 a is below this luminosity cutoff ( @xmath153 ) and has high surface gravity ( @xmath154 ; * ? ? ? * ) , we conclude that surface granulation due to this mechanism should not occur for kepler-13 a. a key question for estimating the effects of macroturbulence upon our results lies with the scales of macroturbulent velocity fields in the stellar atmosphere . if these scales are much smaller than the size of the projected planetary disk during the transit , then this will simply increase the range of radial velocities over which the planet subtracts light from the line profile . the effect will be to `` smear out '' the transit signature , but this should not affect the measured value of @xmath3 . if , however , the macroturbulent velocity field changes on scales of similar or greater size as the planetary disk , then the velocity of the region of the stellar disk covered by the planet will differ from that expected if taking only rotation into account . thus , the planetary transit signal in each spectrum will exhibit a quasi - random shift from the expected velocity . @xcite presented evidence that some of the large number of frequencies seen in the frequency spectra of @xmath155 sct ( early a ) stars observed by corot are in fact due to surface granulation rather than pulsations , as pulsations at these frequencies would be of such high degree @xmath156 that they should not be evident in integrated disk photometry . based upon the inferred granulation frequencies , they find that the granulation properties follow scaling laws derived for solar - type stars . when scaling from such solar models , @xcite make the assumption that the size of granulation cells is proportional to the atmospheric pressure scale height @xmath157 . @xcite use the scaling relation @xmath158 . using these relations and the stellar properties of kepler-13 a from @xcite , we estimate that the size of any surface granulation cells for kepler-13 a should be @xmath159 times that of such cells on the sun , or @xmath160 , comfortably below the size scale of the planetary disk ( using an average solar granule size of 1300 km , from * ? ? ? nonetheless , given the uncertainty in the relations used to derive this estimate , we choose to include `` jitter '' caused by large - scale macroturbulent cells in the stellar atmosphere in our mcmcs ( note that this is not the same as the jitter frequently invoked as a source of noise in radial velocity observations ) . in order to simulate the effect of macroturbulence on the size scale of the planet , we use the following approach . we allow each of the time series line profile residuals to have a small velocity offset from its nominal value . the effect of this is to shift the transit signature in that line profile residual in velocity space . since we have already subtracted off the average line profile shape , this mimics a velocity shift of the transit signature due to large - scale macroturbulence rather than a radial velocity offset for the entire line profile . for computational reasons we apply this velocity shift to the model line profile residuals , not the data . at each mcmc step , we perform a single parameter minimization for each velocity offset using mpfit . similar methodologies have been used by @xcite to deal with jitter and by @xcite to handle stellar pulsations . we limit the velocity offset amplitude to 15 km s@xmath19 in order to prevent the model transit signatures from latching on to the remaining systematics in the data . the mean offset amplitude is 5.7 km s@xmath19 . from these mcmcs , we obtain @xmath161 and @xmath162 . we thus find that including `` jitter '' and differential rotation have opposite systematic effects on our results : large - scale macroturbulence shifts the best - fit parameters to higher @xmath3 and lower @xmath17 , while differential rotation shifts them to lower @xmath3 and higher @xmath17 . thus , we expect that these effects should largely cancel each other out , and our overall result should not be affected , while increasing the uncertainty in our results . in order to remain @xmath112 consistent with both the differential rotation and `` jitter '' mcmc results , we therefore adopt @xmath163 and @xmath164 . additionally , our model assumes an intrinsic line standard deviation of 5 km s@xmath19 . in order to test the impact of this assumption on our results we fit a model with an intrinsic line standard deviation of 10 km s@xmath19 to our data . this did not significantly alter our measured values of @xmath3 and @xmath17 or the @xmath165 value of the model fits , and so we conclude that this has minimal impact on our measurements . cccccc @xmath166 ( km s@xmath19 ) & & @xmath167 & & + @xmath168 ( km s@xmath19 ) & & @xmath169 & & + @xmath3 ( @xmath148 ) & @xmath170 & @xmath171 & @xmath172 & @xmath173 + @xmath17 & @xmath174 & @xmath175 & @xmath176 & @xmath177 in addition to detecting the transit signal of the planet kepler-13 ab , we set upper limits on the mass of the tertiary stellar companion kepler-13 c. we follow @xcite to cross - correlate all het spectra against model spectra of late - type stellar companions and search for significant cross - correlation function ( ccf ) peaks . since the orbit of kepler-13 b is known @xcite , we can predict the velocity of kepler-13 c by assuming some guess mass . we can then shift the ccfs by that velocity and co - add them , amplifying any ccf peak arising from a detection of kepler-13 c if the guess mass is correct . while we do not detect the spectral signature of kepler-13 c for any guess mass from 0.2 - 1.5 @xmath178 , we perform a sensitivity analysis by injecting synthetic companion spectra into the data and repeating the above procedure . the rate of detection is shown as a function of the effective temperature of the companion in fig . [ csensitivity ] . this analysis indicates that we would detect a main sequence companion with effective temperature of @xmath179 k ( corresponding to a mass @xmath180 ) 95% of the time , allowing us to set a mass limit on kepler-13 c of @xmath181 at 95% confidence . combined with the value of @xmath182 found by @xcite , we limit the mass of kepler-13 c to @xmath183 . we note that , as kepler-13 c has not been directly detected , it could in principle be a white dwarf rather than a late - type dwarf ( white dwarfs were not included in our spectral library for cross - correlation due to flux ratio issues ) . our best - fit value for the spin - orbit misalignment for kepler-13 ab , @xmath0 , is in stark disagreement with the value of @xmath1 found by @xcite . even if we fix @xmath17 to the value found by @xcite , we obtain a spin - orbit misalignment of @xmath184 , still in disagreement with the gravity darkening value . we do not have a definitive explanation for the mismatch between our result and that from @xcite . we note , however , that our value relies upon fewer assumptions regarding the physical nature of the star ( e.g. , the gravity - darkening law and gravity - darkening parameter ) , and thus is likely more robust . additionally , @xcite fixed the effective temperature of the pole of kepler-13 a to 8848 k , the temperature from the kepler input catalog ( kic ) , rather than a more accurate spectroscopic value ( @xmath185 , from * ? ? * though these values for the temperature differ by less than 1@xmath71 ) . the fact that kepler-13 is a near - even flux ratio binary is also not accounted for in the kic . @xcite could not account for any effects of the tertiary stellar companion kepler-13 c upon the transit lightcurve , as this companion had not yet been discovered @xcite . kepler-13 c , however , should contribute somewhere between 0.8% and 0.03% of the total flux of the system , given our limits on its mass , insufficient to significantly affect the dilution . variability of either kepler-13 b or c would need to occur on the orbital period of kepler-13 ab , or on a harmonic thereof , in order to systematically affect the light curve shape , which is unlikely . finally , @xcite found a rotation period of 22.0 hours for kepler-13 a by fitting their model to the data , slightly shorter than the likely rotation period of 25.4 hours found by @xcite in the _ kepler _ data . while it is unclear whether the 25.4 hour period is indeed due to stellar rotation , if this is rotation then , given this and the likely too high value of @xmath186 assumed by @xcite , the actual temperature ( and therefore surface brightness ) contrast between the poles and equator of kepler-13 a should be smaller than that assumed by @xcite . the effects of this upon the lightcurve shape and the resulting inferred spin - orbit misalignment , however , are not qualitatively obvious , and a quantitative analysis is beyond the scope of this work . as noted above , there is a great deal of disagreement in the literature as to the value of the impact parameter @xmath17 , with published values ranging from 0.253 @xcite to 0.75 @xcite . as noted earlier , these discrepancies can not be attributed to precession of the planetary orbital plane @xcite . our value of @xmath164 agrees to within @xmath112 only with the published measurement of @xcite , and is in disagreement with other published values by up to @xmath187 . we note that our value of the impact parameter is obtained directly from the spectroscopy , and is thus largely independent of the previous measurements from the _ kepler _ photometry ( although our model requires the assumption of the transit duration from the photometry , as a prior in the mcmcs ) . this suggests a possible reason for the discrepancy between our value of @xmath3 and that from @xcite . the value of @xmath3 derived from gravity darkening is dependent upon the choice of impact parameter ; as the value of @xmath188 used by @xcite differs from the @xmath131 that we measure , it is perhaps unsurprising that the two values of @xmath3 are in disagreement . using the value that @xcite measured for the stellar obliquity with respect to the line of sight ( @xmath189 ; note that @xmath83 was denoted as @xmath190 by * ? ? ? * ) and our measurement of @xmath3 , we calculate a full three - dimensional spin - orbit misalignment of @xmath191 . given the disagreement of our value of @xmath3 with that from @xcite , however , it is unclear whether their measurement of @xmath83 remains applicable . despite the presence of an additional star in the kepler-13 system , @xcite disfavor emplacement of kepler-13 ab via kozai cycles due to the young system age ( @xmath192 myr , determined using isochrones by * ? ? ? * ) , its current circular orbit @xcite or very small eccentricity @xcite , and the long timescale necessary for tidal semi - major axis damping . @xcite estimated that , for an initial kozai - driven eccentric orbit similar to that currently occupied by hd 80606b , the required tidal damping timescale to circularize the orbit at kepler-13 ab s current location is @xmath193 years . @xcite also noted that planet - planet scattering remains viable if it took place early enough that a debris disk sufficiently massive to quickly damp out the planetary eccentricity remained in place . given the characteristics of the kepler-13 and the highly inclined orbit that we find for kepler-13 ab , it seems natural that it could have been emplaced through migration within an inclined disk produced via the mechanism of @xcite . this would require an inclination between the orbital plane of kepler-13 ab and that of kepler-13 bc about kepler-13 a. unfortunately , due to the lack of information about the position angle of kepler-13 ab s transit chord relative to the kepler-13 ab separation , and the long orbital period of kepler-13 bc about a ( the projected separation is @xmath194 au ) , this relative inclination is unlikely to be measured in the foreseeable future . the mechanism proposed by @xcite could also naturally result in an inclined , circular orbit for kepler-13 ab , but would not require the presence of a binary companion . we note , however , that these arguments rest upon the tidal circularization timescale being longer than the age of the system ; as tidal theory continues to be not well understood , the eccentricity damping timescale may be very uncertain . additionally , we note that due to these uncertainties we can not definitively exclude any mechanisms for the emplacement of kepler-13 ab upon its current inclined orbit . a 25.4-hour periodicity is evident in the _ kepler _ lightcurves for kepler-13 . this was suggested to be either stellar pulsations @xcite or rotation @xcite . additionally , @xcite found a 25.5-hour periodicity in their radial velocity measurements of kepler-13 a. they noted that this radial velocity periodicity could also be due to either pulsations or rotation , but preferred the pulsation explanation because their measured radial velocity semi - amplitude of @xmath195 km s@xmath19 is much larger than that expected from starspots and rotation . we folded our stellar radial velocity for kepler-13 ab ( i.e. , the radial velocity offset between the solar and stellar barycentric rest frames discussed earlier ) on the period found by @xcite , and our data appear to exhibit a similar periodicity and phase . in order to quantify this effect , we computed the generalized lomb - scargle periodograms @xcite for the @xcite dataset and our dataset . for the @xcite data we find a best - fitting period of 25.5 hours , and for our dataset , we find a period of 24.7 hours . the false alarm probabilities for these frequencies are 0.9998 and 0.98 , respectively , and so we do not consider the detections of these periodicities in the radial velocity data to be statistically significant . we see no evidence for stellar non - radial pulsations in our data , as are seen for the @xmath155 sct planet host wasp-33 ( * ? ? ? * and [ testing ] ) , although given the short time span of each of our observations ( @xmath196 hour ) such long - period pulsations would not necessarily manifest in our data . in principle we could compare the overall line shape for kepler-13 a between different transit observations , but the moving line profile of kepler-13 b would complicate such an effort , and thus we do not attempt such an analysis . we estimate that @xmath197 dor - like pulsations ( similar in period to the kepler-13 a periodicity , but typically exhibited by cooler stars ) would result in radial velocity shifts of order meters per second ( using the results of * ? ? ? * ) , far too small to be detected in our data or to affect our conclusions . the recently launched _ mission should be capable of further improving the characterization of the kepler-13 system . _ gaia _ is estimated to have an astrometric precision of @xmath198 @xmath199as for stars with @xmath200 @xcite , like both kepler-13 a and b. thus , it should be capable of detecting both the mutual orbit of kepler-13 a and bc ( @xmath196 mas yr@xmath19 for a circular , face - on orbit ) and the orbit of kepler-13 b about c ( total displacement @xmath201 @xmath199as ) . together with the radial velocity observations of @xcite , this will allow the measurement of the true mass of kepler-13 c and its orbital plane . while the orbital period of kepler-13 a around bc is likely too long to obtain a good orbital solution ( @xmath202 yr ) , _ gaia _ should nonetheless be able to place some constraints upon the system parameters . the astrometric orbit of kepler-13 a due to kepler-13 ab is too small to be detectable by _ gaia _ ( total displacement @xmath203 @xmath199as ) . @xcite note that , in principle , the spin - orbit misalignment for kepler-13 ab can be measured using a third mechanism : the photometric rossiter - mclaughlin effect @xcite . unfortunately , given the scatter in the single - quarter @xcite lightcurve of @xmath204 ppm , and that they estimate the amplitude of the photometric rossiter - mclaughlin effect to be @xmath205 ppm , this measurement is probably out of reach of even the full 16-plus quarter _ kepler _ lightcurve . we have constructed a doppler tomography code which now rivals previously established codes in terms of precision . we have validated this code by analyzing data on a transit of wasp-33 b. we have presented doppler tomographic observations for the _ kepler _ planet kepler-13 ab , finding a prograde orbit and measuring a much larger spin - orbit misalignment than that previously found by @xcite via the gravity darkened light curve . given the disagreement between these two techniques , observations of further systems via both techniques will be of interest to determine the reason for the disagreement . we have also suggested that , due to its highly inclined , circular orbit , the ( likely ) long tidal damping timescale of the system , and the presence of a wide binary companion , kepler-13 ab may have been emplaced via migration within an inclined disk . simulations of the system could confirm the viability of this hypothesis for the kepler-13 system , but these are beyond the scope of the current work . thanks to michael endl , edward l. robinson , and chris sneden for guidance during this project ; to andrew collier cameron , for valuable discussions ; to douglas gies and zhao guo , for consultations on velocity fields in the atmospheres of early - type stars ; to michel breger , for consulting on the expected radial velocity pulsation amplitude of kepler-13 a ; and to the anonymous referee , for thorough comments that improved the quality of the paper . m.c.j . gratefully acknowledges funding from a nasa earth and space science fellowship under grant nnx12al59h . this work was also supported by nasa origins of solar systems program grant nnx11ac34 g to w.d.c . work by s.a . was supported by nsf grant no . work by s.d.r . was supported by nsf grant ast-1055910 . was partly supported by the nasa origins program and kepler participating scientist program . this paper includes data taken at the mcdonald observatory of the university of texas at austin . the hobby - eberly telescope ( het ) is a joint project of the university of texas at austin , the pennsylvania state university , stanford university , ludwig - maximilians - universitt mnchen , and georg - august - universitt gttingen . the het is named in honor of its principal benefactors , william p. hobby and robert e. eberly . @xmath206 & @xmath207 & & @xmath208 & & & & & @xmath209 & @xmath210 + & @xmath211 & & @xmath212 & & & & & @xmath213 & @xmath214 + @xmath215 & @xmath216 & & @xmath217 & @xmath218 & @xmath219 & @xmath220 & & & + & @xmath221 & & @xmath222 & @xmath223 & @xmath224 & @xmath225 & & & + @xmath47 ( days ) & @xmath226 & @xmath227 & @xmath228 & & & & @xmath229 & & + & @xmath230 & @xmath231 & @xmath232 & & & & @xmath233 & & + @xmath17 & & @xmath234 & @xmath235 & @xmath236 & & @xmath237 & & 0.31598 & 0.75 + & & @xmath238 & @xmath239 & @xmath240 & & @xmath241 & & & + @xmath242 & & & & & & & & @xmath243 or @xmath244 & + & & & & & & & & @xmath245 & + @xmath246 & & & & & & & & @xmath247 & + & & & & & & & & @xmath245 & + @xmath248 & & & & & & & & @xmath249 & + & & & & & & & & @xmath245 & + @xmath250 & @xmath251 & @xmath252 & @xmath253 & & & & & @xmath254 & + & @xmath255 & @xmath256 & @xmath257 & & & & & @xmath258 & + @xmath24 & & @xmath259 & @xmath260 & @xmath261 & & & & + & & @xmath262 & @xmath263 & @xmath264 & & & & + @xmath265 & & @xmath266 & @xmath267 & @xmath268 & & & & 0.084513 & @xmath269 + & & @xmath270 & @xmath271 & @xmath272 & & & & & @xmath273 + @xmath274 ( hr ) & & & & & & @xmath275 & & 22.0 & + & & & & & & @xmath276 & & & + @xmath277 & & & & & & & & 0.021 & +

the age difference between the two main globular cluster sub populations in the virgo giant elliptical galaxy , ngc 4472 ( m 49 ) , has been determined using hst wfpc2 images in the f555w and f814w filters . accurate photometry has been obtained for several hundred globular clusters in each of the two main sub populations , down to more than one magnitude below the turn over of their luminosity functions . this allows precise determinations of both the mean colors and the turn over magnitudes of the two main sub populations . by comparing the data with various population synthesis models , the age metallicity pairs that fit both the observed colors and magnitudes have been identified . the metal poor and the metal rich globular clusters are found to be coeval within the errors ( gyr ) . if one accepts the validity of our assumptions , these errors are dominated by model uncertainties . a systematic error of up to 4 gyr could affect this result if the blue and the red clusters have significantly different mass distributions . however , that one sub population is half as old as the other is excluded at the 99% confidence level . the different globular cluster populations are assumed to trace the galaxy s major star formation episodes . consequently , the vast majority of globular clusters , and by implication the majority of stars , in ngc 4472 formed at high redshifts but by two distinct mechanisms or in two episodes . the distance to ngc 4472 is determined to be mpc , which is in excellent agreement with six of the seven cepheid distances to virgo cluster spiral galaxies . this implies that the spiral and elliptical galaxies in the main body of virgo are at the same distance .

astro-ph9908203

arguably the two most important open questions about the formation of early type galaxies are when they formed their first stars and when they assembled dynamically . while the latter question is probably best answered by observations at intermediate and high redshift , the answer to the former is likely to be found from observations in the more local universe . a number of studies have estimated the epoch of star formation by comparison of observations of the diffuse stellar component in nearby and low - redshift galaxies to stellar population models ( e.g. renzini 1999 and bender 1997 for reviews ) . however , the use of diffuse star light has two main disadvantages . first , it is very difficult to disentangle different stellar populations and , second , a relatively recent but ( in mass ) unimportant star formation episode can dominate the luminosity weighted quantities used to derive the star formation history . both these problems are by passed when globular clusters are used to study the main epochs of star formation . in nearby galaxies , individual globular clusters appear as point source , and can be characterized by a single age and a single metallicity . distinct populations can easily be identified ( e.g. in a color distribution of the globular cluster system ) and the mean properties of the sub populations can be determined . the relative numbers in each sub population will also indicate the relative importance of each star formation episode . there is strong support for the assumption that globular cluster formation traces star formation . first , major star formation observed today in interacting galaxies is accompanied by the formation of massive young star clusters ( e.g. schweizer 1997 , although their exact nature is still under debate , brodie et al . second , the specific frequency ( i.e. number of globular clusters normalized to the galaxy star light , see harris & van den bergh 1981 ) is constant to within a factor of two in almost all galaxies ( see compilation in ashman & zepf 1998 ) , which points to a link between star and cluster formation . moreover , although this relation seems to fail in giant ellipticals , mclaughlin ( 1999 ) has shown that the number of globular clusters is in fact proportional to the total mass , and the number per unit mass is extremely constant from galaxy to galaxy . finally , in less violent star formation episodes , larsen & richtler ( 1999 ) have shown that the number of young star clusters in spirals is directly proportional to the current star formation rate . to first order , globular clusters are likely to be excellent tracers for the star formation episodes in early type galaxies . there is now evidence for multiple globular cluster sub populations in a number of luminous , early type galaxies ( zepf & ashman 1993 , ashman & zepf 1998 for a recent compilation ) . the origin of these sub populations is still under debate ( e.g. gebhardt & kissler - patig 1999 ) but it seems clear that the main sub populations formed in different star formation episodes , perhaps at different epochs . in the best studied galaxies with multiple populations , the number of clusters in each sub population is roughly equal , suggesting that these galaxies have had more than one _ major _ star formation episode . the overall evidence from photometry and more recently from spectroscopy ( kissler - patig et al . 1998a , cohen et al . 1998 ) suggests that in the cluster giant ellipticals the main sub populations are old . in other non interacting early type galaxies , the non detection of sub populations , either due to their absence or to a conspiracy of age and metallicity , leads to the same conclusion : no major globular cluster ( i.e. star ) formation has occurred since @xmath2 ( kissler - patig et al . 1998b ) . there have been a few previous attempts to determine the relative age difference between globular cluster sub populations . in the s0 galaxy ngc 1380 , kissler - patig et al . ( 1997 ) derived an age difference of around 3 to 4 gyr between the halo and bulge globular clusters . in m 87 , kundu et al . ( 1999 ) estimated the metal rich clusters to be 3 to 6 gyr younger than the metal poor ones . as a comparison , in the milky way , the difference between halo and bulge / disk globular clusters seems to be small ( less than 1 gyr : ortolani et al . 1995 , around 17% : rosenberg et al . 1999 ) . deriving relative ages from photometry alone is complicated by the fact that broad band colors suffer from an age metallicity degeneracy ( e.g. worthey 1994 ) . one possible solution to this problem is to measure several quantities that are affected differently by age and metallicity and to combine the results . for example , most broad band colors are more affected by metallicity than by age . on the other hand , optical magnitudes for a given mass are more affected by age than they are by metallicity . a combination of both can break the age metallicity degeneracy , when the mass distributions are known ( see sect . [ models ] ) . both a mean color as well as a `` mean magnitude '' can be measured for a globular cluster sub population . the mean color is simply the mean color of the globular cluster sub sample , and can be determined from a color distribution of the globular clusters . the turn over ( to ) point of the globular cluster luminosity function ( gclf ) of a given sub population can be used as its `` mean magnitude '' . the absolute to magnitude of the gclf has been found to be fairly constant from one globular cluster system to the next in a wide variety of galaxy types and environments with a peak at @xmath3 ( see whitmore 1997 and ferrarese et al . 2000 for summaries ) . three quantities affect the to value : age , metallicity and mass . the effects of age and metallicity on the magnitude of a single stellar population were quantified by ashman , conti & zepf ( 1995 ) using population synthesis models from bruzual & charlot ( 1995 ) . the mean mass influences the to magnitude because the to of the gclf corresponds to a break in the globular cluster mass distribution ( e.g. mclaughlin & pudritz 1996 ) . that is , if the characteristic mass ( at which the slope of the power law mass function changes ) varies , the to magnitude will be affected . however , the mass distributions seem to be extremely similar from galaxy to galaxy and are expected to be very similar for different sub populations within the same galaxy if destruction processes are the dominant mechanism for shaping the mass distribution , and had time enough to act on both sub populations . we will come back to this point in section [ discussion ] . neither the mean color nor the to magnitude alone can be used to derive the age and metallicity of a globular cluster population because of the above mentioned degeneracy . this paper marks the first time that both the mean color _ and _ the to magnitude of each globular cluster sub population have been simultaneously compared to single stellar population ( ssp ) models ( see sect . [ models ] ) in order to distinguish age and metallicity . deep high quality photometry is required to reliably determine the mean color and the to magnitude ( implying good photometry below this to point ) and clean globular cluster samples are needed , i.e. good discrimination between globular clusters and fore/background contamination is essential . the wide field and planetary camera 2 ( wfpc2 ) on board the hubble space telescope ( hst ) provides the most accurate photometric data available and makes such an analysis feasible . for this first study , we selected a galaxy that hosts a populous globular cluster system which provides good number statistics even after the sample is split into the different sub populations . ngc 4472 ( m 49 ) , the brightest giant elliptical galaxy in the virgo cluster , is located in the center of a sub group within the cluster and is of hubble type e2 . it is one of the brightest early type galaxy within 20 mpc with an absolute magnitude of @xmath4 ( see table 1 for other characteristics ) . it hosts @xmath5 globular clusters ( geisler et al . 1996 ) . the specific frequency of globular clusters is @xmath6 , close to the average for bright cluster ellipticals . previous photometric studies of the globular cluster system of this galaxy include those of cohen ( gunn thuan photometry , 1988 ) , couture et al . ( johnson photometry , 1991 ) , ajhar et al . ( johnson photometry , 1994 ) , and geisler et al . ( 1996 ) and lee et al . ( 1998 , both washington photometry ) . the two most recent studies were the first to obtain a comprehensive enough data set to allow the identification and characterization of the two main globular cluster sub populations . the mean metallicities of the two sub populations were found to be [ fe / h]@xmath7 and @xmath8 dex for the blue and red peak , respectively . the observed metallicity gradient ( @xmath9[fe / h]/@xmath9 log@xmath10 dex / log(arcsec ) ) in the system is mostly due to the radially varying ratio of these two populations . the metal rich component is spatially more concentrated and has the same ellipticity as the galaxy star light while the metal poor component is more extended and spherical . these two components can be loosely associated with the `` halo '' and `` bulge '' of ngc 4472 suggesting different formation mechanisms for these galaxy structures and the associated clusters . the relative ages of these two components is the subject of this paper . the paper is structured as follows : in section [ observation ] we present the observation and data reduction . section [ analysis ] contains details of the data analysis ; color and turn over magnitude determinations , etc . section [ agediffs ] compares the results to stellar evolutionary models to determine the age difference between the sub - populations . in section [ distance ] we derive a new distance to ngc 4472 using the to magnitude method . section [ summary ] summarizes the results . the wfpc2 images of ngc 4472 were obtained from two data sets . the nuclear pointing was taken from westphal et al . ( from hst program go.5236 ) and was complemented by our own pointings to the north and south of the galaxy ( hst program go.5920 , pi : brodie ) which were offset in dec by 2 16 and 2 49 from the center of the galaxy , respectively . the data are summarized in tab . 2 . we used calibrated science images returned by the stsci . the basic reduction was carried out in iraf . the images were combined with the task crrej of the stsdas package . the brodie data set consists of paired , dithered exposures , and was thus registered by shifting the second of each image pair by 0.506 ( i.e. 5 pixels for the wf chips , 12 pixels for the pc chips ) with the task imshift . using the standard aperture radius of holtzman et al . ( 1995 ) of 0.5 , we tested for any magnitude error that might be introduced by incomplete shifting of frames . this error was found to be less than 0.0003 mag for both filters and is therefore negligible . the photometry was carried out using the source extraction software sextractor v2.0.19 by bertin & arnouts ( 1996 ) . for the purpose of object selection only , the images were convolved with the instrumental psf to improve the detection efficiency . the psf for each filter was taken from the hst handbook version 4.0 , using the psfs ( pixel centered ) at 600 nm for f555w and at 800 nm for f814w . the selection criteria were set to 2 connected pixels , 2@xmath11 above the background computed in a @xmath12 pix@xmath13 grid . magnitudes were measured in 2 pixels and 0.5 radius apertures ( see also next section ) . objects in common on the @xmath14 and @xmath15 frames were identified and matched . the positions on the sky of all objects were computed using the task metric in the stsdas package . the calibration and transformation to the johnson cousins @xmath14 and @xmath15 filters followed holtzman et al . ( 1995 ) and corrections for charge transfer efficiency and gain ( @xmath16 for all our data ) were applied . in what follows , the magnitude errors will _ not _ include the @xmath17 systematic error potentially introduced by the calibration . the values measured in the 0.5 radius apertures were used to compute the final magnitudes in order to avoid having to apply an additional correction to the data . globular clusters at the distance of the virgo galaxy cluster are expected to be resolved . milky way globular clusters have typical half light radii between 3 and 10 pc ( e.g. harris 1996 ) , corresponding to 0.04 to 0.13 at 16 mpc distance . indeed , most of our globular cluster candidates have fwhm measurements ( as returned by sextractor ) slightly larger than measured for unresolved objects . this , in turn , implies that the photometric corrections proposed by holtzman et al . ( 1995 ) for unresolved objects will not be valid for converting from the 0.5 radius aperture to the total globular cluster magnitude . new rough corrections were derived as follows . we computed @xmath18 pix@xmath13 normally sampled psfs in @xmath14 and @xmath15 for a wf chip using the tiny tim v4.4 program by krist & hook ( 1997 ) , and convolved these psfs with hubble laws of different core radii ( 0.05 , 0.1 , 0.2 , 0.6 ) . note that these psfs include the convolution with the diffusion kernel ( cf . krist & hook 1997 ) which is not well defined . this kernel introduces an intrinsic smoothing to the optical psf ( as seen at the top of the ccd chip ) due to charge diffusion . the cumulative light distributions of these modeled objects were compared with those of a set of isolated globular cluster candidates of various luminosities and chip positions taken from our data . the data in @xmath15 match almost perfectly the psf convolved with a hubble law of 0.1 core ( corresponding to @xmath19 pc at a distance of 16 mpc ) , while the @xmath14 data lie in between the psf models convolved with 0.05 and 0.1 hubble laws . the same procedure , using 10 times over sampled psfs without applying any diffusion kernel , yielded best fitting hubble core radii of 0.3 for @xmath15 data and 0.1 for the @xmath14 data . until the behavior of the diffusion kernel is better understood ( wavelength dependencies , etc . ) , it is possible that a systematic core radius error of about 0.1 ( @xmath19 pc ) may be present in all radius estimation studies of virgo globular clusters using wfpc2 data . nonetheless , we used the normally sampled and diffusion kernel convolved psfs for our analysis . the modeled light profiles ( see fig . [ ps : wachstum ] ) , unlike the real data , provide high enough signal to noise to determine the extrapolation from an 0.5 radius to the total @xmath14 and @xmath15 magnitudes . the derived corrections are applied to the 0.5 radius aperture measurements in addition to the holtzman point source correction of @xmath20 . these additional corrections are @xmath21 and @xmath22 ( which results in a color correction of @xmath23 ) . the errors reflect the uncertainty in matching the data with the models and the accuracy of the extrapolation to infinity of our fit to the light profile . the errors do not include , for example , the fact that the core and the half light radii of different globular clusters may vary significantly . in general , our data were calibrated applying the correction for point sources of @xmath24 . when additional corrections were included , they are specifically mentioned in the text . note that the corrections are systematic , i.e. they do not affect _ relative _ magnitudes . therefore , the accuracy with which we are able to determine age metallicity _ differences _ should not be seriously affected . artificial star / cluster experiments were carried out using the tasks starlist and mkobjects from the iraf package , artdata . 1000 objects were added per run in both the @xmath14 and @xmath15 images . 400 runs were computed resulting in 400,000 artificial objects per filter and covering the whole parameter space of background values . both objects modeled by a wf chip psf and objects modeled by the same psf but convolved with hubble laws of core radius 0.05 ( @xmath14 ) and 0.1 ( @xmath15 ) were added . all objects had a color of @xmath25 , previous tests with 5000 objects having shown no significant variations among the completeness curves for @xmath26 0.9 , 1.1 and 1.3 . the selection completeness was computed for objects found on both the @xmath14 and @xmath15 images . all detected objects which passed our selection criteria ( see sect . [ selection ] ) and all input objects were sorted by their background value and subdivided in 10 background value bins for @xmath14 and @xmath15 , respectively . a _ skewed fermi function _ @xmath27 was fit to all completeness histograms for each background bin . a smoothing procedure using a low pass filter gives nearly the same results . the 50% completeness limits lie around @xmath28 and @xmath29 for the lowest background value . the 50% completeness magnitude falls off continuously with a growing background ( see fig . [ ps : inc ] ) . the artificial object experiments also allowed us to determine the error in the photometry and to test the accuracy with which the fwhm of objects can be recovered as a function of magnitude . typical fwhm values were deduced for extended objects which are a good match to the fwhm values of our globular cluster candidates . object selection criteria were applied to our extracted data to provide a clean sample of globular cluster candidates for further analysis . only objects which met the color cut of @xmath30 and a magnitude cut of @xmath31 25.5 and @xmath32 25.2 were included so that some of the red unresolved galaxy nuclei were rejected ( cf . fig . [ ps : back ] ) . in addition , a fwhm cut in the range 0.75 pix @xmath33 fwhm @xmath33 2.5 pix was applied . we also calculated the magnitude difference obtained through apertures of 0.5 pix and 3 pix radii which allowed us to distinguish between point source objects and extended objects like globular clusters ( e.g. schweizer et al . 1996 ) . a very rough upper cut of @xmath34 was used . the neural network of sextractor v2.0.19 , i.e. the @xmath35 parameter was used to supplement the above selection criteria . this parameter was defined for well sampled images ( bertin & arnouts 1996 ) . since the wfpc2 provides under sampled images the reliability of the @xmath35 parameter is reduced . we applied only the selection criterion , @xmath36 , to our data . a test showed that for objects brighter than the 50% completeness level the @xmath35 parameter cut reliably eliminates point sources and cosmic rays . our final list of globular cluster candidates includes 705 objects in @xmath14 and @xmath15 . this is 38% of the initially extracted sample . an electronic list of the globular cluster candidates is available from the authors . one significant advantage of hst data over ground based observations is high spatial resolution which allows high confidence rejection of extended objects . however , background contamination is still expected . in order to determine the background fraction in our data we ran the same extraction and selection criteria for the hubble deep field north ( hdf n , williams et al . 1996 ) images . since the hdf n has galactic coordinates of @xmath37 125.89 and @xmath38 + 54.83 while our exposures have @xmath37 286.92 and @xmath38 + 70.20 , we might overestimate the contamination due to foreground stars using this approach , although foreground star contamination is , in any case , expected to be negligible . we obtained from the hst archive images in f606w and in the f814w filter with exposure times of 20200 sec for f814w and 15200 sec for f606w , respectively . these images are far deeper than our own . both magnitudes for f606w and f814w were transformed to johnson @xmath14 and cousins @xmath15 magnitudes . we determined contamination histograms with and without the application of our selection criteria in order to establish their reliability . exponential laws have been fit to the histograms , after applying the rejection criteria , up to the 50% completeness limits . the exponential laws act as input for our improved maximum likelihood code ( see sect . [ tomags ] ) . note the efficiency of our selection criteria . nearly all background objects are removed from the data up to the 50% completeness magnitude . the contamination is less than 5% for both the @xmath39 and @xmath40 filters ( see fig . [ ps : back ] ) . we stress again that all the above corrections were applied to the full sample , i.e. differential values between sub samples remain unaffected even if small systematic effects have been introduced . our goal is to determine the mean colors and turn over magnitudes of the luminosity functions of the two main globular cluster sub populations in both the @xmath14 and the @xmath15 filters . these colors and magnitudes can then be compared to the predictions of population synthesis models for given ages and metallicities . in order to determine the mean colors of the sub populations we selected from the total sample objects with small photometric errors . the application of a magnitude cut at @xmath41 resulted in errors @xmath42 . we further applied an upper magnitude cut at @xmath43 ( corresponding to @xmath44 at the distance of virgo ) in order to exclude any foreground star contamination , and a broad cut in color ( @xmath45 ) to isolate the region where globular clusters are expected . the color magnitude diagram of our sample is shown in fig . [ ps : fhd ] , together with the box ( dotted lines ) of pre selected objects ( see sect . [ selection ] ) . a @xmath46 histogram of those selected objects is shown in fig . [ ps : color ] . formally , a key underlying assumption is that the color ( metallicity ) does not significantly vary with magnitude ( mass ) within a sub population . this appears to be the case as far as can be tested with our data . a kmm test ( ashman , bird & zepf 1994 ) was run on our constrained sub sample . the test determines the confidence with which bimodality can be said to be present , the most likely relative contribution of each population to the sample , and the peaks ( i.e. mean colors ) of each sub population . we find that the color distribution is bimodal at the @xmath47% confidence level . the relative contributions of objects to the selected sample ( cf . sect . [ selection ] ) are 282 and 423 in the blue and red populations respectively . the peaks of the color distribution lie at @xmath48 and @xmath49 for the blue and red sub populations , respectively . formally , the peaks can be determined with an accuracy of 0.005 mag . the error adopted here reflects the range of values derived by varying the selection criteria ( i.e. magnitude and color cuts ) . two fitting modes were used : 1 ) forcing identical dispersions ( homoscedastic fit ) , 2 ) allowing independent dispersions ( heteroscedastic fit ) . we note that substantially better fits , in terms of reduced @xmath50 , are obtained if the dispersions of the blue and red distributions are allowed to differ , and that the red population seems to have a color distribution twice as broad as the blue one ( @xmath51 , @xmath52 ) . we estimated the peak colors with yet another method that avoids the assumption of a gaussian distribution for each color peak ( intrinsic to the kmm code ) . we computed the density distribution in color of the objects with the help of a kernel estimator using the epanechnikov kernel ( e.g. gebhardt & kissler - patig 1999 ) . we used @xmath53 for the smoothing parameter ( see silverman 1986 for detailed discussion ) , an average of the optimal smoothing parameters for the blue and red peak ( @xmath54 and @xmath55 ) estimated from the @xmath11guess of kmm for both modes . the density distribution is overplotted in fig . [ ps : color ] . a biweight fit to each peak returns @xmath56 and @xmath57 , in perfect agreement with the results from the kmm code . to obtain the mean colors of the peaks , the above @xmath46 values need to be corrected for the calibration offset described in sect . [ extended ] ( @xmath58 ) , and for a galactic reddening towards ngc 4472 of e@xmath59 ( schlegel et al . 1998 ) , i.e. e@xmath60 ( following dean et al . 1978 ) . note that the _ relative _ color difference , later used to derive the age difference , is only affected by these corrections in second order . the final , corrected mean colors of the two populations are @xmath61 and @xmath62 . the to magnitudes of the luminosity functions for the blue and red sub population were individually determined in each filter using a maximum likelihood approach ( cf . secker 1992 ) . the changing background counts over our whole sample motivated us to develop a version of secker s ( 1992 ) maximum likelihood estimator which includes background variations . we calculated the completeness as a function of magnitude and background noise as described in sect . [ artstar ] . this improved code , described below , was used to derive the peak of the magnitude distribution for each filter and each sub population . secker ( 1992 ) showed that the gclf of the milky way and m31 was best represented analytically with the student s @xmath63-function @xmath64 we adopted the student s @xmath63function as the most appropriate distribution for the gclf of ngc 4472 since it matches the tails of the distribution better than a gaussian function . the difference in the result using the one or the other function is , however , very small . usually a dataset of @xmath65 independent observations , like a sample of magnitudes @xmath66 , is distributed in an unknown way . assuming a distribution function @xmath67 which describes the distribution of data properly and depends on @xmath68 parameters @xmath69 ( particularly the turn over magnitude @xmath70 and the dispersion @xmath71 ) the likelihood function @xmath72 for a given dataset @xmath73 . the likelihood function is @xmath74 for convenience , during the calculation the logarithmic likelihood function is used . during the evaluation the best set of parameters @xmath75 is found if @xmath76 for a detailed treatment of maximum likelihood theory see , for instance , bevington & robinson ( 1992 ) . our improved maximum likelihood code , based on secker s ( 1992 ) code , takes into account the influence of the background noise level @xmath77 on the completeness function @xmath78 , in addition to the magnitude . similarly , the photometric error function @xmath79 now also depends on these two variables . the distribution function is @xmath80 where @xmath70 is the turn over magnitude of the given distribution and @xmath71 is its dispersion ( note : @xmath81 ) , while @xmath77 is the background noise for the evaluation position on the chip . @xmath82 is the mixing parameter , i.e. the fraction of globular cluster candidates in the whole data set including the background objects . note that , in our case , k is close to 1 . it is necessary to normalize all the functions used , which is implicitly accomplished by @xmath83 and @xmath84 where @xmath85 is the background contamination function obtained from the hdf data . @xmath86 is the photometric error function @xmath87 which is assumed to introduce gaussian errors to the distribution function @xmath88 with a magnitude and background noise dependent dispersion @xmath89 the parameters @xmath90 and @xmath91 can be estimated with sufficient accuracy from a two dimensional fit to the photometric errors obtained from the artificial star experiments . they act as input parameters to our code . the need to account for the variable background , i.e. the need for the improved code , is best seen in fig . [ ps : inc ] . the selection completeness as a function of magnitude is plotted for a few typical values of background counts found in our datasets . the 50% completeness limits shift to brighter magnitudes with increasing background , i.e. photon noise . alternatively one could approximate a `` mean '' completeness by distributing the artificial stars with the same density profile as the globular clusters , without including the background as a parameter in the maximum likelihood analysis . the approach we have adopted is the more accurate solution to the problem . the package described above is available as an iraf task from the authors . the maximum likelihood routine of drenkhahn ( 1999 ) was used to determine the to magnitudes . our completeness functions fit down to the 50% completeness levels ( @xmath92 , @xmath93 ) were used as the input . the data were convolved with the errors as a function of magnitude ( see eq . [ eq : phi ] ) . the input globular cluster data were selected as described in sect . [ selection ] . from kmm mixture modeling we determined that @xmath94 is the best value for separating blue and red clusters . for the analysis we split the full dataset into the central ( westphal ) pointing and the combination of the halo ( brodie ) pointings , since these two datasets had slightly different exposure times , i.e. slightly different limiting magnitudes / completenesses . the to magnitudes for the individual pointings are given in table 3 , together with an average for the whole sample . the globular cluster luminosity functions are shown in fig . [ ps : gclf ] . the values in table 3 still need to be corrected for galactic extinction ( @xmath95 and @xmath96 ) and the additional aperture correction ( see sect . [ extended ] ) , however , these corrections do not affect the _ relative _ difference between the to magnitudes . we varied the input completeness and noise limits to the maximum likelihood code , the resulting to magnitudes all lie within @xmath97 mag . as a consistency check , we attempted to reproduce the mean ( uncorrected ) colors of the sub populations ( @xmath98=0.99@xmath99 and 1.24@xmath99 ) by computing the differences between the @xmath14 and @xmath15 to magnitudes for a given population . we obtained @xmath100 for the blue population and @xmath101 for the red population , where the errors are the errors of the @xmath14 and @xmath15 tos taken in quadrature . the color for the blue population derived in this way is somewhat redder than that derived from the color distribution but the values agree to within the errors for both populations . the error in the to magnitude directly affects the error in the estimation of the age difference between the two sub populations ( see sect . [ models ] ) . we carried out monte carlo simulations to check whether our relatively modest sample sizes ( which were as low as 118 to 232 objects when divided into inner and outer samples ) affect the results significantly and to establish the associated error . the blue and red samples were simulated in the @xmath14 filter using the number of objects in each sample and assuming a gaussian distribution with the parameters derived above . these artificial samples were input into the maximum likelihood code . the 68% confidence levels ( @xmath102 ) on the to magnitudes , from @xmath103 iterations , were 0.10 and 0.13 for the blue and red samples , respectively . these are in good agreement with the @xmath102 errors returned by the maximum likelihood estimation . the derived colors and to magnitudes were compared to previous results from other groups . geisler et al . ( 1996 ) derived metallicity peaks for the blue and red sub populations from washington photometry . they found peaks at [ fe / h]@xmath104 dex and @xmath105 dex for the blue and red sub samples , respectively . our colors , derived in sect . [ meancol ] , can also be translated into metallicities . however , large errors are introduced by the particular choice of a conversion relation of the @xmath46 color into metallicity . using the relation given by kissler - patig et al . ( 1998a ) ( [ fe / h]@xmath106 ) , which is tuned to reflect the non - linear behavior at high metallicity , we obtain metallicities of [ fe / h]@xmath107 dex and @xmath108 dex . these values are somewhat lower than , but within the errors of , the values derived by geisler et al . ( 1996 ) . using the transformation of kundu & whitmore ( 1998 ) , [ fe / h ] @xmath109 , our data yield metallicities of [ fe / h ] @xmath110 dex and [ fe / h ] @xmath111 dex . the error corresponds to our photometric error in @xmath46 only . no calibration error of kundu & whitmore is considered . the metallicity calibration of couture et al . ( 1990 ) , [ fe / h ] @xmath112 , gives the somewhat larger values of [ fe / h ] @xmath113 dex and @xmath114 dex , similar to those derived by geisler et al . ( 1996 ) . to magnitudes have been derived by several groups but only for the whole sample ( i.e. an average of the red and blue populations ) and in different filters . previous results should be compared to the average of our blue and red sample , with the necessary correction for reddening and aperture size : @xmath115 ( including the error on the to magnitude and the error on the aperture correction ) . furthermore , all other groups assumed e@xmath116 towards ngc 4472 ( using burstein & heiles 1982/84 ) while we adopted the schlegel et al . ( 1998 ) map from dirbe / iras whose zero - point is offset by 0.020 in e@xmath117 . our value would translate into @xmath118 if we assumed e@xmath116 . using the washington johnson transformation of geisler ( 1996 ) , @xmath119 , lee et al.s ( 1998 ) result translates into @xmath120 , in excellent agreement with our result . we note that lee et al . investigated their blue and red samples separately , but did not find any significant difference in their to magnitudes to within @xmath121 0.15 mag . however , their analysis was less rigorous in this respect than ours , and their data suffered from more background contamination , as is expected for ground based observations . harris et al . ( 1991 ) derived @xmath122 with a 3parameter gaussian fit to the gclf . using @xmath123 for the total globular clusters population , this translates into @xmath124 . ajhar et al . ( 1994 ) derived @xmath125 ( read off their fig . 14 ) which , for @xmath126 , translates into @xmath127 . these results are in very good agreement with our determination . variation in the to magnitude with radius within a population can occur if the mean age , metallicity or mass changes with radius . large age and metallicity gradients are ruled out by the absence of any significant color gradient in the sub populations ( see fig . [ ps : colorrad ] ) . a systematic change in the mean mass with radius could be caused by destruction processes shaping the globular cluster mass function . these are more important towards the center of the galaxy ( e.g. recent work by vesperini 1997 and gnedin & ostriker 1997 ) , and significant changes are only expected within @xmath128 kpc of the center ( @xmath129 at 16 mpc ) . there the peak of the gclf is expected to be brighter by up to @xmath130 magnitudes , and the peak should be sharper as a consequence of the preferential destruction of low mass clusters . our inner sample ( westphal data ) includes clusters at projected radii between 2 and 127 with a mean of 65 and our outer sample ( brodie data ) includes clusters at projected radii between 76 and 252 with a mean of 160 . differences are therefore expected between these two samples . for the blue population , the to magnitude varies from the inner to the outer regions by @xmath131 and @xmath132 in @xmath14 and @xmath15 respectively . the weighted mean is @xmath133 , i.e. we observed no brightening of the to towards the center , although the uncertainty is on the order of the expected variation . for the red population the to magnitude varies from the inner to the outer region by @xmath134 and @xmath135 in @xmath14 and @xmath15 respectively . formally , the to appears brighter by @xmath136 ( weighted mean ) for the inner red population . we do not see any changes in the dispersion of the gclf to within the errors . this result is probably not yet secure enough to be worth commenting on in depth . however , a difference between the red and blue population could occur , for example , if the blue and red clusters are on different orbits . sharples et al . ( 1998 ) found tentative evidence for the blue clusters in ngc 4472 having a higher velocity dispersion and higher rotation than the red ones . kissler - patig & gebhardt ( 1998 ) found a similar ( but firmer ) result for m 87 , another giant elliptical that hosts enough globular clusters for such an analysis . if indeed the blue clusters are preferentially on tangential orbits , while the red clusters are preferentially on radial orbits , destruction processes would be more efficient for the red population . finally , we note that harris et al . ( 1998 ) and kundu et al . ( 1999 ) also looked for this effect in m 87 . kundu at al . even looked for trends in the blue and red populations separately . neither group found any brighting of the gclf towards the center of m 87 , but note again that the uncertainties in the measurements are still of the order of the expected signal . we looked for color gradients within the sub populations using a weighted least squares fitting routine . in addition to the color cuts ( @xmath137 and @xmath138 ) , the samples were cut at @xmath14 = 24.8 to provide at least 50% completeness even in the central regions . the results are plotted in fig . [ ps : colorrad ] , together with a fit for the total sample . the derived gradients are @xmath139 mag / log(arcsec ) and @xmath140 mag / log(arcsec ) for the blue and red population respectively . different color and magnitude cuts scatter these results within 0.01 mag / log(arcsec ) . any positive gradients in the sub populations are either very small or non existent . for the whole sample , however , we find a clear negative gradient of @xmath141 mag / log(arcsec ) , corresponding to a metallicity gradient of @xmath9[fe / h]@xmath142 dex / log(arcsec ) ( somewhat dependent on the color metallicity relation used ) . we confirm geisler et al.s ( 1996 ) result , that this gradient is caused by the changing ratio of red to blue clusters as a function of radius , rather than an overall decrease in metallicity with radius . quantitatively , we find a somewhat smaller gradient than reported by geisler et al . ( 1996 ) , who found @xmath9[fe / h]@xmath143 dex / log(arcsec ) . this is probably due to our smaller radial coverage and our use of a less metallicity sensitive color . indeed , our sample extends to only 250 while geisler et al.s data extend out to @xmath121 500 . the ratio of blue to red clusters varies by only @xmath121 10% from our inner to outer regions , as derived from a comparison of their cumulative radial distributions . a similar situation is seen in m 87 , for which harris et al . ( 1998 ) report no color gradient within 60 , and kundu et al . ( 1999 ) report a weak gradient of @xmath144 mag / log(arcsec ) within 100 . however , lee & geisler ( 1993 ) report a strong gradient ( @xmath9[fe / h]@xmath145 dex / log(arcsec ) ) over 500 radius . the high resolution of hst allows us to study the _ relative _ sizes ( see sect . [ extended ] ) of the globular cluster candidates using the parameter @xmath146 ( see sect . [ selection ] ) . for extended objects the magnitude difference in two different apertures will appear large as the light profile contributes a non negligible amount of light to the outer aperture . the opposite is the case if the light profile is narrow . for this analysis , we restricted ourselves to clusters found on the wide field ( wf ) camera chips , to avoid the problematic conversion between the wf and the planetary camera ( pc ) . [ ps : sizerad ] presents the histogram of globular cluster sizes for both the blue and red sub populations . the median values for these two distributions point to a systematic size difference between the blue and the red globular clusters . the median values for the blue and red populations are @xmath147 mag and @xmath148 mag , respectively , indicating that the blue globular clusters are substantially larger than the red globular clusters . a maximum likelihood estimation of the mean and its error yields the same results . interestingly , in both ngc 3115 ( kundu & whitmore 1998 ) and m 87 ( kundu et al . 1999 ) , the blue clusters also appear significantly larger than the red ones . the authors suggest this is due to different radial distributions for the red and blue clusters , noting that galactic globular clusters with larger galactocentric distances have larger half light radii ( van den bergh 1994 ) . to study the radial dependence of globular cluster size , we plotted the size parameter ( @xmath146 ) versus radius and found no gradients : @xmath149 where @xmath150 is the radius in arcsec , and @xmath151 , for the blue and red clusters respectively ( see also fig . [ ps : sizerad ] ) . the difference between blue and red clusters is present at all radii . we therefore speculate that the different sizes are a relic of the different formation processes , unless the red and blue clusters are on significantly different orbits . for example , the red clusters could be preferentially on radial orbits and be systematically more affected than the blue clusters by the central potential during their close passage near the galaxy center . we compare our results ( to magnitude and mean color ) to various population synthesis models in order to derive the age _ difference _ between the blue and red globular cluster sub populations . we used the new models from maraston ( 1998 ) as well as the most recent models from the gttingen group ( kurth et al . 1999 ) , and the models from bruzual & charlot ( 1996 ) and worthey ( 1994 ) . in each case we compared our data to every available initial mass function ( imf ) , i.e. usually the salpeter imf and a multi slope imf . the full range of metallicities was used as well as the necessary age range . since we are only interested in age _ differences _ the exact normalization of the absolute magnitude , which is dependent on mass and mass to light ( m / l ) ratio for a given imf , is not important . the grids in fig . [ ps : mar ] to [ ps : wo ] can , therefore , be somewhat arbitrarily shifted in the y direction . in the following , we fixed the grids in the y direction such that the oldest sub population lies on the oldest computed isochrone ( 15 gyr or 16 gyr ) . further , we arbitrarily set the blue population to @xmath152 and @xmath153 in order to compare the different models directly . there is no freedom in the x direction since the color is not mass dependent , and the grid is fixed in @xmath98 for a given stellar population . for the mean colors of the blue and red sub populations we used the colors derived in section [ meancol ] . the to values used were the ones derived for the full sample ( see table [ tab : magcol ] ) . we stress again that additive corrections to both the blue and red to magnitudes ( i.e. reddening correction , potential systematic errors , aperture corrections ) only influence the difference as second order effects , if at all , and are not a concern . however , we make two important underlying assumptions . we assume that the m / l ratios for blue and red clusters are only dependent on age and metallicity ( i.e. the imfs of the blue and red clusters are similar ) . second , we assume that blue and red cluster populations have the same underlying mass distribution ( by which we mean globular cluster masses , not imfs ) . under these assumptions , to magnitudes are influenced only by age and metallicity . we briefly discuss the validity of these assumptions and the errors they potentially introduce . kissler - patig et al . ( 1998b ) review the empirical reasons for expecting the mass distributions and m / l ratios to be similar among various sub populations . briefly , the m / l assumption is essentially the assumption that both blue and red clusters have similar imfs , at least below 1 @xmath154 where it influences our result . observations so far for globular clusters in the local group seem to confirm this picture ( e.g. dubath & grillmair 1997 ) . in the case of the mass distribution , globular cluster formation seems to follow a universal mechanism ( e.g. elmegreen & efremov 1997 ) . the constancy of the to magnitude from galaxy to galaxy ( e.g. the review by whitmore 1997 ) and the observed mass distributions of young clusters ( e.g. brief review by schweizer 1997 ) support this picture . the possibility that size differences might indicate different formation processes for blue and red clusters was discussed in section [ gcsizes ] . it is not clear , however , whether it is only the mean sizes or also the mean masses that are affected . moreover , our assumption would fail if , despite similar mass distributions at formation , blue and red clusters are affected differently by destruction processes . there is a hint ( see section [ radial ] ) that this might be the case in the inner region , where the difference between the blue and the red to magnitudes is smaller by @xmath155 ( combining the @xmath14 and @xmath15 information ) than in the outer region . ( we note for completeness that an age gradient in the population would have the same observable effect as a mass gradient ) . if this effect is real , the to difference of the total sample , as used , could be affected by a 0.1 to 0.2 mag systematic error . we estimate that this empirical determination gives about the right order of magnitude for the potential systematic error in our method and , in what follows , we will assume a potential systematic error of @xmath156 mag , which translates into roughly 4 gyr . note that for similar m / l values of the blue and red clusters , such a difference of 0.2 mag would translate into only a 20% difference in the mass characterizing the to . irrespective of the accuracy needed for our study , blue and red clusters _ have _ roughly similar mass distributions . for completeness , we mention a method for checking the validity of these assumptions and a method for avoiding them . for the chosen filters , age and characteristic mass influence the to far more than metallicity , which is fairly well determined by the color . therefore , one could determine the mean mass difference ( and check the above assumption ) by obtaining an independent estimate of the mean age difference . an independent age estimate could be obtained with spectroscopy of a large number of globular clusters and the measurement of an age sensitive absorption line index . several tens of h@xmath157 measurements in each sub population would allow the mean h@xmath157 value to be determined to within @xmath158 which would translate into an accuracy in the mean age difference of @xmath159 gyr . this in turn would allow us to verify that the characteristic to masses of the red and blue populations lie within 10% to 20% of each other . an alternative method for determining the mean age difference without making any assumptions about the mass distributions would be to use mass independent quantities that depend only on age and metallicity . band colors are such quantities . a plot similar to figures [ ps : mar ] through [ ps : wo ] , but in a color color plane , would remove any uncertainties due to different mass distributions . however , it is impossible to find two colors in the optical and nir spectrum that disentangle age and metallicity as efficiently as the color / to magnitude combination . in figure [ ps : mar ] , we compare our results to the models of maraston ( 1998 ) . she computed single burst population models with a salpeter ( 1955 ) and multi - slope imf ( gould et al . 1998 ) for metallicities z@xmath160 and @xmath161 , and ages up to 15 gyr . we refer the reader to the original paper for further details of the modeling . the derived age difference , which is quoted in the following as ( age@xmath162)@xmath163(age@xmath164 ) , varies between @xmath165 gyr ( gould et al . imf , @xmath15 band ) and @xmath166 gyr ( gould et al . imf , @xmath14 band ) , with errors of about 2.7 gyr ( 1@xmath11 ) . within the errors , the blue and red population appear coeval , with a formal difference of @xmath167 gyr ( straight mean of the four computed differences with their dispersion shown as the error ) . a 3@xmath11 limit on the age difference would be around 6 gyr . in figure [ ps : kurth ] , we compare our results to the recent models of kurth et al . they computed single burst models with a salpeter ( 1955 ) and a miller scalo ( 1979 ) imf for 6 metallicities between z@xmath168 and @xmath169 , and ages up to 16 gyr . the age difference varies between @xmath170 gyr ( salpeter imf , @xmath15 band ) and @xmath171 gyr ( salpeter imf , @xmath14 band ) with a typical error of 2.2 gyr . the mean difference is @xmath172 gyr , i.e. the red population appears somewhat older than the blue one . the 3@xmath11 differences are around @xmath173 and @xmath174 gyr . in figure [ ps : bc ] , we compare our results to the models of bruzual & charlot ( 1996 ) . we used the single burst populations computed with a salpeter ( 1955 ) imf and a scalo ( 1986 ) imf , with metallicities between z@xmath175 and @xmath169 . their models with a miller scalo ( 1979 ) imf were not used . the tracks for their highest metallicity ( z@xmath176 ) are not shown . the oldest modeled populations is 16 gyr . the age differences vary between @xmath177 gyr ( scalo imf , @xmath15 band ) and @xmath178 gyr ( salpeter imf , @xmath14 band ) , with an typical error of 2.8 gyr . the mean difference is @xmath179 gyr . the 3@xmath11 limits lie around @xmath180 and @xmath181 gyr . in figure [ ps : wo ] , we compare our results to worthey s ( 1994 ) single burst models . he computed models with a salpeter ( 1955 ) imf and a miller scalo ( 1985 ) imf for metallicities between z@xmath182 and 0.05 , up to an age of 16 gyr . the age difference varies between @xmath183 gyr ( salpeter imf , @xmath15 band ) and @xmath184 gyr ( miller scalo imf , @xmath14 band ) , with a typical error of 2.5 gyr . the mean difference is @xmath185 gyr , i.e. the red populations appears older than the blue one . the 3@xmath11 limits are around @xmath186 and @xmath177 gyr . we note that the @xmath15 band results give a systematically smaller difference ( in the sense we defined it : age@xmath187age@xmath188 ) than the @xmath14 band results . except for the bruzual & charlot ( 1996 ) models , the results from the @xmath15 band suggest that the red population is marginally older than the blue one . the different models predict mean differences anywhere between @xmath189 gyr and @xmath190 gyr , with a mean of @xmath191 gyr and a dispersion of @xmath192 gyr . it would be somewhat arbitrary to assign more weight to one model than to another , so that we are currently limited in the age determination by the model uncertainties . nevertheless , it seems that both populations are coeval within the errors , and we can exclude at the 99% confidence level one population being half as old as the other . if our results are affected by a systematic error , e.g. if the smaller sizes of the red globular clusters do indeed translate into a lower `` mean '' mass ( such that the red to magnitude is @xmath156 lower see section [ ml ] ) , the age difference would be affected by @xmath193 gyr . the above results suggest that the vast majority of globular clusters and stars formed at high redshift in this cluster galaxy . as already mentioned in section [ intro ] , several other lines of evidence lead to similar results for cluster galaxies . interestingly , the globular cluster results imply that two major star formation events took place at high redshift . those two events appear to have happened moderately closely in time but from differently enriched gas . the blue clusters appear to have formed from gas with less than @xmath194th solar metallicity while the red clusters formed out of gas with close to solar metallicity . furthermore , the mean sizes of the blue and red clusters appear to be significantly different . this suggests different conditions at the formation epoch , unless they are directly related to the metallicity and associated cooling processes in the molecular clouds in which the clusters originated . it is unclear whether the blue or the red clusters are the older ones . at face value it seems odd that the more metal rich clusters would be the older ones . however , in scenarios where the blue clusters formed in halo clouds ( e.g. kissler - patig et al . 1998b ) or were accreted with dwarf galaxies ( e.g. ct et al . 1998 , hilker et al . 1999a , b ) , this could be understood in terms of the large structures ( e.g. bulges ) collapsing first , followed shortly thereafter by the smaller structures ( clouds / dwarf galaxies ) . this , however , would force a revision of the merger scenario which clearly predicts that the red clusters will be younger than the blue ones . alternatively , if two sub populations of clusters are coeval or if the blue clusters are the older ones , age differences of a few gyr between the different sub populations could be explained in all globular cluster / galaxy formation scenarios , i.e. the above mentioned scenarios , merging ( ashman & zepf 1992 ) , or in situ formation of distinct populations ( e.g. forbes et al . 1997 , harris et al . 1998 , harris et al . 1999 ) . we note only that , in the merger scenario , merging would have to have taken place at early times ( @xmath195 at the 3@xmath11 level ) which is somewhat at odds with the recent theoretical predictions from hierarchical clustering models ( e.g. kauffmann 1996 , baugh et al . these tend to predict that the main star formation in cluster environments occurred at redshifts @xmath196 . the to magnitude ( @xmath197 ) of the gclf appears to be essentially universal and , therefore , can be used as a distance indicator ( e.g. the review by whitmore , 1997 ) . @xmath197 depends on the age and metallicity of the globular clusters ( ashman , conti & zepf 1995 , see also sect . [ models ] ) , and weakly on galaxy type ( harris 1997 ) which may be a consequence of the age / metallicity dependence . the derived to magnitude for the globular clusters in ngc 4472 can be used to obtain a distance to the galaxy . the method is best calibrated using the to magnitude of the milky way gclf , since it involves the minimum number of steps on the distance ladder . the m31 gclf serves as a strong check on this calibration ( cf . ferrarese et al . 2000 ; kavelaars et al . 1999 ) . the milky way system is dominated by old globular clusters of mean metallicity [ fe / h]@xmath198 . it is , therefore , well suited to fitting the to magnitude of the blue population in ngc 4472 ( see sect . [ models ] ) and no correction for age metallicity dependence need be applied . however , we have to assume that age metallicity dominates any dependence on galaxy type , and that globular cluster mass functions are intrinsically similar , which seems justified to first order ( e.g. kissler - patig et al . 1998b ) . della valle et al . ( 1998 ) recently re derived @xmath197 for the milky way system . to derive individual distances to all galactic globular clusters , they adopted a horizontal branch ( hb ) magnitude metallicity relation derived from globular cluster main sequence fitting using distance measurements to sub dwarfs from hipparcos ( in this case by gratton et al . 1997 ) . from the resulting gclf they obtained @xmath199 noting that , depending on the adopted hb magnitude metallicity relation , this result could suffer from a 0 to @xmath200 mag systematic error . drenkhahn & richtler ( 1999 ) used a different approach . they assumed the lmc distance to be the fundamental zero point ( here @xmath201 ) , and derived an rr lyrae magnitude metallicity relation . the remaining steps were similar to those of della valle et al .. drenkhahn & richtler found @xmath202 and @xmath203 for the milky way gclf . the apparent to magnitude derived for the blue population is @xmath204 and @xmath205 ( see tab . [ tab : magcol ] ) . these values need to be corrected for the photometric term discussed in sect . [ extended ] ( @xmath206 , @xmath207 ) , as well as for galactic extinction ( @xmath95 , @xmath96 ) . using the absolute to magnitudes discussed above , we derive distance moduli of @xmath208 and @xmath209 ( adding all errors in quadrature ) , i.e. a mean @xmath210 ( where the error does not include any possible systematic error in the calibration ) . the corresponding distance is @xmath211 mpc . a quick comparison with distances derived for ngc 4472 by other methods shows very good agreement ( despite completely different calibrations ) . jacoby et al . ( 1990 ) derived @xmath212 from the planetary nebulae luminosity function . tonry et al . ( 1990 ) and neilsen ( 1999 ) derived @xmath213 and @xmath214 , respectively , from surface brightness fluctuations . cepheid distances now exist for 7 spiral galaxies in the virgo cluster ( cf . pierce et al . 1994 and ferrarese et al . 2000 ) . with the exception of ngc 4639 ( saha et al . 1996 ) , which appears to be background to the main body of spirals , the mean distance to the other 6 galaxies is @xmath215 or [email protected] mpc . our gclf distance to ngc 4472 is in excellent agreement with this distance , which supports the view that the virgo cluster spirals and ellipticals are at essentially the same distance . the age difference between the two major globular cluster sub populations in an early type galaxy has been determined . we have developed an age differentiating method which uses the precise to magnitudes and colors of the sub populations . we have applied this method to ngc 4472 , matching our observables to several ssp synthesis models . an extended object analysis showed that at the distance of virgo ( i.e. @xmath21716 mpc ) hst is capable of resolving single globular clusters . thus it is necessary to make an additive correction ( of @xmath218 mag and @xmath22 mag at the distance of virgo ) to the mean correction of @xmath20 mag ( for point sources ) when extrapolating hst photometry obtained in 0.5 radius apertures to total luminosities . this corresponds to a color correction of @xmath219 . we found the globular cluster color distribution in ngc 4472 to be clearly bimodal ( at the 99.99% confidence level ) with two peaks located at @xmath220 and @xmath221 corresponding to @xmath61 and @xmath62 when corrected for reddening and finite aperture . we determined that the red globular clusters have a color distribution almost twice as broad as the blue ones . we divided the whole globular cluster population into an inner and an outer sample , as defined by the hst different pointings . we used an improved maximum likelihood estimator , which accounts for varying background values , to determine the to magnitudes for each sub population in each radial sample ( see tab . [ tab : magcol ] ) . the to of the blue sub population is the same for the inner and outer sample . by contrast , the red sub population appears to be marginally brighter by @xmath136 mag in the inner region than in the outer region . this difference is of the expected order of magnitude if destruction processes act on the red sample within 5 kpc . the radial color distributions of each sub population show almost no gradients . however , we detected an overall gradient of @xmath222 mag / log(arcsec ) in the system , which is mostly an effect of the changing ratio of red to blue globular clusters with radius ( cf . geisler et al . 1996 ) . taking advantage of hst s high spatial resolution , we estimated the mean sizes of the globular clusters using the @xmath146 parameter . the blue globular clusters were found to be larger than the red ones , as was already observed in m 87 ( kundu et al . 1998 ) and ngc 3115 ( kundu & whitmore 1997 ) . the most likely explanations are that the size difference reflects different formation processes , or that red clusters are on preferentially radial orbits , while blue clusters are on preferentially tangential orbits . we used 4 different ssp models to derive the mean age difference between the red and blue sub population . our underlying assumption was that both blue and red populations have similar mass distributions which appears to be valid at least at the 10% to 20% level ( which , nevertheless , corresponds to a potential systematic error of @xmath223 gyr ) . the mean age difference between the blue ( metal poor ) and the red ( metal rich ) sub populations is age@xmath224 age@xmath225 gyr , where the error is the dispersion introduced by the different models ( for a given model the error on our method is @xmath2172 gyr ) . the images as well as electronic lists of the photometry are available from the authors . also available is the maximum likelihood code as an iraf task used to derive the globular cluster luminosity functions taking into account the background variations over the field . we are very thankful to claudia maraston for customizing part of her models for our purposes and providing electronic files , as well as to georg drenkhahn for his help in implementing his maximum likelihood code . we also thank sandy faber , duncan forbes , karl gebhardt , and klaas de boer for interesting discussions . thp gratefully acknowledges support from the uco / lick observatory . mkp was partly supported by the alexander von humboldt foundation . this research was funded by hst grants go.05920.01 - 94a and go.06554.01 - 95a and faculty research funds from the university of california at santa cruz . this research was partly supported by nato collaborative research grant crg 971552 . l ll ll & ra(2000)@xmath226 & @xmath22712h 29 m 46.79s @xmath228s + & dec(2000)@xmath226 & @xmath22908@xmath230 00 01.50 @xmath228 + & @xmath231 @xmath232 & @xmath233 mag + & @xmath98 @xmath234 & @xmath221 mag + & @xmath235 @xmath232 & @xmath236 mag + & mtype @xmath232 & e2/s0(2 ) + & @xmath237 @xmath232 & @xmath238 km s@xmath239 + & distance @xmath240 & @xmath1 mpc + [ tab : ngc4472 ] l ll ll go.5236 westphal & 12 29 48.73 & @xmath22907 59 55.85 & 1800 sec & 1800 sec + go.5920 brodie ( north ) & 12 29 45.22 & @xmath22908 02 34.38 & 2200 sec & 2300 sec + go.5920 brodie ( south ) & 12 29 45.20 & @xmath22907 57 30.11 & 2200 sec & 2300 sec + [ tab : observation ] ll ll ll l total sample & & & & & & + & & & & & & + @xmath14 & blue & 23.62 & 0.09 & & + & red & 24.13 & 0.11 & & + @xmath15 & blue & 22.48 & 0.07 & & + & red & 22.90 & 0.11 & & + & & & & & & + westphal ( inner ) sample & & & & & & + & & & & & & + @xmath14 & blue & 23.55 & 0.15 & 1.21 & 0.10 & 118 + & red & 23.97 & 0.19 & 1.42 & 0.10 & 180 + @xmath15 & blue & 22.62 & 0.21 & 1.24 & 0.14 & 129 + & red & 22.83 & 0.19 & 1.39 & 0.10 & 174 + & & & & & & + brodie ( outer ) sample & & & & & & + & & & & & & + @xmath14 & blue & 23.68 & 0.11 & 1.36 & 0.09 & 175 + & red & 24.30 & 0.13 & 1.39 & 0.09 & 232 + @xmath15 & blue & 22.33 & 0.08 & 1.14 & 0.07 & 178 + & red & 22.97 & 0.13 & 1.39 & 0.08 & 213 + & & & & & & + [ tab : magcol ]

effective complexity measures the information content of the regularities of an object . it has been introduced by m. gell - mann and s. lloyd to avoid some of the disadvantages of kolmogorov complexity , also known as algorithmic information content . in this paper , we give a precise formal definition of effective complexity and rigorous proofs of its basic properties . in particular , we show that incompressible binary strings are effectively simple , and we prove the existence of strings that have effective complexity close to their lengths . furthermore , we show that effective complexity is related to bennett s logical depth : if the effective complexity of a string exceeds a certain explicit threshold then that string must have astronomically large depth ; otherwise , the depth can be arbitrarily small . effective complexity , kolmogorov complexity , algorithmic information content , bennett s logical depth , kolmogorov minimal sufficient statistics , shannon entropy .

0810.5663

is complexity ? a great deal of research has been performed on the question in what sense some objects are `` more complicated '' than others , and how this fact and its consequences can be analyzed mathematically . one of the most well - known complexity measures is _ kolmogorov complexity _ @xcite , also called _ algorithmic complexity _ or _ algorithmic information content_. in short , the kolmogorov complexity of some finite binary string @xmath0 is the length of the shortest computer program that produces @xmath0 on a universal computer . so kolmogorov complexity quantifies how well a string can in principle be compressed . this notion of complexity has found various interesting applications in mathematics and computer science . despite its usefulness , kolmogorov complexity does not capture the intuitive notion of complexity very well . for example , random strings without any regularities , say strings that are constructed bitwise by repeated tosses of a fair coin , have very large kolmogorov complexity . but those strings are not `` complex '' from an intuitive point of view those strings are completely random and do not carry any interesting structure at all . _ effective complexity _ is an attempt by m. gell - mann and s. lloyd @xcite,@xcite to define some complexity measure that is closer to the intuitive notion of complexity and overcomes the difficulties of kolmogorov complexity . the main idea of effective complexity is to split the algorithmic information content of some string @xmath0 into two parts , its random features and its regularities . then , the effective complexity of @xmath0 is defined as the algorithmic information content of the regularities alone . in this paper , we are interested in the basic properties of effective complexity , and how it relates to other complexity measures . in particular , we give a more precise formal definition of effective complexity than has been done previously . we use this formal framework to give detailed proofs of the properties of effective complexity , and we use it to show an unexpected relation between effective complexity and bennett s logical depth @xcite . since there are now so many different complexity measures @xcite , our result contributes to the clarification of the interrelations within this `` zoo '' of complexity measures . moreover , we hope that our more formal approach helps to find applications of effective complexity within mathematics , in a similar manner as this has been done for kolmogorov complexity . we now describe our main results and give a synopsis of this paper : * after some notational preliminaries in section [ secpreliminaries ] , we motivate and state the main definition of effective complexity in section [ seceffectivecomplexity ] . * in section [ sec : basicproperties ] , we analyze the basic properties of effective complexity . in particular , we show in theorem [ thesimple ] that effective complexity indeed avoids the disadvantage of kolmogorov complexity that we have explained above : random strings are effectively simple . + although the existence of effectively complex strings has been mentioned in @xcite , it has not been conjectured explicitly . based on the notion of algorithmic statistics as studied by gcs at al . @xcite we provide a formal existence proof , see theorem [ thecomplex ] . * section [ secdepth ] contains our main result ( theorem [ theeffdepth ] and theorem [ theeffdepth2 ] ) , the relation between effective complexity and logical depth . in short , it states that if the effective complexity of some string exceeds a certain explicit threshold , then the time it takes to compute that string from a short description must be astronomically large . this threshold is in some sense very sharp , such that the behavior of logical depth with respect to effective complexity is comparable to that of a phase transition ( cf . fig . [ figphasetransition ] on page ) . * in section [ sec : ec - and - kmss ] we show how effective complexity is related to the notion of kolmogorov minimal sufficient statistics . * finally , in the appendix , we give an explicit example of a computable ensemble on the binary strings that has non - computable entropy . this illustrates the necessity of the details of our definition in section [ seceffectivecomplexity ] . we start by introducing notation . we denote the finite binary strings @xmath1 by @xmath2 , where @xmath3 is the empty string , and we write @xmath4 for the length of a binary string @xmath5 . an _ ensemble _ @xmath6 is a probability distribution on @xmath2 . all logarithms are in base @xmath7 . we assume that the reader is familiar with the basic concepts of kolmogorov complexity ; a good reference is the book by li and vitnyi @xcite . there is a `` plain '' and a `` prefix '' version of kolmogorov complexity , and we will use both of them in this paper . the plain kolmogorov complexity @xmath8 of some string @xmath0 is defined as the length of the shortest computer program that outputs @xmath0 if it is given as input to a universal computer @xmath9 , @xmath10 prefix kolmogorov complexity is defined analogously , but with respect to a universal _ prefix computer _ @xmath11 . a prefix computer @xmath11 has the property that if @xmath12 is defined for some string @xmath13 , then @xmath14 is undefined for every string @xmath15 that is not the empty string . so @xmath16 there are different possible choices of @xmath11 and @xmath9 ; we fix one of them for the rest of the paper . several variations of kolmogorov complexity can be easily defined and will be used in this paper , for example , the complexity of a finite list of strings , or the complexity of an integer or a real number . with a few exceptions below , we will not discuss the details of the definitions here and instead refer the reader to ref . @xcite . the first exception that deserves a more detailed discussion is _ conditional complexity_. there are two versions of conditional complexity , a `` naive '' one and a more sophisticated one . the naive definition is @xmath17 that is , the complexity of producing string @xmath0 , given string @xmath18 as additional `` free '' information . a more sophisticated version due to chaitin @xcite reads @xmath19 that is , the complexity of producing @xmath0 , given a minimal program @xmath20 for @xmath18 . the advantage of @xmath21 compared with @xmath22 is the validity of a chain rule @xmath23 for all strings @xmath0 and @xmath18 . here we make use of a well - known notation @xcite which helps to suppress additive constants : suppose that @xmath24 are functions on the binary strings , and there is some @xmath25 independent of the argument value @xmath13 such that @xmath26 for every @xmath27 , i.e. the inequality holds _ uniformly _ for @xmath27 . then we write @xmath28 we use the notation @xmath29 if both @xmath30 and @xmath31 hold . note that the `` naive '' form of conditional complexity as defined in ( [ eqdefnaiveconditional ] ) does not satisfy the chain rule ( [ chain - rule ] ) . only the weaker identity @xmath32 holds in general . we will often use obvious identities like @xmath33 or @xmath34 without explaining in detail where they come from ; we again refer the reader to the book by li and vitnyi @xcite . another important prerequisite for this paper is the definition of the prefix kolmogorov complexity @xmath35 of some ensemble @xmath6 . in contrast to bit strings , there are several inequivalent notions of a `` description '' , and we can learn from ref . @xcite the lesson that it is very important to exactly specify which of them we will use . our definition of @xmath35 for ensembles @xmath6 is as follows . first , a _ program that computes @xmath6 _ is a computer program that expects two inputs , namely a string @xmath27 and an integer @xmath36 , and that outputs ( the binary digits of ) an approximation of @xmath37 with accuracy of at least @xmath38 . then , our preliminary definition of @xmath35 is the length of the shortest program for the universal prefix computer @xmath11 that computes @xmath6 . obviously , not every ensemble @xmath6 is computable there is a continuum of string ensembles , but there are only countably many algorithms that compute ensembles . another unexpected difficulty concerns the entropy of a computable ensemble , defined as @xmath39 . contrary to a first naive guess , the entropy of a computable ensemble does not need to be computable ; all we know for sure is that it is enumerable from below . to illustrate this , we give an explicit example of a computable ensemble with a non - computable entropy in example [ exnonentropy ] in appendix [ secappendix ] . thus , for the rest of the paper , * we assume that all ensembles are computable and have computable and finite entropy @xmath40 * , unless stated otherwise . even when one restricts to the set of ensembles with computable entropy , the map @xmath41 is not necessarily a computable function . hence the approximate equality @xmath42 is not necessarily true uniformly in @xmath6 . thus , from now on we replace the preliminary definition @xmath35 by @xmath43 * i.e. we assume that computer programs for ensembles @xmath6 carry additionally a subprogram that computes the entropy @xmath40*. to define the notion of effective complexity , we follow the steps described in one of the original manuscripts by m. gell - mann and s. lloyd @xcite . first , they define the _ total information _ of an ensemble as the sum of the ensemble s entropy and complexity . to understand the motivation behind this definition , suppose we are given some data @xmath0 ( a finite binary string ) which has been generated by an unknown stochastic process . we would like to make a good guess on the process that generated @xmath0 , even if we only have one sample of the process . this is similar to a scientist that tries to find a ( probabilistic ) theory of physics , given only the present state of the universe . to make a good guess on the probability distribution or ensemble @xmath6 that produced @xmath0 , we make two natural assumptions : * the explanation should be simple . in terms of kolmogorov complexity , this means that @xmath35 should be small . * the explanation should not allow all possible outcomes , but should prefer some outcomes ( including @xmath0 ) over others . for example , the uniform distribution on a billion different possible physical theories is `` simple '' ( i.e. @xmath35 is small ) , but it is not a `` good explanation '' of our physical world because it contains a huge amount of arbitrariness . this arbitrariness can be identified with the `` measure of ignorance '' , the entropy of @xmath6 . thus , it is natural to demand that the entropy @xmath40 shall be small . putting both assumptions together , it is natural to consider the sum @xmath44 which is called the `` total information '' @xmath45 . a `` good theory '' is then an ensemble @xmath6 with small @xmath45 . ' '' '' for every ensemble @xmath6 with entropy @xmath46 , we define the _ total information _ @xmath45 of @xmath6 as @xmath47 note that the total information is a real number larger than or equal to @xmath48 . if @xmath6 is computable and has finite entropy , as always assumed in this paper , then @xmath45 is finite . in the subsequent work @xcite by m. gell - mann and s. lloyd , it has been pointed out that @xmath49 . it follows that @xmath50 this has a nice interpretation : the total information gives the average complexity of computing a string with the detour of computing the ensemble . the next step in @xcite is to explain what is meant by a string being `` typical '' for an ensemble . going back to the analogy of a scientist trying to find a theory @xmath6 explaining his data @xmath0 , a good theory should in fact predict that the appearance of @xmath0 has non - zero probability . even more , the probability @xmath51 should not be too small ; it should be at least as large as that of `` typical '' outcomes of the corresponding process . what is the probability of a `` typical '' outcome of a random experiment ? suppose we toss a biased coin with probability @xmath52 for heads and @xmath53 for tails @xmath54 times , and call the resulting probability distribution @xmath6 . then it turns out that typical outcomes @xmath0 have probability @xmath51 close to @xmath55 , where @xmath56 , and @xmath57 is the entropy of @xmath6 . in fact , the probability that @xmath51 lies in between @xmath58 and @xmath58 for @xmath59 tends to one as @xmath54 gets large . in information theory , this is called the `` asymptotic equipartition property '' ( cf . an appropriately extended version of this result holds for a large class of stochastic processes , including ergodic processes . this motivates to define that a string @xmath0 is _ typical _ for an ensemble @xmath6 if its probability is not much smaller than @xmath60 . ' '' '' let @xmath6 be an ensemble , @xmath5 a string and @xmath61 . we say that _ @xmath0 is @xmath62-typical for @xmath6 _ , if @xmath63 we return to the scenario of the scientist who looks for good theories ( ensembles @xmath6 ) explaining his data @xmath0 . as discussed above , it is natural to look for theories with small total information @xmath45 . moreover , the theory should predict @xmath0 as a `` typical '' outcome of the corresponding random experiment , that is , @xmath0 should be @xmath62-typical for @xmath6 for some small constant @xmath62 . how small can the total information of such a theory be ? the next lemma shows that the answer is `` not too small '' . [ lemmintotalinfo ] ' '' '' it uniformly holds for @xmath5 and @xmath64 that @xmath65 * remark . * the upper bound @xmath66 and the computability of @xmath6 show that the set is finite , and the infimum is indeed a minimum . fix some @xmath64 and some @xmath67 . clearly , @xmath0 is @xmath62-typical for the singlet distribution @xmath68 , given by @xmath69 and @xmath70 for every @xmath71 . this ensemble has entropy @xmath72 . thus , the total information @xmath73 equals the complexity @xmath74 . we also have @xmath75 as describing the ensemble @xmath68 boils down to describing the string @xmath0 . furthermore , the corresponding additive constant does not depend on @xmath0 or @xmath62 . it follows that @xmath76 . to prove the converse , suppose @xmath6 is any ensemble such that @xmath0 is @xmath62-typical for @xmath6 . then we have the chain of inequalities @xmath77 the first two inequalities follow from general properties of prefix kolmogorov complexity , while the third inequality is due to the upper bound @xmath78 which follows from coding every string @xmath0 with @xmath79 into a prefix code word of length @xmath80 ( such a code exists due to the kraft inequality ) . moreover , ( [ ineq : typ_ensemble_estimate ] ) is a consequence of @xmath62-typicality of @xmath13 for @xmath6 , and ( [ ineq : eff_compl_def ] ) uses the definition of the total information @xmath81 . ' '' '' @xmath82 the ultimate goal of effective complexity is to assign a useful complexity measure @xmath83 to strings @xmath0 . in our analogy , this means that the scientist wants to assign a natural number to his data @xmath0 saying how `` complex '' @xmath0 is . simply taking the kolmogorov complexity @xmath66 as this value has important drawbacks : it does not at all capture the intuition that `` complexity '' should measure the `` amount of structure '' of an object . in fact , if @xmath0 is uniformly random ( i.e. the result of fair coin tossing ) , then @xmath66 is large , while the string possesses almost no structure at all . the strategy of s. lloyd and m. gell - mann @xcite is instead to take that complexity @xmath35 of `` the best '' theory @xmath6 that explains the data @xmath0 . what is `` the best '' theory ? as already discussed , a good theory should have small total information @xmath45 , and the data @xmath0 should be `` typical '' for @xmath6 in the sense that the probability @xmath51 is not much smaller than @xmath60 . given some data @xmath0 , there are always many `` good theories '' which satisfy these requirements . which one is `` the best '' ? to think about this question , it is helpful to look at a graphical representation of `` good theories '' and their properties as described in @xcite and depicted in fig . [ figmindomain ] . ( -5,-2)(22,22 ) ( 0,0)(20,0 ) ( 0,0)(0,20 ) ( 22,0)@xmath40 ( 0,21.3)@xmath35 ( 15,0)(0,15 ) ( 18,0)(0,18 ) ( 0,17)0.8 ( -2,15)@xmath66 ( 15,-1.5)@xmath66 ( -1.5,17)@xmath84 ( 10,10)0.5 ( 8,9)0.5 ( 8,8)0.5 ( 15,5)0.5 ( 17,8)0.5 ( 3,13)0.5 ( 7,12)0.5 ( 14,8)0.5 ( 5,12)0.5 ( 14,13)0.5 ( 12,10)0.5 ( 16,11)0.5 ( 10,14)0.5 ( 18,15)0.5 ( 6,13)0.5 ( 8,14)0.5 ( 0,8)(8,8 ) ( -2,8 ) suppose we plot the set of theories in the entropy - complexity plane . that is , for every computable ensemble @xmath6 ( with finite and computable entropy ) , we plot a black dot at the plane , where the @xmath0-axis labels the entropy @xmath40 and the @xmath18-axis labels the kolmogorov complexity @xmath35 . the kolmogorov complexity @xmath35 is integer - valued , and if @xmath36 is small , there are only few ensembles @xmath6 with @xmath85 ( in fact , the number of such ensembles is upper - bounded by @xmath86 ) . thus , there are only few black dots at small values of the @xmath18-axis . going up the @xmath18-axis , the number of ensembles and hence the density of the black dots increases . the total information @xmath45 is the sum of the ensemble s entropy and complexity . thus , ensembles with constant total information correspond to lines in the plane that are parallel to the tilted line in fig . [ figmindomain ] . suppose we fix some data @xmath0 and plot only those ensembles @xmath6 such that @xmath0 is @xmath62-typical for @xmath6 for some fixed constant @xmath64 . this was one of our two requirements that a `` good theory '' @xmath6 for @xmath0 should fulfill . that is , we dismiss all the ensembles for which @xmath0 is not a typical realization . then lemma [ lemmintotalinfo ] tells us that all the remaining ensembles must , up to an additive constant , have total information larger than @xmath87 . graphically , this means that all those ensembles must approximately lie right of the straight line with @xmath88 . one of these ensembles is the dirac measure @xmath84 , the ensemble with @xmath89 and @xmath90 for @xmath91 : it has kolmogorov complexity @xmath92 and entropy @xmath93 , hence minimal total information . it corresponds to the circle on the @xmath18-axis at the left end of the line . we also have discussed a second requirement for a `` good theory '' : the total information should be as small as possible . according to lemma [ lemmintotalinfo ] , this means that @xmath45 should not be much larger than the kolmogorov complexity @xmath66 . we identify the `` good '' theories as those ensembles that are not too far away from the line in fig . [ figmindomain ] ; say , we consider those ensembles as `` good '' that are below the dotted line with @xmath94 . among the remaining good theories , which one is `` the best '' ? the convincing suggestion by m. gell - mann and s. lloyd is that the best theory is the simplest theory ; that is , the ensemble @xmath6 with the minimal kolmogorov complexity @xmath35 . the complexity @xmath35 of this minimizing ensemble is then defined as the effective complexity of @xmath0 . in other words , the effective complexity of @xmath0 is defined as the smallest possible kolmogorov complexity of any `` good theory '' ( satisfying the two requirements ) for @xmath0 . this suggests the following preliminary definition ( we discuss an important modification below ) : [ defeffectivecomplexity ] [ def : effcompli ] ' '' '' given parameters @xmath95 , the _ effective complexity _ @xmath96 of any string @xmath5 is defined as @xmath97 or as @xmath98 if this set is empty . we refer to the set on the right - hand side as the * minimization domain * of @xmath0 for effective complexity , and denote it by @xmath99 . thus @xmath100 note that ensembles @xmath6 of the minimization domain @xmath99 of @xmath101 satisfy @xmath102 this notion of effective complexity is closely related to a quantity called `` kolmogorov minimal sufficient statistics '' . we explain this fact in more detail in definition [ defminsuffstat ] and lemma [ lemsuffstat ] in section [ sec : ec - and - kmss ] below . as pointed out by m. gell - mann and s. lloyd , it is often useful to extend this definition of effective complexity by imposing additional conditions ( `` constraints '' ) on the ensembles that are allowed in the minimization domain . there are basically two intuitive reasons why this is useful . to understand those reasons , we go back to the scenario of a scientist looking for good theories to explain his data @xmath0 . recall the interpretation of the minimization domain of effective complexity as the set of `` good theories '' for @xmath0 . reasons for considering constraints on the ensembles are : * given the string @xmath0 , there might be _ certain properties _ of @xmath0 that the scientist _ judges to be important_. those properties should be explained by the theories in the sense that the properties are not just simple random coincidences , but necessary or highly probable properties of each outcome of the corresponding process . + for example , suppose that a scientist wants to find good theories that explain the present state of our universe . in addition , that scientist finds it particularly important and interesting that the value of the fine structure constant is about @xmath103 and would like to find theories that explain why this constant is close to that value . then , he will only accept ensembles @xmath6 that have expectation value of this constant not too far away from @xmath103 . + in terms of effective complexity , this scientist views the appearance of a fine structure constant of about @xmath103 as an important structural ( non - random ) property of @xmath0 , the encoded state of our universe . thus , he considers this property as part of the regularities of @xmath0 . effective complexity is the kolmogorov complexity of the regularities of @xmath0 , thus , this scientist tends to find a larger value of effective complexity than other scientists who consider the fine structure constant as unimportant and random . * the scientist might have _ additional information _ on the process that actually created @xmath0 . this situation is often encountered in thermodynamics . suppose that @xmath0 encodes some microscopic properties of a gas in a container that a scientist has measured . in addition to these measurement results , the scientist typically also has information on several _ macroscopic observables _ like the temperature or the total energy in the box at least , crude upper bounds are usually given by basic properties of the laboratory physics . then , a `` good theory '' consistent with the actual physical process within the lab must obey the additional constraints given by the macroscopic observables . + in terms of effective complexity , the macroscopic observables respectively the additional information contributes to the regularities of @xmath0 and enlarges its effective complexity . [ def : effcompl ] ' '' '' given parameters @xmath104 , the _ effective complexity _ @xmath105 of a string @xmath0 constrained to a subset @xmath106 of all ensembles is @xmath107 or as @xmath98 if this set is empty . the set on the right - hand side is called the _ constrained minimization domain _ of @xmath0 for effective complexity and equals @xmath108 according to the notation of definition [ def : effcompli ] . thus , @xmath109 . note that we allow that @xmath106 depends on @xmath0 . this makes it possible , for example , to introduce the constraint that @xmath0 is an element of an ensemble of strings _ that all have the same length _ : just take @xmath106 as the set of probability distributions on @xmath110 , where @xmath111 . in general restricting the set of ensembles will increase the value of effective complexity , i.e. @xmath112 this is in agreement with our intuition . indeed such restrictions give a way to demand explicitly that some regularities of the considered string @xmath0 appear as a consequence of regularities of the generating process . as such , they contribute to the effective complexity . if the constrained set @xmath106 or the constant @xmath113 are too small such that the ( constrained ) minimization domain @xmath114 is empty , then effective complexity is infinite , according to definition [ def : effcompl ] . furthermore note that * @xmath105 is decreasing in @xmath62 and @xmath113 , * if @xmath105 is finite and @xmath115 , then @xmath116 in many situations in physics , the constrained sets @xmath106 appearing in the definition of @xmath105 consist of those ensembles that have expectation values of observables within certain intervals . that is , we have real - valued functions @xmath117 , the observables , and the set of ensembles @xmath106 consists of those ensembles @xmath6 with @xmath118 where the sets @xmath119 are intervals or possibly fixed real numbers . sometimes it even makes sense to allow different intervals @xmath120 for different ensembles @xmath6 ; say , the intervals may all be centered around the same fixed expectation value , but may have a width which grows with the standard deviation of @xmath6 with respect to the observable @xmath117 . this is explored in more detail in the following example . [ ex : constraints ] ' '' '' fix some string @xmath5 . let @xmath121 be an index set and @xmath122 a family of real - valued constraint functions on @xmath2 ( the observables ) . we would like to define a constrained set of ensembles @xmath123 with the following property : @xmath123 shall contain all those ensembles which have expectation values of the observables @xmath124 that are `` not too far away from '' the actual values @xmath125 of the observables evaluated on the string @xmath0 . this is done in the following way : to each observable @xmath117 and ensemble @xmath6 , associate a corresponding interval @xmath126 . the choice of those intervals is somewhat arbitrary we only demand that they contain the expectation values of the corresponding observables , i.e. @xmath127 then define the constrained set of ensembles @xmath123 by @xmath128 that is , @xmath123 consists of those ensembles @xmath6 such that the corresponding interval ( centered around the corresponding expectation value ) contains the `` correct '' value of the observable evaluated at @xmath0 . the corresponding effective complexity value @xmath129 has a natural interpretation as the effective complexity of @xmath0 if the observable properties @xmath117 of @xmath0 are judged to be important ( or are fixed as macroscopic observables ) . compare the discussion before definition [ def : effcompl ] . to illustrate the notation introduced in the previous example , we look at the situation when we would like to define the effective complexity of strings under the constraint of fixed length . that is , suppose that we consider the length @xmath4 of our string @xmath0 as an important regularity or that we have additional side information that the unknown random process generates strings of fixed length only . in this case , it makes sense to look at the effective complexity @xmath130 , where @xmath131 [ ex : fixed - length ] ' '' '' consider the effective complexity notion @xmath130 as explained above . instead of using equation ( [ eqcx ] ) , we can also use the notation of example [ ex : constraints ] : we have only one constraint , so the index set @xmath121 satisfies @xmath132 , for example @xmath133 . our observable @xmath134 is then the characteristic function on the strings of length @xmath4 , that is , @xmath135 to every ensemble @xmath6 , we associate an interval @xmath136 which only consists of the single real number that equals the corresponding expectation value of @xmath134 , i.e. @xmath137 it is then easy to see that the set @xmath123 defined in example [ ex : constraints ] above equals the set in equation ( [ eqcx ] ) . due to linearity , the constrained sets @xmath123 in example [ ex : constraints ] are always convex . this property , together with a computability condition , will be useful in the following . a set @xmath106 of ensembles on the binary strings is called * _ convex _ , if for every finite set of ensembles @xmath138 , every computable convex combination @xmath139 with @xmath140 and @xmath141 , is also in @xmath106 , * _ decidable _ , if there exists some algorithm that , given some string @xmath5 as input , decides in finite time whether the dirac measure on @xmath0 is an element of @xmath106 or not , that is , whether the measure @xmath142 satisfies @xmath143 or not . in this sense , we may define @xmath144 as the length of the shortest computer program that computes the corresponding decision function . we proceed by analyzing some basic properties of effective complexity . we have remarked that the effective complexity @xmath105 can be infinite , for example , if the constant @xmath113 is too small or the constrained set of ensembles @xmath106 is too restrictive such that the minimization domain satisfies @xmath145 . thus , we start by proving a simple sufficient condition that guarantees that effective complexity is finite . [ lem : finiteness ] there is a constant @xmath146 such that @xmath147 for all strings @xmath0 with dirac measure @xmath143 , @xmath64 and @xmath148 . due to ( [ eqmx ] ) , we only have to prove that @xmath105 is finite . according to definition [ def : effcompl ] , it remains to prove that @xmath149 under the conditions given above , where @xmath99 is the minimization domain of @xmath0 . to this end , we show that @xmath150 . this follows from * @xmath151 for every @xmath64 , so @xmath0 is @xmath62-typical for @xmath84 , * @xmath152 , where @xmath146 is a constant . + ' '' '' + @xmath82 our first result resembles the example on p. 51 in @xcite . suppose that we have a random binary string @xmath0 of length @xmath54 , maybe a string which has been determined by tossing a fair coin @xmath54 times . the kolmogorov complexity of such a string typically satisfies @xmath153 , that is , structureless random strings have maximal kolmogorov complexity . this was one of reasons for s. lloyd s and m. gell - mann s criticism of kolmogorov complexity and for their attempt to define effective complexity as a more useful and intuitive complexity measure . the following theorem proves that random strings indeed have small effective complexity , which supports the point of view that effective complexity measures only the complexity of the non - random structure of a string . before we state that theorem , we have to explain in detail what we mean by a `` random '' string . it is well - known that most strings are incompressible , which is what we mean by `` random '' at this point . in more detail , if @xmath154 is some fixed parameter , the number of strings @xmath0 of length @xmath54 with prefix complexity @xmath155 does not exceed @xmath156 ( cf . 3.3.1 ) ) . that is , most strings are incompressible in the sense that @xmath157 . we call such strings _ @xmath158-incompressible_. [ thesimple ] there exists some global constant @xmath159 such that @xmath160 for all @xmath158-incompressible strings @xmath0 of length @xmath54 , @xmath64 and @xmath161 . moreover , if @xmath106 is a convex and decidable constrained set of ensembles , then for all @xmath158-incompressible strings @xmath0 of length @xmath54 with the property that the dirac measure @xmath143 , we have @xmath162 whenever @xmath64 and @xmath163 . note that the @xmath164-term does not depend on @xmath106 . with a suitable choice of @xmath165 , the first part of the theorem is a special case of the second part ( with @xmath166the set of all ensembles ) , so it is sufficient to prove the second part . suppose that @xmath106 and @xmath0 satisfy the conditions of the theorem . let @xmath167 be the uniform distribution on @xmath168 then @xmath169 , and @xmath170 , so @xmath0 is @xmath62-typical for @xmath167 . moreover , @xmath171 thus , @xmath172 . the strings @xmath0 which are @xmath158-incompressible satisfy by definition @xmath173 . denoting by @xmath174 the corresponding degree of incompressibility of @xmath0 gives @xmath175 i.e. there is a global constant @xmath25 such that @xmath176 . now if @xmath177 , it follows from definition [ defeffectivecomplexity ] and @xmath178 that @xmath179 note that according to the theorem above , every string @xmath101 becomes effectively simple if @xmath113 is large enough . indeed , for every @xmath61 , relation ( [ rel : thesimple ] ) is satisfied by @xmath0 if @xmath113 is larger than the @xmath0-dependent threshold @xmath180 . ( here @xmath165 is the global constant appearing in theorem [ thesimple ] . ) for strings @xmath0 of fixed length @xmath54 , one can give an @xmath54-dependent threshold @xmath181 such that @xmath182 if @xmath183 . on the other hand , due to lemma [ lem : finiteness ] , to ensure @xmath184 for @xmath101 , one should choose @xmath113 not too small , namely @xmath185 , where @xmath186 was a global constant depending on the reference universal computer only . these simple observations show that a discussion of the dependence of effective complexity for arbitrary but fixed strings @xmath67 on the parameter @xmath113 should be useful for a deeper understanding of the concept . in a forthcoming paper @xcite , we investigate in more detail the behavior of effective complexity of long strings generated by stochastic processes for different choices of @xmath187 . for the rest of this paper , we assume @xmath113 to be a fixed constant ( not depending on @xmath54 or @xmath0 ) that is larger than the aforementioned constant @xmath186 . in what follows we strengthen the result in theorem [ thesimple ] in a way that will be interesting later in section [ secdepth ] : [ corollary7 ] there exists some global constant @xmath159 such that uniformly @xmath188 for all @xmath158-incompressible strings @xmath0 , @xmath64 and @xmath161 . if @xmath106 is a convex and decidable set of ensembles , then for all @xmath0 that additionally satisfy @xmath143 we have @xmath189 whenever @xmath163 . it follows from the proof of theorem [ thesimple ] that @xmath190 according to the definition of @xmath158-incompressibility , we also have @xmath191 , thus @xmath192 . moreover , it holds @xcite @xmath193 and @xmath194 . ' '' '' @xmath82 the fact that incompressible strings have small effective complexity and most strings are incompressible raises the question if there exists any string with _ large _ effective complexity at all . fortunately , the answer is `` yes '' ; otherwise , the notion of effective complexity would be an empty concept . on the one hand , we might drop a requirement posed in theorem [ thesimple ] . there , we restricted to constrained sets @xmath106 of ensembles that contain the dirac measure @xmath84 , which basically means that the string @xmath0 should itself fulfill the constraints that are used to define @xmath106 . the effective complexity of strings @xmath0 that do not meet this requirement might possibly be large . on the other hand , even among strings that fulfill this requirement , there are still strings that are effectively complex , namely those strings that are called `` non - stochastic '' in the context of algorithmic and kolmogorov minimal sufficient statistics @xcite . suppose we have a finite subset @xmath195 of the finite binary strings . there are elements @xmath196 that are easy to specify , once the set @xmath197 is given , in the sense that @xmath198 is small . for example , the smallest element of @xmath197 in lexicographical order has very small conditional complexity @xmath198 . we call such elements _ atypical_. on the other hand , most of the elements of @xmath197 will not be special in such a way , such that we can specify them only by giving their `` index '' within the set @xmath197 , which takes about @xmath199 bits . thus , most elements @xmath196 will have @xmath200 we can call such elements _ typical _ for @xmath197 . there is a lemma by gcs , tromp , and vitnyi @xcite , stating that there exist strings which are atypical for _ every _ simple set @xmath197 . they are called _ non - stochastic _ : [ lemgacs ] there are constants @xmath201 such that the following holds true : suppose @xmath36 is fixed . for every @xmath202 , there is some string @xmath115 with @xmath203 , such that @xmath204 for every @xmath205 with @xmath206 . we want to use this result to prove that for every @xmath54 , there are binary strings of length @xmath54 that have effective complexity of about @xmath54 . therefore , we have to show that basically all we do with _ ensembles of strings _ can as well be accomplished with _ equidistributions on sets _ : [ lemensemblessets ] ' '' '' let @xmath5 and @xmath207 be arbitrary , and let @xmath106 be a set of decidable and convex constraints such that @xmath208 . moreover , let @xmath209 be an arbitrary set of constraints such that the effective complexity @xmath210 is finite , and let @xmath6 be the corresponding minimizing ensemble . then , for every @xmath211 , there is a set @xmath195 containing @xmath0 with @xmath212 such that the equidistribution on @xmath197 is in @xmath106 , where @xmath159 is a global constant . * remark . * the most interesting case is @xmath213 , showing that the minimizing ensembles in the definition of effective complexity can `` almost '' ( up to the additive terms above ) be chosen to be equidistributions on sets even in the presence of decidable and convex constraints . the minimizing ensemble @xmath6 in the definition of @xmath210 has the following properties : @xmath214 we would like to write a computer program that , given a description of @xmath6 , computes a list of all strings @xmath215 that satisfy the constraints @xmath106 and the inequality @xmath216 . such a program could search through all strings @xmath18 , decide for every @xmath18 whether this equation and the constraints hold for @xmath18 , and do this until the sum of the probabilities of all the previously listed elements exceeds @xmath217 . but there is a problem of numerics : it is in general impossible for the program to decide with certainty if this inequality holds , because of the unavoidable numerical error . instead , we can construct a computer program that computes a set @xmath195 with the following weaker properties : @xmath218 that is , the program computes a set @xmath197 which definitely contains all strings @xmath18 with @xmath216 that satisfy the constraints , but it may also contain strings which slightly violate this inequality as long as they still satisfy the constraints . however , the numerical methods are chosen good enough such that we are guaranteed that every element of @xmath197 has probability of at least @xmath219 . this set @xmath197 contains @xmath0 and has the desired properties ( [ eqensemblessets1 ] ) and ( [ eqensemblessets2 ] ) . this can be seen as follows . by definition , @xmath220 , so @xmath196 . since every element @xmath221 has probability @xmath222 , it holds @xmath223 . finally , the description length of the corresponding computer program can be estimated via @xmath224 . ' '' '' @xmath82 now we are ready to prove the existence of effectively complex strings . first , what should we expect from `` effectively complex '' strings how large could effective complexity @xmath96 of some string @xmath0 of length @xmath54 possibly be ? if @xmath6 is the minimizing ensemble in the definition of @xmath96 , then @xmath225 thus , the best result we can hope for is the existence of strings of length @xmath54 that have effective complexity close to @xmath54 . the next theorem shows exactly this . [ thecomplex ] ' '' '' for every @xmath104 and @xmath36 , there is a string @xmath0 of length @xmath54 such that @xmath226 as effective complexity is increased if constraints are added , the same statement is true for @xmath105 if @xmath106 is an arbitrary constrained set of ensembles . * remark . * an explicit lower bound is @xmath227 where @xmath228 is a global constant . fix @xmath229 , @xmath230 , and @xmath115 . let @xmath68 be the minimizing ensemble associated to @xmath96 , i.e. @xmath231 choose @xmath59 arbitrary . according to lemma [ lemensemblessets ] , there is a set @xmath232 such that @xmath233 and @xmath234 where @xmath159 the sum of a global constant and @xmath235 and @xmath236 . let @xmath237 be the best constant for our universal computer @xmath11 such that @xmath238 and at the same time @xmath239 using ( [ eqm3 ] ) , ( [ eqm4 ] ) and ( [ eqm5 ] ) , we conclude that @xmath0 is almost typical for @xmath240 : @xmath241 it also follows @xmath242 now we get rid of the term @xmath243 . first note that ( [ eqm5 ] ) , ( [ eqm1 ] ) and ( [ eqmx ] ) yield @xmath244 where @xmath245 is some constant such that @xmath246 for every @xmath27 , and @xmath247 for every @xmath248 . by elementary analysis , it holds @xmath249 if @xmath250 . hence there is some constant @xmath251 which does not depend on @xmath54 , @xmath62 , @xmath113 , or @xmath0 , such that for @xmath252 , @xmath253 using the same argument with @xmath254 , for get for all @xmath252 @xmath255 going back to ( [ eqm9 ] ) , it follows @xmath256 where @xmath257 note that @xmath0 was arbitrary , so this equation is valid for every @xmath115 . let now @xmath258 , and @xmath259 where @xmath260 is the constant from lemma [ lemgacs ] . if @xmath54 is large enough , then @xmath261 holds , and lemma [ lemgacs ] applies : there is a string @xmath262 such that @xmath263 for every set @xmath264 with @xmath206 , where @xmath265 is another global constant . first note the following inequality : @xmath266 now suppose that @xmath267 . consequently , @xmath268 but this is a contradiction to ( [ eqcontrak ] ) . hence our assumption must be false , and we must instead have @xmath269 . thus , using ( [ eqm1 ] ) , ( [ eqm5 ] ) , and @xmath270 , @xmath271 applying relation ( [ eqmx ] ) to the case of effectively complex strings @xmath272 constructed here , we obtain a lower bound on @xmath273 : @xmath274 where @xmath275 . for large @xmath54 , where the constant @xmath276 becomes negligible , this bound is non - trivial and in particular for @xmath277 remarkably close to the maximal value @xmath278 . on the other hand , from the previous proof and lemma [ lemgacs ] we deduce the following upper bound on the complexity of @xmath272 : @xmath279 where @xmath280 is a global constant . this implies the following relation between the degree of incompressibility @xmath281 and the constant @xmath113 : @xmath282 where the second inequality holds for sufficiently large @xmath54 and the last one uses the definition ( [ def : lambda ] ) of @xmath283 . indeed , this relation prevents @xmath272 from falling into the domain of theorem [ thesimple ] , which would force it to have small effective complexity . note that all effectively complex strings must , as long as effective complexity is finite , have large kolmogorov complexity , too . this follows from ( [ eqmx ] ) . in this section , we show that effectively complex strings have very large computation times . in more detail , it takes a universal computer an astronomically large amount of time to compute such a string from its minimal program , or from an almost minimal program . the time it takes to compute a string from its minimal program is discussed by c. bennett @xcite in the context of `` logical depth '' . the notion of logical depth formalizes the idea that some strings are more difficult to construct than others ( say , a string describing a proof of the riemann hypothesis is harder to construct than a uniformly random string ) . however , the computation time of a string s minimal program is not a very stable measure for this difficulty : there might be programs that are `` almost minimal '' , i.e. only a few bits longer , but run much faster than the minimal program . thus , logical depth is defined as the shortest time to compute a string from its almost - minimal program . `` almost - minimal '' means that the program is only a few bits longer than the minimal program , and the maximum length overhead is called the `` significance level '' . [ defdepth ] ' '' '' let @xmath5 be a string and @xmath284 a parameter . a string s _ logical depth _ at significance level @xmath285 , denoted @xmath286 , will be defined as @xmath287 where @xmath288 denotes the halting time of the universal computer @xmath9 on input @xmath52 . note that we have used plain kolmogorov complexity @xmath289 here , that is , the universal computer @xmath9 is not assumed to be prefix - free in contrast to the original definition @xcite . as plain and prefix kolmogorov complexity are closely related , this modification will not result in large quantitative changes . however , for us it has important technical advantages as we will see below . logical depth is sometimes defined in a different manner with reference to algorithmic probability , cf . however , if computation times are large ( which will be the case in our context , cf . the remark after theorem [ theeffdepth ] ) , then those different definitions are essentially equivalent ( * ? ? ? * claim 7.7.1 ) . clearly , it takes a computer ( i.e. a turing machine ) at least @xmath4 steps to print a string @xmath0 on its tape . thus , the depth of a string must be lower - bounded by its length : @xmath290 for every @xmath285 . following @xcite , strings that have a depth almost as small as possible , i.e. @xmath291 , will be called _ shallow_. the notion of depth depends on the choice of the universal reference computer but not too much . as explained in @xcite , there is a universal @xmath7-tape turing machine that simulates @xmath15 steps of an arbitrary @xmath292-tape turing machine in time @xmath293 , where @xmath165 is a constant that only depends on the machine . in particular , this turing machine can implement obvious tasks like copying of @xmath54 bits in time @xmath54 from one tape to the other . therefore , we will fix this `` hennie - sterns machine '' ( cf . @xcite ) as our universal reference machine for this section . to state the first example , a string @xmath0 will be called _ @xmath186-random _ if @xmath294 . [ exshallow ] random strings are shallow . that is , there are constants @xmath295 such that for every @xmath186-random string @xmath0 it holds @xmath296 as @xmath297 is decreasing in @xmath285 , it also follows that @xmath298 for all @xmath299 . there is always a computer program @xmath52 of length @xmath300 that sequentially lists the bits of @xmath0 , producing @xmath301 in the most trivial way . such a program will have a running time of @xmath4 , plus potentially some overhead @xmath302 resulting from initialization . if @xmath303 , then @xmath304 , and the claim follows from the definition of depth . ' '' '' @xmath82 it will be interesting in the following that this conclusion carries over to `` most '' strings that are @xmath158-incompressible as defined in theorem [ thesimple ] . this will be proved in the next example . moreover , we give a technical result which will be useful below . note that @xmath305 and we will be interested in strings for which some kind of converse holds . for this purpose , we say that a string @xmath0 is _ @xmath292-well - behaved _ for some @xmath248 if @xmath306 in fact , it is stated in @xcite that most strings are @xmath292-well - behaved if @xmath292 is large enough . the next example shows that incompressible random strings are well - behaved . [ lemincompshallow ] for every @xmath54 , there are at least @xmath307 strings of length @xmath54 that are @xmath158-incompressible and @xmath186-random , where @xmath159 is a constant . they satisfy @xmath308 where @xmath295 are constants . moreover , those strings are @xmath292-well - behaved , where @xmath248 is a constant that only depends on @xmath158 and @xmath186 , but not on @xmath0 . recall two basic incompressibility facts that are listed in @xcite : * there is a constant @xmath165 such that there are at least @xmath309 strings of length @xmath54 which are @xmath158-incompressible . * for every @xmath146 , there are at least @xmath310 strings of length @xmath54 with @xmath311 ( we call such strings `` @xmath186-random '' ) . a simple application of the union bound gives that there are at least @xmath307 strings of length @xmath54 that are at the same time @xmath158-incompressible and @xmath186-random . the upper bound on the logical depth follows then from example [ exshallow ] . if @xmath0 is @xmath186-random , then @xmath312 , and a converse inequality holds trivially . the well - known continuity property @xcite of @xmath313 then guarantees the existence of a constant @xmath314 ( that depends only on @xmath186 but not on @xmath0 ) such that @xmath315 . moreover , if @xmath316 is a constant such that @xmath317 for all strings @xmath13 , the @xmath158-incompressibility property yields @xmath318 it follows that @xmath0 is @xmath319-well - behaved . ' '' '' @xmath82 we will now show that effectively complex strings must have very large logical depth if they are well - behaved . this is in contrast to the fact that incompressible strings ( which have small effective complexity according to theorem [ thesimple ] ) are always shallow , that is , have very small depth as shown in lemma [ lemincompshallow ] . the main idea is as follows : suppose that an almost - minimal program for the string @xmath0 has a rather short halting time @xmath320 . then we can consider the ensemble @xmath6 , that is defined by equidistribution over all strings that have a short program ( of length less or equal to @xmath8 ) with some halting time less than or equal to @xmath320 . such an ensemble is simple , @xmath0 is typical for it , and it has small total information . thus , it forces the effective complexity of @xmath0 to be small . [ theeffdepth ] there is a global constant @xmath228 with the following property : suppose that @xmath321 is a strictly increasing , computable function . furthermore , suppose that @xmath0 is a @xmath292-well - behaved string . if the effective complexity of @xmath0 satisfies @xmath322 for some arbitrary @xmath64 and @xmath323 , then @xmath324 * remark . * before we prove this theorem , we explain its meaning and implications . first , since @xmath194 , it follows that @xmath325 is of the order @xmath326 or less , which is quite small . thus , the inequality for @xmath327 in this theorem is a very weak condition . the function @xmath328 can be chosen to be simple ( such that @xmath329 is small ) , but extremely rapidly growing , for example of the form @xmath330 thus , the consequence of this theorem is that the depth must be astronomically large . as @xmath331 is decreasing in @xmath62 , it is sufficient to prove the theorem for the case @xmath332 . for @xmath333 and every computable function @xmath321 , define the set @xmath334 where @xmath288 denotes the halting time of the ( plain , not prefix ) universal computer @xmath9 on input @xmath52 . clearly @xmath335 , and if @xmath336 is the uniform distribution on @xmath337 , then @xmath338 . moreover , there is some constant @xmath159 such that @xmath339 . hence the total information satisfies @xmath340 now let @xmath5 be a @xmath292-well behaved string , and let @xmath323 be arbitrary . suppose that @xmath341 , then @xmath0 is @xmath342-typical for @xmath343 . since there exists some global constant @xmath344 such that @xmath345 we can estimate , using ( [ eqtotaldep ] ) , @xmath346 by definition of effective complexity , we get @xmath347 in summary , we have so far shown the following : if there is a program @xmath348 with @xmath349 and @xmath350 such that the corresponding halting time satisfies @xmath351 , then the effective complexity is as small as in ( [ eqecdepth ] ) . the negation of this statement , together with @xmath352 and @xmath353 , proves the theorem . ' '' '' @xmath82 interestingly , incompressible strings just slightly fail to fulfill the inequality of the previous theorem . according to corollary [ corollary7 ] , @xmath158-incompressible strings @xmath0 have effective complexity @xmath354 . thus , we can not conclude that those strings have large depth fortunately , because most @xmath158-incompressible strings are in fact _ shallow _ according to example [ lemincompshallow ] . thus , it follows that the expression @xmath325 sharply marks the `` edge of depth '' , in the sense that strings with larger effective complexity always have extremely large depth , but strings with smaller effective complexity can have arbitrarily small depth . in some sense , a phenomenon similar to a `` phase transition '' occurs at effective complexity of @xmath325 ( apart from additive constants that get less and less important in the `` thermodynamic limit '' @xmath355 ) . this behavior is schematically depicted in figure [ figphasetransition ] . the previous theorem says that if the effective complexity exceeds @xmath325 ( omitting all additive constants here ) , then the logical depth must be astronomically large . on the other hand , if effective complexity is smaller , different values of depth seem possible . in particular , if @xmath0 is @xmath158-incompressible , then we know from corollary [ corollary7 ] that effective complexity is ( possibly only up to a few bits ) smaller than @xmath325 , while the logical depth is as small as possible ( of the order @xmath4 ) due to lemma [ lemincompshallow ] . ( -5,-2)(22,22 ) ( 0,0)(20,0 ) ( 0,0)(0,20 ) ( 22,0)@xmath356 ( 0,21.3 ) ( 13,1)(13,-1 ) ( 15,-2)@xmath111 ( -1,0)(1,0 ) ( -5,0 ) ( 3,1)(3,-1 ) ( 3,-2)@xmath325 ( 3,19)(3,0 ) ( 13,19)(13,0 ) ( 3,17)(13,17 ) ( -1,14.5)(1,14.5 ) ( -4,16 ) ( -4,14 ) ( 2,0)0.8 ( 1.5,2.5 ) this theorem can easily be extended to the case of effective complexity with constraints as long as the constrained sets of ensembles satisfy the usual technical conditions : [ theeffdepth2 ] there is a global constant @xmath228 with the following property : suppose @xmath321 is a strictly increasing , computable function . furthermore , suppose that @xmath0 is a @xmath292-well behaved string , and the constrained set @xmath106 is decidable and convex and contains the dirac measure @xmath84 . if the effective complexity of @xmath0 satisfies @xmath357 for some @xmath358 with @xmath64 and @xmath323 , then @xmath324 * remark . * for an explanation and interpretation of this theorem , see the remarks after theorem [ theeffdepth ] above . the proof is almost identical to that of theorem [ theeffdepth ] above . the only modification is that the set @xmath337 has to be replaced by a set @xmath359 the convexity condition then ensures that the uniform distribution on @xmath360 , called @xmath361 , is contained in @xmath106 . this construction enlarges every additive constant by a term @xmath144 , i.e. the complexity of a computer program that is able to test for strings if the corresponding dirac measures are contained in @xmath106 . ' '' '' @xmath82 now we study the relation between effective complexity without constraints and kolmogorov minimal sufficient statistics ( kmss ) . for more information on kolmogorov minimal sufficient statistics and related notions , see @xcite . for strings @xmath0 and integers @xmath248 , we can define a ( version of ) the _ kolmogorov structure function _ @xmath362 by @xmath363 i.e. @xmath364 is the logarithm of the minimal size of any subset of strings of length @xmath54 which contains the string @xmath0 and has complexity upper bounded by @xmath292 , given @xmath54 . the corresponding minimal set will be called @xmath365 ( if there are several minimizers , we take the first set in some canonical order ) . hence @xmath366 and @xmath367 . [ defminsuffstat ] ' '' '' let @xmath0 be a string of length @xmath54 and denote by @xmath368 the minimal natural number @xmath292 satisfying @xmath369 a minimal program @xmath370 for @xmath371 is called _ kolmogorov minimal sufficient statistics _ of the string @xmath0 . originally , the kolmogorov structure function as well as kolmogorov minimal sufficient statistics were defined by using plain kolmogorov conditional complexity @xmath372 in ( [ def : entropy - of - suffstat ] ) and ( [ def : length - of - kolmminsuffstat * ] ) instead of chaitin s prefix version @xmath373 . it holds @xmath374 moreover , @xmath375 can equivalently be written as @xmath376 this formula looks very similar to the definition of ( unconstrained ) effective complexity , as given in definition [ defeffectivecomplexity ] . hence the kolmogorov minimal sufficient statistics is approximately the minimal program of the minimizing set within the minimization domain of effective complexity . we will explore this observation in more detail in the following lemma . [ lemsuffstat ] there is a constant @xmath159 such that for all @xmath104 it holds @xmath377 uniformly in @xmath5 , where @xmath378 . moreover , there is a constant @xmath379 such that for all @xmath104 @xmath380 uniformly in @xmath5 . let @xmath381 and @xmath378 . by definition , @xmath382 let @xmath383 be the uniform distribution on @xmath365 . it follows @xmath384 thus , there is some constant @xmath159 such that @xmath385 then ( [ ecminsuff1 ] ) follows from ( [ minsuffineq ] ) . in order to show ( [ eff - compl - with - c ] ) , we use lemma [ lemensemblessets ] . let @xmath6 be the minimizing ensemble in the definition of @xmath96 . in particular , it holds @xmath386 fix any @xmath59 . due to lemma [ lemensemblessets ] , there is a set @xmath387 containing @xmath0 such that @xmath388 let now @xmath389 . it still holds @xmath390 and @xmath391 . since @xmath392 , we get the chain of inequalities @xmath393 using ( [ eqdefkdelta ] ) and ( [ eqkec ] ) , it follows that @xmath394 then ( [ eff - compl - with - c ] ) follows again from ( [ minsuffineq ] ) ' '' '' @xmath82 we have given a formal definition of effective complexity and rigorous proofs of its basic properties . in particular , we have shown that there is an interesting relation between effective complexity and logical depth : the depth of a string @xmath0 is astronomically large if the effective complexity exceeds @xmath325 ; otherwise , it can be arbitrarily small . this statement is true up to a few technical conditions and up to certain additive constants . these constants become less and less important for longer and longer strings this is comparable to the `` thermodynamic limit '' in statistical mechanics , and the behavior can be compared to that of a phase transition . so how useful is effective complexity for the study of natural systems ? we do not yet know the answer to this question , but we hope that our mathematically rigorous approach gives the first steps towards possible applications within mathematics or theoretical computer science . at least , we have shown that effective complexity has interesting properties , for example there are strings that have effective complexity close to their lengths . those strings are rare events only `` most '' strings are in fact effectively simple ; this follows from theorem [ thesimple ] . but this property is unavoidable : recall that a major motivation for the definition of effective complexity was that random strings should be simple ( in contrast to kolmogorov complexity ) . now since almost all strings are random , it follows that almost all strings must be effectively simple . most of our results concerning the constrained version of effective complexity @xmath105 were derived under the assumption that the dirac measure @xmath84 is an element of @xmath106 . although this is a natural assumption , the behavior of effective complexity might as well be completely different if it is dropped . investigating such situations in more detail could be useful in order to get better insight into the concept of effective complexity and its limitations . finally , a possible field of application of effective complexity might be statistical mechanics , where the notion of entropy and algorithmic complexity have both already led to interesting conclusions . after all , the constraints given by the set @xmath106 can be interpreted as macroscopic observables as discussed in section [ seceffectivecomplexity ] , and we have already compared our result on logical depth with certain notions of statistical mechanics . as promised in the introduction , we give an explicit construction of a computable ensemble @xmath6 with the property that the entropy @xmath40 is finite but not computable : [ exnonentropy ] ' '' '' for every @xmath36 , let @xmath395 be the set of strings that start with exactly @xmath396 zeroes , such that the @xmath54-th bit either does not exists or is a one . that is , @xmath397 and so on . clearly , this is a computable partition of @xmath2 into countably - infinite , mutually disjoint subsets : @xmath398 , @xmath399 if @xmath400 . we now construct an ensemble @xmath6 such that every set @xmath395 has weight @xmath38 , that is @xmath401 . we distribute the weight @xmath38 among the members of @xmath395 in a way , such that the resulting ensemble @xmath6 is computable , but has non - computable entropy . this is done as follows : let @xmath402 be a real number which is not computable , but enumerable from below . that is , there exists a computable sequence @xmath403 with @xmath404 which is increasing , i.e. @xmath405 , and which converges to @xmath406 , i.e. @xmath407 . for example , we may use chaitin s omega number@xcite @xmath408 where the sum is over all strings @xmath5 such that the universal prefix computer @xmath11 halts on input @xmath0 . the number @xmath406 gives the probability that the computer @xmath11 halts on randomly chosen input . it is a real number between zero and one , and it is obviously enumerable from below , but it is not computable . given some weight @xmath409 $ ] and finitely many positive real numbers @xmath410 such that @xmath411 , the resulting entropy sum @xmath412 will always be larger than or equal to @xmath413 . the converse is also true : given some fixed entropy value @xmath414 , we can always find finitely many positive real numbers @xmath410 with @xmath411 such that @xmath415 . such a list of real numbers can be found in an obvious systematic way that can be implemented as a computer program . thus , to every @xmath36 , we may systematically distribute the weight @xmath38 to finitely many strings in @xmath416 such that the corresponding probabilities @xmath417 have entropy sum @xmath418 , i.e. @xmath419 the resulting ensemble is obviously computable , and the entropy is @xmath420 this is not a computable number . the authors would like to thank eric smith for helpful discussions . this work has been supported by the santa fe institute .

we have observed a small sample of hot post - agb stars with the infrared array camera ( irac ) and the infrared spectrograph ( irs ) on - board the spitzer space telescope . the stars were selected from the literature on the basis of their far - infrared excess ( i.e. , post - agb candidates ) and b spectral type ( i.e. , close to the ionization of the envelope ) . the combination of our irac observations with 2mass and iras catalog data , along with previous radio observations in the cm range ( where available ) allowed us to model the seds of our targets and find that in almost all of them at least two shells of dust at different temperatures must be present , the hot dust component ranging up to k. in several targets grains larger than 1 m are needed to match the far - ir data points . in particular , in iras 17423 - 1755 grains up to 100 m must be introduced to match the emission in the mm range . we obtained irs spectra to identify the chemistry of the envelopes and found that more than of the sources in our sample have mixed chemistry , showing both mid - ir bands attributed to polycyclic aromatic hydrocarbons ( pah ) and silicate features . the analysis of the pah features indicates that these molecules are located in the outflows , far away from the central stars . we consider the larger than expected percentage of mixed - chemistry targets as a selection bias towards stars with a disk or torus around them . our results strengthen the current picture of mixed chemistry being due to the spatial segregation of different dust populations in the envelopes .

0908.0638

planetary nebulae ( pne ) evolve from intermediate mass stars . during their evolution they go through the asymptotic giant branch ( agb ) and then into the pn phase , ending their days as white dwarfs . the formation and early evolution of pne are a poorly understood phase of stellar evolution . in particular , it is not clear yet how the spherically symmetric agb circumstellar shells transform themselves into the non spherical symmetries observed in the envelopes of evolved pne . during their agb phase stars lose most of their initial mass , their mass loss rates ranging from 10@xmath3 up to 10@xmath4 m@xmath5 yr@xmath6 . the star develops a wind that carries away from its photosphere the enriched material produced by nucleosynthesis and dredged up from the core to the surface . this ejected material eventually cools down and creates a circumstellar envelope of dust and gas , whose chemistry is determined by the c / o ratio in the stellar outer regions @xcite . when the star has evolved past the tip of the agb , the dust and gas circumstellar envelope is typically so thick that the star can not be detected at optical wavelengths : the expansion and subsequent dilution of the envelope leads to the optical detection of post - agb stars . the end of the post - agb evolution can be identified as the onset of the ionization in the envelope , when the central star reaches temperatures around 2030@xmath7 k. the ionization is in fact considered as the beginning of the pn phase . studies of the infrared ( ir ) properties of agb and post - agb stars have determined that in o - rich envelopes silicate features at 10 and 18 @xmath1 m are first seen in emission during the agb , then in absorption when the envelope becomes optically thick , and then again in emission , after the envelope becomes optically thin to the mid - ir radiation @xcite . a similar evolutionary scheme is observed in c - rich envelopes , although with different spectral features . low - mass o - rich stars are not expected to develop an optically thick envelope , therefore the silicate features are never seen in absorption in these stars @xcite . a small fraction of post - agb stars and pne show both o - rich and c - rich features and are therefore classified as _ mixed - chemistry _ objects . the origin of their mixed chemistry is still uncertain . for example , in the red rectangle this has been explained as due to the presence of a circumbinary disk that was formed when the star was still o - rich and then served as a silicate reservoir , while the c - rich features originate in the dredged - up outflow @xcite . a different scenario has been proposed for high mass stars , in which _ hot bottom burning _ would convert @xmath8c into @xmath9n and change a c - rich envelope back into an o - rich one @xcite . many observational programs have been carried out to find new planetary and proto - planetary nebulae among unidentified iras sources with far infrared colors similar to those of known pne @xcite , with the final aim of understanding pn formation through the discovery and analysis of new transition objects ( to ) . in these sources the physical processes associated with pn formation , such as dynamical shaping , are still occurring and , as the number of known to is extremely small , the identification and study of new samples is relevant for further testing current models of stellar evolution . to detect new to , we have selected a sample of stars classified in the literature as post - agb on the basis of their iras colors and optical spectra . we have selected our stars from lists of post - agb candidates available in the literature @xcite , with the constraint of a b - type spectral classification and far - ir excess . the b spectral type implies the central star can be hot enough to have started to ionize the circumstellar envelope . this maximizes the chance that the objects are right in the transition from post - agb to pn , although misclassifications can not be ruled out . radio observations of this sample were carried out to check for the presence of free - free emission from ionized shells . these observations have been reported in @xcite and @xcite . to further characterize our sample , we performed observations with the spitzer space telescope . mid - ir observations probe the extended envelopes that surround post - agb stars and young pne , which are in most cases only partly ionized . the main goals of such observations are to assess the morphology of the neutral gas and dust components and to characterize their mineralogy . comparing the results of these observations to the properties of the ionized gas , as derived from previous radio observations , could in principle provide important clues on the shaping of the nebulae . as a follow - up to our radio detections , we observed 26 objects with the spitzer space telescope ( spitzer ) . infrared imaging was performed with the infrared array camera ( irac ) @xcite at 3.6 , 4.5 , 5.8 , and 8.0 @xmath1 m to search for faint extended emission . the observations were carried out in several spitzer campaigns in 2006 and 2007 ( program i d : 30036 ) . the irac observations were performed in high dynamic range mode . for each target we obtained 8 dithered frames , for a total on source time of about 96 s per irac channel . basic calibrated data ( bcd ) were retrieved from the spitzer archive ( pipeline versions : s14.0.0 , s15.0.5 , and s16.1.0 ) , cleaned to correct such artifacts as mux - bleeding and banding and then coadded using iracproc @xcite . the photometry was performed by psf subtraction in the final mosaics . this technique has been specifically developed to perform photometry even in heavily saturated objects @xcite . the method derives vega magnitudes of saturated stars by fitting the low level features of the saturated psf ( diffraction spikes and psf wings ) with a model of the irac psf obtained from a sample of bright stars @xcite . the psf is directly normalized to the irac observation of vega , therefore , by fitting it to the observed sources , magnitudes can be calculated with an accuracy within 13% independently of the standard irac flux calibration @xcite . observations with the infrared spectrograph ( irs ; * ? ? ? * ) were performed within the same program in 2006 and 2007 . the observations were carried out in low spectral resolution mode ( r@xmath1060120 ) , covering the 538 @xmath1 m range , to assess the chemistry of the envelopes . each target was observed in all the available low resolution modules ( sl and ll ) with a ramp time of 6 s per cycle . two cycles per module were typically performed . those targets with iras 12 @xmath11 flux below 1 jy were observed for 3 cycles in the 5.214.5 @xmath12 modules , to keep a high enough s / n ( typically @xmath1050 ) . the irs pipeline version s17.2.0 ( partial ) bcds were retrieved from the archive . the sky subtraction was performed taking advantage of the way the data are acquired by default . each module has two sub - modules ( sl2 and sl1 , ll2 and ll1 ) observing at the same time , but not in the same direction . since the two sub - modules do not point the same target at the same time , when one is on - target , the other is performing a sky observation . the data taken by sl1 ( ll1 ) when sl2 ( ll2 ) is on target can then be used as a sky frame for a subsequent sl1 ( ll1 ) pointing and vice versa . for each bcd , we have averaged all the off - target observations and subtracted this from the single on - target bcds . the spectrum was extracted in every bcd with spice using the optimal point source extraction option . for each target , all its extracted spectra were then averaged and combined into one using smart @xcite . irac has proved to be a very sensitive instrument , revealing previously unseen structures in the outer shells of expanded pne @xcite . since the nebulae in our sample are typically very compact and ir - bright , which is normal for their evolutionary status , their core emission dominates the inner few arcsec near the central source . this unfortunately usually prevents us from detecting any weak extended emission near the core , but , by psf subtraction , does not prevent us from measuring the fluxes of our sources . given the size of the psf , an upper limit of @xmath105@xmath13 to the sizes of the nebulae can be estimated ( marengo , private communication ) . in table [ tab : iracflux ] we list the photometric results obtained . for a comparison of our results to previous observations in the literature ( i.e. , @xcite ) , we show in figure [ fig : irac_distr ] four color - color diagrams based on irac data . as already shown in @xcite , pne do not seem to follow any specific track on the irac @xmath14-[4.5]$ ] versus @xmath15-[8.0]$ ] diagram . we find our targets to be generally located in an area ranging from 0 to 3 in @xmath15-[8.0]$ ] and from 0 to 1 in @xmath14-[4.5]$ ] . this region is included in the area where pne are found in the two works mentioned above . a comparison of the @xmath14-[4.5]$ ] versus @xmath16-[8.0]$ ] diagram to the analogous one in @xcite shows that the stars in our sample have on average lower values of @xmath14-[4.5]$ ] , which may be explained if pah emission features are present ( see [ spectra ] ) ; the spread in @xmath16-[8.0]$ ] indicates that different dust temperatures can be found in these objects , with the radio - detected ones ( i.e. , hotter central stars ) being redder in @xmath16-[8.0]$ ] . lccccc 01005 + 7910 & 175.6 @xmath17 9.1 & 189.2 @xmath17 9.8 & 460.0 @xmath17 24.0 & 2004.1 @xmath17 104.4 & 18.1 @xmath17 0.9 + 06556 + 1623 & 759.2 @xmath17 39.5 & 898.5 @xmath17 46.8 & 1045.5 @xmath17 54.5 & 1210.0 @xmath17 63.0 & 2.8@xmath17 0.2 + 09470 - 4617 & 70.2 @xmath17 3.7 & 89.9 @xmath17 4.7 & 191.7 @xmath17 10.0 & 1233.3 @xmath17 64.3 & 3.5 @xmath17 0.2 + 17074 - 1845 & 17.6 @xmath17 0.9 & 12.0 @xmath17 0.6 & 10.5 @xmath17 0.5 & 47.5 @xmath17 2.5 & 10.7 @xmath17 0.6 + 17203 - 1534 & 18.7 @xmath17 1.0 & 12.8 @xmath17 0.7 & 11.5 @xmath17 0.6 & 24.7 @xmath17 1.3 & 8.9 @xmath17 0.5 + 17364 - 1238 & 10.8 @xmath17 0.6 & 8.2 @xmath17 0.4 & 7.7 @xmath17 0.4 & 9.9 @xmath17 0.5 & 1.1 @xmath17 0.1 + 17381 - 1616 & 4.3 @xmath17 0.2 & 3.6 @xmath17 0.2 & 2.9 @xmath17 0.1 & 33.1 @xmath17 1.7 & 4.4 @xmath17 0.3 + 17423 - 1755 & 3425.6 @xmath17 178.4 & 4492.5 @xmath17 233.8 & 5227.3 @xmath17 272.5 & 5344.2 @xmath17 278.5 & 24.2 @xmath17 1.2 + 17542 - 0603 & 936.3 @xmath17 48.8 & 1057.1 @xmath17 55.0 & 1045.5 @xmath17 54.5 & 1105.7 @xmath17 57.6 & 5.3 @xmath17 0.3 + 18040 - 1457 & 312.1 @xmath17 16.3 & 211.4 @xmath17 11.0 & 209.1 @xmath17 10.9 & 142.5 @xmath17 7.4 & [email protected] + 18062 + 2410 & 7.4 @xmath17 0.4 & 9.0 @xmath17 0.5 & 13.5 @xmath17 0.7 & 557.7 @xmath17 29.1 & 16.3 @xmath17 0.9 + 18070 - 2346 & 187.3 @xmath17 9.8 & 119.8 @xmath17 6.2 & 82.1 @xmath17 4.3 & 55.8 @xmath17 2.9 & 1.0 @xmath17 0.1 + 18367 - 1233 & 6532.6 @xmath17 340.3 & 5990.0 @xmath17 311.8 & 5750.0 @xmath17 299.8 & 7125.6 @xmath17 371.3 & + 18371 - 3159 & 7.0 @xmath17 0.4 & 6.0 @xmath17 0.3 & 7.7 @xmath17 0.4 & 22.1 @xmath17 1.2 & 5.9 @xmath17 0.3 + 18379 - 1707 & 41.3 @xmath17 2.2 & 38.2 @xmath17 2.0 & 82.1 @xmath17 4.3 & 320.7 @xmath17 16.7 & 29.9 @xmath17 1.0 + 18442 - 1144 & 16.5 @xmath17 0.9 & 18.0 @xmath17 0.9 & 71.9 @xmath17 3.7 & 320.7 @xmath17 16.7 & 12.1 @xmath17 0.6 + 19200 + 3457 & 18.7 @xmath17 1.0 & 13.8 @xmath17 0.7 & 38.3 @xmath17 2.0 & 142.5 @xmath17 7.4 & 1.6 @xmath17 0.1 + 19306 + 1407 & 40.1 @xmath17 2.1 & 30.0 @xmath17 1.6 & 88.5 @xmath17 4.6 & 583.0 @xmath17 30.4 & 47.5@xmath17 2.9 + 19336 - 0400 & 9.4 @xmath17 0.5 & 9.0 @xmath17 0.5 & 10.5 @xmath17 0.5 & 53.4 @xmath17 2.8 & 6.7 @xmath17 0.4 + 19399 + 2312 & 280.9 @xmath17 14.6 & 179.7 @xmath17 9.4 & 164.3 @xmath17 8.6 & 142.5 @xmath17 7.4 & [email protected] + 19590 - 1249 & 14.0 @xmath17 0.7 & 11.2 @xmath17 0.6 & 11.5 @xmath17 0.6 & 25.7 @xmath17 1.3 & 7.9 @xmath17 0.4 + 20462 + 3416 & 24.4 @xmath17 1.3 & 16.3 @xmath17 0.9 & 17.7 @xmath17 0.9 & 27.9 @xmath17 1.5 & 11.1 @xmath17 0.6 + 20572 + 4919 & 936.3 @xmath17 48.8 & 898.5 @xmath17 46.8 & 884.6 @xmath17 46.1 & 1733.2 @xmath17 90.3 & 8.5 @xmath17 0.4 + 21289 + 5815 & 61.1 @xmath17 3.2 & 64.2 @xmath17 3.3 & 92.0 @xmath17 4.8 & 246.7 @xmath17 12.9 & 1.1 @xmath17 0.1 + 22023 + 5249 & 18.7 @xmath17 1.0 & 16.3 @xmath17 0.9 & 37.7 @xmath17 2.0 & 229.0 @xmath17 11.9 & 19.0 @xmath17 1.1 + 22495 + 5134 & 10.4 @xmath17 0.5 & 9.5 @xmath17 0.5 & 12.1 @xmath17 0.6 & 80.2 @xmath17 4.2 & 9.7 @xmath17 0.5 + we also plotted our targets on a h - k versus k-@xmath14 $ ] diagram , which can be compared to the jhk diagram in figure [ fig:2mass_distr ] . a similar trend can be found in both plots . as shown in @xcite , the jhk diagram allows us to distinguish different types of sources . many of our targets are concentrated in the _ hot star _ and _ nebula @xmath18 star _ regions , as expected for their b spectral type classification . other targets fall within regions of objects with a hot star and strong dust emission . the presence of a large amount of dust around the central stars is also evident in the @xmath14-[8.0]$ ] versus @xmath19-[24]$ ] diagram . since we do not have mips observations of our targets but we have 538 @xmath1 m irs spectra , we calculated the expected 24 @xmath1 m mips measurement by convolving our irs spectra with the mips spectral response function and then integrating over the bandpass . performing this calculation on literature targets with both irs and mips data has produced results compatible , within errors , with the observations , therefore we are confident that our expected mips values are reasonable approximations to the actual values . interestingly , among the radio detected objects in this diagram we can distinguish a group of stars clustered around ( 4,2 ) , namely iras 06556 + 1623 and 17423 - 1755 , while the others seem to be found in a stripe roughly going from ( 5,6 ) to ( 9,1 ) . after noticing these possible different distributions , we checked the positions of the central targets in the other diagrams , thus finding that although the overall distributions of our targets match those observed in @xcite , iras 06556 + 1623 and 17423 - 1755 seem not to follow the trend of the other radio detected objects . in @xmath14-[4.5]$ ] versus @xmath15-[8.0]$ ] and @xmath14-[4.5]$ ] versus @xmath16-[8.0]$ ] , these stars are found respectively around ( 0.7,0.7 ) and ( 1.3,0.7 ) , while in h - k versus k-@xmath14 $ ] they are the two top - right radio detected objects ( highest h - k and k-@xmath14 $ ] ) . , the sources are represented as squares if detected at radio wavelengths , as diamonds if not . radio detected targets tend to concentrate in what has been identified as the _ hot star _ region in @xcite.,width=302 ] the reduced irs spectra of the objects in our sample are shown in figures [ fig : spectra_crich][fig : spectra_mixed2 ] . they are split into three groups : c - rich , o - rich , and mixed chemistry . pne are usually classified as c - rich or o - rich . this classification is linked to the evolution of the central star , because it depends on the possibility for the star to go through a third dredge - up . it is this event that alters the chemistry in the envelope turning it from o - rich into c - rich during the agb phase @xcite . because of the stability of the co molecule , if c is less abundant than o , all c is trapped in co and then the envelope shows features of molecules containing oxygen ( besides co ) and vice versa if o is less abundant than c. it is known however that a minority of stars shows mixed chemistry , with c - bearing and o - bearing molecules . more than @xmath2 of the stars in our sample show both polycyclic aromatic hydrocarbons ( pah ) and silicate features , while a smaller percentage ( @xmath1010% ) is expected . among these mixed - chemistry objects about 70% are radio detected . we distinguished o - rich and c - rich envelopes on the presence of amorphous silicate ( 10-@xmath1 m bump ) or pah features ( mainly 6.2 and 7.7 @xmath1 m ) . objects with both a 10-@xmath1 m silicate feature and pah transitions were classified as mixed chemistry . it must be noticed that the sl and ll irs modules overlap around 1415 @xmath1 m , therefore this region of the spectra is subject to higher noise . shifts between the flux levels are also possible . in a few cases , we have scaled the sl spectra to the ll flux level . all the targets classified as c - rich show 6.2 , 7.7 , 8.5 , and 11.2 @xmath1 m features , commonly attributed to pah . the shape and strength of the features depend on the excitation status and size of the molecules . we address the properties of these features in the next section . we do not detect the so - called 21 @xmath1 m feature in any of our targets . this feature has in fact been found in 15 post - agb stars with f - g spectral types and in two pne @xcite . therefore , besides being a rare feature , the chances for its detection are even lower in b stars as those in our sample , because of the harder radiation field from the central star . the sources iras 01005 + 7910 and 19200 + 3457 show a strong broad feature around 30 @xmath1 m that has been previously observed in other post - agb stars and pne and is attributed to mgs . the peak position and width of this feature change as the temperature of the underlying continuum changes , corresponding to the evolution from post - agb to pn . the peak is found in the 2635 @xmath1 m range @xcite . it is likely that such emission is present in iras 18442 - 1144 around 34 @xmath1 m , although weaker and with a smaller width . this target and iras 06556 + 1623 show an excess of emission peaking around 17.7 @xmath1 m ( figure [ fig:17micron_crich ] ) that can be explained as due to vibrationally excited c - h bonds @xcite . this emission is also detected in 01005 + 7910 and 19200 + 3457 as a very broad feature . iras 06556 + 1623 also shows a weak broad emission underlying the 6.2 , 7.7 and 11.2 @xmath1 m features , which is due to pah molecules as well . in o - rich targets , as we can see in figures [ fig : spectra_orich1 ] and [ fig : spectra_orich2 ] , the 10 @xmath1 m broad feature from amorphous silicates is clearly seen , although it often appears to be structured . in iras 17074 - 1845 , 17203 - 1534 , and 20572 + 4919 we clearly distinguish the presence of single features contributing to the overall bump . the shift in shape from the smooth ism - like feature ( clearly peaked at 9.8 @xmath1 m ) to the structured plateau - like feature of the material in the solar system ( peaked at 11.3 @xmath1 m ) has been explained as due to grain processing , i.e. coagulation of small ( @xmath100.1 @xmath1 m ) grains into large ( 12 @xmath1 m ) grains and annealing of amorphous silicates into crystalline magnesium silicates and silica @xcite . a correlation of the shape and strength of the 10-@xmath1 m silicate emission feature has been observed in herbig aebe and t tauri stars and it has been interpreted as evidence of grain processing in the circumstellar disks of those stars @xcite . the shape of the feature can be estimated by comparing the continuum - subtracted flux ratio at 11.3 and 9.8 @xmath1 m to the peak - to - continuum ratio . we have performed this comparison for our o - rich targets . mixed - chemistry sources were not included because of the pah 11.2 @xmath1 m feature , which would affect the flux estimates . iras 18435 - 0052 was not included either , because it shows the 9.8 @xmath1 m feature in absorption , which would clearly make it an outlier . the plot obtained is shown in figure [ fig : silicate_plot ] . the data points are sparse in the diagram : though a similar trend as in young stars might be present , we can not conclude that this is the case . the absence of linear correlation may be explained as due to non - linear effects in highly processed dust ( 11.3/9.8 flux ratio @xmath20 ) , as pointed out by @xcite . our results match with those found by @xcite , who inspected disks around post - agb stars . the similarity of the results leads us to the conclusion that in spite of the absence of correlation , the dust in our targets shows a high degree of processing , which points to the presence of a stable structure where the grains can have time to grow and anneal . this is a hint for the presence of circumstellar / circumbinary disks in our sources . m ) as a function of the 10-@xmath1 m silicate emission feature strength ( peak flux over continuum flux).,width=302 ] as noticed above , iras 18435 - 0052 is the only target in which the silicate bump shows an absorption feature on top of the emission . since silicates are seen in absorption early after the agb , the presence of both absorption and emission features may be interpreted as an earlier evolutionary stage for this source . although this target has been classified as a bii star @xcite , its irs spectrum indicates major differences to the rest of the sample . almost all of these targets also show an emission feature around 18 @xmath1 m that can be attributed to amorphous silicates . in figure [ fig:17micron_orich ] we show two examples of this feature . in such targets as 17381 - 1616 , 17460 - 3114 , 19157 - 0247 , 18371 - 3159 , 19336 - 0400 , 19590 - 1249 , 20462 + 3416 , and 22495 + 5134 crystalline silicate features around 23.7 , 27.6 and 33.6 @xmath1 m are visible . the presence of crystalline silicates in post - agb stars has been previously reported although different hypotheses exist about the production mechanism of these crystals @xcite . in figures [ fig : spectra_mixed1 ] and [ fig : spectra_mixed2 ] the spectra of those sources with both silicate and pah features are shown . the 10 @xmath1 m silicate bump is typically much stronger than the 6.2 and 7.7 @xmath1 m pah features . the 11.2 @xmath1 m feature is blended with the silicate emission and detected on top of it . six of these stars also show the [ ] 12.8 @xmath1 m line , which proves their ionized status . four of them also have the [ ] 7 @xmath1 m emission line . the presence of pah features in post - agb envelopes is well - known , although it is not yet clear how these molecules are produced @xcite . as a first step in the analysis of these features , we have classified our targets according to the scheme developed in @xcite , with the exception of iras 06556 + 1623 and 17542 - 0603 because of their s / n ratio in the 511 @xmath1 m range . almost all of our sources have their peak of the 6.2 @xmath1 m emission feature beyond 6.235 @xmath1 m , which puts them in class _ b_. if we also consider the 7.7 @xmath1 m and 8.5 @xmath1 m features , the stars in our sample show a more complicated picture . while the three _ a _ class sources are also classified as _ a@xmath21 _ and _ a@xmath22 _ , class _ b _ sources spread into a group of intermediate subclasses . tables [ tab : pah_summary1 ] and [ tab : pah_summary2 ] report the classification for every target . the wavelengths calculated do not correspond to the peaks of the features , but to their central ( median - flux ) wavelengths , which better accounts for the different components within each complex . the 7.7 @xmath1 m features have been classified as a@xmath21 or b@xmath21 if their central wavelength was respectively smaller or larger than 7.7 @xmath1 m , since the complex appears to be mainly due to two features peaking around 7.6 and 7.8 @xmath1 m . lccccc 01005 + 7910 & 6.231 @xmath17 0.005 & 7.58 @xmath17 0.08 & 8.59 @xmath17 0.03 & 11.21 @xmath17 0.07 & aa@xmath21a@xmath22 + 09470 - 4617 & 6.227 @xmath17 0.005 & 7.66 @xmath17 0.03 & 8.60 @xmath17 0.06 & 11.29 @xmath17 0.01 & aa@xmath21a@xmath22 + 18442 - 1144 & 6.238 @xmath17 0.001 & 7.64 @xmath17 0.01 & 8.60 @xmath17 0.07 & 11.205 @xmath17 0.004 & ba@xmath21a@xmath22 + 19200 + 3457 & 6.24 @xmath17 0.04 & 7.61 @xmath17 0.07 & 8.58 @xmath17 0.02 & 11.250 @xmath17 0.006 & ba@xmath21a@xmath22 + 18371 - 3159 & 6.28 @xmath17 0.01 & 7.68 @xmath17 0.09 & 8.61 @xmath17 0.09 & 11.2 @xmath17 0.1 & ba@xmath21a@xmath22 + 18379 - 1707 & 6.266 @xmath17 0.004 & 7.62 @xmath17 0.04 & 8.57 @xmath17 0.06 & 11.03 @xmath17 0.06 & ba@xmath21a@xmath22 + 19306 + 1407 & 6.240 @xmath17 0.003 & 7.74 @xmath17 0.04 & 8.6 @xmath17 0.1 & 11.30 @xmath17 0.07 & bb@xmath21b@xmath22 + 19336 - 0400 & 6.25 @xmath17 0.01 & 7.61 @xmath17 0.02 & 8.70 @xmath17 0.02 & 11.30 @xmath17 0.03 & ba@xmath21b@xmath22 + 19590 - 1249 & 6.22 @xmath17 0.06 & 7.58 @xmath17 0.02 & 8.62 @xmath17 0.03 & 11.21 @xmath17 0.03 & aa@xmath21a@xmath22 + 20462 + 3416 & 6.25 @xmath17 0.01 & 7.66 @xmath17 0.06 & 8.58 @xmath17 0.04 & 11.26 @xmath17 0.03 & ba@xmath21a@xmath22 + 21289 + 5815 & 6.251 @xmath17 0.001 & 7.74 @xmath17 0.03 & 8.64 @xmath17 0.08 & 11.25 @xmath17 0.07 & bb@xmath21b@xmath22 + 22023 + 5249 & 6.23 @xmath17 0.01 & 7.69 @xmath17 0.03 & 8.55 @xmath17 0.09 & 11.23 @xmath17 0.07 & ba@xmath21a@xmath22 + 22495 + 5134 & 6.25 @xmath17 0.01 & 7.58 @xmath17 0.02 & 8.5 @xmath17 0.1 & 11.45 @xmath17 0.05 & ba@xmath21a@xmath22 + lcccc 01005 + 7910 & 24 @xmath17 1 & 38 @xmath17 2 & 9 @xmath17 1 & 29 @xmath17 3 + 09470 - 4617 & 11.5 @xmath17 0.6 & 19 @xmath17 1 & 2.6 @xmath17 0.5 & 14 @xmath17 1 + 18442 - 1144 & 4.34 @xmath17 0.05 & 8.2 @xmath17 0.3 & 2.2 @xmath17 0.2 & 2.41 @xmath17 0.04 + 19200 + 3457 & 1.77 @xmath17 0.05 & 3.3 @xmath17 0.1 & 0.64 @xmath17 0.06 & 1.10 @xmath17 0.03 + 18371 - 3159 & 0.12 @xmath17 0.02 & 0.25 @xmath17 0.05 & 0.04 @xmath17 0.01 & 0.13 @xmath17 0.05 + 18379 - 1707 & 4.0 @xmath17 0.2 & 5.6 @xmath17 0.4 & 0.5 @xmath17 0.1 & 1.1 @xmath17 0.3 + 19306 + 1407 & 3.98 @xmath17 0.03 & 8.6 @xmath17 0.4 & 0.2 @xmath17 0.1 & 4.9 @xmath17 0.8 + 19336 - 0400 & 0.17 @xmath17 0.03 & 0.74 @xmath17 0.05 & 0.06 @xmath17 0.01 & 0.33 @xmath17 0.04 + 19590 - 1249 & 0.17 @xmath17 0.05 & 0.37 @xmath17 0.05 & 0.03 @xmath17 0.01 & 0.18 @xmath17 0.06 + 20462 + 3416 & 0.49 @xmath17 0.02 & 1.5 @xmath17 0.1 & 0.14 @xmath17 0.03 & 0.28 @xmath17 0.07 + 21289 + 5815 & 1.20 @xmath17 0.02 & 2.2 @xmath17 0.2 & 0.18 @xmath17 0.05 & 0.35 @xmath17 0.07 + 22023 + 5249 & 2.4 @xmath17 0.1 & 4.7 @xmath17 0.2 & 0.6 @xmath17 0.1 & 1.0 @xmath17 0.2 + 22495 + 5134 & 0.31 @xmath17 0.04 & 0.52 @xmath17 0.04 & 0.13 @xmath17 0.03 & 0.6 @xmath17 0.06 + @xcite propose that objects in class _ b _ contain fresh pah molecules , since the post - agb stars and pne they observed fall in this class . class _ a _ objects , which include pdrs and reflection nebulae ( ism - like pahs ) , are sources with more processed pahs . in the attempt to characterize the pah population in our targets , we have followed @xcite in their inspection of pah features in herbig aebe stars . we fit a spline to the continuum in the 514 @xmath1 m region of the spectra , including in the spline the 10 @xmath1 m bump from amorphous silicates , when present . the continuum - subtracted spectra are shown in figure [ fig : pah_spectra ] . we notice that despite the spline subtraction , which should remove the 10 @xmath1 m silicate feature , residual emission in the 911 @xmath1 m range remains . as noted above , these features may be due to hot crystalline silicates . after calculating the central wavelengths as in their study , we compare our results to @xcite , who found a correlation between the effective temperature of the central star and the central wavelength of both the 11.2 and the 7.7 @xmath1 m features . although the astrophysical targets are quite different ( young and post - agb stars ) , the properties of the pahs can be compared to determine how these molecules behave in different environments . in figure [ fig:7_11 ] , we show the correlation between the central wavelengths of the features at 6.2 , 7.7 , and 11.2 @xmath1 m . a trend with increasing temperature from red to blue wavelengths is clear . with the exception of two outliers ( 22495 + 5134 in the top and 18379 - 1707 in the bottom of the 7.7 vs 11.2 @xmath1 m plot ) , our data follow the trend in @xcite , being located at the high - temperature limit , as expected for stars with an ionized envelope ( t@xmath23 20000 k ) . although the plots show a clear correlation , the position occupied by the hot pn ngc 7027 ( t@xmath24 k ) - shown as a dark green dot just as an example of hot pn - points out that other factors play an important role . this complicated dependence is evident when inspecting the pah ionization fraction . since the intensity of the features in the 69 @xmath1 m range relative to that of the 11.2 @xmath1 m one is an order of magnitude higher for ionized than for neutral pahs @xcite , in figure [ fig : ion_frac ] we plot i@xmath25i@xmath26 vs i@xmath27i@xmath26 ratios . while @xcite could not conclude that a correlation was present in this diagram , the inclusion of our sample allows us to find a remarkable correlation , but no clear link between the ionization fraction and t@xmath28 . our sample of post - agb stars contains hotter stars than the haebe stars in @xcite ( figure [ fig:7_11 ] and the presence of radio continuum ) but it has a somewhat lower fraction of ionized pahs . this trend for a decrease of ionization fraction in the post - agb has been reported in @xcite for 3 objects . we find this as a general trend within our sample . since the ionization fraction does not seem to correlate with the temperature of the central star , we tried to link it to the local physical conditions experienced by the pahs , by plotting the i@xmath25i@xmath26 ratio versus the ionization factor g@xmath29n@xmath30t@xmath31 , where g@xmath29 is the radiation field in habing units , n@xmath30 the electron density in @xmath32 , and t the gas temperature in 10@xmath33 k. we have calculated this parameter by applying the empirical formula in @xcite that links it to i@xmath27i@xmath26 in its lorentzian formulation , since it extends to lower values of i@xmath27i@xmath26 . in figure [ fig : ion_frac ] , we show the result for @xmath34g@xmath29n@xmath30t@xmath35 ( limits of reliability of the formula ) . although the data are spread in the plot , a weak correlation can be identified , indicating that the ionization fraction increases with the ionization factor . once again , no clear correlation with t@xmath28 is observed , indicating that the local conditions in the envelope play a major role . the results of our analysis of the pah features are somewhat conflicting . on one side we see that the central wavelengths of the features are shifted toward the blue , as expected in photo - processed molecules ; on the other side , we find a low ionization fraction , which implies low processing . we conclude that a higher effective temperature implies a harder field and stronger processing in terms of shift of the central wavelength toward the blue , but not necessarily a higher ionization fraction , this being dependent on the available amount of ionizing photons , thefore on the distance to the central star . our plots indicate that the pah molecules in our stars are located in the outflows , where they are subject to a more diluted field . although the probability of interaction between photons and molecules far from the star is smaller , the hard photons from the central star - when interacting - determine a major damage to the molecules , resulting in the shift in central wavelength . in this scenario , the closer molecules must have been previously destroyed by the radiation field . since pah features with red central wavelengths are found in post - agb stars ( rafgl 2688 is one of the two class _ c _ objects in @xcite ) , this disruption should occur during the few 10@xmath33 yr between the early post - agb and the development of an ionized shell . high - angular resolution observations in the pah emission ranges are necessary to support such an interpretation , which confirms the idea of c - bearing molecules being located in the outflows of mixed - chemistry nebulae . finally , it must be noticed that the wavelength shifts observed can be determined not only by photo - processing : nitrogenation can in fact be another explanation @xcite . as shown in @xcite , the mass that goes into pahs can be linked to the ratio of the flux emitted in the pah features to the continuum flux in the far infrared range ( fir ) , nominally between 40 and 500 @xmath1 m . we have therefore taken advantage of our sed models ( see [ seds ] ) to calculate the fir flux by integrating our curves in the mentioned interval . the fraction of carbon trapped in pahs can be estimated as @xcite @xmath36 where @xmath37 is the average uv absorption cross section of pahs per c atom and @xmath38 is the pah to continuum flux ratio . we assume @xmath39 @xmath40 as a typical value for @xmath37 @xcite . the fraction of c in pahs can then be estimated as : @xmath41 table [ tab : abundances ] lists the results of the calculation , along with the estimates of the number abundance of pah molecules to h nuclei , assuming that the average pah molecule has 50 c atoms . the errors in the table have been calculated by standard error propagation , including a 15% uncertainty in the model estimation of the underlying continuum . lcccc _ c - rich _ + 01005 + 7910 & 6 & 2 & 51 & 19 + 09470 - 4617 & 43 & 37 & 336 & 293 + 18442 - 1144 & 1.0 & 0.3 & 8 & 2 + 19200 + 3457 & 2.7 & 0.9 & 21 & 7 + _ mixed _ + 18371 - 3159 & 0.07 & 0.02 & 0.55 & 0.18 + 18379 - 1707 & 1.1 & 0.4 & 8.7 & 2.8 + 19306 + 1407 & 0.48 & 0.15 & 3.7 & 1.1 + 19336 - 0400 & 0.12 & 0.04 & 0.9 & 0.3 + 19590 - 1249 & 0.10 & 0.03 & 0.7 & 0.2 + 20462 + 3416 & 0.12 & 0.03 & 0.9 & 0.3 + 21289 + 5815 & 5.2 & 1.9 & 41 & 15 + 22023 + 5249 & 0.5 & 0.1 & 3.8 & 1.2 + 22495 + 5134 & 0.12 & 0.04 & 1.0 & 0.3 + the results show a trend for larger values of @xmath42 and pah abundance in c - rich stars . in these sources , the values of c fraction range from 1% to 43% ( 3% is a typical value in the ism ) , while those of the pah abundance are between 7@xmath43 and 336@xmath43 . with the remarkable exception of iras 21289 + 5815 , stars with both pah and silicate features show lower values : 0.11% for @xmath42 and 0.58@xmath43 for the pah abundance . we notice here that since in our modeling we lack observations in the sub - mm and mm ranges , our estimation of the fir flux and the values derived from it should be taken with caution . to model our targets , we have collected data from the 2mass @xcite and iras archives . we have dereddened the 2mass magnitudes following @xcite . the intrinsic _ b - v _ color indexes have been estimated from @xcite on the basis of the spectral classifications of the targets , then the magnitude correction was calculated with the derived color excess . we have used the dusty code @xcite to solve the problem of radiation transfer in the dust envelopes of our targets . our typical first cycle of dusty had the following input : * central source : blackbody curve with temperature matching the spectral type of the target ; * chemistry of the envelope : either 100% silicates or 100% amorphous carbon , depending on irs features ; * grain size distribution : power law @xmath44 , @xmath45 , @xmath46 , @xmath47 being the grain radius in @xmath1 m ; * density distribution : power law @xmath48 , @xmath49 being the shell radius ; * shell relative thickness : @xmath50 ; * dust temperature at @xmath51 : adapted to each sed ; * optical depth at 60 @xmath1 m : calculated for each star from iras data ; we calculated the opacity at 60 @xmath1 m for each object as @xmath52 where @xmath53 is the dust temperature , @xmath54 the angular radius of the nebula in arcsec , @xmath55 the flux density at the frequency @xmath56 in erg @xmath40 s@xmath6 hz@xmath6 , @xmath57 the planck function at the temperature @xmath53 in erg @xmath40 s@xmath6 hz@xmath6 sr@xmath6 @xcite . this gives us a first estimate of the opacity , based on the 5@xmath13 upper limit to the envelope sizes derived from our irac images . this value of opacity was used in the first cycle of dusty and then changed in the following cycles to better match the data . in several targets we have found that it is not possible to model the sed with only one dust component . dusty does not allow the user to include in the model several dust shells at the same time , but it is possible to use the output of a run as an input to a second run , thus approximately reproducing a multi - shell structure . we have successfully applied this technique to the sources that clearly showed a near - ir excess of radiation , pointing out the presence of hot dust . a summary of the input parameters used in our modeling is given in table [ tab : models ] . lcccccc _ c - rich spectra _ & & & & & & + 01005 + 7910 & 20300 & am - c & 0.0050.26 & 700 & 1000 & 0.16 + & & & 0.0052.5 & 145 & 20 & 7 + 06556 + 1623 & 20000 & sil - ow 15% , am - c 85% & 0.0050.26 & 900 & 1000 & 0.61 + & & sil - dl & 0.0052 & 135 & 1000 & 3 + 09470 - 4617 & 20000 & am - c & 0.0050.26 & 900 & 1000 & 0.2 + & & & & 220 & 20 & 2 + 17423 - 1755 & 20000 & am - c & 0.005100 & 1150 & 445 & 135 + & & & & 132 & 2 & 80 + & & & & 80 & 2 & 100 + 18442 - 1144 & 20000 & am - c & 0.0050.26 & 280 & 2 & 0.4 + & & sil - oc & & 110 & 1000 & 20 + 19200 + 3457 & 7000 & am - c & 0.0050.26 & 380 & 1000 & 0.27 + & & am - c 80% , sil - oc 20% & & 110 & 5 & 3.2 + _ o - rich spectra _ & & & & & & + 17074 - 1845 & 17100 & sil - ow & 0.0050.26 & 320 & 5 & 0.15 + & & sil - oc & & 120 & 5 & 4.3 + 17203 - 1534 & 20000 & sil - oc & 0.0050.26 & 470 & 2 & 0.04 + & & sil - oc & & 110 & 4 & 2.75 + 17364 - 1238 & 7500 & sil - ow 50% , am - c 50% & 0.0050.26 & 450 & 2 & 0.27 + & & sil - oc & & 100 & 2 & 20 + 17381 - 1616 & 24000 & sil - ow & 0.0050.26 & 250 & 3 & 0.15 + & & sil - oc & & 120 & 3 & 4 + 17460 - 3114 & 34700 & sil - ow & 0.0050.26 & 500 & 1000 & 0.02 + & & sil - oc & & 140 & 2 & 0.017 + 18062 + 2410 & 24000 & sil - ow & 0.0050.26 & 320 & 2 & 0.32 + & & sil - oc & & 160 & 2 & 4 + 18435 - 0052 & 3000 & sil - ow 95% , am - c 5% & 0.0050.26 & 1300 & 1000 & 1400 + 19157 - 0247 & 24000 & am - c & 0.0050.26 & 950 & 1000 & 0.32 + & & am - c 70% , sil - oc 30% & & 330 & & 0.17 + 20572 + 4919 & 12000 & 70% sil - dl , 30% am - c & 0.0050.26 & 1300 & 1000 & 4.5 + & & & & 150 & 1000 & 3.75 + _ mixed spectra _ & & & & & & + 17542 - 0603 & 21000 & am - c & 0.0050.26 & 1050 & 100 & 1 + & & sil - ow & & 145 & 3 & 1.45 + 18371 - 3159 & 24000 & sil - ow 85% , am - c 15% & 0.0050.26 & 350 & 2 & 0.05 + & & sil - oc & & 105 & 5 & 2.4 + 18379 - 1707 & 24000 & sil - ow & 0.0050.26 & 590 & 3 & 0.1 + & & sil - oc & & 120 & 2 & 1.7 + 19306 + 1407 & 16000 & sil - ow 95% , am - c 5% & 0.0050.26 & 440 & 10 & 1.5 + & & sil - oc & & 120 & 2.5 & 30 + 19336 - 0400 & 24000 & sil - ow 85% , am - c 15% & 0.0050.26 & 320 & 2 & 0.4 + & & sil - oc & & 105 & 2 & 4.3 + 19590 - 1249 & 24000 & sil - ow 85% , am - c 15% & 0.0050.26 & 300 & 2 & 0.035 + & & sil - oc & & 100 & 2 & 1.5 + 20462 + 3416 & 24000 & sil - ow 80% , am - c 20% & 0.0050.26 & 300 & 1000 & 0.034 + & & sil - oc & & 100 & 1000 & 1.5 + 21289 + 5815 & 15000 & sil - ow & 0.0050.26 & 1300 & 100 & 4.8 + & & sil - oc & & 150 & 100 & 4.6 + 22023 + 5249 & 20000 & 90% sil - ow , 10% am - c & 0.0050.26 & 440 & 3 & 0.25 + & & sil - oc & & 105 & 2 & 6 + 22495 + 5134 & 20000 & 30% sil - ow , 70% am - c & 0.0050.26 & 360 & 5 & 0.05 + & & sil - oc & & 115 & 100 & 7 one of the issues encountered in fitting these data involves the actual shape of the 10 @xmath1 m bump . our models , though often reproducing the overall shape of the sed quite well , fail in reproducing the correct shape of the 10 @xmath1 m bump especially in its longer wavelength part , where it overlaps with the underlying rising continuum . the shape of the feature depends on several factors , including dust temperature , grain size , and composition . geometrical effects can also play a role , if different dust species are segregated in different regions of the envelopes . another issue concerned the fit of the near - ir emission . in several instances dusty underestimates the continuum in this range . this can point out the presence of additional hot dust , but another possibility is the presence of a free - free contribution from the ionized envelope , which is neglected in dusty . our radio observations prove to be very useful then , since by modeling the radio emission we can check if the hypothesis of a free - free contribution in the near - ir is valid or not . we have modeled the radio continuum spectra assuming the central star is surrounded by a shell of ionized gas with constant density . the usual approximation of the gaunt factor at radio wavelengths typically results into a quick cutoff of the free - free emission in the far - ir , where , compared to the dust emission , such contribution is indeed negligible . nevertheless , in the near - ir , where the central star and dust fluxes are at a minimum , the free - free contribution to the observed flux may not be much smaller than the contributions from the other emitting components . for example , in the evolved pn ic 4406 a near - ir excess , previously supposed to indicate a hot - dust component , has been explained as a free - free contribution @xcite . therefore , we decided to estimate the gaunt factor following @xcite for a 10@xmath58 k electron temperature , extending the calculation of the emission curve from the cm range to 0.8 @xmath1 m . for details about the radio modeling , see @xcite . where radio data were available , we summed up the free - free and dust model contributions . this often results into a substantial correction to the dust - only model , not only in the cm range , as expected , but also in the near - ir . in figures [ fig : dusty1 ] and [ fig : dusty2 ] we show examples of the performed dusty and radio fits for about half of our sample . during the various trials performed to match the data , we have found that in some sources it is necessary to add an extra - component that does not match chemically with the classification of the target . specifically , in half of the o - rich targets ( iras 17364 - 1238 , 17542 - 0603 , 18435 - 0052 , 19157 - 0247 , 20572 + 4919 ) we had to include an am - c component to match the observational points . similarly , in two of the c - rich sources ( iras 06556 + 1623 and 19200 + 3457 ) we had to include a silicate component . it must be considered that this may be due to limits in the modeling code , in particular the current knowledge of the opacity coefficients in the mid - ir , and does not necessarily represent a mixed - chemistry nature of the targets . iras 17423 - 1755 has been modeled with amorphous carbon grains , although it does not show any pah features . this target is reported to show c@xmath59h@xmath59 absorption feature @xcite , which indicates it has a c - rich chemistry . such a chemistry also matches with its featureless mid - ir spectrum . in our attempts to model it with silicates , we have found that if silicates were present they would determine a distinct 10-@xmath1 m emission feature , because of their high temperature ( peak in irac bands ) . in two targets showing both pah and silicate features ( iras 18379 - 1707 and 21289 + 5815 ) a silicate - only chemistry is sufficient to fit the continuum spectrum . iras 17423 - 1755 is our only target with mm measurements @xcite . we have assumed as size of the dust region the angular size of its co shell as reported in @xcite , @xmath102@xmath13 , with a distance of 5.8 kpc and luminosity 12600 l@xmath5 . when using in dusty the opacity calculated at 60 @xmath1 m from the iras data , we find that the model highly overestimates the far - ir emission . changes in the opacity input value do not improve the model , nor do different density distributions ( we tried r@xmath6 and r@xmath60 ) . a good match can be obtained if a maximum grain size of 100 @xmath1 m is introduced ( figure [ fig:17423 ] ) . this is remarkable , since hints for such large grains have been found so far only in a handful of stars , such as the red rectangle @xcite . it is interesting to notice that in another target , iras 20462 + 3416 , our point at 1 cm seems not to follow the free - free curve , being well above it ( see figure [ fig : dusty1 ] ) . if we combine this to the shape given by the iras long - wavelength data points , we can speculate that this target might have an excess of radiation in the sub - mm as well . we could not classify iras 18367 - 1233 because the iras coordinates turned out to be off - set respect to the actual position of the target , as we can see in our irac images . since the irac imaging was not available at the time of the irs observations , the target resulted to be out of the irs slits . modeling our sources allows us to estimate the mass of the envelopes of our targets . we follow here the procedure in @xcite . the mass is estimated as @xmath61 in eq . [ eq : dustmass ] @xmath51 is the inner radius of the envelope , @xmath62 is the outer - to - inner radius ratio , @xmath63 and @xmath64 are the optical depth and the absorption coefficient at 100 @xmath1 m , and @xmath65 is the dust - to - gas mass ratio . no distance information is available for most of our targets , therefore we assume as @xmath51 a typical value of 10@xmath66 cm . @xmath62 is the value given as input in dusty and @xmath63 is given by the code as an output . for @xmath65 , whose value depends on the evolutionary stage , we consider a typical value of 200 . @xmath64 depends on the chemistry of the envelope : we use 92 g @xmath40 for amorphous carbon @xcite and 34 g @xmath40 for silicates @xcite . we have used the values for silicates not only in the targets that we have classified as o - rich , but also in mixed - chemistry environments , neglecting any contribution from c - bearing molecules . as already noticed by @xcite , the code seems to underestimate the amount of cold dust in the envelope or anyhow not to be very sensitive to the contribution from such grains . if we also consider that we lack measurements in the sub - mm / mm range - which would help constrain the cold - dust emission - the values listed in table [ tab : masse ] are to be regarded as lower limits to the circumstellar mass . lc _ c - rich spectra _ + 01005 + 7910 & 20 + 06556 + 1623 & 300 + 09470 - 4617 & 5 + 17423 - 1755 & 100 + 18442 - 1144 & 7 + 19200 + 3457 & 0.7 + _ o - rich spectra _ + 17074 - 1845 & 20 + 17203 - 1534 & 700 + 17364 - 1238 & 100 + 17381 - 1616 & 2 + 17460 - 3114 & 0.008 + 18062 + 2410 & 3 + 18435 - 0052 & 700 + 19157 - 0247 & 5 + 20572 + 4919 & 4 + _ mixed spectra _ & + 17542 - 0603 & 2 + 18371 - 3159 & 10 + 18379 - 1707 & 400 + 19306 + 1407 & 100 + 19336 - 0400 & 5 + 19590 - 1249 & 1 + 20462 + 3416 & 0.7 + 21289 + 5815 & 3 + 22023 + 5249 & 3 + 22495 + 5134 & 200 for iras 18040 - 1457 , 18070 - 2346 , and 19399 + 2312 the iras measurements must evidently include emission from nearby dust , since in our irac images we clearly detect a diffuse nebulosity around these targets ( see as an example figure [ fig:18040 ] ) . these sources have quite different irs spectra , compared to the other targets ( figure [ fig : various ] ) . since iras data are likely contaminated by ambient dust , the location of these targets in the iras color - color diagram is probably wrong , therefore their classification as post - agb candidates appears dubious . in iras 18070 - 2346 , the 18 @xmath1 m silicate feature is detected , although it might be due to the interstellar medium or diffuse nebulosity . iras 19399 + 2312 shows an emission feature around 11.2 @xmath1 m that can be attributed to neutral pah molecules . m in logarithmic scale . the field is about @xmath67 and the units are mjy / sr . north is up and east is left.,width=302 ] the number density of post - agb / proto - pn is estimated to be around 0.4 kpc@xmath68 . as a comparison , the density of agb stars is 15 kpc@xmath68 and of main sequence stars is @xmath102@xmath69 kpc@xmath68 @xcite . knowledge about this phase is mostly derived from a handful of objects , such as the c - rich source rafgl 618 and the o - rich object oh231.8 + 4.2 . because of the paucity of known to , the identification and study of new to is very important for testing models of stellar evolution . among to candidates , we have selected hot targets ( b spectral type ) , therefore searching for recently - ionized or close - to - ionize envelopes . 2mass and iras data were combined with our irac and radio observations to construct the seds of the sample stars . dusty modeling shows the presence of more than one emitting component in the dust envelopes and indicates that dust temperatures range from @xmath101000 k to @xmath10100 k. the modeling also gives hints about the presence of large ( @xmath70 1 @xmath1 m ) grains , although sub - mm and mm observations are necessary to constrain the seds in these ranges , where such grains would give a major contribution . as an example , in iras 17423 - 1755 it is necessary , to account for its mm emission , to have three dust components ( at 1150 , 132 , and 80 k ) with grain size up to 100 @xmath1 m ( see fig [ fig:17423 ] ) . the irs spectra we obtained allow us to classify the to candidates as c - rich , o - rich , or mixed - chemistry . we have classified as mixed - chemistry objects those stars with both a 10 @xmath1 m silicate bump and pah features in the 611 @xmath1 m range . a possible explanation for the mixed chemistry involves the presence of a circumbinary disk / torus , where o - bearing molecules would be preserved from the third dredge up , while elsewhere in the outflow features from c - bearing molecules would arise . this picture appears to be confirmed by our analysis of the pah features , which indicates that pah molecules in our targets are not located close to the central star . since these objects are typically compact ( @xmath102@xmath13 ) , sub - arcsec imaging may evidence if different dust species have different spatial distributions . in most of our targets with o - rich dust , the 10 @xmath1 m silicate emission feature often appears with a structured shape and/or a peak shift to larger wavelengths . our analysis of the shape and strength of this feature indicates a high degree of dust processing . such processing , which typically consists in dust growth and crystallization , indicates again the presence of large dust grains and/or of crystalline silicates and more interestingly is a further hint for the presence of a circumstellar / circumbinary disk : a stable structure where the grains can have the time to grow and crystallize . whereas the expected fraction of mixed - chemistry envelopes is less than 10% , we find that about 40% of the stars in our sample show both pah and silicate features . our sample has been selected on the basis of b optical spectral type and far - ir excess . these two conditions are indeed typical of hot post - agb stars . nevertheless , selections based on infrared thermal emission ( iras fluxes , for instance ) introduce a bias towards far - ir bright objects . among post - agb stars , objects hosting a disk or torus are stronger ir emitters . these circumstellar ( or circumbinary ) structures are highly stable and store the dust around the central star / s on a much longer time - scale than in stellar envelopes without a disk / torus @xcite . as these envelopes are brighter and live longer , selections based on far - ir emission are biased towards stars surrounded by a disk or torus . the higher percentage of mixed - chemistry envelopes in our sample can therefore be connected to a higher percentage of disks ( or tori ) . this confirms the general picture of a disk / torus as a reservoir preserving o - bearing molecules from the third dredge up . if we refer to the red rectangle as an example , mixed chemistry and large grains are linked together by the presence of a disk , where the particles can survive enough orbits to grow by coagulation to sizes as large as 1 mm @xcite . with only a few examples of this phenomenon , several hints for it found in this project provide us with candidate targets to test the current understanding of dust processing in pn envelopes . l. cerrigone acknowledges funding from the smithsonian astrophysical observatory through the sao predoctoral program . this work is based in part on observations made with the spitzer space telescope , which is operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . support for this work was provided by nasa through an award issued by jpl / caltech . we would like to thank an anonymous referee for his criticism , which led to a substantial improvement of the paper . + _ facilities : _

this article explores the agreement between the predictions of modified newtonian dynamics ( mond ) and the rotation curves and stellar velocity dispersion profiles measured by the diskmass survey . a bulge - disk decomposition was made for each of the thirty published galaxies , and a mond poisson solver was used to simultaneously compute , from the baryonic mass distributions , model rotation curves and vertical velocity dispersion profiles , which were compared to the measured values . the two main free parameters , the stellar disk s mass - to - light ratio ( ) and its exponential scale - height ( ) , were estimated by markov chain monte carlo modelling . the average best - fit k - band stellar mass - to - light ratio was . however , to match the diskmass survey data , the vertical scale - heights would have to be in the range to pc which is a factor of two lower than those derived from observations of edge - on galaxies with a similar scale - length . the reason is that modified gravity versions of mond characteristically require a larger to fit the rotation curve in the absence of dark matter and therefore predict a stronger vertical gravitational field than newtonian models . it was found that changing the mond acceleration parameter , the shape of the velocity dispersion ellipsoid , the adopted vertical distribution of stars , as well as the galaxy inclination , within any realistic range , all had little impact on these results . cosmology : dark matter ; galaxies : kinematics and dynamics ; methods : numerical

1505.05522

[ sec : intr ] understanding the dynamics of disk galaxies is essential to the vetting process of theories of galaxy formation and cosmology ( @xcite ) . disk galaxies moderately inclined to the line of sight ( 50 - 80@xmath5 ) can provide an hi rotation curve from which the dark matter ( dm ) content and distribution can be deduced in the context of newtonian dynamics ( @xcite ) . however , there exist degeneracies between the dm halo density profile , the stellar mass - to - light ratios of the stellar components and the scale - height of the stellar disk ( @xcite ) . the vertical stellar distribution is assumed to be a declining exponential on both sides of the mid - plane . thus , the scale - height of the stellar disk is simply the exponential rate at which the stellar luminosity density drops , with increasing height above or below the mid - plane of the disk . it is assumed to be constant with radius , which is supported by observations ( e.g. @xcite ) . the stellar mass - to - light ratio ( @xmath0 ) is the ratio between the mass of a stellar population and the luminosity , as observed through a particular bandpass . it can not be observed directly , so it is ordinarily inferred by numerical stellar population synthesis models . this modelling crucially depends on the star - formation and chemical enrichment history considered and on the initial mass function ( imf ; @xcite ) . the imf is the spectrum of stellar types formed given a molecular cloud of certain initial mass , metallicity and other relevant properties . it therefore has the scope to vary from galaxy to galaxy . in addition to the imf , stellar population synthesis models have other sources of uncertainty ( @xcite ) . in standard fitting of rotation curves of highly inclined disk galaxies , it is common to invoke the `` maximum disk hypothesis '' , ( @xcite ) i.e. that the stellar disk contributes maximally to the rotation curve . this hypothesis is supported by observations that deduce the microlensing optical depth in the milky way ( e.g. , @xcite ) , the baryonic tully - fisher relation ( @xcite ) and measurements that place the co - rotation radius of barred galaxies just beyond the end of the bar ( e.g. , @xcite ) . it yields values for the @xmath0 that are typically in accordance with the predictions of stellar population synthesis models . this , however , does not demonstrate that the hypothesis is correct ( see e.g. @xcite ) and it would be ideal to have a robust , independent measurement that breaks the disk mass degeneracy . this is theoretically possible because the @xmath0 can be determined dynamically through the vertical velocity dispersions of stars over the full projected area of the galactic disk , if the scale - height is known ( @xcite ) . for close to edge - on disk galaxies , however , one can not measure the vertical velocity dispersion and thus the technique is limited to only those galaxies with moderate inclinations to the line of sight ( @xmath6 ) , where @xmath7 is seen as optimal ( @xcite ; hereafter dmsii ) . in order to break the degeneracy between the dm halo and @xmath0 , the diskmass survey ( dms ; @xcite ; hereafter dmsi ) made observations of the line of sight velocity dispersion profiles of 46 nearly face - on disk galaxies ( dmsi ) : 30 of which have been published and a further 100 are part of the larger survey . they also measured their surface brightness profiles and rotation curves . there were two further , essential ingredients in the analysis that come from scaling relations . the first is the luminous tully - fisher ( tf ) relation between the absolute magnitude of a galaxy and a measure of its outer rotation speed . this helps to isolate the inclination of the galaxy , which for nearly face - on galaxies can otherwise be obtained using the tilted disk method of @xcite . the second is the correlation between the disk stellar scale - height , a measure of the thickness of the disk , and scale - length , a measure of the radial extent . this relationship is explored in detail by dmsii ( their 2.2 ) , and is derived from the studies of @xcite ( hereafter k02 ) and @xcite . this allowed @xcite and @xcite ( hereafter dmsvi and dmsvii ) to infer the @xmath0 of the disk . the data imply that the stellar disks are `` sub - maximal '' ( k - band @xmath8 or lower , see @xcite ) which means they do _ not _ contribute maximally to the rotation curve in the central regions , contrary to the value found from population synthesis models that assume a kroupa imf ( @xmath9 ; @xcite ) . this leaves more room for dm in the central regions . modified newtonian dynamics ( mond @xcite , see @xcite for a recent review ) is a theory which proposes a modification of dynamics whose impact is most apparent in regions of low acceleration . most current working versions of mond consist of an actual modification of gravity , i.e. , at the classical level , a modification of the newtonian poisson equation ( but see also @xcite ) . this modification occurs due to the hypothesised existence of a new constant of physics with dimensions of acceleration , @xmath10 . for accelerations much stronger than this threshold , @xmath11 , there is no discerned deviation from newtonian gravity . however , far below the threshold the true acceleration perceived by a test mass is found from @xmath12 - where @xmath13 is the expected newtonian gravitational field . @xcite and @xcite made studies of the vertical dynamics of the galaxy in mond , showing that it could be possible to distinguish between mond and the dm paradigm with data from the milky way . the extra constraint on the dynamics from vertical velocity dispersions in nearly face - on disks of external galaxies provides a new test of this hypothesis . this article addresses whether mond can simultaneously account for the measured vertical velocity dispersions and rotation curves , while keeping in line with galaxy scaling relations . in section 2 the framework is presented for the joint modelling of galaxy rotation curves and stellar vertical velocity dispersions in the mond context . in section 3 the methods are discussed and this includes a discussion of the bulge - disk decomposition , the accuracy of the poisson solver , and the observational error budget for the main data . in section 4 the primary results are presented , this includes a discussion of the fits to the vertical dynamics and rotation curves , the confidence ranges of the fitted parameters , and how well the fitted parameters mesh with other observations . in section 5 possible scenarios that could alter the results are discussed . in section 6 conclusions are drawn and their implications are explored . [ sec : dyn ] [ sec : rcf ] the following reviews how to fit the measured rotation curve of a disk galaxy in an idealised case . this is done in order to expose the free parameters and it is generalised for both dm and mond . the total model rotation speed is required from the total model potential in order to compare with the measured rotation curve . this can be found from [ eqn : vrot ] v_tot(r)^2 r= d_tot dr(r ) . here , @xmath14 is the cylindrical radius in the disk mid - plane and @xmath15 is the total gravitational potential . next , the total potential is required from the total mass distribution . this is computed via the poisson equation , which in newtonian dynamics is [ eqn : newtpois ] ^2_n , tot(r , z)=4g_tot(r , z ) , where @xmath16 is newton s gravitational constant , @xmath17 is the total mass density from all sources ( see the following and [ sec : bardens ] ) and @xmath18 is the total _ newtonian _ potential . in newtonian dynamics , @xmath18 is fully equivalent to @xmath19 , but @xmath20 consists of @xmath21 and @xmath22 . in mond , @xmath20 is fully equivalent to @xmath21 , but a second step is made to find @xmath15 from @xmath23 , which is ( @xcite ) [ eqn : mondpois ] ^2_tot(r , z)= , where @xmath24 is the acceleration threshold of mond and @xmath25 is an interpolating function , chosen here to depend on its argument as [ eqn : nualp ] _ ( y)=^1/ , where @xmath26 is the simple @xmath25-function and @xmath27 is the standard @xmath25-function ( @xcite eqs . 51 and 53 ) , or likewise by [ eqn : nudel ] _ ( y)=^-1/+(1-^-1)e^-y^/2 . [ sec : bardens ] the last question here is the distribution of @xmath20 . the derivation of the contribution from atomic and molecular gas is described in dmsvii . both are assumed to have non - smooth , axisymmetric , radial surface densities . both gas components are included in the modelling , but they are considered fixed in mass and distribution . they are given nominal scale - heights of 200 pc , to which reasonable variations are inconsequential . there are typically two stellar components , a bulge and a disk . below , the bulge and disk surface brightnesses are given to expose the free parameters in their fitting . the bulge is assumed to be spherical and to follow a profile [ eqn : sersic ] i_b(r)=i_e\{-k[(r / r_e)^1/n-1 ] } , where the effective surface brightness ( @xmath28 ) , the effective projected radius ( @xmath29 ) and the index ( @xmath30 ) can be fitted to the observed surface brightness distribution . is fixed at 7.67 . ] simultaneously , the disk luminosity density can be expressed with a simple form [ eqn : jdisk ] j_d(r , z)=l_d 4h_r^2h_z(-r h_r)(-|z| h_z ) , where @xmath31 is the total luminosity of the disk , @xmath32 is the scale - length and @xmath1 is the scale - height . the vertical distribution is characterised by the exponential function , but the commonly used @xmath33 function ( @xcite ) would be equally appropriate . here @xmath34 is the @xmath35 scale - height which corresponds to @xmath36 at large @xmath37 . using the @xmath35 vertical distribution does not change the conclusions ( see [ sec : sech ] ) . in general , the surface density can be found by integrating along the line of sight . for a face - on galaxy , the luminosity density of eq [ eqn : jdisk ] can be projected to give the surface brightness [ eqn : sigdisk ] i_d(r)=l_d 2h_r^2(-r h_r ) . here the density profiles are always assumed to be smooth . to find the mass density of the bulge and disk , the luminosity densities of the bulge and the disk must be multiplied by their respective @xmath0 so that [ eqn : massdens ] _ * = _ bj_b+_dj_d , where @xmath38 and @xmath39 are the stellar @xmath0 values of the bulge and disk respectively . the total density of baryons is then @xmath40 , where @xmath41 is the atomic and molecular gas density . as stated previously , in mond , @xmath21 is equivalent to the total mass density , @xmath20 ( eq [ eqn : newtpois ] ) , because there is no dm in mond galaxies . for newtonian gravity @xmath42 . for the baryonic mass models , there are 8 parameters : @xmath43 . of those 9 , the surface brightness parameters are either directly observed or unambiguously fitted to the surface brightness profile . this leaves only @xmath44 . since the dms sample is chosen so that the total bulge luminosity to total disk luminosity is low , @xmath38 is relatively insignificant and is never independently varied in the modelling performed here ( @xmath45 ) . therefore , @xmath46 and @xmath1 are the only two free parameters that are relevant to a theory like mond . in principle , the inclination of the galaxy also has a small amount of freedom but it is strongly curtailed by the luminous tf relation . the aforementioned free parameters are fitted for through a simultaneous comparison of the model vertical velocity dispersions and rotation curves with the observed ones , as is described in [ sec : errors ] . [ sec : inc ] the derivation of the rotation curve of a moderate or high - inclination disk galaxy from the measured 2d velocity field , permits the fitting of tilted rings ( see @xcite ) . these tilted rings allow us to model the variation in the inclination and position angle of numerous concentric annuli at different galactocentric radii . the inclinations of the various rings , as a function of radius , can vary by 10@xmath5 ( e.g. @xcite ) depending on the quality of the data , the regularity of the velocity field and characteristic inclination of the disk . the dms galaxies have low inclinations ( they are close to face - on ) , thus the inclination has a lot of leverage on the inferred rotation speed because of the shape of the sine function . it is possible to derive accurate kinematic inclinations for nearly face - on disks using the tilted disk ( as opposed to tilted ring ) technique of @xcite . it is also possible to infer the inclination using the luminous tf relation ( @xcite ) . this relates the absolute k - band magnitude of the galaxy , @xmath47 , to a measure of the outer rotation speed , @xmath48 such that [ eqn : tfr ] v_f=0.5 10^(5.12-m_k)/11.3 . anderson et al . ( 2015 , in prep ) have determined that the kinematic and tf inclinations for the diskmass galaxies generally agree well , although there are some outliers . relating the measured , inclined outer rotation velocity @xmath49 with the expected outer velocity from the tf relation ( eq [ eqn : tfr ] ) allows the expected inclination to be deduced . this inclination is only that expected for the outer parts of the rotation curve , and thus the inner parts can vary somewhat due to a warp . in addition to the luminous tf relation , there is a baryonic tf relation ( @xcite ) which relates the total baryonic mass of a galaxy to its outer , flat rotation speed , where @xmath50 . this relation is fundamental to mond and the exponent @xmath51 and the constant of proportionality @xmath11 are predictions which agrees well with the observed relation ( @xcite ) . thus , the mond baryonic tf relation can be written in a similar form to eq [ eqn : tfr ] as [ eqn : btfr ] v_f^4 ga_0=m_g+(_bf_b+_df_d)10^(m_k ,- m_k)/2.5 . here , @xmath52 and @xmath53 are the fractions of the total luminosity contributed by the bulge and disk respectively . the absolute magnitude of the sun in the k - band is @xmath54 ( @xcite ) and @xmath55 is the gas mass . it was found that the inclinations from eq [ eqn : btfr ] are typically between 5 - 15% larger than those found with eq [ eqn : tfr ] , depending on the @xmath0 used ( here the @xmath0 was taken to be between 0.6 and 1 ) : a smaller @xmath0 implies a smaller corresponding rotation velocity , hence a larger inclination . for a large enough sample there should , in principle , be no systematic deviation from either tf relation . when modelling rotation curves in general , it is not always clear when the rotation curve has reached the terminal velocity , so some margin of error must be granted . since eq [ eqn : tfr ] has no dependence on @xmath0 , and to make it easier to compare with the dms results , the luminous tf relation inclinations ( eq [ eqn : tfr ] ) are used in this article . [ sec : svvd ] [ sec : csve ] in addition to the measured rotation curve , the dms also measured the line of sight velocity dispersion profile of the stars over the full projected area of the disk . this is then azimuthally averaged to give a 1d line of sight velocity dispersion , which can be converted to a vertical velocity dispersion ( @xmath56 ) through the equation ( @xcite ) [ eqn : vellip ] _ z^2=_los^2 ^2i^-1 . here , the inclination of the galaxy to the line of sight is again , @xmath57 , and @xmath58 & @xmath59 provide information about the stellar velocity dispersion ellipsoid . generally , @xmath60 and @xmath61 are expected to take on certain values from measurements in the solar neighbourhood ( @xcite ) , and @xmath61 is presumed to take on specific values from the epicycle approximation . however , beyond the milky way their variation is not empirically well known ( see discussion in sect 2.1 of dmsii , and also @xcite ) . by choosing galaxies that are nearly face - on , the dms reduces their importance ( cf . fig [ fig : vvdvar ] ) . the statistical variation of these parameters is discussed in dmsii 2.1 , and the dms analysis establishes @xmath62 and @xmath63 . the mean values used by the dms are chosen for the default values in this article . [ sec : modsvvd ] in order to compare with the observations , the model vertical velocity dispersions of the galaxies must be computed . the vertical velocity dispersion at a height @xmath37 above the mid - plane , at a radial distance @xmath14 from the centre of the disk galaxy is found from ( see @xcite ) [ eqn : rho * ] _ * ( r , z)_z(r , z)^2=_z^_*(r , z)d_tot(r , z ) dzdz , and the equivalent of the observed vertical velocity dispersion at any radius , r , weighted by the local stellar surface density is given by [ eqn : sig * ] _ * ( r)_z(r)^2=_-^_*(r , z)_z(r , z)^2 dz . these equations effectively reduce to [ eqn : sigred ] _ z(r)^2=1 h_z_0^dz . in the newtonian gravity framework , eq [ eqn : sigred ] depends mainly on 2 fitted parameters : @xmath39 and @xmath1 . as stated previously , inclination could also be varied but only in a tight range around the tf relation values . thus , combining simultaneous fits to the observed rotation curves and vertical velocity dispersions is a strong test of the mond paradigm . it is worth noting that the infinite potential well of isolated galaxies in mond is irrelevant here since the vertical gravitational field in eq [ eqn : rho * ] is convolved with the exponentially declining stellar density , and thus the mond gravity is only relevant where there are stars . [ sec : dmsmethod ] in the analysis of dmsvi , the measured rotation curve of each galaxy is used to fit the dm halo parameters and then the vertical velocity dispersion is used to directly give the mass surface density using @xmath64 , where @xmath65 is assumed to be 1.5 to describe an exponential vertical stellar distribution . from this surface density , the gas disk was subtracted . this left the stellar disk surface density , @xmath66 . this also includes an unknown component of dm . the stellar @xmath0 as a function of radius was given by @xmath67 . an average of this @xmath0 out to a given radius then defines the quoted disk @xmath0 . this approach , although straight - forward , is only accurate when the derivative of the rotation curve is small . it is also not transferable to mond because of the non - linearity of the theory and that the @xmath0 in mond affects both the rotation curve and vertical velocity dispersion . it is more secure to make the reverse calculation and go from observed surface brightness , sample a @xmath0 to give surface density , then use eq [ eqn : sigred ] to give a model vertical velocity dispersion which can be compared with the observed vertical velocity dispersions . however , this requires calculation of the full three dimensional potential from a model galaxy - which is performed here . [ sec : bdd ] the inclination corrected surface brightness profiles presented by dmsvi were analysed with a markov chain monte carlo ( mcmc ) approach to fit bulge plus disk surface brightness models using eqs [ eqn : sersic ] and [ eqn : sigdisk ] . the seeing was simultaneously accounted for with a gaussian convolution of appropriate radius ( see tables 3 and 4 of dmsvii ) . all 5 surface brightness parameters ( 2 disk and 3 bulge ) were fitted , but the bulge surface brightness parameters were found to be degenerate with each other due to seeing effects and the lack of data points at low radii . thus , the bulge parameters were fixed at the maximum likelihood values , which allowed better sampling of the more important disk parameters . in figs [ fig : like0]-[fig : like2 ] the posterior probability distributions are presented for the two disk surface brightness parameters of each galaxy . for the disk scale - length , the range found by dmsvi ( red curve ) is also plotted . the best fits to the surface brightness profiles are given in figs [ fig:1like0]-[fig:1like2 ] and are generally good . both linear radius and log radius are plotted on the @xmath68-axis to expose the quality of the fits to the bulge and the outer disk . the fitted scale - lengths found here ( see figs [ fig : like0]-[fig : like2 ] ) are generally consistent with those fitted by the dms , although theirs have smaller error ranges . once the bulge and disk surface brightness profiles were established , an abel transform was used to de - project the bulge surface brightness , which was stored numerically . the disk has cylindrical symmetry , so it does not require de - projection . following this , rejection sampling was used to generate n - particle representations of the galaxies , with half of all particles representing the stellar disk and the other half split evenly between the stellar bulge and the atomic and molecular gas disks . the fitted scale - lengths are used for all the following analysis , but they are not used to generate the n - particle representations of the stellar disk . instead , the `` observed '' surface brightness of the disk is sampled after subtracting the fitted bulge - since the fit to the surface brightness can be poor at large radii . in this analysis the static n - particle representations are only used to find the potential and gravitational field , there is no n - body evolution of the simulated galaxies . the masses recovered after generating the n - particle representations of each component of each galaxy generally agreed very well with the masses reported by the dms . the mond poisson solver described in @xcite is used to compute the radial and vertical gravitational fields of the n - particle galaxy models . the code solves the modified poisson equation of the quasi - linear version of mond ( qumond ) given by eq [ eqn : mondpois ] . the code uses a 3d grid and the cloud - in - cell technique to numerically discretise the 3d density of an n - particle distribution . it then employs finite differencing and multigrid methods to iterate from a test potential to the final potential which accurately reflects the density . [ sec : galmods ] the poisson solver is used to compute the gravitational field of the baryons in the radial direction , @xmath69 , as a function of radius , to find the model rotation speed ( eq [ eqn : vrot ] ) . simultaneously , the gravitational field of the baryons is solved for in the vertical direction , @xmath70 , as a function of height above the disk , at several discrete radii : @xmath71 to 9.5 kpc in steps of 1 kpc and then @xmath72 , 14.5 , 18.5 and 25.5 kpc . in mond , both gravitational fields noted above are equivalent to the gradient of the total potential ( i.e. @xmath73 ) used in eqs [ eqn : vrot ] and [ eqn : rho * ] and thus the rotation speed can be straight - forwardly calculated and eqs [ eqn : rho * ] and [ eqn : sig * ] can be integrated to find @xmath56 at different radii , @xmath14 . the most important , and difficult to compute , quantity is the vertical gravity profile at large radii . for an isolated , double exponential disk , like that introduced in eq [ eqn : jdisk ] , the newtonian vertical gravity profile at a given radius can be calculated numerically - as described in detail by @xcite . in the top panel of fig [ fig : grav ] the newtonian vertical gravity profile is plotted for three discrete cylindrical radii @xmath74 , 8.5 and 18.5 kpc . the left and right hand panels correspond to two different double exponential disk galaxies ( as per eq [ eqn : jdisk ] ) . the left hand panel has scale - length @xmath75 kpc , scale - height @xmath76 kpc and total mass @xmath77 . the right hand panel has scale - length @xmath78 kpc , scale - height @xmath79 kpc and total mass @xmath80 . the computation from the poisson solver using newtonian dynamics is the thick black line and the theoretical value using eq 27 of @xcite is the red line . the agreement is generally very good , except for small heights above the disk . this is due to a combination of the limit of spatial resolution at large radii , owing to the centrally refining mesh , and limited particles at large radii . the average percentage error for these discrete radii ( 3.5 , 8.5 , 11.5 , 14.5 , 18.5 and 25.5 kpc ) between the numerical calculation of @xmath56 ( using eq [ eqn : sigred ] ) and the theoretical values are 0.05 , 0.8 , 0.7 , 5.8 , 5.2 and 4.4% . given that only 6 of the 30 galaxies have vertical velocity dispersion data points beyond 11 kpc and the errors on those data points are typically more than 10% , this is more than adequate . in the bottom two panels of fig [ fig : grav ] the newtonian ( black lines ) and mond ( @xmath26 ; turquoise lines ) vertical gravity profiles are compared , both of which were computed with the poisson solver . the mond profiles are strongly boosted relative to the newtonian profiles , however , the higher surface density galaxy ( left hand panel ) receives less of a boost than the other . [ sec : errors ] the extra information that allows the dms to close their set of equations is that , in general , disk scale - heights are observed to correlate with their scale - lengths . these scale - heights and scale - lengths have been fitted to edge - on galaxies and certain assumptions are made in their modelling ( like constant inclination , that the disks are well described by exponential or @xmath35 distributions ) . the dms made an analysis of literature measurements and derived a simple relationship between @xmath32 and @xmath1 such that in units of kpc ( see dmsii ) [ eqn : kreg ] h_z ~0.2h_r^0.633 . this relation has a 1@xmath81 scatter of roughly 25% ( dmsii ) . in the analysis presented here , the scale - height and disk @xmath0 are fitted to the observed vertical velocity dispersions and rotation curves of the sample of 30 galaxies . after the distribution of fitted scale - height and scale - lengths is known , they will be compared for consistency with the direct observations of @xmath32 and @xmath1 for the sample of edge on galaxies compiled by the dms . therefore , a simultaneous fit must be made to the observed vertical velocity dispersions and rotation curves . a critical concern is the contribution of each of the two datasets to the overall likelihood . since the vertical velocity dispersions are the primary data set , the errors on each data point as taken as computed by the dms . given that there is the potential for some systematic errors from disk warping , the rotation curve should not be given too large a weighting , especially the inner parts . the decision was made to increase the error bars on the rotation curve data points to 10 @xmath82 . lastly , the two separate reduced @xmath83s from fitting the vertical velocity dispersion profile and the rotation curve are combined . therefore , the likelihood is given by [ eqn : like ] -2log_e&=&n_rc^-1_i=1^n_rc(v_mod(r_i)-v_obs(r_i ) 10 km / s ) ^2 + & + & n_vvd^-1_j=1^n_vvd(_z , mod(r_j)-_z , obs(r_j ) _ z , obs , error(r_j ) ) ^2 , where @xmath84 and @xmath85 are the number of relevant data points in the rotation curve and vertical velocity dispersion profile respectively . the prior then multiplies @xmath86 to give the un - normalised posterior probability . [ sec : primary ] [ sec : parlike ] mcmc sampling was used to find the posterior probability distribution for each of the free parameters : disk scale - height ( @xmath1 ) , disk @xmath0 ( @xmath46 ) and inclination ( @xmath57 ) . the scale - height was varied by considering the ratio of the fitted scale - height to the one derived from observations ( eq [ eqn : kreg ] ) . a broad gaussian prior of width 1.5 was placed on this ratio . the prior placed on the disk @xmath0 was centred on @xmath87 and had a width 0.5 @xmath88 . the ratio between the fitted inclination and the luminous tf relation inclination ( from eq [ eqn : tfr ] ) received a fairly tight gaussian prior of 0.15 ( or 15% ) , since there should be little deviation from the tf relation . the main reason for including inclination as a free parameter is just in case there is a sharp increase in posterior probability for a small change in inclination . the full expression for the likelihood is given by eq [ eqn : like ] . this defines how the goodness of fit for the two data sets is combined : rotation curve and vertical velocity dispersion . the re - normalised posterior probability distributions of each of the three parameters ( scale - height , inclination and @xmath0 ) is plotted in figs [ fig:2like0]-[fig:2like2 ] for each galaxy individually . there are two main lines in each of the three columns : mond with @xmath26 and 2 ( see eq [ eqn : nualp ] ) black lines , solid and dashed respectively . the second column ( inclination ) also has the curve of the prior ( turquoise ) . the third column ( m / l ) has a green line which represents the newtonian gravity ( with dm halo ) @xmath0 found by the dms ( dmsvi ) , shown only for reference . all the lines are re - normalised to have the same maximum posterior probability . the two mond fits vary little in terms of goodness of fit . from the left hand column it is clear that the majority of the fits require substantially lower scale - heights than those derived from observations ( eq [ eqn : kreg ] ) , which in those panels are unity . there is a preference with most galaxies to have a higher inclination because that decreases the amplitude of the rotation curve . a lower rotation curve requires a lower @xmath0 . a lower @xmath0 allows a larger scale - height - which allows better agreement with the observations of edge - on galaxies ( eq [ eqn : kreg ] ) . however , the fairly tight prior on the inclination prevents it from changing substantially . usually it changes by less than 10% . in table [ tab : sims ] various 1@xmath81 confidence ranges for the fitted scale - heights and stellar mass - to - light ratios are presented using mond with varying constraints ( from interpolating functions and which parameters are left free ) . the preferred scale - heights , those from eq [ eqn : kreg ] and used by the dms , are given for each galaxy ( column 2 ) . for reference ( column 3 ) the @xmath0 found by the dms when using newtonian gravity and dm halos is shown , followed by the best fit mond scale - heights ( as a ratio between the fitted scale - height and the one derived from observations of edge - on galaxies - eq [ eqn : kreg ] ) and the @xmath0 for three different interpolating functions ( the two used above and one other ) . these fits are for the scenario where the scale - height and @xmath0 are barely constrained , but the inclination is tightly constrained by its prior . in figs [ fig:3like0]-[fig:3like2 ] the best fits to the vertical velocity dispersion ( first and third columns ) and rotation curve ( second and fourth columns ) are plotted in the mond case with @xmath26 . the two interpolating functions are not plotted together because different best fit inclinations mean the data points vary in each case . there is little difference between the two mond cases in terms of quality of fit . most fits to the vertical velocity dispersion are good , but the quality of fits to the rotation curves range from poor to good . rotation curves with good fits include the galaxies ugc 448 , 1081 , 1087 , 1908 , 3091 , 4036 , 4368 , 4380 , 7244 , 6918 , 8196 , 9965 , 11318 , 12391 . quite good fits are found for ugc 463 , 1529 , 1635 , 1862 , 3701 , 4256 ; whereas the fits to ugc 3140 , 3997 , 4107 , 4458 , 4555 , 4622 , 6903 , 7917 , 9177 , 9837 are poor . mond fits to the rotation curves of nearly face - on galaxies are very sensitive to the inclination ( see e.g. @xcite ) . in the outer parts , they depend on the fourth power of @xmath89 . warping of the disk might be invoked to improve certain rotation curve fits , however , in some cases the dms galaxies have already been corrected for warps . to compare with models that use the scale - height from eq [ eqn : kreg ] , the vertical velocity dispersion ( red line ) is over - plotted for each galaxy using that larger scale - height , but the @xmath0 and inclination of the best fit are kept . in many cases , the red line has a far greater amplitude than the data points and best fit ( black line ) . there are , however , a few cases where the red and black lines are close together , such as ugc 1862 , ugc 4368 and ugc 4458 - although the latter only has a single relevant data point . from fitting the observed surface brightness profiles of the dms galaxies , a value for the radial scale - length was estimated for each galaxy disk . as can be seen in figs [ fig : like0]-[fig : like2 ] ( second and fourth columns ) the radial scale - length fitted by dmsvi is generally slightly larger than the best fit found here , but the majority are consistent within the errors . this fact makes the analysis presented here slightly more favourable to mond . the difference in fitted scale - lengths is in part due to the fact that in dmsvi a finite central region was used to fit the scale - length , whereas the full extent of the galaxy is used here . thus , for any galaxies whose disk surface brightness is not well fit by a single exponential there is a difference . for example , ugc 4036 has a shallow inner surface brightness profile and then a steep outer decline , thus the scale - length found here naturally has a smaller scale - length . recall that the n - particle realisations ( [ sec : bdd ] ) use the observed surface brightness , not the fitted exponential disk . similarly , by fitting the vertical velocity dispersion profile and the rotation curve of each dms galaxy , a confidence range for the scale - height of each galaxy has been derived . in principle , this combination of @xmath1 and @xmath32 parameters can be compared to the measurements of scale - lengths and scale - heights of a sample of edge - on galaxies . in fig [ fig : simp ] the @xmath1 vs @xmath32 diagram for the mond fits ( @xmath26 , @xmath24 ) is plotted along with the observations of these parameters for edge - on galaxies made by k02 . in order to visualise this more plainly , contours in the @xmath1 vs @xmath32 plane were added to fig [ fig : simp ] . here , the re - normalised posterior probability density of all individual data points are co - added on a fine grid , with each galaxy receiving equal weighting . a point with larger error bars will spread its probability density over a larger range . the blue contours represent the co - added mond probability densities and the black triangles give the location of each individual galaxy point . the error bars are not included to minimise clutter . the measurements by k02 have red shaded contours with green circles . the mond contours are significantly inclined relative to the k02 contours . the equivalent plot for the @xmath27 interpolating function is almost identical . [ sec:2dks ] to demonstrate statistically the the large offset in the @xmath1 vs @xmath32 plane between the k02 data points in fig [ fig : simp ] and the fitted mond points , a 2d kolmogorov - smirnov ( ks ) test ( see @xcite ) chapter 14.7 ) was employed . in this test , the first k02 data point was taken and made the origin of four quadrants . then the fraction of k02 points in each quadrant and the fraction of mond points in the same quadrants . the largest difference between the two fractions in a single quadrant is stored and then the rest of the k02 points are cycled through , making each one the origin in turn . the largest of the largest differences is stored and then the procedure is repeated centring on the mond points . the largest difference in fractions is ordinarily the same regardless of whether the k02 or mond points are central . this number is the statistic of merit for the 2d ks test and provides a significance level that the two samples originate from the same parent population . since the data points have fairly large error bars , if the mean of a particular data point lies in the first quadrant , it is not ideal to add the whole point to that quadrant . instead , the fraction of the point inside each quadrant determined by the point s error bars is added . according to the 2d ks test , the significance levels that the mond fits with @xmath26 and 2 come from the same parent distribution as the k02 sample are @xmath90 and @xmath91 respectively . using the interpolating function @xmath92 the significance level is @xmath93 . comparing with dm models , it is easy to obtain significance levels of order unity , so the method seems robust . as a separate test , a straight line was fitted to each data set of @xmath1 vs @xmath32 with a fixed intercept . the resulting gradients differ at the 2@xmath81 level , regardless of interpolating function . using a variable intercept only improves the agreement slightly . [ sec : superpos ] in the approach presented here to modelling the dms data the @xmath0 is fitted for as well as the scale - height . at near infrared wavelengths a smaller variation of the @xmath0 is expected from one galaxy to the next than at optical wavelengths ( @xcite ) . to find the posterior probability distribution of @xmath0 across the full sample of galaxies , the summation of the re - normalised posterior probability density for the @xmath0 of all 30 galaxies is plotted in fig [ fig : cumfits ] . this is done for three mond interpolating functions and the dms computation of @xmath0 in the dm scenario . this means the same weighting is given to each galaxy s @xmath0 posterior probability density in the summation . recall that in the fitting process a gaussian prior is used with width 0.5@xmath88 centred on @xmath94 . the @xmath26 mond interpolating function ( solid black line ) has an average best fit @xmath2 , whereas the @xmath27 mond interpolating function ( dashed black line ) requires significantly larger values , @xmath95 . bear in mind that the @xmath27 produces less of a boost to the gravity than the @xmath26 interpolating function at intermediate gravities around @xmath96 . the @xmath92 interpolating function ( which uses eq [ eqn : nudel ] ) has a similar @xmath0 distribution as @xmath26 . there is no impact on the derived @xmath0 values from the weak prior . the green line , given only for reference , is the @xmath0 derived from the dms analysis with newtonian gravity and dm . mond fits for the interpolating functions using eq [ eqn : nualp ] with @xmath26 and 2 are represented by black solid and dashed curves respectively and @xmath92 ( from eq [ eqn : nudel ] ) has a dotted curve . the co - added @xmath0 found by the diskmass survey in dmsvi has a green curve.,width=321 ] the results presented here suggest that mond requires disks that have roughly half the vertical scale - height as those inferred from observations of edge on galaxies ( dms ii or eq [ eqn : kreg ] ) . there are two key differences between the mond vertical velocity dispersions and those using dm . the first effect is related to the required @xmath0 of the galaxies . comparing a disk with the same surface brightness in mond and newtonian gravity , the newtonian disk is surrounded by a dark matter halo . this dark matter halo can account for the radial gravitation field that produces the observed rotation speed . this means the @xmath0 of the newtonian disk is only weakly constrained by the rotation curve ( @xcite ) . in mond , the @xmath0 is essentially fixed by the rotation speed because it is the only parameter that can be varied to allow a match between the observed and expected rotation speeds . based on the results presented here , the required @xmath0 of the disk is a factor of between 1.5 and 2.5 higher in mond ( cf . fig [ fig : cumfits ] ) , depending on the interpolating function used . since the @xmath0 is typically much larger in mond , the resulting vertical gravitational field is larger than the dark matter equivalent . moreover , even if the disks in mond and newtonian gravity had the same @xmath0 , the mond gravitational field is still boosted relative to the newtonian one as seen in fig [ fig : grav ] . this follows because the nearly spherical dark matter halo typically has a weak influence on the vertical gravitational field within the disk . the influence of mond on the vertical gravitational field , relative to newton , for the same disk is of course greater when the surface density is low . this is demonstrated by comparing the vertical gravitational field of a high surface brightness galaxy ( fig [ fig : grav ] bottom left panel ) with a lower surface brightness one ( fig [ fig : grav ] bottom right panel ) . in relation to the equations of [ sec : dyn ] , it can be seen that increasing @xmath46 or @xmath1 increases the amplitude of the model @xmath56 . however , in mond only increasing @xmath46 increases the model rotation curve at intermediate to large radii . thus , once @xmath46 is fixed by the rotation curve , only @xmath1 can be varied to fit the observed vertical velocity dispersion . since the observed vertical velocity dispersions are typically much lower than predicted by mond , @xmath1 must be decreased to fit them . it is for these reasons , the fitted scale - heights in mond are significantly lower than those expected from observations ( eq [ eqn : kreg ] ) . @xcite used the dms data to present a correlation between the extrapolated disk central surface luminosity density , from the photometry , and the extrapolated disk dynamical central surface mass density ( see [ sec : dmsmethod ] ) . the majority of galaxies are consistent with the relation @xmath97 , but at the low surface luminosity density end a clump of 6 galaxies lie above the relation . in the newtonian context , this can be attributed to greater relative dm dominance for low surface luminosity density galaxies ( @xcite ) , but this is also reminiscent of the mond transition in rotation curves of spiral galaxies , where the effect becomes pronounced at low surface densities . thus , a prediction for dynamical surface density versus central surface luminosity density in mond would be valuable . to this end , a representative galaxy from the dms was selected - ugc 4036 . the stellar bulge and all gas was removed from the n - particle realisation , thus it has the same stellar disk surface brightness profile as ugc 4036 , and the same vertical profile . for this model , the central vertical velocity dispersion , @xmath98 from eq [ eqn : sigred ] , can be found for a chosen @xmath0 . from here , the surface brightness of the stellar disk can be scaled to compute @xmath98 at a broad range of central surface luminosity densities which allows a comparison with the data from @xcite , which is done in fig [ fig : dmdvar ] . the data points come from @xcite fig 1 ( left hand panel ) and both coordinates are extrapolated from exponential fits to the observed projected disk surface luminosity density and observed vertical velocity dispersion . the red line is the relation @xmath97 . the black solid line is the mond prediction with @xmath99 and the scale - height expected from observations of edge - on galaxies ( dmsii ; eq [ eqn : kreg ] in this paper ) . the black dashed line has the same @xmath0 , but used disks that are twice as thin , which mond has been shown here to require in order to mesh with the dms data . the two blue dashed lines also use disks twice as thin as observations of edge - on galaxies , with the upper line using @xmath100 and the lower line using @xmath101 . thus , mond provides a natural explanation for the deviation of low surface luminosity density galaxies from the central dynamical surface density versus central surface luminosity density relation . . the other lines show mond predictions : the dashed lines use scale - heights that are twice as thin as those derived from fits to observations of edge - on galaxies ( eq [ eqn : kreg ] ) , and the solid black line uses the scale - heights derived from fits to observations of edge - on galaxies . the black lines all use @xmath99 . the upper blue dashed line uses @xmath100 and the lower one uses @xmath101 . the apparent flattening of the central velocity dispersions at low surface brightnesses is consistent with mond.,width=321 ] in this section the impact of other variables on the results from [ sec : primary ] is considered . [ sec : vsve ] as discussed in [ sec : svvd ] , the stellar velocity dispersion ellipsoid ( sve ) parameters are fixed to those used by the dms i.e. @xmath102 and @xmath103 . to demonstrate the degree to which a reasonable variation in these parameters might affect the results , in fig [ fig : vvdvar ] the vertical velocity dispersion ratio between three combinations of parameters and the default values are plotted against inclination . using different values for @xmath60 in eq [ eqn : vellip ] can make a sizeable difference to the derived @xmath56 , especially for large inclinations . the dotted line of fig [ fig : vvdvar ] shows that increasing @xmath60 to 0.9 , which is 2@xmath81 above the default value used by dmsvi , would increase the derived @xmath56 by around 10% for standard dms inclinations of @xmath104 . on the other hand , decreasing @xmath60 to 0.48 and using @xmath105 ( dot - dashed line in fig [ fig : vvdvar ] ) , which was the value derived by @xcite for ugc 463 , would _ decrease _ the derived @xmath56 by around 15% for @xmath106 . found from eq [ eqn : vellip ] , using certain @xmath60 and @xmath61 parameter combinations , and the default values used by the dms ( which are @xmath102 and @xmath103 as per dmsvi ) against inclination . the combination of @xmath60 and @xmath61 used for each line is made clear in the figure . varying @xmath60 can have a significant impact on the derived velocity dispersion for increasingly inclined galaxies.,width=321 ] to test the impact of these sve parameters on the fits to the vertical velocity dispersions and rotation curves , mcmc modelling was used to simultaneously fit @xmath0 , @xmath60 and @xmath61 , with inclination fixed by the tf relation and @xmath107 set by observations of edge - on galaxies ( eq [ eqn : kreg ] ) . here , only flat priors are used such that @xmath108 and @xmath109 . the @xmath0 has a gaussian prior of width 0.5 centred on 0.3 . the most relevant parameter is @xmath60 , due to its impact on eq [ eqn : vellip ] . in the top panel of fig [ fig : alpbet ] the co - addition ( for all 30 galaxies ) of the re - normalised posterior probability is plotted as a function of @xmath60 . each of the 30 galaxies are given equal weighting , regardless of their individual posterior probability . this shows the impact of increasing @xmath60 from the nominal value of 0.6 used here and by the dms . the posterior probability increases roughly a factor of two and a half from @xmath102 to 1 . this increase does not necessarily mean that the fit qualities are good . to demonstrate this , it is worth comparing the fits where @xmath60 is a free parameter ( but scale - height is fixed ) to the fits where the scale - height is free . these fits where scale - height is free represent , for the majority of galaxies , a good quality fit to the vertical velocity dispersion . for the remainder of this section , this fit is referred to as the _ benchmark _ fit . @xmath60 does not influence the fit to the rotation curve . in the bottom panel of fig [ fig : alpbet ] the @xmath110-axis is the ratio between the posterior probability of the fit with fixed scale - height ( eq [ eqn : kreg ] ) and inclination ( luminous tf relation ) , @xmath111 and best fit @xmath61 and @xmath0 to the benchmark fit . this is plotted against the inclination of the galaxy derived from the luminous tf relation to show that increasing @xmath60 may help with some ( but not all ) of the high inclination galaxies , but the majority of the sample would have extremely poor fits to the dms data even with @xmath111 . . ( bottom panel ) ratio between the posterior probability of the mond fit to the dms data with fixed scale - height and inclination , but @xmath111 ( along with best fit @xmath61 and @xmath0 ) , to the benchmark fit . this is plotted against the inclination of the galaxy derived from the luminous tf relation . the mond parameters @xmath26 and @xmath112 are used.,width=321 ] a relevant question is how far the fitted inclination would have to deviate from the luminous tf relation before a good fit to the dms data could be achieved , whilst remaining consistent with the @xmath1 values derived from fits to the photometry of edge - on galaxies . to do this , the scale - height was fixed to the aforementioned value ( eq [ eqn : kreg ] ) for each galaxy and fitted only for the inclination and @xmath0 . the parameters @xmath27 and @xmath112 were used . for the sake of brevity , in fig [ fig : fixhz ] only the co - added , posterior probability densities for the two parameters are plotted . the top panel shows the @xmath0 , which is shifted to values roughly three times smaller ( cf . dashed line of fig [ fig : cumfits ] ) than the original models with scale - height free . in the last column of table [ tab : sims ] the @xmath0 range is given for each galaxy in this scenario individually . the @xmath0 values are much smaller than the other models where inclination is constrained . in the bottom panel of fig [ fig : fixhz ] the inclination , which is given as a ratio of the fitted value ( @xmath113 ) to the luminous tf relation value ( @xmath114 ) , is plotted . with an unchanged inclination ( @xmath115 ) , the quality of the fits to the majority of the dms galaxies are poor , but by increasing the fitted inclination to @xmath116 the fits are of a high quality . this is partly owed to the resulting rotation speeds generally having lower amplitudes . such significant deviations from v01 s luminous tf relation seems unlikely . to demonstrate this last point , in fig [ fig : tfrv01 ] the luminous tf relation from v01 ( eq [ eqn : tfr ] ) is plotted , which is used for the dms and this study . the solid black line is the best fit tf relation and the black data points are the values of the dms galaxies . the blue , green and red points are the positions of those galaxies if their inclinations were increased by a factor 1.1 , 1.2 or 1.4 respectively . the dashed and dotted lines are the 1@xmath81 and 2@xmath81 scatter in the observed tf relation . clearly even changing the inclinations by a factor of 1.1 would be in stark conflict with the tf relation . . these models use fixed scale - heights ( eq [ eqn : kreg ] ) , but the inclination is left completely unconstrained . ( top panel ) co - added posterior probability density as a function of @xmath0 , for all 30 galaxies , required for high quality fits to the dms data when the inclination is left free . ( bottom panel ) co - added posterior probability density against the ratio of the fitted inclination to the inclination derived from the luminous tf relation ( v01 ) . the inclinations required for good fits to the dms data are far larger than the inclinations derived from the luminous tf relation.,width=321 ] ) showing k - band magnitude against outer rotation velocity . the solid black line is the best fit luminous tf relation and the black data points represent the dms galaxies . the blue , green and red points are the positions those galaxies would take if their inclinations were increased by a factor 1.1 , 1.2 or 1.4 respectively . the dashed and dotted lines are the 1@xmath81 and 2@xmath81 scatter in the observed luminous tf relation.,width=321 ] the dms data seem to be at odds with mond when using the default acceleration parameter @xmath112 . to test this parameter s influence , models were tested where the scale - height and inclination are fixed , but the mond acceleration parameter is variable - along with the @xmath0 . although varying the mond acceleration parameter from galaxy to galaxy is not an acceptable solution , this could identify a unique value for @xmath11 that is more consistent with the dms data . in the top panel of fig [ fig : a0 ] another co - added , posterior probability density is plotted , this time against the mond acceleration parameter . this is done for three different interpolating functions ( @xmath26 , 4 and 8 - black , red and green lines respectively ) . the parameters @xmath117 and 8 are trialled to see if the sharper transition from the mond regime to the newtonian regime would affect the ability to fit the dms data . it appears that little is gained overall by varying @xmath11 and the same is true on a case by case basis . this can be demonstrated by comparing the fit with @xmath11 free ( and scale - height fixed ) to the benchmark fit ( similar to what was done in [ sec : vsve ] and fig [ fig : alpbet ] bottom panel ) . in the bottom panel of fig [ fig : a0 ] , for each galaxy individually , the ratio is plotted between the posterior probability of the best fit using a variable @xmath11 ( with fixed scale - height ) and the benchmark fit . the vast majority of the galaxies would have poor fits with a variable @xmath11 and fixed scale - height . varying @xmath11 suffers from the same drawbacks as varying the @xmath0 . increasing @xmath11 can allow a fit to the rotation curve with a lower @xmath0 , but this simultaneously increases the vertical gravitational field , which in turn requires the scale - height to be reduced . in the second ( with @xmath118 ) and third last ( with @xmath117 ) columns of table [ tab : sims ] the @xmath0 range is displayed for each galaxy individually when the mond acceleration parameter is variable . . three interpolating functions were used : @xmath26 , 4 and 8 ( black , red and green lines respectively ) . the default value for the mond acceleration parameter is @xmath112 . ( bottom panel ) ratio between the posterior probability of the best fit using a variable @xmath11 and the benchmark fit . the three interpolating functions are represented by the same colours as the lines in the top panel . varying @xmath11 does not allow good fits to the dms data.,width=321 ] [ sec : sech ] regarding whether the poor results were caused by using an exponential function to describe the vertical scale - height , the dms data was modelled using a @xmath33 function ( @xcite ) to describe the vertical distribution of the stellar disk in the n - particle models . the scale - height used in the @xmath35 function is @xmath34 . in fig [ fig : sech ] the posterior probability densities are plotted for the ratios of the fitted @xmath32 to @xmath34 for the @xmath26 and 2 interpolating functions . these are compared with data from @xcite and @xcite . from fig 5d of @xcite the indicative range of observed @xmath120 is between 12 and 22 . a solid vertical red line is placed in fig [ fig : sech ] to mark the @xmath120 value where their histogram begins to decline and a dashed vertical red line to mark where their data effectively terminates . the 153 galaxies of @xcite are also used to produce a probability density ( green curve in fig [ fig : sech ] ) to compare with the posterior probability density found here . the @xmath120 distributions ( solid and dashed black lines in fig [ fig : sech ] ) fitted here peak beyond where @xcite and @xcite have any evidence of such values . the distribution found here also extends to much larger values . to demonstrate the mismatch statistically , a 2d ks test was performed to compare the @xmath32 and @xmath34 distributions found here with the data of @xcite . the confidence levels found for the @xmath26 and 2 interpolating functions were @xmath121 and @xmath122 . evidently , the @xmath35 vertical distribution does little to aid the agreement between mond and the dms data . function to describe the vertical distribution of the stellar disk , for the 30 galaxies in the diskmass sample . the black solid curve corresponds to the @xmath26 interpolating function and the black dashed curve to @xmath27 . the distribution found by the dms , with an exponential vertical distribution , uses a turquoise curve and is just shown for reference . ( bottom panel ) superposition of the re - normalised posterior probability density of fitted scale - length to vertical scale - height , using a @xmath35 distribution , for the 30 galaxies in the diskmass sample . the black solid curve corresponds to the @xmath26 interpolating function and the black dashed curve to @xmath27 . the green curve is the probability density of the sample of 153 galaxies from bizyaev and mitronova ( 2002 ) . the vertical solid red line represents the value of @xmath120 at which the sample of mosenkov et al . ( 2010 ) begins to decline from a previously flat value . the dashed red line represents where the probability density of mosenkov et al . ( 2010 ) drops to zero . using a @xmath35 distribution does not improve the ability of mond to fit the dms without decreased scale - heights.,width=321 ] by assuming the scale - heights of face - on stellar disks are similar to the measured scale - heights of edge - on galaxies ( dmsii ) , the dms found that stellar disks are sub - maximal , with k - band mass - to - light ratios of the order of @xmath8 ( dmsvi ) . in this paper the measured vertical velocity dispersions and rotation curves from dmsvi were analysed in the context of mond . the problem is that the velocity dispersions measured by the diskmass survey can only allow for thick enough disks if the vertical gravitational field is weak . in newtonian gravity this can be accommodated by lowering the @xmath0 . in mond , the @xmath0 must be larger to account for the rotation curve in the absence of dm . this means the vertical gravitational field is strong . moreover , it is amplified by mond relative to the vertical gravitational field of a disk with the same @xmath0 in newtonian gravity . thus , the most straight - forward way to reduce the model vertical velocity dispersions in mond is to decrease the stellar scale - height . if these were reduced from the values derived from observations of edge - on galaxies by roughly a factor of 2 then the dms vertical velocity dispersions and rotation curves would be compatible with mond . regardless of stellar scale - length , these disks would have scale - heights between 200 and 400 pc . according to a two - dimensional k - s tests , such thin disks are strongly at odds with observations . all other key parameters involved in analysing the dms data in mond were modelled , such as the stellar velocity dispersion ellipsoid parameters , inclination , the choice of stellar vertical distribution profile , the mond acceleration parameter and interpolating function . no clear way was found to reconcile mond with the data . if the derived rotation velocities were lowered , by imposing 30 - 40% higher inclinations , it became possible to simultaneously fit the low vertical velocity dispersions and low rotation speeds with low values for the stellar @xmath0 . by design , since the outer rotation speeds are fitted , this is in accordance with the mondian baryonic tf relation ( with logarithmic slope fixed to 4 , eq [ eqn : btfr ] ) . despite this , these circular velocities are inconsistent with the _ luminous _ tf relation which is quite well established for high - inclination galaxies . to this end , it would be interesting to make a fully self - consistent analysis that models the effect of the varying inclination on the two dimensional velocity fields and the surface brightness profiles . a prediction was made for the scaling of the central dynamical surface density versus central surface luminosity density in mond and compared with the dms galaxies ( see @xcite ) . mond correctly predicts that the low surface luminosity density galaxies deviate from the simple relation @xmath97 . in addition to viewing mond as a modification of the law of gravity , @xcite has maintained that a modification of inertia might be better positioned to explain all available galactic dynamics data . in such a theory , the orbital trajectories of particles would have different inertias meaning the interpolating function in the radial direction would differ from the vertical directions . this is an intriguing possibility since the problem with mond outlined here is that for the radial force defined by a galaxy , the corresponding vertical force is too large . it is interesting therefore that the vertical velocity dispersions of disk galaxies must increase to agree with mond , but the observed radial velocity dispersions of some dwarf spheroidal galaxies ( @xcite ) must decrease . this anecdotal evidence should be followed up with more rigorous study . something worth further investigation is the impact from super - thin disks , much thinner than the observed scale - heights represented here ( @xcite ) , on the measurement of vertical velocity dispersions by the dms survey . specifically , how prominent are super - thin disks in all disk galaxies , and are the vertical velocity dispersions that the dms measures meaningfully influenced by them ? the super - thin disk found by @xcite was truncated at around 3 kpc , suggesting the influence is minimal . as discussed in dmsi , it is expected that stars in or near the mid - plane diffuse over time . the youngest stars may contribute to the light because of the ob stars , but they contribute little to the absorption lines , because those ob stars have weak spectral features ( except for h and he ) . it is possible to make even stronger claims about the status of mond - like theories , as well as conventional dynamics , with certain desiderata . from a theoretical point of view , this would be a detailed study of the impact of the neglected cross - term ( tilt of the velocity ellipsoid ) in eq [ eqn : sigred ] . from an observational point of view , a larger sample of edge - on galaxies ( like k02 ) with near - infrared photometry , to more precisely confirm the correlation between scale - length and scale - height . it would also be beneficial to have an increased sample of nearly face - on galaxies , with measured stellar velocity dispersions and rotation curves and with greater sensitivity to be able to measure the stellar velocity dispersions to larger radii . gwa is a postdoctoral fellow of the fwo vlaanderen ( belgium ) . ad acknowledges partial support from the infn grant indark , the grant progetti di ateneo to call 2012 0011 ` marco polo ' of the university of torino and the grant prin 2012 `` fisica astroparticellare teorica '' of the italian ministry of university and research . the authors thank m. bershady , k. westfall and f. lelli for valuable comments and are deeply indebted to the diskmass survey team , and thomas martinsson in particular , for taking and sharing their data . the reviewer also suggested several important improvements to the article . j. m. , terlouw j. p. , begeman k. g. , zwitser w. , roelfsema p. r. , 1992 , in astronomical society of the pacific conference series , vol . 25 , astronomical data analysis software and systems i , worrall d. m. , biemesderfer c. , barnes j. , eds . , p. 131 .this table presents various scale - heights and mass - to - light ratios with 1@xmath81 errors for each of the 30 galaxies in the dms sample . column ( 1 ) gives the galaxy name , ( 2 ) the scale - height used by the dms , ( 3 ) has the @xmath0 fitted by the dms using newtonian gravity and a dm halo . the remaining columns all pertain to mond fits . columns ( 4)&(5 ) , ( 6)&(7 ) and ( 8)&(9 ) give the best fit ratio of fitted scale - height to the one used by the dms and the fitted @xmath0 for three distinct interpolating functions respectively . these models come from analyses where the @xmath0 and scale - height are almost unconstrained , and the inclination has a limited amount of freedom . the next two columns ( 10 ) and ( 11 ) are for models where only the mond acceleration parameter and @xmath0 were free to vary ( inclination and scale - height were fixed ) . these two columns use different interpolating functions and show the @xmath0 of the best fit . the last column is the best fit @xmath0 found when the inclination is unconstrained , but scale - height is fixed . [ cols="<,^,^,^,^,^,^,^,^,^,^,^ " , ]

the evolution of the schrdinger - cat states in a dissipative parametric amplifier is examined . the main tool in the analysis is the normally ordered characteristic function . squeezing , photon - number distribution and reduced factorial moments are discussed for the single- and compound - mode cases . also the single - mode wigner function is demonstrated . in addition to the decoherence resulting from the interaction with the environment ( damped case ) there are two sources which can cause such decoherence in the system even if it is completely isolated : these are the decay of the pump and the relative phases of the initial cat states . furthermore , for the damped case there are two regimes , which are underdamped and overdamped . in the first ( second ) regime the signal mode or the idler mode collapses " to a statistical mixture ( thermal field ) . * key words : * cat states , parametric amplifier , decoherence and dissipation

1108.0127

the linear superposition principle is at the heart of quantum mechanics since , using it , one can control the properties of the single states making them more or less pronounced . the most significant example reflecting the power of such a principle are the schrdinger - cat states @xcite , which exhibit various nonclassical effects such as squeezing , sub - poissonian statistics and oscillations in photon - number distribution , even if the original states are close to the classical ones @xcite . such type of states ( of a @xmath0th mode for convenience ) can be represented as @xmath1 is a coherent state of the @xmath0th mode with complex amplitudes @xmath2 , @xmath3 is a relative phase and @xmath4 is the normalization constant having the form @xmath5}.\label{2}\ ] ] specifically there are three values of @xmath3 , namely , @xmath6 and @xmath7 , for which ( [ 1 ] ) reduces to even coherent ( ecs ) , odd coherent ( ocs ) and yurke - stoler ( yss ) states , respectively . these three states will be frequently used in this paper . there are several proposals for generating states ( [ 1 ] ) in various nonlinear processes , e.g. see . further , the evolution of cat states in several quantum systems has been intensively studied @xcite@xcite . for convenience of the present work we refer to the interaction of these states with environment , where the common goal of such studies is the comprehension and description of the decoherence processes ( decoherence is the rapid transformation of a pure linear superposition state into the corresponding statistical mixture state ) in the system under observation , and the energy loss due to the interaction with the reservoir . most of these papers adopt the heisenberg langevin approach or the markov master equation approach . although the non - markovian dynamics is important in quantum physics ( theory ) , a unified and compact treatment of the non - markovian reservoirs dates as late from @xcite , where also the decay and revival of the schrdinger - cat states are characterized . on the other hand , parametric amplifier ( nondegenerate ) takes a considerable interest in quantum optics since it can perfectly generate two - mode squeezing . recently , this device has been supported with the fast progress of new nonlinear crystals and improved laser sources , especially femtosecond lasers @xcite and it has been employed in experiments , e.g. in the interference experiments . an investigation of the statistical properties of the parametric amplifier with losses @xcite and without losses for compound - mode case have been performed using different techniques when the modes are initially prepared in the coherent states . the anticorrelation in this model is an interesting effect @xcite , where also under certain conditions the variance of the photon number is less than the average photon number and the photocounting distribution becomes narrower than the corresponding poisson distribution for a coherent state with the same mean photon number . moreover , the sum photon - number distribution can exhibit collapses and revivals @xcite ( and references therein ) similar to those typical in the jaynes - cummings model ( jcm ) @xcite ; the former is in the photon number domain rather than the time domain . so the question we would like to address in this paper is the following : what would be the influence of the interference in phase space on these phenomena ? in other words , we study the evolution of the cat states ( [ 1 ] ) through the parametric amplifier and take into account the interaction with environment . the system alone may be described by tracing a more complete description over the environment . actually , the interaction considered here is quite different from that considered in @xcite@xcite because there are two operations controlling its behaviour , which are the interference in phase space and the entanglement ( correlation ) between the signal and idler modes as well as between the signal - idler system and the environmental oscillators ( reservoirs ) to which the system may be connected . such entanglements lead to the increase of the marginal entropies of the subsystems ( as is known , if the density operator describes pure states , then the entropy ( @xmath8 ) is zero , otherwise @xmath9 ) @xcite ) . for the present system we treat the single- and compound - mode cases . the main tools in our treatment are the normally ordered characteristic functions . the hamiltonian which governs the interaction is @xmath10\\ \\ + \hbar \sum\limits_{j=1}^{2}\sum\limits_{l=1}^{\infty}[\varphi_{jl}\hat{b}_{jl}^{\dagger } \hat{b}_{jl}+k_{jl}\hat{b}_{jl}\hat{a}^{\dagger}_{j } + k^{*}_{jl}\hat{b}_{jl}^{\dagger}\hat{a}_{j } ] , \label{3 } \end{array}\end{aligned}\ ] ] where @xmath11 , are the annihilation ( creation ) operators assigned to the signal and idler modes , respectively ; @xmath12 are the natural frequencies of oscillations of the uncoupled modes with @xmath13 being a pump frequency ( under the assumption of resonance frequency ) ; @xmath14 is the initial phase of the pump and h.c . means the hermitian conjugate term to the previous one ; @xmath15 is the real coupling constant including the amplitude of the pump ( gain coefficient ) ; @xmath16 and @xmath17 are the boson annihilation and creation operators of the reservoir oscillators , respectively , with the frequencies @xmath18 , and @xmath19 are the coupling constants of the system - reservoirs interaction . hamiltonian ( [ 3 ] ) represents three wave mixing in a nonlinear crystal described by the second - order susceptibility @xmath20 . specifically , a pump beam with frequency @xmath21 travelling through the crystal creates photons of the signal and idler modes with different frequencies @xmath12 , @xmath22 , such that @xmath23 . the whole system is considered to be interacting with its surroundings ( e.g. , via imperfect cavity mirrors ) , which are modelled as reservoirs . for completeness , the well - known solution of the heisenberg langevin equations for the hamiltonian ( [ 3 ] ) is @xmath24,\\ \\ \hat{a}_{2}(t ) = f_{3}(t)\hat{a}_{2}(0)+f_{2}(t)\hat{a}^{\dagger}_{1}(0 ) + \sum\limits_{l}^{\infty}[\hat{b}_{1l}^{\dagger}(0)\gamma_{2l}(t ) + \hat{b}_{2l}(0)\gamma^{'}_{2l}(t ) ] , \label{4 } \end{array}\end{aligned}\ ] ] where @xmath25 ; @xmath26 and @xmath27 are the initial operators of the modes and reservoirs , respectively . the explicit forms for the dynamical coefficients @xmath28 and @xmath29 are given in the appendix a. further , it is worth mentioning that the langevin forces have the forms @xmath30 these forces satisfy the following commutation rule @xmath31=\gamma_{j } \delta_{jj^{'}}\delta(t - t')\hat{1 } , \label{6}\ ] ] where @xmath32 is the cavity decay rate of the @xmath0th mode ( for more details about the properties of the reservoir oscillators , see ) . finally it is worth mentioning that including lossy mechanism , the difference mean - photon number between the signal and idler modes in the system becomes nonconservative . this paper is organized as follows : in section 2 the basic equations and relations , such as two - mode normally ordered characteristic function , quadrature squeezing , wigner function , reduced factorial moments and photon - number distribution are given when the modes are initially prepared in the cat states ( [ 1 ] ) . in sections 3 and 4 discussion of the results of the single - mode and compound - mode cases are performed , respectively , and the conclusions are summarized in section 5 . in this section we give the main relations which will be used to investigate the properties of the system under consideration . actually , the calculations are lengthy , but straightforward , and for this reason , we give briefly general calculations in such a way that particular results can be obtained using suitable choice of parameters . we start by discussing the initial density matrix of the system . we assume that the system and reservoirs are initially independent and noninteracting before switching on the interaction , and the interaction between them starts at @xmath33 . this means that the density operator could be written initially as a direct product @xcite : @xmath34 where @xmath35 is the density matrix of the system of damping oscillators , which we assume to have flat ( i.e. constant function of frequency ) and broad - band reservoir spectra , so that the mean number of reservoir quanta ( phonons ) in the mode @xmath36 is @xmath37 independently of @xmath36 , where the subscript @xmath38 denotes broad - band reservoir . also we assume that the reservoirs form chaotic systems with mean photon numbers @xmath39^{-1},$ ] where the reservoirs - oscillators are at the temperature @xmath40 , with the frequency @xmath41 , @xmath42 is boltzmann s constant and @xmath43 planck s constant divided by @xmath44 . we proceed assuming that @xmath45 is the field density operator , which in our case , where the modes are assumed to be initially in superposition states ( [ 1 ] ) , has the form @xmath46 . \label{8 } \end{array}\end{aligned}\ ] ] the concise form in the second line of ( [ 8 ] ) means that the values of @xmath47 and @xmath48 should be specified to obtain the exact form of the density operator in the first line . here @xmath49 denotes the statistical mixture part of the density operator as @xmath50 @xmath51 denotes symmetric interference part of the density operator in which the two modes are in off - diagonal basis of coherent states and can be written as @xmath52 |-\alpha_{1}\rangle \\\\ + \exp[-i(\phi_{1}+\phi_{2 } ) ] |\alpha_{1}\rangle |\alpha_{2}\rangle \langle-\alpha_{2}| \langle-\alpha_{1}|\bigr\}+ { \rm h.c . } \end{array } \label{10}\end{aligned}\ ] ] and @xmath53 denotes asymmetric interference part of the density operator in which one of the modes is in a diagonal coherent states basis while the other is in off - diagonal basis or vice versa and reads : @xmath54 \\\\ + \exp(-i\phi_{2 } ) \bigl [ \alpha_{1}| + |-\alpha_{1}\rangle |\alpha_{2}\rangle \langle -\alpha_{2}| \langle -\alpha_{1}|\bigr]\bigr\ } + { \rm h.c.}. \end{array } \label{10ad}\end{aligned}\ ] ] the quantum properties of the system can be traced via the evolution of the normally ordered characteristic function , which for the two - mode case has the form @xmath55\exp [ - \zeta_{j}^ { * } \hat{a}_{j}(t ) ] \right\ } , \label{11}\ ] ] where , for the system under consideration , @xmath56 are given in ( [ 4 ] ) and @xmath57 is the initial density operator of the system ( [ 7 ] ) . it is obvious that the characteristic function here includes 16 elements @xcite . we give only the calculations related to the element @xmath58 ) into ( [ 11 ] ) and calculating the expectation values , we arrive at @xmath59 \\ \\ \times\exp \left\ { \zeta _ { 1}\zeta _ { 2}d(t)+ \zeta^ { * } _ { 1}\zeta^ { * } _ { 2}d^{*}(t ) -[\sum\limits_{j=1}^{2}|\zeta _ { j}|^{2}b_{j{\cal n}}(t ) ] + \sum\limits_{j=1}^{2}[\zeta _ { j}\bar{\alpha}_{j}(t ) -\zeta^ { * } _ { j}\bar{\alpha}'_{j}(t ) ] \right\ } , \label{12 } \end{array}\end{aligned}\ ] ] the explicit forms of the quantities @xmath60 and @xmath61 are given in the appendix b. it is worth mentioning that throughout the calculation of ( [ 12 ] ) , the standard commutator of the langevin forces ( [ 6 ] ) and the usual wigner - weisskopf approximation have been used , which leads to a replacement of the coupling constants by the cavity decay rates @xmath62 . more details about the calculation of ( [ 12 ] ) can be found in . on the other hand , the characteristic function of the single - mode case can be obtained from that of the two - mode case by simply setting the parameter related to the absent mode equal to zero , e.g. the characteristic function of the signal mode ( first - mode ) can be obtained by setting @xmath63 in ( [ 11 ] ) and ( [ 12 ] ) . the different moments of the bosonic operator system can be determined by differentiation of the normally ordered characteristic function through the relation @xmath64 in order to investigate squeezing property for the compound - mode case we can define the two quadrature operators @xmath65 , \quad \hat{y}(t)=\frac{1}{2i}\sum\limits_{j=1}^{2}[\hat{a}_{j}(t)-\hat{a}_{j}^{\dagger}(t)]$ ] , where @xmath66 = i\hat{1}$ ] and then the uncertainty relation reads @xmath67 , where , e.g. @xmath68 . therefore , we can say that the system is able to yield two - mode squeezing if the squeezing factor @xmath69 or @xmath70 . similar quantities can be defined for the single - mode case . we proceed to the two - mode normal generating function which is defined as @xmath71 this function may be used in studying the sum photon - number distribution and the reduced factorial moments for compound modes . for the general term ( [ 12 ] ) , relation ( [ 14 ] ) can be calculated to obtain @xmath72\\ \\ \times\exp[i\phi_{1}+i\phi'_{2 } -\frac{1}{2}\sum\limits_{j=1}^{2}(|\alpha_{j}|^{2 } + |\alpha'_{j}|^{2}-2\alpha^{*}_{j}\alpha'_{j } ) ] , \label{15 } \end{array}\end{aligned}\ ] ] where @xmath73 \pm \frac{1}{2 } \sqrt{[b_{1{\cal n}}(t ) -b_{2{\cal n}}(t)]^{2}+4|d(t)|^{2}},\\ \\ a_{\pm}=\frac{\pm 1}{\lambda_{-}-\lambda_{+}}\left\ { \bar{\alpha}'_{1}(t)\bar{\alpha}'_{2}(t)d(t ) + \bar{\alpha}_{1}(t)\bar{\alpha}_{2}(t)d^{*}(t)\right.\\ \\ -\left . \bar{\alpha}_{1}(t)\bar{\alpha}'_{1}(t)[b_{2{\cal n}}(t)-\lambda_{\pm } ] -\bar{\alpha}_{2}(t)\bar{\alpha}'_{2}(t)[b_{1{\cal n}}(t)-\lambda_{\pm}]\right\}. \label{16 } \end{array}\end{aligned}\ ] ] expression ( [ 15 ] ) shows that each term of the generating function of the system is the two - fold generating function for laguerre polynomials . further , in each term the quantities @xmath74 and @xmath75 play the role of the mean numbers of coherent photons and mean numbers of chaotic photons , respectively . the sum photon - number distribution @xmath76 and the reduced factorial moments @xmath77 for the compound - mode case can be defined by means of the derivative of @xmath78 via the relations @xmath79 on using ( [ 15 ] ) , these quantities can be deduced to become @xmath80\\ \\ \times\exp [ i\phi_{1}+i\phi'_{2 } -\frac{1}{2}\sum\limits_{j=1}^{2}(|\alpha_{j}|^{2 } + |\alpha'_{j}|^{2}-2\alpha^{*}_{j}\alpha'_{j})]\\ \\ \times\sum\limits_{l=0}^{n } \frac{1}{(n - l)!l ! } \left(\frac{\lambda_{-}}{1+\lambda_{-}}\right)^{n - l } \left(\frac{\lambda_{+}}{1+\lambda_{+}}\right)^{l } { \rm l}_{n - l}[\frac{a_{-}}{\lambda_{-}(1+\lambda_{- } ) } ] { \rm l}_{l}[\frac{a_{+}}{\lambda_{+}(1+\lambda_{+ } ) } ] , \label{18 } \end{array}\end{aligned}\ ] ] @xmath81\\ \\ \times\sum\limits_{l=0}^{k } { k \choose l } \lambda^{k - l}_{- } \lambda^{l}_{+ } { \rm l}_{k - l}(\frac{a_{-}}{\lambda_{- } } ) { \rm l}_{l}(\frac{a_{+}}{\lambda_{+ } } ) , \label{19 } \end{array}\end{aligned}\ ] ] where the subscript @xmath82 in ( [ 18 ] ) and ( [ 19 ] ) means that these quantities have been calculated for the general term ( [ 12 ] ) . also in ( [ 18 ] ) and ( [ 19 ] ) @xmath83 are the laguerre polynomials of order @xmath84 having the form @xmath85 we should mention that the phase terms , e.g. the term in the second line of ( [ 18 ] ) , exist only in the off - diagonal elements indicating that these elements are suppressed faster than the diagonal elements . in other words , the greater the distance " between the components of the cat states , the more rapidly the off - diagonal elements are damped . on the other hand , the corresponding quantities for the single - mode case can be obtained with the help of the single - mode normally ordered characteristic function and applying the same procedures as before . to be more specific , a general term @xmath86 of the first - mode case can be obtained from ( [ 12 ] ) by setting @xmath63 and consequently the corresponding photon - number distribution and the reduced factorial moments are @xmath87\\ \\ \times \left[\frac{b_{1{\cal n}}(t)}{1+b_{1{\cal n}}(t)}\right]^{n_{1 } } \frac{1}{1+b_{1{\cal n}}(t ) } { \rm l}_{n_{1}}[\frac { -\bar{\alpha}_{1}(t)\bar{\alpha}'_{1}(t)}{b_{1{\cal n}}(t)(1+b_{1{\cal n}}(t ) ) } ] , \label{19b } \end{array}\end{aligned}\ ] ] @xmath88\\ \\ \times b^{k}_{1{\cal n}}(t ) { \rm l}_{k}[\frac { -\bar{\alpha}_{1}(t)\bar{\alpha}'_{1}(t)}{b_{1{\cal n}}(t ) } ] . \label{20 } \end{array}\end{aligned}\ ] ] finally , the quantum properties of the system can be visualized well by analyzing its wigner function ( @xmath89 ) . actually , this function is sensitive to the interference in phase space and therefore it is helpful to be examined . the single - mode wigner function is defined as @xmath90 where @xmath91 is the single - mode normally ordered characteristic function . the wigner function for the single - mode general term is @xmath92 } \exp [ i\phi_{1}+i\phi'_{2 } -\frac{1}{2}\sum\limits_{j=1}^{2}(|\alpha_{j}|^{2 } + |\alpha'_{j}|^{2}-2\alpha^{*}_{j}\alpha'_{j})]\\ \\ \times\exp\left\{-2\frac { [ \bar{\alpha}^{*}_{1}(t)-z^ { * } ] [ \bar{\alpha}'_{1}(t)-z]}{1 + 2b_{1{\cal n}}(t)}\right\}. \label{22 } \end{array}\end{aligned}\ ] ] in this expression @xmath93 is always positive and represents the sum of quantum noises in the field , related to quantum fluctuations in the interaction and the mean value of reservoirs oscillators ( the number of photons or phonons transferred from the reservoirs to the quantum system ) . this shows that the width of this part of @xmath89 function becomes broader when the coupling with environment is considered . in the following sections we analyze the behaviour of the system under discussion on the basis of the results of the present section . we investigate the properties of both the single - mode and compound - mode cases . in our analysis for the dissipative case we consider symmetrical losses that is @xmath94 and @xmath95 for @xmath22 . we analyze two cases : underdamped case when @xmath96 and overdamped case when @xmath97 . the origin of these conditions can easily be recognized by careful examination of the time dependent coefficients in appendix a. for example , it holds that @xmath98 + \exp [ -(g+\frac{\gamma}{2})t]\right\}. \label{23}\ ] ] in the following we discuss the behaviour of @xmath89 function , photon - number distribution , squeezing and reduced factorial moments for the single - mode case . investigation of the single - mode @xmath89 function is important since it is informative and sensitive to the interference in phase space , and also it gives a prediction to the possible occurrence of the nonclassical effects in the system . furthermore , this function can be obtained experimentally using the optical homodyne tomography @xcite . in general , the @xmath89 functions of ecs , ocs and yss ( at @xmath33 ) are consisting of two gaussian bells , resulting from statistical mixture of individual composite states and interference fringes in between ( signature of the nonclassical effects ) originating from the superposition between different components of the states . nevertheless , the locations of the extreme values of the interference fringes of ecs , ocs and ycs are quite different . there are several papers , e.g. , that have been devoted to these fringes making them less or more pronounced by allowing the cat states to evolve in different quantum optical systems . for example , the decay rate of the interference fringes of the @xmath89 function of cat states of a mode which is coupled to a phase - sensitive reservoir ( squeezed reservoirs ) @xcite , @xcite p. 114 , can be smaller or larger than that for ordinary thermal heat bath reservoir @xcite . furthermore , in these systems the fringes can be washed out completely for a specific choice of the interaction parameters . in the language of the density matrix this means that the contribution of the off - diagonal elements is small compared with the contribution of the diagonal elements but it is not necessarily zero . of course , such diagonalization produces a statistical mixture of coherent states , which are close to classical wavepackets . now if we consider the @xmath89 function of the signal mode for the system under discussion , the situation becomes more complicated than before owing to the entanglement between the signal and idler modes . the structure of the density matrix ( [ 8 ] ) carries more information , e.g. , when the signal and idler modes are initially prepared in a distinguishable macroscopic cat state and after switching on the interaction , the initial two gaussian bells of the @xmath89 function of the signal mode can be transformed , in principle , into four - fold form as indicated in ( [ 9 ] ) , and the initial interference fringes of the cat will be dramatically changed during the interaction as a result of the competition between the different components of @xmath53 and @xmath51 . moreover , during evolution , the rate of movement of the centres of peaks of the @xmath89 function is rather different and this leads to irregularity in its behaviour . so that the evolution of the cat states in the present interaction yields different types of multicomponent cat states @xcite . all these facts can be seen in figs . c , where we have displayed the @xmath89 function for the signal mode ( excluding losses ) for the given values of the parameters . in these figures ( and throughout this paper ) we have taken @xmath99 , where @xmath100 is the phase of the initial amplitude of the @xmath0th mode and also we have defined @xmath101 in ( [ 21 ] ) as @xmath102 where the quadrature variances @xmath103 and @xmath104 of the signal mode are associated with the real part @xmath105 and imaginary part @xmath106 , respectively . it is worth mentioning that the @xmath89 function curves contain most of their information within one plane , in particular , for constant values of @xmath106 . for this reason and for simplicity we have plotted a @xmath89 function in one plane considering @xmath107 , where the @xmath89 function illustrates more pronounced nonclassical effects ( e.g. , by means of negative values ) for the case @xmath108 and also it includes a representative information for the other cases . 1a shows that the initial nonclassical negative values of the @xmath89 function have been smeared out and the system collapses to four - component statistical mixture state . further , such behaviour reveals that decoherence can be established via the entanglement between different modes in the parametric processes . in these cases the single - mode behaviour undergoes amplification resulting from the spontaneous pump photon decay @xcite . it should be borne in mind that in the present case the system is completely isolated and then such type of decoherence can be called nondissipative decoherence @xcite . on the other hand , the nonclassical negative values can be recovered by controlling the distance " between the initial cat states of the signal and idler modes . to be more specific , when @xmath108 these negative values can be realized , however , when @xmath109 they will disappear as indicated in figs . 1b and c , respectively . in fig . 1c a three - peak structure is dominant . comparison of figs . 1a , b and c is instructive . so , excluding the influence of the environment the system can decohere and recohere by adjusting the initial distance between the components of the cat of the input modes . however , for the dissipative case the decoherence process is related not only to the amplification of the pumping field but also to damping of radiation caused by the flux of coherent energy from radiation to the reservoirs and noise from reservoirs to the radiation . under these circumstances the decoherence can be achieved in a time shorter than that for the undamped case . we displayed the @xmath89 function for the overdamped and underdamped cases in figs . 2a and b , respectively , for the same situation as in fig . these figures show overall distortion due to the dissipative nature of dynamics . it is clear that the origin of the main contribution is in the diagonal elements of the density matrix and then the negative values of the @xmath89 function of the undamped case ( fig . 1b ) are washed out . from fig . 2a one can observe that the @xmath89 function exhibits the well - known shape for the thermal light , i.e. the system exhibits bose - einstein statistics ( super - classical light ) . however , from fig . 2b ( underdamped case ) we can observe that the two - peak structure is dominant . this situation is similar to that of the single harmonic oscillator interacting with the thermal bath in which a double gaussian structure with missing oscillatory behaviour occurs . the explanation of the behaviour of @xmath89 function in fig . 2 can be understood by analyzing the behaviour of the @xmath89 function for the single - mode general term ( [ 22 ] ) . let us restrict ourselves to the overdamped case for which the values of the parameters @xmath110 are exponentially decaying ( this can be easily checked ) when the interaction is going on and accordingly the centres of the peaks move toward the origin and eventually the peaks merge with each other . the opposite situation occurs for the underdamped case . it is worth mentioning that in these cases ( overdamped and underdamped cases ) the contribution of the off - diagonal elements corresponding to @xmath53 is suppressed faster than that of @xmath51 . this point will be discussed in the sum photon - number distribution in the following section . in general , if the interaction between different components in the system occurs ( regardless whether the interaction with environment is considered or not ) , the decoherence gradually increases and thus the system evolves into a mixture ( @xmath111 is the reduced density matrix of the signal mode ) , i.e. there occurs a destruction of the inherent nonclassical effects . the rate of destruction is sensitive to the nature of both the reservoir and parametric processes . furthermore , we should mention that the @xmath112 function possesses somewhat similar behaviour as the @xmath89 function in such a case , i.e. it can take on negative values in certain regions for certain values of the parameters and therefore it can not behave like a classical probability distribution function . such an effect is independent of the type of cat states , which are initially used in the interaction . finally , we should stress that it is difficult to analyze the behaviour of the off - diagonal elements to obtain concrete information , so we have basically concentrated on the computer simulation . the concept of photon is an integral part of the modern description of light and the discrete nature of light can be demonstrated by a photon detector based on the photoelectric effect . on the other hand , one of most interesting nonclassical effects emerging from the superposition principle is the oscillatory behaviour of the photon - number distribution . in general , such behaviour is closely related to the behaviour of the @xmath89 function , however , this is necessary but not sufficient condition . ( solid curve ) , @xmath113 ( dashed curve ) . , width=377 ] for example , the photon - number distributions of ecs , ocs and yss are completely different ; whereas those of ecs and ocs exhibit pairwise oscillations in phase space ( even number of photons can be observed for ecs and odd numbers for ocs ) , the distribution of yss is a poissonian even though the behaviour of the @xmath89 function for these states is qualitatively similar . in the interaction under discussion for undamped case the oscillatory behaviour in the photon - number distribution can be established even if the initial cat states exhibit poissonian statistics . this is of course based on the values of the interaction parameters . the origin of such behaviour is in the interference in phase space where the photon - number distribution of input coherent light is always displaying a single - peak structure , which is broader than the corresponding poisson distribution with the same mean photon number . we start our discussion by investigating the behaviour of the undamped case . we consider here the photon - number distribution @xmath114 of the signal mode when the signal and idler modes are initially prepared in the yss . we have seen that this quantity can exhibit oscillatory behaviour after switching on of the interaction by a suitable time provided that @xmath115 , as indicated in fig . 3 for shown values of the parameters . in this figure @xmath116 , where @xmath100 is the phase of the initial @xmath0th mode ( signal or idler ) and @xmath14 is the phase of the pump as before . in this case @xmath117 represents the phase mismatch . furthermore , one can see that when @xmath117 changes from @xmath7 to @xmath118 the parity of oscillations changes ( compare dashed and solid curves ) . the reason for taking @xmath119 can be found in @xcite . the origin of the oscillatory behaviour in the photon - number distribution @xmath114 here is the competition between the contributions of @xmath49 and of @xmath51 , as we will see in the sum photon - number distribution . for @xmath120 ( undamped case - solid curve ) ; @xmath121 ( underdamped case - short - dashed curve ) ; @xmath122 ( overdamped case - long - dashed curve ) . , width=377 ] further , it is worth mentioning that the behaviour of @xmath114 in fig . 3 is similar to that of initially cat state of a mode coupled to a phase - sensitive reservoir @xcite , @xcite p. 114 , however , the source of the oscillations in @xcite is the phase information included in the reservoir , which can be transferred to the field . in fig . 4 we have displayed @xmath114 for the same situation as given by the solid curve in fig . 3 , but for @xmath123 ( solid curve ) ; further the damped cases are considered : underdamped case ( short - dashed curve ) and overdamped case ( long - dashed curve ) . comparing the solid curves in figs . 3 and 4 , we see from fig . 4 that the oscillations in @xmath114 are smoothed out . further , the comparison of various curves in fig . 4 shows that the behaviour of @xmath114 for underdamped and undamped cases is similar in the sense that they include smooth oscillations . these smooth oscillations are completely washed out for the overdamped case , as is expected . actually , for @xmath109 coherence is lost and the main contribution is related to the energy of the field mode ( diagonal terms ) . as we can see the behaviour of the photon - number distribution is in a good agreement with that of @xmath89 function . finally , it has been verified that the behaviour of @xmath114 for the damped case when @xmath108 and @xmath123 is quite similar . this is connected with the fact that the oscillatory behaviour in the photon - number distribution is highly sensitive to the dissipation dynamics . squeezing is one of the most important phenomena in quantum optics because of its applications in various areas , e.g. in optics communication , quantum information theory , etc . squeezed light can be measured by a homodyne detection in which the signal is superimposed on a strong coherent beam of the local oscillator . furthermore , quite recently it has been shown experimentally that there is an evidence of squeezed light in the biological systems @xcite . so an analysis of squeezing phenomenon in quantum optical systems is an important topic . we start our investigation by determining the behaviour of the single - mode squeezing . generally , in the system under consideration the output mode loses its initial squeezing feature during the interaction @xcite as a result of an amplification process and the interaction with the environment , which accelerates the loss in quantum fluctuations . here we analyze the influence of different types of cat states on the behaviour of quadrature squeezing . to do so , considering @xmath2 to be real , we write down the quadrature variance of the @xmath124-component of(which is expected to yield squeezing ) for the signal mode and various initial input cat states as follows : @xmath125 \right\},\;\ ; \label{24}\ ] ] @xmath126 \right\ } , \;\ ; \label{25}\ ] ] @xmath127 \right\ } .\label{26 } \end{array}\end{aligned}\ ] ] in these expressions the subscripts ee , oe and ey stand for the initial ( signal , idler ) modes which are in ( ecs , ecs ) , ( ocs , ecs ) and ( ecs , yss ) , respectively . furthermore , one can note that the significant value of squeezing can be obtained when the pump phase is @xmath128 . also it can be mentioned that the correlation between signal and idler modes does not occur since cross terms such as @xmath129 are absent . actually , this is connected with the two - photon nature of ecs and ocs where @xmath130 , when @xmath131 is an odd integer . it is worth mentioning that the origin of losses caused by reservoirs in expressions ( [ 24])([26 ] ) is the mean photon number . further , it should be reminded that ecs and ycs can exhibit normal squeezing , which is more pronounced for ecs , however , ocs are unsqueezed states @xcite . further , for the undamped case with @xmath132 , expression ( [ 25 ] ) can be rewritten as @xmath133 + \alpha_{1}^{2}\cosh^{2}(gt)(\coth\alpha_{1}^{2 } -1 ) \right\}.\label{27}\ ] ] the last term in this expression is non - negative and can become zero by appropriately choosing the value of @xmath134 . in order to get squeezing in the signal mode , expression ( [ 27 ] ) must be negative and this depends on the behaviour of the function @xmath135 . in other words , squeezing can be established if @xmath136 , however , @xmath137 , and consequently squeezing can not be obtained . more specifically , to obtain single - mode squeezing from this device , the mode under consideration should be prepared initially in a squeezed - cat state regardless of the type of the cat state in the free port ( if it is squeezed or not ) . furthermore , one can easily estimate how long the single - mode squeezing of the initial light can survive if this light is imposed at the input of the parametric amplifier . restricting ourselves to the input ecs and using the fact that squeezing is surviving if @xmath138 , the time range over which such a situation occurs is @xmath139 where @xmath140 has the same expression as before . for @xmath141 ( i.e. ( [ 25 ] ) ) the procedure shows that the required time is a complex number and this agrees with the previous remark . from ( [ 24 ] ) and ( [ 26 ] ) one can easily verify that the amount of squeezing available in @xmath142 is much larger than that in @xmath143 provided that @xmath144 is finite ; nevertheless , when the value of @xmath144 is zero or large enough , both quantities are typical . this implies that the larger the degree of squeezing in the free port is , the more squeezing will be available in the mode under consideration . for the signal mode being initially in ecs and the idler mode being in ecs ( solid curve ) , yss ( short - dashed curve ) and in ocs ( long - dashed curve ) for @xmath145 and @xmath146 . triangle- and star - centred curves are given for the squeezing factors of a zero - temperature heat bath ( @xmath147 ) and a nonzero - temperature heat bath ( @xmath148 ) , respectively , for the underdamped case with @xmath149 and for the same situation as represented by the solid curve . , width=377 ] fig . 5 has been displayed to illustrate the behaviour of the @xmath150-squeezing factor for the undamped case when the signal mode is prepared initially in ecs and the idler mode in ecs ( solid curve ) , yss ( short - dashed curve ) and ocs ( long - dashed curve ) in dependence on @xmath151 for the given values of the parameters . furthermore , the triangle- and star - centred curves represent the underdamped case with a zero - temperature ( @xmath152 ) and a nonzero - temperature heat baths ( @xmath153 ) , respectively , when the two modes are in ecs . the choice of @xmath154 is related to the fact that the initial ecs gives maximum squeezing at this value . now the analytical facts discussed above are remarkable in fig . 5 . further , from this figure we can also see that for the ocs - idler - mode squeezing is minimum when @xmath155 is close to zero ( indeed there is a singularity at @xmath156 related to the nature of the ocs ) , increasing monotonically as @xmath144 increases , then it stagnates at large values of @xmath144 yielding its maximum value . actually , the behaviour of this case is quite different from the behaviour of the ecs - idler- and yss - idler - mode cases ( compare long - dashed curve with solid and short - dashed curves in fig . 5 ) , where the maximum value of the former is the minimum value for the latter . this means that for large values of @xmath155 the signal mode can produce the same value of squeezing regardless of the type of the input cat states in the free port . also this figure shows how one can control the single - mode squeezing relying on the type of initial cat states in the free port . we now turn our attention to the damped case . as we mentioned earlier @xmath93 in expressions ( [ 24])([26 ] ) is always positive and therefore the coupling of the system with the environment degrades the amount of the single - mode squeezing . we have given two examples in fig . 5 for the underdamped case ( triangle- and star - centred curves ) . comparing these curves with the solid one in the same figure , the conclusion becomes clear . moreover , from the comparison between triangle - centred curve and star - centred curve we can conclude that in nonzero - temperature heat bath case the quantum coherence ( i.e. nonclassical effects ) is lost much faster than in zero - temperature heat bath case @xcite . on the other hand , the occurrence of squeezing in the damped case has been verified also in the behaviour of the @xmath89 function where for the same values of the parameters we obtained that it always exhibits noise - ellipse forms of cuts together with single - peak or two - peak structure according to @xmath144 is small or large , respectively . it is worth mentioning that for the cat states interacting with heat bath the authors of @xcite claimed that squeezing is much more robust with respect to damping than the oscillations in the photon - number distribution or the interference in phase space described by @xmath89 function . they obtained this conclusion by analyzing these quantities graphically . nevertheless , in their analysis , squeezing has been obtained for a specific region of @xmath157 ( in particular when @xmath157 is small ) whereas the behaviour of the photon - number distribution and the @xmath89 function has been analyzed in a different region ( when @xmath158 , where squeezing does not occur , see fig . 8a in @xcite ) . this leads to the incorrect conclusion . we have examined the behaviour of both the photon - number distribution and the @xmath89 function for these cases using the same values of the parameters as in @xcite , however , we have taken @xmath159 ( where squeezing is noticeable ) and found that the behaviour of these quantities reflects equally the properties of squeezed light . for more details reader can consult @xcite . finally , for the single - mode reduced factorial moments , which can be measured by a set of photodetectors , we have found that the behaviour of these quantities is in a good agreement with the behaviour of the single - mode squeezing . in this section we demonstrate the two - mode properties for the system under discussion by determining the sum photon - number distribution , two - mode squeezing and reduced factorial moments . the oscillations in the joint photon - number distribution might be observed in experiments that generate two - mode squeezed light in which the two modes may be distinguished by frequency or by propagation direction . having separated the two modes , as was done by using a polarizing beamsplitter @xcite , one can send them directly onto separate photocounters and then build up the joint photon - number distribution from the photocount statistics of a sequence of pulses @xcite . it has been shown @xcite that the sum photon - number distribution @xmath160 of the nondegenerate parametric amplifier , when the modes are initially prepared in coherent state and under certain conditions , displays two regimes , which exhibit either single - peak structure or oscillatory behaviour . the single - peak structure can be narrower ( or broader ) than that of the corresponding poissonian distribution showing nonclassical ( or classical ) effects . the striking feature is that the @xmath160 can exhibit , for certain choice of the interaction parameters , collapses and revivals in the photon - number domain somewhat similar to those known in the jcm . for the present system the situation is rather complicated regarding to the evolution of the density matrix ( [ 8 ] ) where the interference in phase space is established . excluding dissipation and using @xmath119 as in @xcite , we have seen generally that @xmath160 exhibits always oscillatory behaviour irrespective of which type of cat states has been considered initially . moreover , @xmath160 can yield pairwise oscillations identifying that even or odd photons are being observed . figs . 6 and 7 have been plotted to show such a phenomenon for given values of the parameters . from fig . 6 we observe that the long scale oscillations in the behaviour of @xmath160 with @xmath161 are somewhat similar to those of squeezed states @xcite . actually for squeezed states the pairwise oscillations are explained as a direct consequence of the quadratic , or two - photon nature of the squeeze operator @xmath162 , i.e. @xmath163 $ ] @xcite . however , here the origin of the pairwise oscillations in @xmath160 is in the competition between the processes described by three parts of the density matrix of the field . figures 7a c give insight into this point , i.e. they show the manner in which the photon - number distributions of the three parts of the density matrix ( [ 8 ] ) compete . more precisely , figs . 7a , b and c represent the sum photon - number distributions @xmath164 and @xmath165 associated with @xmath166 and @xmath53 , correspondingly . 7a shows revival - collapse pattern , which is resulting from the statistical mixture part ( [ 9 ] ) . furthermore , from this figure one can observe that when the number of photons increases , the amplitude of the revivals diminishes , but the revival periods extend . it is worth mentioning that such a quantum collapse - revival phenomenon has been seen for the photon - number distribution of single - mode @xcite and two - mode @xcite squeezed coherent states with complex squeeze and displacement parameters . associations with the quantum phases of the modes may be considered @xcite , @xcite p. 281 . from fig . 7b it is clear that @xmath167 oscillates between negative and positive values . comparison of fig . 7a , fig . 7b and fig . 7c shows that the values of @xmath165 are approximately negligible compared with those of @xmath168 and @xmath167 . now if we turn our attention back to fig . 6 we can recognize that the competition between @xmath168 and @xmath167 leads to the destruction of collapse - revival phenomenon ( which appeared in fig . 7a ) , however , the distribution evolves " in an interesting manner similarly as that for squeezed states . in fact , the situation here is in contrast with that of the evolution of cat states in the jcm where the interference in phase space makes the revival - collapse phenomenon in the atomic inversion more pronounced when the revival time equals the half of that for the standard jcm with initial coherent light @xcite . the reason is that for the latter case the distribution of the spectral components of the atomic inversion is one dimensional distribution since only one mode is involved , whereas here we have two - dimensional superimposed distributions . further , we have obtained that the oscillations in @xmath160 increase if either @xmath15 or @xmath169 or both are increased . this fact is clear if we look at the problem as evolution of cat states under the action of two - mode squeeze operator , where the squeezing parameter in this case is @xmath170 and the oscillations become more pronounced for the large values of @xmath171 . in conclusion , for the system under consideration the origin of the oscillations in the sum photon - number distribution is two - fold : ( i ) the interference in phase space . ( ii ) the strong coupling between the signal and idler modes of the system in the course of the interaction time . on the other hand , we have found by an explicit algebraic calculation for all quantities studied in this paper excepting quadrature squeezing that the contributions associated with elements of @xmath51 and @xmath53 involve @xmath172 and @xmath173 , respectively . this fact together with the information included in figs . 7 show that @xmath160 can collapse and yield decoherence in relation to distribution for statistical mixture part using specific types of cat states initially . more illustratively , preparing one of the modes initially in ecs ( or ocs ) and the other in yss or vice versa , the @xmath160 evolves as described by fig . 7a ( for the same values of the parameters ) . this shows how one can decohere the system apart from the amplification nature of the system and without coupling it to the environment . we call such a type of decoherence a phase decoherence . actually , for some quantities the contribution of off - diagonal elements of the density matrix is responsible for the nonclassical effects , e.g. as we will see below in the compound - mode reduced factorial moments . then we can make a good estimation of these phenomena using such a property . the final remark is that the nonclassical effects for the compound - mode case are much richer than those for the single - mode case . this results from the strong quantum correlation between the signal and idler modes , which manifests itself as the summation in expressions ( [ 18 ] ) and ( [ 19 ] ) . to make this point clear , we give fig . 8a for the @xmath160 and for the same situation as that of the solid curve in fig . the comparison between these two figures is instructive . further , fig . 8b includes information on the damped case where one can observe smooth oscillations for the underdamped case ( solid curve ) and two - peak structure for the overdamped case ( dashed curve ) . specifically , for the overdamped case the off - diagonal elements are completely suppressed . firstly , we will discuss two - mode squeezing , which can be measured by heterodyne detection where squeezing is carried jointly by two modes of different frequencies and the local oscillator has a frequency midway in between @xcite . we proceed taking into account that the nondegenerate parametric amplifier is well described by the two - mode squeeze operator . actually , two - mode squeezing phenomenon is essentially connected with the states upon which such operator acts . in other words , squeezing may not exist for some specific states even if they include two - mode squeeze operator in their structures . to show this we give the form of the two - mode squeezing factor @xmath150 when the signal and idler modes are initially prepared in cat states , excluding losses , considering @xmath174 , are real and @xmath175 . in this case the squeezing factor @xmath150 takes the form @xmath176 , \label{29}\ ] ] where @xmath177 , represent the corresponding single - mode squeezing factors of the signal and idler modes , respectively . from expression ( [ 29 ] ) it is clear that the correlation between the signal and idler modes does not exist since @xmath178 , where @xmath179 , are the expectation values of the @xmath124-quadrature of the signal and idler modes . the existence of correlation between these two modes is important to obtain squeezing in the compound modes even if the individual modes are not themselves squeezed @xcite . expression ( [ 29 ] ) shows that interference in phase space can destroy squeezing and also that the rates of degradation of the two - mode and single - mode squeezing are on the same level for symmetrical losses . also it is obvious that the two - mode squeezing can be realized if at least one of the two modes ( signal or idler ) can exhibit single - mode squeezing . this is necessary but not sufficient condition . further , the maximum squeezing may be produced when both the signal and idler modes exhibit maximum single - mode squeezing . this reflects the importance of the choice of the type of the initial cat states . from the discussion in section 3 it is obvious that when the two modes are initially in ocs , the interaction can not generate two - mode squeezing . we now discuss the case that one of the modes is initially squeezed ( ecs ) and the other is unsqueezed ( ocs ) for the same situation as that in ( [ 29 ] ) . the required time to obtain squeezing in the compound - mode case ( @xmath180 ) is @xmath181 this inequality gives relation between the squeeze time and the distances " between the states , which are forming the cats . so that the length of the device can be adjusted to obtain squeezed light . an analysis to the right - hand side of ( [ 29a ] ) gives that the maximum scaled time ( effective time ) for squeezing is @xmath182 at @xmath183 . it is worth mentioning at this point that the @xmath124-component quadrature squeezing of ecs takes on its maximum , whereas that of ocs vanishes . against @xmath184 for @xmath185 , and for the undamped case when the signal and idler modes are prepared initially in ( ocs , ocs ) ( solid curve ) and ( ocs , ecs ) ( long - dashed curve ) . the damped case is considered for the same situation as the solid curve , but for the underdamped case ( short - dashed curve ) @xmath186 for the overdamped case ( star - centred curve ) @xmath187 . the straightline shows the antibunching bound . , width=377 ] on the other hand , when the phase information is included ( i.e. @xmath188 ) the situation is improved in such a way that squeezing can be detected in both quadratures provided that one of the modes is initially prepared in a squeezed - cat state , i.e. in ecs or yss . the role of the phase is analyzed in figs . 9a and b where we have plotted the squeezing factors corresponding to @xmath189-component and @xmath124-component , respectively , when the signal and idler modes are initially prepared in ecs . from these figures we can see that maximum squeezing occurs in @xmath189-component and locates at @xmath190 . the opposite situation can be observed for the @xmath124-component where squeezing also exists but it is less pronounced and locates at @xmath191 , i.e. the extreme values are exchanged . such a behaviour in the two quadratures indicates that the uncertainty relation holds . further , for @xmath192 we obtained similar behaviour , however , the maximum squeezing values would be available in the @xmath124-component . further similar behaviour can be seen if the other types of cat states are used . it is remarkable that the behaviour of squeezing in the present interaction is quite different from that for both ecs @xcite and two - mode squeezed coherent states @xcite where for both squeezing exists only in one of the two quadrature components ; perfect squeezing ( i.e. @xmath193 squeezing ) occurs only for the two - mode squeezed coherent states . in conclusion the squeezing phenomenon discussed here is a manifestation of co - operative effects among the phases of the different components of the system . furthermore , by controlling the phases in this device , squeezing can be amplified and switched between the two quadratures . we would like to conclude this subsection by shedding the light on the behaviour of the reduced factorial moments for the compound - mode case . in this case the antibunched light can be measured if both the signal and idler modes are simultaneously detected by means of two photodetectors and then their outputs are correlated . generally , reduced factorial moments can evolve to produce nonclassical negative values based on the types of initial cat states . further , we have checked that the contribution of @xmath49 can not produce antibunched light independently . these two facts reflect the role of the phase decoherence property discussed above . on the other hand , similar argument as that for the single - mode case can be given here except the case when the two modes are in yss where a slight antibunching can be observed , but only for certain values of the interaction parameters . 10 shows the reduced factorial moments of the compound - mode case for the given values of the parameters . in this figure we have used the normalized factor @xmath194 , where the nonclassical effect occurs when @xmath195 . from this figure , it is interesting to observe that the behaviour of the reduced factorial moments when the ( signal , idler ) modes are in ( ocs , ocs ) is very close to the behaviour of the second - order correlation function of ocs @xcite provided that @xmath196 , are finite . in this paper we have analyzed the properties of the dissipative parametric amplifier when the signal and idler modes are initially prepared in schrdinger - cat states . needless to say the present interaction as described by the hamiltonian ( [ 3 ] ) and the density matrix ( [ 8 ] ) is much more complicated than that in the simple case of a harmonic oscillator coupled with the heat bath @xcite@xcite . after obtaining the solution of the heisenberg langevin equations quantum statistics of interacting modes have been investigated based on the normally ordered characteristic functions . the system of damping oscillators has been assumed to have a flat spectrum and a chaotic distribution . in general , there are two operations controlling the behaviour of the interaction which are the interference in phase space and the entanglement . intuitively , the initial macroscopic cat states can not be preserved in the system . furthermore , for long - time interaction the initial nonclassical effects of the cats are degraded by the amplification dynamics inherent in the system and the cumulative effects of dissipation . excluding dissipation we have shown that if the input to the device are squeezed ( sub - poissonian ) cat states , then the output may be squeezed ( antibunched ) too provided that the interaction time and gain are finite . furthermore , the device could be used to amplify squeezing in the compound - mode case . on the contrary , the well - known role of the parametric amplifier as a source of perfect squeezing in the compound - mode case rather than in single - mode case fades out for input cat states . the photon - number distribution can exhibit oscillatory behaviour , which is more pronounced in the compound - mode case than in the single - mode case , regardless of the types of the initial cat states . also we have shown that the decoherence can arise in the system from the decay of the pump and the phase control . for dissipation case we have considered in detail only two cases which are related to underdamped and overdamped regimes . in these cases the interaction tends to eliminate the off - diagonal elements of the density matrix and to affect the diagonal elements . it has been found that the dynamical subsystems ( signal or idler modes ) collapse to the statistical mixture state or thermal state according to whether the underdamped regime or overdamping regime is considered . further , we have also shown that the single - mode squeezing at nonzero temperature of the environment decreases much faster than that at zero temperature , however , this situation is generally valid for all quantities studied in this paper . in conclusion , the losses uniformly distributed over the device deteriorate its operational characteristics . of course , such deterioration is more pronounced for the overdamped case . finally , the inclusion of lossy mechanics in the system is of a great interest for accurate measurements . j. p. , v. p. and f. a. a. e - o acknowledge the partial support from the project ln00a015 of the czech ministry of education . one of us ( m. s. a. ) is grateful for the financial support from the project math 1418/19 of the research centre , college of science , king saud university . schrdinger e 1935 _ nature _ * 23 * 844 . yurke b and stoler d 1986 _ phys . lett . _ * 57 * 13 ; hillery m 1987 _ phys . rev . a _ * 36 * 3796 ; xia y and guo g 1989 _ phys . lett . a _ * 136 * 281 ; gerry c c 1992 _ opt . commun . _ * 91 * 47 ; janszky j and vinogradov a v 1990 _ phys . lett . _ * 64 * 2771 ; sun j , wang j and wang c 1992 _ phys . a _ * 46 * 1700 . schleich w , pernigo m and le kien f 1991 _ phys . a _ * 44 * 2172 . brune m , haroche s , raimond j m , davidovich l and zagury n 1992 _ phys . a _ * 45 * 5193 . buek v , vidiella - barranco a and knight p l 1992 _ phys . rev . a _ * 45 * 6570 . milburn j and holmes c a 1986 _ phys . a _ * 56 * 2237 ; milburn j 1986 _ phys . a _ * 33 * 674 ; mecozzi a and tombesi p 1987 _ phys . lett . _ * 58 * 1055 ; tombesi p and mecozzi a 1987 _ j. opt . b _ * 4 * 1700 ; wolinsky m and carmichael h j 1988 _ phys . rev . lett . _ * 60 * 1836 ; sanders b c 1989 _ phys . a _ * 39 * 4284 ; savage c m and cheng w a 1989 _ opt . commun . _ * 70 * 439 ; miranowicz a , tana r and kielich s 1990 _ quant . opt . _ * 2 * 253 . la porta a , slusher r e and yurke b 1989 _ phys . lett . _ * 62 * 26 ; song s , caves c m and yurke b 1990 _ phys . a _ * 41 * 5261 ; yurke b 1986 _ j. opt . b _ * 3 * 732 ; yurke b , schleich w and walls d f 1990 _ phys . rev . a _ * 42 * 1703 . slosser j j , meystre p and wright e m 1990 _ opt . lett . _ * 15 * 223 ; slosser j j and meystre p 1990 _ phys . rev . a _ * 41 * 3867 ; wilkens m and meystre p 1991 _ phys . rev . a _ * 43 * 3832 . schaufler s , freyberger m and schleich w p 1994 _ j. mod . _ * 41 * 1765 . vidiella - barranco a , moya - cessa h and buek v 1992 _ j. mod . opt . _ * 39 * 1441 . filip r 2001 _ j. opt . b : quant . semiclass . opt . _ * 3 * s1 . disi l , gisin n and strunz w t 1998 _ phys . rev . a _ * 58 * 1699 . zhang s k , fujita m , yamanaka m , nakatsuka m , izawa y and yamanaka c 2000 _ opt . commun . _ * 184 * 451 . kwiat p g , vareka w a , hong c k , nathel h and chiao r y 1990 _ phys . a _ * 41 * 2910 ; ou z y , zou x y , wang l j and mandel l 1990 _ phys . * 65 * 321 . zou x y , wang l j and mandel l 1991 _ phys . lett . _ * 67 * 318 . mita l and peina j 1978 _ czech j. phys . b _ * 28 * 392 . mollow b r and glauber r j 1967 _ phys . rev . _ * 160 * 1076 ; ibid . 1097 . mita l and peina j 1977 _ acta phys . pol . 52(3 ) * 425 . el - orany f a a , peina j and abdalla m s 2001 _ opt . * 187 * 199 . jaynes e t and cummings f 1963 _ proc . * 51 * 89 ; eberly n h , norozhny n b and sanchez - mondragon j j 1980 _ phys . lett . _ * 44 * 1323 . phoenix s j d and knight p l 1988 _ ann . phys . _ * 186 * 381 . peinov v 1981 _ opt . act . _ * 28 * 747 . peinov v and peina j 1981 _ opt . act . _ * 28 * 769 . peina j : _ quantum statistics of linear and nonlinear optical phenomena _ , 2nd edition ( kluwer , dordrecht 1991 ) . mandel l and wolf e : _ optical coherence and quantum optics _ ( univ . press , cambridge 1995 ) . smithey d t , beck m , cooper j and raymer m g 1993 _ phys . rev . a _ * 48 * 3159 ; beck m , smithey d t and raymer m g 1993 _ phys . rev . a _ * 48 * 890 ; beck m , smithey d t , cooper j and raymer m g 1993 _ opt . lett . _ * 18 * 1259 ; smithey d t , beck m , cooper j , raymer m g and faridani m b a 1993 _ phys . t _ * 48 * 35 . agarwal g s 1999 _ phys . a _ * 59 * 3071 . mogilevtsev d and kilin s y 1996 _ opt . _ * 132 * 452 . bandilla a and ritze h - h 1980 _ opt . commun . _ * 34 * 190 . bonifacio r , olivares s , tombesi p and vitali d 2000 _ j. mod . _ * 47 * 2199 . braunstein s l and kimble h j 1998 _ phys . _ * 80 * 869 ; milburn g j and braunstein s l 1999 _ phys . a. _ * 60 * 937 ; ralph t c 2000 _ phys . rev . a _ * 61 * 010303(r ) ; hillery m 2000 _ phys . rev . a _ * 61 * 022309 . popp f a , chang j j , herzog a , yan z and yan y 2002 _ phys . lett . a _ * 293 * 98 . el - orany f a a 2002 _ phys . a _ * 65 * 043814 . aytr o and kumar p 1990 _ phys . lett . _ * 65 * 1551 . caves c c , zhu c , milburn g j and schleich w 1991 _ phys . a _ * 43 * 3854 . yuen h p 1976 _ phys . * 13 * 2226 . dutta b , mukunda n , simon r and subramaniam a 1993 _ j. opt . soc . b _ * 10 * 253 . selvadoray m , kumar m s and simon r 1994 _ phys . a _ * 49 * 4957 . selvadoray m , kumar m s 1997 _ opt . commun . _ * 136 * 125 .

we construct a class of codimension-2 solutions in supergravity that realize t - folds with arbitrary monodromy and we develop a geometric point of view in which the monodromy is identified with a product of dehn twists of an auxiliary surface fibered on a base . these defects , that we call t - fects , are identified by the monodromy of the mapping torus obtained by fibering over the boundary of a small disk encircling a degeneration . we determine all possible local geometries by solving the corresponding cauchy - riemann equations , that imply the equations of motion for a semi - flat metric ansatz . we discuss the relation with the f - theoretic approach and we consider a generalization to the t - duality group of the heterotic theory with a wilson line . mpp-2015 - 185 + lmu - asc 50/15 * the monodromy of t - folds and t - fects * 1.5 cm dieter lst , stefano massai and valent vall camell + 0.8 cm _ arnold sommerfeld center for theoretical physics , + theresienstrae 37 , 80333 mnchen , germany _ + 0.3 cm _ max - planck - institut fr physik + fhringer ring 6 , 80805 mnchen , germany _ 0.3 cm

1508.01193

the geometrization of duality groups has been the source of many insights both in string and field theories . perhaps the simplest example is that of montonen - olive duality in four - dimensional @xmath6 sym , where the discrete @xmath7 group acting on the complexified coupling @xmath8 is interpreted as the modular group of a @xmath9 . the torus is physical : one can obtain the four - dimensional theory with parameter @xmath8 by compactifying a six - dimensional @xmath10 theory on @xmath11 @xcite . a related example is s - duality in @xmath12 theories , which can be identified with the mapping class group of an @xmath13-punctured genus-@xmath14 riemann surface @xmath15 @xcite . in type iib string theory , identifying the axio - dilaton @xmath8 with the modular group of an auxiliary torus leads to f - theory @xcite . even if we do not have a complete twelve dimensional picture , the torus is again physical and this is best understood via duality to m - theory . similar attempts have been made to geometrize the full u - duality group @xcite . recently , investigations along these lines have been carried out for the t - duality group in the context of heterotic / f - theory duality in @xcite , following earlier works @xcite . one of the most interesting outcomes of this analysis is the description of inherently quantum bundles , namely spaces in which different patches are glued together with t - dualities . these spaces are usually referred to as `` t - folds '' @xcite or more generally as `` non - geometric backgrounds '' @xcite . while string dualities may help to understand such backgrounds , a direct study beside the most simple realizations such as asymmetric orbifolds has proved quite challenging . a clearly fascinating question is how the notion of differentiable manifolds should be changed to accommodate such stringy geometries . one might think that t - duality covariant formalisms such as generalized complex geometry @xcite and double field theory @xcite are relevant tools for such investigations . these approaches provide a natural framework for the inclusion of the b - field ( as well as rr - fields ) , but there are also attempts to describe non - geometric backgrounds in this context , see for example @xcite . in view of these motivations , it is of interest to obtain approximate solutions in which there is a non - trivial duality symmetry between various patches , and which provide a concrete realization of t - folds . this will be the subject of the present paper . a useful approach is to consider string theory compactified on a torus , and fiber the resulting t - duality group over a base @xmath2 . we develop a geometric point of view in which the duality group is identified with the mapping class group @xmath16 of a given surface @xmath1 fibered over @xmath2 . we then classify all possible torus fibrations with geometric and non - geometric twists in terms of a classification of elements in @xmath16 . in particular , we identify t - duality elements with dehn twists of @xmath1 and we categorize them according to the type of surface diffeomorphisms . most of the paper will deal with the simplest example of a two torus @xmath9 , in which case a partial geometrization of the t - duality group @xmath17 is obtained by simply taking @xmath18 , where @xmath8 is the complex structure of @xmath9 , @xmath19 is the complexified khler form and @xmath20 is identified with the compactification torus . as we will discuss shortly , in some cases there is a physical picture for the auxiliary @xmath21 . we first consider the case in which the base @xmath2 is a circle @xmath22 . if we only vary @xmath8 , the fibration is specified by a monodromy @xmath23 which is used to identify the fibers after encircling the base @xmath22 . the monodromy can be factorized as a product of dehn twists , and the total space of the fibration is a 3-manifold @xmath24 , known as the mapping torus for @xmath25 , whose geometry is determined by the type of torus diffeomorphism @xmath26 ( which can be either reducible , periodic or pseudo - anosov , respectively matching parabolic , elliptic and hyperbolic @xmath7 conjugacy classes ) and is either an euclidean space , a nil or a sol geometry . by replacing @xmath8 with @xmath19 we obtain a classification of possible t - fold metrics . we can think of the monodromy in @xmath19 as a product of dehn twists of the auxiliary torus @xmath21 fibered over @xmath22 . we then consider a fibration of the torus over a two dimensional base @xmath27 . this is the familiar situation of stringy cosmic strings @xcite and the parameters @xmath8 and @xmath19 need to be holomorphic functions on the base where a collection of points , at which the fiber torus degenerates , has been removed . this situation can be described by two elliptic fibrations @xcite and the corresponding degenerations , classified by kodaira @xcite , are associated to stringy cosmic five - branes . for each degeneration is associated a monodromy , namely an element of the mapping class group of @xmath20 or @xmath21 . non - trivial elements of @xmath28 , with the exception of a perturbative shift @xmath29 which is associated to a ns5 brane @xcite , will give non - geometric solutions . the fibration data for the auxiliary @xmath21 can be determined in the heterotic theory by a fibration of heterotic / f - theory duality @xcite , and the non - geometric backgrounds are mapped to geometric k3-fibered calabi - yau compactifications on the f - theory side . we will generically refer to such degenerations as t - duality defects , or _ t - fects_. it is interesting to ask to what extent one can invert this reasoning , namely if it is possible to determine , from a classification of elements in the mapping class group , which monodromies of the fibration on the boundary @xmath30 of a small disk @xmath31 correspond to a family of degenerating surfaces at a point in @xmath31 . as a result of an elaborate theorem @xcite , one can make this approach precise and basically obtain a topological classification of singular fibers . in order to construct approximate solutions for such backgrounds we use a semi - flat approximation @xcite where the fields do not depend on the fiber coordinates . in our case this means to preserve the @xmath32 isometry of the torus . for the geometric fibrations this is a very good approximation of the full metric , up to exponentially suppressed terms close to the degenerations . we then construct all possible local geometries for a fibration on a small disk @xmath31 with a given monodromy @xmath33 $ ] at the boundary , for each conjugacy class @xmath34 $ ] of @xmath35 , by solving the corresponding cauchy - riemann equations . some of the geometric solutions are identified with a smeared version of known branes , such as the kk monopole . solutions for the parabolic conjugacy class in @xmath36 give , other then a smeared ns5 brane solution , a class of non - geometric backgrounds in which patching requires @xmath37-transformations . in particular , we recover the solution for the @xmath38 exotic brane recently discussed in @xcite . we obtain a class of solutions for elliptic @xmath19-fects which are related to an asymmetric orbifold description @xcite . we also construct solutions with hyperbolic monodromy , whose corresponding mapping torus is a sol - geometry . the volume of the fiber torus in such backgrounds is a highly oscillating function near the would - be degeneration point , in line with the fact that hyperbolic monodromies can not be obtained from colliding degenerations of elliptic fibrations . however , such solutions can approximate a distribution of branes . when both @xmath8 and @xmath19 fibrations are allowed to degenerate at the same point , we obtain non - geometric solutions which are not t - dual to geometric ones . in particular we discuss the example of double elliptic monodromies and the relation with the corresponding asymmetric orbifold studied in @xcite . the solutions that we construct give approximate local geometries around degenerations of a given global model . for the case of elliptic fibrations , these local solutions will be approximated by an expansion of the solutions for @xmath8 and @xmath19 determined by the appropriate combination of hypergeometric functions that invert klein s @xmath39-invariant . the same result holds for the axio - dilaton profile in f - theory . finally , we describe from the geometric point of view fibrations of a genus 2 surface @xmath40 , that provides a geometrization of the heterotic t - duality group @xmath41 @xcite . degenerations of surface fibrations over @xmath42 correspond to heterotic t - fects . we describe a classification of such defects in terms of elements in @xmath43 and we give the monodromy factorization for some of them . let us mention few possible applications of our results . having explicit solutions which require patching by arbitrary elements of the t - duality group should be useful to investigate the correct geometric description of such backgrounds , as in the approach of @xcite , as well to understand general non - geometric microstate solutions for black holes @xcite . it would be interesting to see if solutions with the monodromies considered in this paper can arise via brane polarization as in @xcite , as this would be an additional mechanism to obtain a globally well defined solution , since only dipole exotic charges would be nonzero . we also note that exotic branes should be relevant for cosmological billiards @xcite and it would be nice to clarify the relation with the t - fects described here . we mention that a particular kind of elliptic @xmath19-fects ( described in terms of a gauged linear sigma model ) were used in @xcite as an ingredient to `` uplift '' ads / cft dual pairs to ds ones . the semi - flat solutions we are considering incorporate the backreaction of such objects . lastly we note that , as mentioned in @xcite , there is a close relation between the codimension-2 t - fects that we study in this paper and the theory of non - abelian anyons and topological quantum computation @xcite . the approach we take here , by identifying the monodromy ( or charge ) of the t - fects in terms of dehn twists , makes this analogy even more compelling , since in both situations braid groups play a central role . in the case of abelian anyons , there is also a close relation with quantum groups @xcite . it would be interesting to see if there is a connection with the approach we develop in this paper . this paper is organized as follows . in section [ sec : torusbundles ] we recall a number of elementary facts about torus fibrations and the @xmath7 mapping class group , in a way that makes easy the generalization to fibrations of arbitrary genus . in section [ sec : tfolds ] we classify torus fibrations over a circle by means of monodromy elements in the mapping class group and we write explicit metrics for the total spaces . we then apply the same technique to fibrations in which the khler modulus undergoes non - trivial monodromies , obtaining a classification of t - fold metrics . in section [ sec : t - walls ] we discuss a way to obtain explicit solutions for domain walls that realize t - folds backgrounds . we dub such solutions t - walls . in section [ sec : exoticbranes ] we consider fibrations over a two - dimensional base and we derive local solutions around a degeneration with arbitrary monodromy in @xmath35 , by solving the corresponding cauchy - riemann equations for @xmath8 and @xmath19 . we refer to such degenerations as t - fects . some of these t - fects are identified with a semi - flat approximation of known brane solutions , while some of them are new exotic solutions . in section [ sec : hyperellipticfibrations ] we briefly discuss the generalization of our results to fibrations of genus 2 surfaces and their relation to heterotic t - fects . we present our conclusions and a list of open questions in section [ sec : conclusions ] . we relegate various materials in the appendices . in appendix [ app : mcg ] we recall briefly basic results regarding mapping class groups and hyperbolic maps , in appendix [ app : braid ] we discuss the relation between the braid action on the monodromy factorization in terms of dehn twists and the familiar technique of moving branch cuts and the abc factorization used in f - theory . in appendix c and d we give some details on the semi - flat geometries and the embedding of the mapping class groups in @xmath0 . we begin by considering a @xmath9 fibration over a base @xmath2 . we will give a presentation of the results that can be readily generalized to higher genus fibrations . the fiber torus can be either part of the ten dimensional space - time , or be an auxiliary space that geometrizes an @xmath7 duality group , as in f - theory @xcite . we will be mainly interested in fibering the t - duality group @xmath44 over the base @xmath2 . the complex structure @xmath8 and the complexified khler form @xmath19 are defined as : @xmath45 where @xmath46 is the metric on @xmath9 , @xmath47 and @xmath48 is the component of the @xmath48-field . the actions of the two @xmath7 groups on @xmath8 and @xmath19 are mbius transformations : @xmath49 \equiv \frac{a \tau + b}{c \tau + d } \ , , \qquad m_{\tau } = \begin{pmatrix } a & b\\ c & d\end{pmatrix } \in sl(2,\mathbb{z})_{\tau } \ , , \\ \rho & \rightarrow m_{\rho}[\rho ] \equiv \frac{\tilde a \rho + \tilde b}{\tilde c \rho + \tilde d } \ , , \qquad m_{\rho } = \begin{pmatrix } \tilde a & \tilde b\\ \tilde c & \tilde d\end{pmatrix } \in sl(2,\mathbb{z})_{\rho } \ , .\end{aligned}\ ] ] note that the kernel of this action is @xmath50 . the two @xmath51 factors are the mirror symmetry @xmath52 and a reflection @xmath53 . a partial geometrization of the duality group is obtained by identifying the two @xmath7 factors with the group of large diffeomorphisms of two tori , @xmath20 ( the compactification torus ) and @xmath21 . we will first consider torus bundles with @xmath2 a circle , and later we will study fibrations over a two dimensional base , where the circle becomes contractible . the first situation has been discussed many times in the context of scherk - schwarz dimensional reduction and as toy model for non - geometric backgrounds ( see for instance @xcite ) , but our discussion will focus more on the role of monodromy . the case @xmath27 was introduced in @xcite and it can be understood , for the heterotic theory , from a duality with f - theory @xcite . this is the most interesting situation , since if one describes the auxiliary @xmath21 fibration as an elliptic fibration , the corresponding line bundles on @xmath2 that specify the fibration can be mapped explicitly to geometric compactifications on the f - theory side . reversing this map , one can determine the conditions on the non - geometric @xmath54 fibration to obtain sensible string vacua . we will make precise the relation between these two approaches , and we will provide a comprehensive classification of local geometries that arise in this context . we will later generalize our results beyond torus fibrations , motivated by the heterotic t - duality group with a single wilson line . we note that some of the results we obtain can be applied in more general contexts as well . let us review some elementary facts about monodromies of torus fibrations . we leave some of the details to appendix [ app : mcg ] . we identify the @xmath7 group with the mapping class group @xmath55 , the group of large diffeomorphisms of a two torus @xmath9 , by the standard bijective map @xmath56 it will be useful to consider a fibration @xmath57 of the @xmath9 over a circle constructed as @xmath58}{(x , 0)\sim ( \phi(x),2\pi ) } \ , , \ ] ] where @xmath59 . @xmath24 is known as the mapping torus for @xmath26 and we call @xmath26 the monodromy of the fibration . we will often use identification by @xmath60 and write @xmath61 . it is a well known result that the geometry of the mapping torus is completely determined by the trace of @xmath26 in @xmath7 , or equivalently by the kind of induced diffeomorphism of @xmath9 . let us briefly recall how the group @xmath55 is generated . a presentation of the mapping class group of the torus is given in terms of dehn twists along two closed curves @xmath62 with intersection number one , that we denote by @xmath63 , @xmath64 . we then have : @xmath65 note that the first relation is a braid relation . a simple choice for the curves is the standard basis for the homology , see figure [ fig : twistt2 ] . , @xmath66 . right : the action on @xmath66 of a dehn twist along the @xmath67-cycle , represented by the matrix @xmath63 in . ] this gives the following matrix representation of the two dehn twists : @xmath68 which indeed satisfy . we now study a useful classification of elements in @xmath55 . this comes from the characterization of elements in @xmath7 inherited by the classification of orientation - preserving isometries of the hyperbolic space , from the identification @xmath69 given by mbius transformation . for a given isometry @xmath26 represented by a matrix @xmath70 , the following mutually exclusive possibilities are given : 1 . _ elliptic type _ : @xmath26 has one fixed point in @xmath71 , or equivalently @xmath72 . parabolic type _ : @xmath26 has no fixed points in @xmath71 and exactly one fixed point on @xmath73 , or equivalently @xmath74 . hyperbolic type _ : @xmath26 has no fixed points in @xmath71 and exactly two fixed points on @xmath75 , or equivalently @xmath76 . + a geometric picture of this classification for @xmath77 is shown in figure [ fig : isometries ] . note that elliptic elements are necessarily of finite order . we also remark that this trichotomy does not exhaust all conjugacy classes in @xmath7 . indeed , there are 6 conjugacy classes of elliptic type , given by the matrices @xmath78 respectively of order 3 , 4 , 6 , together with their inverses . note that from the relations , the inverses are given by @xmath79 the only other finite order element is @xmath80 . for parabolic elements , which are of infinite order , there are infinite conjugacy classes . note that both the @xmath63 and @xmath64 dehn twists are in the same conjugacy class ; a general representative is @xmath81 for hyperbolic elements there is a conjugacy class for each value of the trace , plus additional sporadic classes , whose number can be computed by studying equivalence classes of binary quadratic forms , as explained for example in @xcite . two examples are given by @xmath82 we also provide a dehn twist decomposition for the sporadic elements listed in @xcite ( some of the following matrices are conjugated to their representative matrices ) : @xmath83 the subscript indicates the trace . we note that the decomposition @xmath84 seems not to be conjugate to the element with the same trace , @xmath85 , nor to its inverse for @xmath86 so we conjecture that this gives a list of sporadic conjugacy classes for hyperbolic elements of even trace @xmath87 , @xmath88 . case . however , we could not find any element that conjugates the matrix @xmath89 to @xmath90 or its inverse . ] the classification obtained from the trace of @xmath7 elements is reflected in a trichotomy of the corresponding torus diffeomorphisms . these are periodic , reducible and anosov maps . we refer to appendix [ app : mcg ] for a brief review of mapping class groups . a standard result in the theory of 3-manifolds is that the class of torus diffeomorphism @xmath26 determines the geometry of the corresponding mapping torus @xmath24 , which can be either euclidean , nil or sol . we will see an example of this in the next section . we can apply the discussion of the previous section to string theory compactified on a two torus @xmath20 . an additional fibration on a circle generates a class of 3-manifolds which provides the simplest examples of nil- and sol - manifolds . if we also include monodromies in @xmath36 , classified by an auxiliary @xmath21 fibration over the circle , the resulting 3-spaces will be in general non - geometric and are usually referred to as t - folds @xcite ( see for example @xcite for a discussion of such spaces in various contexts ) . usually , one arrives to such spaces by performing a chain of t - duality transformations @xcite . the simplest example is to start with a three torus @xmath91 and a b - field @xmath92 . busher rules on @xmath93 give a metric @xmath94 and @xmath95 . to make sense of the metric globally one identifies @xmath96 . note that later in this section and in section [ sec : exoticbranes ] we will find more convenient to use conventions for which the period of @xmath97 is @xmath98 . the resulting 3-manifold is a nil geometry and it is identified with a mapping torus @xmath24 for an element @xmath99 that acts by @xmath100 . a further application of busher rules on @xmath101 gives a metric @xmath102 and a b - field @xmath103 , with @xmath104 . the monodromy is now an element of @xmath36 and acts by @xmath105 . this has a non - trivial action on the volume , and the corresponding 3-space is a t - fold . the three torus with b - field obviously is not a string vacuum , so this simple example should be considered as an illustrative toy model . an important caveat is the fact that the nilmanifold is not a principal torus fibration , and the lack of a global @xmath106 isometry makes the application of busher rules a priori not obvious . for a detailed discussion on such `` obstructed '' t - duality see for example @xcite , and @xcite for a discussion on how to possibly overcome this problem . our logic will be different and we do not rely on t - duality . in this section , we will classify all possible geometric and non - geometric 3-spaces from the corresponding classification of mapping class groups of the @xmath20 and @xmath21 fibrations . in the following sections , we will construct explicit solutions for such t - folds and we will use the results of this section to classify all possible local geometries around defects arising from fibration of the t - duality group on @xmath42 . the mapping tori we are considering will arise at a boundary of a fibration on a small disk encircling such t - fects . in particular , a subset of mapping class group elements arises from degenerations of elliptic fibrations dual to geometric f - theory compactifications . we will first briefly review geometric torus bundles with twists in @xmath107 and we will then discuss spaces with non - geometric twists . the monodromy of the fibration , that determines the geometry of the total space , is a product of dehn twists . the simplest example is a twist along one cycle of the homology basis : the total space is a nilmanifold . ] we now consider a torus fiber with @xmath108 fixed and with a complex structure @xmath109 varying on @xmath110 , as shown in figure [ fig : mappingtorus ] . this example is elementary and has been discussed in some details in @xcite but it serves well to illustrate the role of monodromy . in order to construct the torus fibration , we need to determine the function @xmath109 , given a monodromy @xmath111 . the metric of the total space is then : @xmath112 where the torus metric is , from , @xmath113 we set @xmath114 and we fix the radius of the base circle to @xmath115 . through this section we will set @xmath116 , @xmath117 . the action of a given element @xmath70 on @xmath8 is by mbius transformation . we fix @xmath118 and we demand that @xmath8 has the monodromy @xmath119 $ ] . to find such a function we consider the element @xmath120 of the lie algebra @xmath121 . then , we construct the path @xmath122 in @xmath77 given by the exponential map @xmath123 . by construction , @xmath124 . we define then @xmath125 \ , , \ ] ] which obviously has the desired monodromy properties : @xmath126 , @xmath127 $ ] . we will discuss each of the three classes of @xmath107 monodromies , and the corresponding torus diffeomorphisms , separately . we first consider a monodromy of parabolic type , namely an element @xmath128 with @xmath129 . the simplest example is a shift @xmath100 : @xmath130 this monodromy implements a @xmath131-th power of a dehn twist along the @xmath66 cycle of the torus once encircling the base and the complex structure is given by @xmath132 , where @xmath133 is an arbitrary complex parameter . the corresponding total space is a nilmanifold . this can be seen from the metric which ( fixing @xmath134 ) becomes @xmath135 with the coordinate identification discussed above , that implements the desired monodromy . the metric can be written explicitly as a left - invariant metric on a nilpotent lie group @xmath136 as follows . we introduce the maurer - cartan forms @xmath137 and @xmath138 , which satisfy : @xmath139 the generators @xmath140 of the corresponding lie algebra @xmath141 then satisfy @xmath142 = -(\mathbf{m}_{\tau})_{a}^{\phantom{a } b}\ , t_b \ , , \quad [ t_a , t_b ] = 0 \ , .\ ] ] we see from and that @xmath141 is the heisenberg algebra and the torus fibration can also be seen as a principal circle bundle over a torus . the global identification that makes the space compact is a quotient of the lie group @xmath136 by a discrete subgroup . in the notation of @xcite this space is referred to as @xmath143 . the existence of a compactification of the lie group can be inferred from the condition that the lie algebra element @xmath144 is traceless . the previous example is the simplest example of a nil geometry and , as we discussed above , it can be obtained by t - duality from a three torus with non - vanishing b - field . we can also consider a monodromy which implements a dehn twist around the @xmath67 cycle of the fiber @xmath9 : @xmath145 which corresponds to the following action on @xmath8 : @xmath146 note that this corresponds to a shift of @xmath147 instead of a shift of @xmath8 : @xmath148 . more generally , for each @xmath131 the conjugacy class of a parabolic monodromy is labeled by two coprime integers @xmath149 : @xmath150 with @xmath151 . the corresponding solution for @xmath109 is given by : @xmath152 = \frac{(2\pi+n pq \theta)\tau_0 + n p^2 \theta}{-n q^2 \theta \ , \tau_0 + ( 2\pi - n pq \theta ) } \ , .\ ] ] the corresponding metric on the total space is : @xmath153 where @xmath154 this is of course the expected result , and the identification of the coordinates now corresponds to gluing the @xmath9 after performing a @xmath131-th power of a dehn twist along a cycle represented by @xmath155 + q [ u]$ ] . it is easy to see from and that the lower central series , namely the sequence of ideals @xmath156 , \ , \mathfrak{g}^2 = [ \mathfrak{g},\mathfrak{g}^1 ] , \dots$ ] , terminates and the algebra is nilpotent . one might think that different monodromies in the same conjugacy class are uninteresting , however we will see by repeating the previous analysis for twists in @xmath36 , that parabolic elements in the same class give rise to very different local metrics . we now consider an elliptic conjugacy class represented by the order 4 monodromy : @xmath157 from we then find the following solution for @xmath109 ( see also @xcite ) : @xmath158 the corresponding total space is a compactification of the lie group iso(2 ) , namely the symmetries of the euclidean 2-plane . this can be seen from the local coordinates representation of the left - invariant forms : @xmath159 which satisfy @xmath160 . the left - invariant metric is then @xcite @xmath161 . analogous results apply to the remaining finite order conjugacy classes . for example , for the conjugacy classes of order 6 and 3 , represented by @xmath162 , @xmath163 , we have @xmath164 and the complex structure is : @xmath165 all 3-manifolds obtained in this way are compactifications of lie groups of a unimodular solvable ( but not nilpotent ) algebra . note that if the modulus @xmath133 is chosen to be the fixed point of the monodromy element , namely @xmath166 the complex structure does not depend on the circle coordinate and we have @xmath167 . these fixed points are the minima of the potential for the moduli obtained by a reduction with duality twsits and at those points one has a cft description in terms of symmetric orbifolds . more details can be found in @xcite . lastly , we consider hyperbolic monodromies , corresponding to anosov diffeomorphisms of the fiber torus . let us first give a simple example with monodromy in @xmath77 : @xmath168 the corresponding modulus is given by @xmath169 and the metric is that of a sol - manifold . for example , the metric on the total space assuming @xmath133 is an imaginary parameter is : @xmath170 from the relations we see that the corresponding algebra is solvable . however , the lower central series generated by @xmath171 $ ] for the @xmath121 element do not terminate and thus the algebra is not nilpotent . the compact torus bundle is obtained by a quotient of the lie group sol by a discrete subgroup ( see for example @xcite for more details ) . note that the background was discussed in @xcite and since then sol - manifolds have played an important role in the search for de sitter vacua in string theory . some of the hyperbolic @xmath7 conjugacy classes , labeled by an integer @xmath131 are given by : @xmath172 note that the matrix @xmath173 representing arnold s cat map on the torus , is conjugate to the matrix above for @xmath174 . the monodromy matrix can be diagonalized with two eigenvalues ( @xmath175 , @xmath176 ) , with @xmath177 powers of the anosov diffeomorphism stretch and contract exponentially the two eigenspaces . for this reason , the repackaging of the torus image under powers of an hyperbolic map into the fundamental domain results in a complicate behavior that share many similarities with chaotic systems . we refer to appendix [ app : mcg ] for a discussion on the dynamics of hyperbolic maps . in terms of @xmath175 the algebra element is given by @xmath178 and the corresponding modulus is then given by : @xmath179 this expression again determines a metric on the total space of the torus bundle . analogous solutions can be found for the remaining conjugacy classes . given the peculiar nature of anosov diffeomorphisms , it would be interesting to revisit phenomenological applications of string models based on hyperbolic torus bundles . we now discuss the case of a torus bundle in which we fix the complex structure @xmath180 of the fiber but we allow @xmath181 to vary along the base @xmath110 . a classification of such spaces can be obtained formally by applying the results of the previous section to an auxiliary @xmath21 fibration on the circle , that geometrizes the @xmath36 factor of the t - duality group . a local expression for the three dimensional metric and the b - field can be then obtained from : @xmath182 where @xmath183 . with the exception of gauge transformations for the b - field , the monodromy mixes the volume and the b - field , and there is an obstruction to glue the ends of a mapping cylinder with @xmath19 monodromy : the resulting total space will not be a manifold . there exists particular limits in moduli space at which we have a string description of such spaces from asymmetric orbifold cfts , but otherwise there is no clear string description of such t - folds . our discussion for now will be formal , but as we will discuss in the next sections , the non - geometric monodromies that we describe do arise in fibrations over @xmath42 , and in some cases one could obtain evidence for the existence of the corresponding string vacuum from string dualities . let us consider the parabolic monodromy @xmath184 this corresponds to a shift of @xmath19 , giving @xmath185 and can be understood geometrically as a power of a dehn twist along the @xmath66 cycle of the auxiliary @xmath21 . the corresponding total space is just a three torus @xmath186 with @xmath131 units of @xmath187-flux , and can be obtained by t - duality from the nilmanifold along the global @xmath106 isometry as we discussed at the beginning of this section . if we consider the conjugate monodromy @xmath188 we obtain the following solution for @xmath19 : @xmath189 the gluing condition now mixes the volume of the fiber torus with the b - field , as follows from the action on the volume : @xmath190 we could consider a fibration of the torus on a interval , but we encounter an obstruction to glue the endpoints to obtain a torus bundle . note that for @xmath191 , from we obtain the following metric : @xmath192 which coincides with the t - fold metric we obtain from an obstructed t - duality on the nilmanifold . we can also consider the general @xmath149 parabolic conjugacy class as in , thus obtaining a class of t - fold metrics that interpolate between the 3-torus and the t - fold . we can now repeat the analysis of finite order elements done for the geometric @xmath8 monodromies . the corresponding solutions for @xmath181 , for monodromies of order 3 , 4 and 6 , are obtained by and by a fiberwise mirror symmetry @xmath193 . the corresponding non - geometric t - folds have been discussed in the context of generalized scherk - schwarz reduction within double field theory in @xcite . the corresponding potential @xmath194 admits a minimum at the fixed point of the monodromy matrix . at such minimum , there exists a description in terms of an asymmetric orbifold cft @xcite . we note that in the cft description ( as well as in double field theory ) we see the presence of both @xmath187-flux and @xmath195-flux . these fluxes should be identified by the corresponding monodromy , the perturbative shift @xmath196 and the non - geometric twist @xmath197 . the simultaneous presence of both fluxes fits well with the monodromy decomposition of the elliptic monodromy of order 4 , @xmath198 . as we will discuss in the next section , such decompositions are defined up to some redundancy , which basically follows from a braid relation of the dehn twist decomposition . in this case we have for example , from , @xmath199 , so it is not simple to match the monodromy decomposition with the corresponding flux parameters . we will come back to this point when we will study fibrations on a @xmath42 base and we will identify the sources for the fluxes . the last case is that of a hyperbolic monodromy in @xmath19 . a simple example is again obtained by taking the @xmath77 monodromy , which for a choice of parameters gives a metric @xmath200 . let us discuss the action on the volume for an @xmath7 representative element @xmath201 the action on @xmath19 is given by @xmath202 we note that the action on the volume given by is the same as for the elliptic case : @xmath203 however , iterations of the map are not periodic and act in a complicated way . for example , successive iterations act as : @xmath204 and for @xmath205 the continued fraction does not terminate since the fixed points of the transformation are quadratic irrational numbers . the exact expression can be found by setting @xmath206 in with @xmath207 . in terms of the largest eigenvalue @xmath175 we find the following action on the volume : @xmath208 again , we refer to appendix [ app : mcg ] for more details on the dynamics of hyperbolic diffeomorphisms of surfaces . finally , we briefly discuss the situation in which the full t - duality group is fibered over the circle , and both @xmath109 and @xmath181 vary . if such configurations are allowed , we can get non - geometric backgrounds which are not fiberwise mirror symmetric to any geometric background . we will discuss the consistency of such configurations in the next sections in relation to heterotic / f - theory duality . we point out that a torus fibration with both @xmath8 and @xmath19 elliptic monodromies has been studied recently in @xcite . an example is the order 4 solution : @xmath209 here the parameters @xmath210 and @xmath14 should be identified with geometric and non - geometric fluxes . such configuration , being not t - dual to a geometric space , can be described in double field theory at the price of relaxing the strong constraint for the generalized killing vectors . the potential for the moduli @xmath211 has a minimum at the fixed points of the monodromy @xmath212 . at this minimum , an asymmetric orbifold description was studied in @xcite . we note that the presence of both kind of geometric and non - geometric fluxes should be related in the present context to the factorization of the monodromy in terms of dehn twists . as we will discuss in more details in the following sections , we should really think in terms of mapping class group of a genus 2 surface obtained by gluing the two tori @xmath20 , @xmath21 with a trivial map . if we then embed the representative matrices @xmath213 , @xmath214 in @xmath215 matrices acting on the homology of the genus-2 surface , we can think about the double elliptic monodromy as being decomposed into @xmath216 . this would corresponds to having two kind of geometric fluxes , as well as @xmath187 and @xmath195-fluxes , in agreement with the analysis of @xcite . we will discuss more about genus 2 fibrations in the last section . we will now turn to the construction of explicit solutions that provide a concrete realization of the spaces described in this section . before studying in details fibrations over two - dimensional bases , we mention here a way to embed the 3-spaces discussed in the previous section into backgrounds that solve the supergravity equations of motion . this construction was described in a different context by hull in @xcite and further discussed in @xcite . let us consider the mapping torus with a parabolic monodromy @xmath217 , namely a nil - manifold , whose metric is given in . the latter is determined by the complex modulus @xmath218 ( for simplicity in this section we take a periodicity @xmath219 ) . we can now make an ansatz for a ten - dimensional metric of the form @xmath220 , where the mapping torus is fibered over a line parametrized by a coordinate @xmath221 . we then assume that @xmath133 is a function of @xmath221 : @xmath222 and we further assume that @xmath223 and @xmath224 @xcite . the ricci curvature is then given by @xmath225 it is easy to see that a flat solution which also solves the ricci - flatness condition is given by @xmath226 , with @xmath227 an arbitrary constant ( see figure [ fig : twall ] ) . to obtain a sensible solution we can then take the following functions : @xmath228 \ , , \quad \tau = \left[n\theta(s ) - m\theta(-s)\right ] ( \theta + i s ) + i c\ , , \ ] ] where @xmath229 is the step function . for @xmath230 the background becomes : @xmath231 + \frac{1}{h(s)}\left [ n \theta(s ) \theta dx+ dy\right]^2 \ , .\ ] ] note that at @xmath232 the solution breaks down and there is a curvature singularity @xmath233 . as we will discuss in more details in the next section , the background corresponds to a kk monopole smeared on two of the @xmath234 directions , and so we should accept the singularity at @xmath232 . alternatively , one can obtain the same solution by first smearing a ns5 brane on a transverse @xmath235 and then t - dualize . the solutions clearly make no sense at infinity , much like a d8 brane , and should be regarded as local approximations . for general @xmath236 , the @xmath232 degeneration acts much like the domain - wall d8 brane : it separates two regions with mapping tori of different monodromy . it is easy to construct the analogous solution with parabolic monodromies in @xmath19 . the geometric monodromy @xmath237 gives indeed the ns5 brane smeared on a @xmath235 : @xmath238 \ , , \\ b & = n \theta \theta(s ) dx \wedge dy \ , , \quad e^{2\phi } = h(s ) \ , , \end{aligned}\ ] ] where @xmath239 and we made a gauge choice for the @xmath48-field . a t - fold solution can be obtained starting from the solution . since the monodromy is now a perturbative shift in @xmath240 , we take the following ansatz : @xmath241 . we then find the following solution : @xmath242 + \frac{h(s)}{1 + n^2 h(s)^2}\left[dx^2+dy^2\right ] \ , , \\ b & = \frac{n \theta}{h(s)^2 + n^2\theta^2 } dx\wedge dy \ , , \quad e^{2\phi } = \frac{h(s)}{1 + n^2 h(s)^2 } \ , .\end{aligned}\ ] ] this solution coincides with the t - dual along @xmath101 of the smeared kk monopole . again one could glue t - folds with different monodromies on the two sides of the domain wall . for this reason such defects could be called _ t - walls_. for the domain wall considered in the text . ] in principle it should be possible to generalize this construction to obtain t - walls with arbitrary t - folds ( see @xcite for an example with hyperbolic monodromy ) . however , embedding such local solutions into a global consistent background is not simple , and unless one can understand the degenerations as t - dual of known ones , it is not clear how to discern good and bad singularities . in the next section we will study codimension-2 defects , where a systematic construction of local supergravity solutions is possible for arbitrary t - duality monodromies . moreover , in some cases the non - geometric fibrations can be understood from string dualities , and this makes possible to eventually put the existence of such backgrounds on more firm ground . in this section we generalize the previous discussion to the case of a fibration of the t - duality group @xmath0 over a two dimensional base @xmath2 . this configuration is well known from stringy cosmic strings @xcite and was discussed in the context of non - geometric backgrounds in @xcite . a complementary point of view , recently described in @xcite , is to describe the fibrations of the complex structure @xmath8 and the complexified khler form @xmath19 in heterotic string theory as elliptic fibrations , and map the corresponding line bundles to an f - theory compactification via heterotic / f - theory duality . a surprising result of this analysis is that even spaces with non - geometric @xmath19 monodromies are mapped to geometric compactifications of f - theory . in such configurations @xmath8 and @xmath19 are meromorphic functions on the base . there is a set of points , that we denote @xmath243 , at which the fibration degenerates and a defect sits ( see figure [ fig : torusfibration2d ] ) . these points are branch points for @xmath8 and @xmath19 and there is a nontrivial monodromy around them . possible degenerations of the elliptic fibrations are described by the kodaira list as it is familiar in f - theory . since such defects have t - duality monodromy around them , we will call them generically _ t - fects_. here we take a slightly different point of view , with the goal of understanding the relation between the t - folds described in the previous sections and the defects that arise in the case of fibrations over @xmath42 . we will consider all possible monodromies in the mapping class group of the compactification torus @xmath244 and the auxiliary torus @xmath21 and classify the corresponding local solutions . in principle , this classification is more general then the kodaira classification , since not all of the @xmath7 conjugacy classes arise from degenerations of elliptic curves . an illustration of this in f - theory is the case of non - collapsible singularities studied in @xcite . we will clarify the relation between these two approaches in a way that can be easily generalized to higher genus fibrations . the latter situation , for genus @xmath245 , is of interest in the case of heterotic compactification on a @xmath9 with a wilson line @xcite and we outline an interesting geometric perspective on the algebraic classification of genus 2 degenerations @xcite , as well as a generalization to include non - collapsible defects in this case . another outcome of our analysis is the list of all possible geometries arising in a neighborhood of a given @xmath8 and @xmath19 degeneration . in the case of a geometric fibration , some of such geometries can be interpreted as a semi - flat approximation @xcite of a class of known ten - dimensional brane solutions , such as the ns5 brane and the kk monopole . in the case of non - geometric fibrations , we recover the exotic brane solution recently discussed in @xcite . our list contains new exotic solutions corresponding to t - fects described by different @xmath7 conjugacy classes . as in the case of torus fibrations over a circle , when both @xmath8 and @xmath19 degenerations collide , we found solutions which are not t - dual to a geometric one . a crucial point that need to be emphasized is about the viability of such local solutions . the solutions alone can not be taken as automatic evidence that the corresponding degeneration exists in string theory . put in another way @xcite , one has to supplement the semi - flat solutions with the knowledge of the microscopic description of a given degeneration , at which the semi - flat approximation breaks down . once this is understood , the semi - flat torus fibration is a powerful tool to understand physical properties of the corresponding t - fects . one possible point of view is that heterotic / f - theory duality can provide the evidence , even if quite indirect , that non - geometric degenerations should be allowed and in the situations in which enough supersymmetry is preserved , the supergravity solutions should provide a qualitatively correct description . . the mapping torus at the boundary of a small disk @xmath31 that encircle a degeneration point determines the local geometry . ] [ [ relation - to - previous - works ] ] relation to previous works + + + + + + + + + + + + + + + + + + + + + + + + + + we note that some of the exotic brane solutions , which correspond to a subset of the solutions for our t - fects , have been described in various places , from different point of views and with a plehtora of different names . we give a short guide to the literature in order to facilitate the comparison with our discussion . as we already mentioned , a parabolic exotic brane usually referred to as @xmath246-brane has been studied in details in @xcite and in @xcite , where it was named q - brane . early works include @xcite . in @xcite , parabolic , elliptic and hyperbolic branes were called n , k and a - branes respectively ( following an iwasawa decomposition ) . solutions for elliptic and hyperbolic branes for the s - duality @xmath7 group have been discussed in @xcite for the 7-branes of f - theory , where they are referred to as q7-branes , and in @xcite as solutions of five dimensional supergravity . elliptic branes were used in @xcite as an ingredient to uplift holography and they were referred to as sc5-branes ( stringy cosmic five - branes ) . an elliptic brane was studied in @xcite and called @xmath247-brane . there have been also more general investigations of codimension-2 solutions , sometimes referred to as `` defect branes '' ( see for example @xcite ) . here we will simply refer to all the codimension-2 defects with t - duality monodromies as t - fects , emphasizing the relation with the t - folds described in section [ sec : tfolds ] . let us briefly describe torus fibrations from the point of view of the mapping class group . the following discussion is in part inspired by the theory of lefschetz fibrations @xcite , and we stress that it can be applied to fibrations of surfaces of arbitrary genus . we consider a torus fibration on the punctured sphere @xmath248 , with @xmath249 . let us suppose that at a point @xmath250 the fiber degeneration is such that a cycle @xmath251 + q [ v]$ ] , where @xmath67 and @xmath66 is the basis used in section [ sec : torusbundles ] , shrinks to zero . the monodromy of the torus fiber around @xmath250 is a positive dehn twist @xmath252 along the vanishing cycle and the corresponding action on the homology is given by the picard - lefschetz formula ( we refer to appendix [ app : mcg ] for details ) . consider now a disk that encircles some of the degeneration points . if the disk contains the whole @xmath253 , the monodromy around the disk should be homotopic to the identity in order to have a globally defined fibration ; otherwise we will have a monodromy @xmath70 , that we can factorize as the product of dehn twists around each degeneration point : @xmath254 , as shown in figure [ fig : monodromy ] . this factorization has some gauge redundancy . locally , we make a choice for defining the monodromy around each points , for example by choosing an arc that connects @xmath250 to a given smooth reference point @xmath255 . the choice is arbitrary , and this results in a equivalence class of monodromy factorizations obtained from the list of vanishing cycles by the action of a braid group @xmath256 : @xmath257 this action is usually called an _ hurwitz _ move and it clearly preserves the global monodromy . we illustrate this in more details in appendix [ app : braid ] , where we use it to derive the relation between our dehn twist factorization for the @xmath7 monodromies , and the abc factorization commonly used in f - theory ( in this context the hurwitz moves arise by moving branch cuts across branes , as in @xcite ) . there is an additional freedom in the choice of the smooth reference fiber over @xmath255 , which is just a global conjugation on @xmath70 . we arrive at the conclusion that the torus fibration with a fixed number of elementary degenerations is defined by the factorization of the global monodromy up to hurwitz moves and global conjugation . in general , the degeneration of a torus fiber can be worse then a simple shrinking cycle . the classification of such degenerations are well known from the kodaira classification of singular fibers for elliptic fibrations that we briefly review in appendix [ app : braid ] . however , the generalization to fibrations of higher genus is much more involved , see @xcite for a discussion of the genus 2 case . given the geometric group approach we described above , it is natural to ask whether one can give a more geometric description of such classification from the point of view of fiber diffeomorphisms . note that for the torus fibration a classification of periodic diffeomorphisms , i.e. elliptic @xmath7 conjugacy classes , is in correspondence with the kodaira list with the exception of @xmath258 fibers , which instead corresponds to the parabolic class , but no hyperbolic ( anosov ) monodromy arises from degenerating elliptic fibers . it turns out that such a geometric approach exists as a result of an elaborated theorem @xcite . for genus @xmath259 , there is a bijection between degenerations of a family of riemann surfaces ( up to an appropriately defined topological equivalence relation ) and a subset of conjugacy classes of the mapping class group of the fiber , given by pseudo - periodic elements . for the torus , the map is surjective . , @xmath260 . ] we refer to appendix [ app : mcg ] for a definition of pseudo - periodic mapping classes , but this subset basically contains parabolic and elliptic elements . we anticipate that this geometric point of view will be useful to generalize our presentation to the genus 2 case , relevant for non - geometric backgrounds of the heterotic theory with a wilson line . note that for lower genus fibrations , known facts about mapping class groups can be also helpful to determine possible global models . for example , one can show that for torus fibrations it is not possible to factorize the identity with less then 12 dehn twists . this corresponds to the defining relation for the presentation of the mapping class group @xmath55 . we can think of a word in the @xmath261 alphabet that factorizes the identity as defining a global fibration in which the total charge is canceled . some of the elementary constituents can then be collapsed into one of the kodaira degenerations if their total monodromy is elliptic or parabolic . however , one can still imagine situations in which a group of non - collapsed branes gives rise to a hyperbolic monodromy . we now turn to a classification of all possible local geometries that arise in our model , both for the geometric @xmath20 and the auxiliary @xmath21 fibrations . the complex and khler moduli @xmath8 and @xmath19 should really be thought as functions on an appropriate multi - sheeted riemann surface , where the transition between sheets is determined by a given @xmath7 element , that we identify with an element in @xmath55 . elements , but our discussion regarding mapping class groups is general . ] we construct the local solutions by taking the mapping torus @xmath24 for a given @xmath262 in each conjugacy class , and we then solve the cauchy - riemann equations to obtain a meromorphic function with the desired monodromy . this determines a semi - flat approximation of the corresponding t - fect . for parabolic and elliptic conjugacy class , we will recover in this way the known local solutions that already appeared in the work of kodaira @xcite , see table [ tab : localsolution ] . these can be obtained by inverting klein s @xmath39 modular function that specify a given global elliptic fibration . inserted in the appropriate semi - flat ansatz , we obtain the corresponding geometric and non - geometric local solutions . note however that interesting solutions can arise from different monodromies in the same conjugacy class , so we also need to determine the general form of the meromorphic functions . solutions for hyperbolic monodromies are substantially more complicated and can not be found by other simple methods . .local solutions around degenerations of kodaira type @xmath263 , @xmath264 , @xmath265 , @xmath266 . [ cols="^,^,^,^",options="header " , ] we give a factorization of the monodromy in terms of the two generators @xmath63 , @xmath64 of the mapping class group of the fiber torus , corresponding to a dehn twist along the @xmath67 cycle and a twist along the @xmath66 cycle , see figure [ fig : twistt2 ] . note that the starred types have a monodromy which is the inverse of the corresponding un - starred type . this follows trivially from the relations which present @xmath7 , namely from @xmath267 and from the braid relation @xmath268 . we note that the degenerations of the table precisely correspond to parabolic elements ( fibers @xmath263 and @xmath269 ) and to the elliptic conjugacy classes listed around . more precisely , all the mapping tori with parabolic and elliptic monodromies ( where we restrict to a given chirality ) arise at the boundary of small disks encircling the given degenerations , in line with the general theorem discussed in section [ subsec : monodromy ] . do not arise since they correspond to left - handed twists . this offers a simple explanation of why singularities of the kind @xmath270 with @xmath271 or type @xmath272 ( see for instance @xcite ) represent non - collapsible singularities . ] let us discuss now the hurwitz moves on the above factorizations . we define elementary moves that consist in shifting the @xmath64 components one step from right to left . this gives @xmath273 , where @xmath274 ( see figure [ fig : braidtn ] ) . analogously , we get @xmath275 . note that in terms of the more familiar abc branes used in the f - theory literature ( see for instance @xcite ) , we have @xmath276 , @xmath277 and @xmath278 . and @xmath227 are defined as @xmath279 , @xmath280 . for some factorizations this makes no difference since @xmath281 is independent on @xmath13 . ] we can use the elementary move on @xmath64 and the braid relation to bring the uv factorization in table [ tab : kodaira ] to the corresponding abc factorization . as an example , we show in figure [ fig : braidd4 ] the case for the @xmath282 singularity . from the braid relation ( applied to the underlined twists ) and the hurwitz moves we obtain @xmath283 which is the standard factorization for the @xmath284 monodromy . for the exceptional singularities we have ( see figure [ fig : braiden ] ) @xmath285 where the last relation follows from a global @xmath7 conjugation . factorization . ] the uv factorization is useful to understand colliding singularities and the corresponding symmetry enhancements . for example , when two @xmath265 fibers collide we get a monodromy @xmath286 giving an @xmath287 gauge group . another example is the branch ii at constant @xmath8 . there are generically twelve type @xmath264 degenerations , each with monodromy @xmath288 . to understand enhanced symmetries on this branch , we see that when @xmath13 singularities coincide , @xmath289 , we get a monodromy @xmath290 , corresponding to the gauge groups @xmath291 , @xmath287 , @xmath292 , @xmath293 . this reproduces the list in @xcite . we summarize some details about the geometry of the ansatz . the relevant equations of motion are : @xmath294 it is easy to check that the cauchy - riemann equations for the functions @xmath8 , @xmath19 and @xmath26 imply that the ansatz solves the equations of motion . the ricci scalar ( in the string frame ) for such ansatz is : @xmath295 ^ 3\tau_2(r,\theta)}\left [ r^2 \left(\frac{\partial \rho_2(r,\theta)}{\partial r}\right)^2+\left(\frac{\partial \rho_2(r,\theta)}{\partial \theta}\right)^2\right ] \ , .\ ] ] for completeness we also give the components of the ricci tensor along the torus and the base directions : @xmath296+\frac{e^{-2\varphi_1}(\tau_1 ^ 2-\tau_2 ^ 2)}{2r^2\rho_2\tau_2 ^ 3}\left[\partial_{\theta}\rho_2\partial_{\theta}\tau_2+r^2\partial_r\rho_2\partial_r\tau_2\right]\ , , \nonumber\\ r_{8,9 } & = \frac{e^{-2\varphi_1}}{2r\rho_2\tau_2 ^ 2}\left[\partial_{\theta}\rho_2\partial_r\tau_2-\partial_r\rho_2\partial_{\theta}\tau_2\right]+ \frac{e^{-2\varphi_1}\tau_1}{2r^2\rho_2\tau_2 ^ 3}\left[\partial_{\theta}\rho_2\partial_{\theta}\tau_2 + r^2\partial_r\rho_2\partial_{r}\tau_2\right]\ , , \nonumber\\ r_{9,9}&= \frac{e^{-2\varphi_1}}{2r^2\rho_2\tau_2 ^ 3}\left[\partial_{\theta}\rho_2\partial_{\theta}\tau_2 + r^2 \partial_r\rho_2\partial_r\tau_2\right ] \ , . \end{aligned}\ ] ] we recall well known facts about the action of an element of the t - duality group @xmath0 ( see for example @xcite ) . if we define the background matrix @xmath297 the action of an element @xmath298 on @xmath299 is @xmath300 where the following relations hold : @xmath301 if we introduce the generalized matrix @xmath302 the action is @xmath303 . in terms of the @xmath7 action on @xmath8 and @xmath19 , defined as @xmath304 the element @xmath305 is determined as follows : @xmath306 in particular , the action of @xmath107 is given by @xmath0 matrices of the form : @xmath307 which is just a basis change @xmath308 . the action of @xmath36 is embedded as @xmath309 where @xmath310 . perturbative monodromies in @xmath19 given by the @xmath7 element @xmath311 in , are identified with the upper right corner . we see then that upper triangular matrices represent a geometric subgroup @xmath312 which is a product of diffeomorphisms and b - transformations of the form : @xmath313 a diffeomorphism combined with a @xmath314 element of the parabolic conjugacy classes in @xmath36 , namely @xmath315 which acts as : @xmath316 is represented by a lower triangular matrix @xmath317 the action given by the transformation forms an abelian subgroup of @xmath0 usually referred to as @xmath37-transforms @xcite since it acts naturally on bivectors @xmath318 . such elements implement tst transformations and are very useful in holography to construct gravity duals of marginal deformations , such as @xmath37-deformations of @xmath6 sym @xcite .

we study the out - of - equilibrium nonlinear dynamics of fields after post - inflationary preheating . during preheating , the energy in the homogeneous inflaton is exponentially rapidly transfered into highly occupied out - of - equilibrium inhomogeneous modes , which subsequently evolve towards equilibrium . the infrared modes excited during preheating evolve towards a saturated distribution long before thermalization completes . we compute the equation of state during and immediately after preheating . it rapidly evolves towards radiation domination long before the actual thermal equilibrium is established . the exact time of this transition is a non - monotonic function of the coupling between the inflaton and the decay products , and it varies only very weakly ( around s ) as this coupling changes over several orders of magnitude . this result is applied to refine the relation between the number of efoldings n and the physical wavelength of perturbations generated during inflation . we also discuss the implications for the theory of modulated perturbations from preheating . we finally argue that many questions of the thermal history of the universe should be addressed in terms of pre - thermalization , illustrating this point with a calculation of perturbative production of gravitinos immediately after chaotic inflation . we also highlight the effects of three - legs inflaton interactions on the dynamics of preheating and thermalization in an expanding universe .

hep-ph0507096

according to the inflationary scenario , the universe at early times expands quasi - exponentially in a vacuum - like state without entropy or particles . during this stage of inflation , all energy is contained in a classical slowly moving inflaton field @xmath1 . the fundamental lagrangian @xmath2 contains the inflaton part with the potential @xmath3 and other fields which give subdominant contributions to gravity . in chaotic inflationary models , soon after the end of inflation the almost homogeneous inflaton field @xmath4 coherently oscillates with a very large amplitude of the order of the planck mass around the minimum of its potential . due to the interactions of other fields with the inflaton in @xmath5 , the inflaton field decays and transfers all of its energy to relativistic particles . if the creation of particles is sufficiently slow ( for instance , if the inflaton is coupled only gravitationally to the matter fields ) the decay products simultaneously interact with each other and come to a state of thermal equilibrium at the reheating temperature @xmath6 . this gradual reheating can be treated with the perturbative theory of particle creation and thermalization as long as the couplings are sufficiently small @xcite . however , the particle production from a coherently oscillating inflaton for a wide range of couplings occurs in a non - perturbative regime of parametric excitation @xcite . this picture , with variation in its details , is extended to other inflationary models . for instance , in hybrid inflation inflaton decay proceeds via a tachyonic instability of the inhomogeneous modes which accompany the symmetry breaking @xcite . one consistent feature of preheating non - perturbative copius particle production immediately after inflation is that the process occurs far away from thermal equilibrium . the energy of the inflaton zero mode is transferred to particles in an out - of - equilibrium state with very large occupation numbers within a very short time interval of about @xmath0 sec , which can be much shorter than the time needed for relaxation towards thermal equilibrium . one can expect the initial conditions for many parameters of the subsequent cosmological thermal history to be settled during or around preheating . it is often thought that the details of the transition between inflation and the hot radiation dominant stage are not relevant , except for the so - called reheating temperature @xmath6 , the temperature of the ultra - relativistic plasma at the time when it reaches thermal equilibrium . definitely , this is an important parameter of the early universe . however , a precise understanding of how thermal equilibrium is reached is crucial since partial thermal distributions can be responsible for cosmological baryo / leptogenesis , the possible creation of dangreous cosmological relics , etc . the out - of - equilibrium character of preheating opens the possibility for equally relevant phenomena associated with non - equilibrium physics , including phase transitions , non - thermal production of heavy particles , etc . precise knowledge of the expansion of the universe @xmath7 is also required to connect a physical wavelength of the observed cosmological perturbation , @xmath8 , to the number of e - foldings @xmath9 at which that wavelength exited the horizon during inflation . finally , we can have an alternative mechanism of generation of ( almost ) scale free adiabatic metric perturbations from preheating , based on spatial modulation of couplings @xcite . small spatial fluctuations in the couplings lead to small fluctuations of the rates of physical processes , which in turn generate adiabatic perturbations of the energy density and of the metric . analyzing the details of this mechanism require precise knowledge of the evolution of the equation of state ( eos ) . to understand the early post - inflationary period we have to understand the dynamics of interacting fields in an out - of - equilibrium state with large occupation numbers evolving towards a state of ultimate thermal equilibrium . this is a complicated problem of non - equilibrium quantum field theory , which is by itself a very interesting topic . the theory of the transition from inflation to thermalization has been investigated with fully non - linear numerical lattice simulations plus different techniques of classical field dynamics @xcite . classical field dynamics is adequate as long as occupation numbers of field excitations are large . there has also been progress in understanding out - of - equilibrium qft dynamics in @xmath10 sigma models beyond the hartree approximation @xcite , taking into account crucial effects of rescattering after preheating @xcite . from all of these methods , the following picture emerges . immediately after preheating , either from parametric resonance or tachyonic , one or more bose fields are excited in an out - of - equlibrium state with large occupation number . backreaction of these fields terminates their production . interaction ( rescattering ) of these modes between each other , and with the remaining inflaton field after the first stage of preheating is violent and non - perturbative . during this very short stage a large amount of entropy is generated , chaotic ( turbulent ) wave dynamics is established , and the fields not directly excited during preheating get excited in out - of - equilibrium states due to rescattering ; residual inflaton oscillations are still present . a next , longer stage then takes place , characterized by lower ( but still large ) occupation numbers such that rescattering is perturbative , and the occupation spectra gradually move towards a saturated state by cascading towards ultra - violet ( uv ) and infra - red ( ir ) modes ( in the spirit of kolmogorov wave turbulence ) . the last and the longest stage will be the stage of proper thermalization , when the distributions evolve towards thermal equilibrium . quantum physics is important at the end of this stage , when the occupation numbers are small and the classical approximation breaks down . one may say that reheating is completed when this last , longest stage is completed , and its timing defines @xmath6 and @xmath11 . in this paper we suggest a new look at the transition between inflation and thermalization . we focus on calculation and understanding of the effective equation of state @xmath12 throughout all of these stages after inflation . we use numerical lattice simulations @xcite to calculate @xmath13 in the simple chaotic model with a quadratic potential and a simple four - legs interaction @xmath14 . in particular , we notice that @xmath13 approaches ( but does not necessarily reach ) the equilibrium radiation - dominated value @xmath15 while the fields are far from ultimate thermal equilibrium . we compare this with the analysis of @xcite , devoted to thermalization of the @xmath16 sigma model in minkowski spacetime . in that case , the equation of state was found to evolve sharply towards @xmath17 long before thermalization completes , which prompted the authors of @xcite to describe the state of the system as a pre - thermalized state . although we also see a somewhat similar effect , the expansion of the universe and the presence of a residual massive inflaton component , prevent the equation of state from being exactly the one of radiation at this very early stage . moreover , we enphasize that the exact microphysics of the system is described by a turbulent state , which is still very far from a thermalized one . ( this difference can be important when discussing the cosmological effects of decay products , as for instance production of dangerous relics from nonthermal distributions . ) indeed , our simulations show that the occupation numbers of excited infra - red modes evolve towards a steady state , related to turbulence of classical interacting waves . the plan of the paper is the following : in section [ sec : model ] we describe preheating in a simple chaotic inflationary model . in section [ sec : numerics ] we present the results of lattice simulations for time - dependent variables , including the equation of state @xmath18 during different stages of preheating , occupation numbers and fluctuations . in section [ sec : implementations ] we discuss some immediate applications of our results for cosmology . we consider the application of the eos evolution to modulated cosmological fluctuations and to the @xmath19 formula . we note several qualitatively important issues in the dynamics of thermalization . the interaction @xmath20 describes ( at the perturbative level ) scattering between inflaton quanta , rather than inflaton decay . scattering between massive @xmath1 soon becomes inefficient , and does not lead to a complete depletion of the inflaton , which eventually ends up dominating over the light degrees of freedom @xmath21 . this problem is automatically avoided if three - legs bosonic interactions @xmath22 are also present . we illustrate this with an example motivated by supersymmetry . we also discuss the rapid saturation of the ir modes excited during preheating , which occurs long before the ultimate thermal equilibrium ( of all modes ) is reached . one consequence of the excitation of infrared modes is that it can lead to an overproduction of gravitinos or of other dangerous gravitational relics . this generation is different from direct nonperturbative production during preheating @xcite . instead , it is analogous to the perturbative thermal production first considered in @xcite . the main difference with @xcite is that we compute the production right after the first preheating stage , and not only after thermalization has completed @xcite . although the distributions formed at preheating / rescattering are far from thermal , the high particle occupation numbers can lead to a significant generation of gravitinos . the above results are summarized in section [ sec : summary ] . in this section we briefly describe the model of chaotic inflation with a quadratic potential and additional interaction terms and the methods we use to investigate the fields dynamics . the quadratic form of the potential is a good generic approximation around the minimum . for simplicity , we take the potential [ pot ] v()= m^2 ^2 also throughout inflation . the inflaton mass is taken to be @xmath23 gev , which , in the simplest cases , reproduces the correct amplitude for the cmb anisotropies . after inflation , the background inflaton field oscillates with its amplitude diluted by the expansion as @xmath24 , where @xmath25 is the initial amplitude , and @xmath26 is the scale factor of the frw flat universe . more accurately , the inflaton evolution is given by @xmath27 . the energy density of the inflaton oscillations evolves as @xmath28 , while the pressure is @xmath29 . since the frequency of the oscillations is higher than the rate of expansion @xmath30 , we deal with a nearly vanishing averaged effective pressure , @xmath31 . therefore , immediately after inflation , the equation of state ( eos ) is effectively the one of matter domination [ eos1 ] w = p/0 . this is what we expect for inflaton oscillations interpreted as the coherent superposition of heavy inflaton ( quasi)-particles at rest . inflaton oscillations decay due to the coupling between the inflaton and other fields . one may think about the inflaton coupling to other bosons through four - legs or three - legs interactions , for instance coupling to another scalar @xmath21 as @xmath32 or @xmath33 . yukawa - type couplings to fermions will be of the form @xmath34 . during preheating the leading channel of inflaton decay will be parametric resonance decay into bosons . a three legs interaction between bosons @xmath35 requires a dimensionful scale @xmath36 , which emerges naturally in susy theories . however , in most cases the amplitude of the inflaton is sufficiently high during the early stages of preheating that the four - legs interaction dominates @xcite . for this reason , at this stage we consider only the four - legs interaction [ int ] l_int=- ^2 ^2 . three - legs interactions are important for the completion of the decay of massive inflatons , and we will discuss them further in section [ sec : implementations ] , and in detail in @xcite . before proceeding with the generation of @xmath21 particles from the inflaton oscillations , we need to comment on the range of the coupling @xmath37 . at the level of qft we can obtain an upper limit for @xmath37 . indeed , in non supersymmetric models , the interaction ( [ int ] ) leads to radiative corrections and can thus be cancelled by a counter - term . ] to the effective potential ( [ pot ] ) , of the form @xmath38 . for inflaton values @xmath39 , which correspond to the scales where cosmological fluctuations are observed , radiative corrections do not alter the the potential ( [ pot ] ) as long as @xmath40 . in supersymmetric theories radiative corrections from bosons and fermions cancel each other out so that the value of @xmath41 can be even higher . there is also a lower limit on @xmath37 . during inflation the field @xmath21 has the effective mass @xmath42 , where @xmath43 is its `` bare mass '' : we take @xmath44 and thus neglect it . it is known @xcite that if we have two scalar fields @xmath1 and @xmath21 , the latest stage of inflation will be driven by the lightest scalar ( light scalars are always generated during inflation ) . a consistent setting where the end of inflation is driven by the @xmath1 field requires @xmath45 , i.e. @xmath46 . this lower limit can be evaded , however , if @xmath21 is conformally coupled to gravity . a quantum scalar field @xmath21 in a flat frw background has the eigenfunctions @xmath47 with comoving momenta @xmath48 . the temporal part of the eigenfunction @xmath49 obeys the equation x_k + + g^2 ^2 x_k = 0 [ modesofchi ] with positive - frequency initial conditions in the far past . in the effective frequency @xmath50 we neglect small terms proportional to @xmath51 . this is an oscillator - like equation with a periodic frequency , which contains exponentially unstable solutions corresponding to parametric resonance . it is convenient to use a new time variable @xmath52 . then , the essential dimensionless coupling parameter which characterizes parametric amplification is @xmath53 . the limits on the coupling @xmath37 we have discussed above correspond to @xmath54 . a large portion of this interval is in the broad parametric resonance regime ( @xmath55 ) of production of @xmath21 particles @xcite . what is important , however , is that even one bosonic field with a large parameter @xmath56 is sufficient to have the leading channel of inflaton decay occur through broad parametric resonance . indeed , if one such field @xmath21 is exponentially amplified in this way then other fields that are coupled to it directly or indirectly will tend to be amplified as well @xcite . we are interested in the dynamics of the fields @xmath1 and @xmath21 after inflation . although this toy model is simple , its dynamics are rather rich and complex . the first stage is the stage of parametric excitation of the quantum fluctuations of the field @xmath21 , where the light @xmath21 particles are copiously produced from the classical inflaton oscillation . this process is described by the theory of broad stochastic parametric resonance in an expanding universe @xcite . as soon as the occupation number of created particles [ particles ] n_k = + |x_k|^2 - becomes large , @xmath57 , one can approximate the dynamics with that of the classical interacting fields @xmath1 and @xmath21 . fully non - linear wave equations for coupling classical fields are rather complicated . we use the latticeasy numerical simulations @xcite to solve the equations @xmath58 and @xmath59 in an expanding universe ( @xmath60 being the covariant wave operator in the frw geometry ) . the scale factor @xmath26 is self - consistently calculated from the friedmann equation . the initial conditions correspond to gaussian initial fluctuations of @xmath1 and @xmath21 , which arise from vacuum fluctuations . the novel element of our current study is the focus on new diagnostics of the numerical experiments . in particular , we investigate the evolution of the equation of state @xmath18 during different stages , and for various couplings @xmath37 . many previous numerical studies of preheating concentrated on inflation with a quartic potential @xmath61 , where the expansion of the universe can be `` scaled away '' . the inflaton oscillations in this model have the eos @xmath17 , which corresponds to that of radiation domination , so @xmath13 does not change after the inflaton decay . by contrast , @xmath13 changes during preheating in the more general model with a quadratic potential . the total energy density @xmath62 and pressure @xmath63 contain contributions from kinetic and gradient energies of the fields @xmath1 and @xmath21 , plus interaction terms . it is instructive to calculate separate contributions from all of these terms . we find that the interaction term , except for a transient moment between the end of preheating and onset of turbulent regime , is sufficiently small that we can interpret the results in terms of different particles @xmath64 and @xmath21 coupled to each other . in particular , this allows us to interpret the energy density in each component in terms of the integral [ energy ] d^3k _ k n_k , where the comoving frequency is @xmath65 and the co - moving occupation number @xmath66 is given in ( [ particles ] ) . to study @xmath13 we have to monitor the relationship between the effective masses of the fields [ mass1 ] m_,eff^2=m^2 + g^2 ^2 , [ mass2 ] m_,eff^2=g^2 ^2 + g^2 ^2 . and the typical momenta . the variances of the fluctuations [ fluc1 ] ^2= d^3k |_k|^2 [ fluc2 ] ^2= d^3k |_k|^2 are also an important diagnostic of the non - equilibrium physics . let us rewrite the wave function @xmath67 in the wkb form @xmath68 . while the occupation numbers in terms of @xmath69 coefficients are simply @xmath70 , the exact expression for the variances is more complicated [ fluc3 ] ^2= ( n_k+ 2 re ( _ k _ k^ * e^2 dt i_k ) ) . particles @xmath21 produced at preheating are created in a squeezed state . in particular , this is reflected in the presence of the oscillatory second term in the integral ( [ fluc3 ] ) , where the phases of @xmath71 are present . the total co - moving number densities of @xmath21 and @xmath1 particles are [ part1 ] n_= d^3k n^_k , [ part2 ] n_= d^3k n^_k , respectively . it is instructive to monitor the time evolution of the total number of particles @xmath72 . at leading order in @xmath37 the total number of particles is conserved , @xmath73 . however , there is a brief period immediately after preheating where @xmath74 is poorly defined due to the strong interactions between the waves . indeed , during preheating the occupation number of @xmath21 particles can be as large as @xmath75 . consider the collision integral in the kinetic equation for the particle . if we formally represent it as a series with respect to the coupling @xmath37 , the smallness of @xmath37 will be compensated by the large occupation numbers . this means that the next terms of the collision integral associated with the higher order diagrams ( for example , the diagram with four incoming , two outgoing legs , and two vertices ) are as important as the first term . in fact , the non - perturbative character of the interactions during this short intermediate stage makes it difficult to investigate the whole dynamics analytically . after this period ends , the spectrum of the occupation numbers formed at preheating is weighted towards the infra - red relative to a thermal distribution of the same energy density . hence , particle fusion is more favored than particle dissociation . once rescattering leads to a decrease of @xmath66 , only the leading four - legs diagram with one vertex becomes important , and @xmath74 decreases on the way to thermalization , on a timescale which is much longer than the one which can be observed in our lattice simulations . we performed three - dimensional lattice simulations for the model of section [ sec : model ] . our grid was a @xmath76 cube with a comoving edge size @xmath77 , which corresponds to a comoving grid spacing of @xmath78 . as energy flows towards the uv end of the spectrum the simulations eventually reach a point where the grid spacing is too large to capture the important uv physics . by monitoring the spectra of the fields , however , we can verify that these simulation parameters were adequate to capture the relevant ir and uv physics well past the end of preheating . the time step was @xmath79 and the inflaton mass was @xmath80 . we used values of the coupling near @xmath81 this value is optimal because it is large enough to produce highly efficient preheating , but small enough that the occupation numbers @xmath75 produce strong rescattering . the results should be qualitatively similar for a wide range of values of @xmath37 , but would require more ir and/or more uv to simulate accurately . to probe later times and wider ranges of the couplings it will be necessary to extend the lattice simulations . this can be done with a parallelized version of the simulation code latticeeasy ( currently under construction ) , or by combining the straightforward lattice simulations with other methods , like the equations for a large number of weakly coupled oscillators @xcite . we intend to pursue both of these approaches in subequent work . in the rest of this section we present the results of our simulations . the time evolution of the eos @xmath18 for different couplings is shown in figure [ fig : w1 ] . each point plotted on this figure represents the value of @xmath13 averaged over a complete inflaton oscillation . this represents one of the main results of our study . = immediately after inflation , the eos averaged over inflaton oscillations is @xmath82 . it sharply changes at the end of preheating . there are at least three important points worth enphasizing about the evolution of @xmath13 . \i ) _ first , the transition of the eos from @xmath82 to the value @xmath83 occurs very sharply , within a time interval @xmath84 sec . _ indeed , recall that the unit of time on the plots is @xmath85 , where @xmath86 is the inflaton mass , i.e. @xmath87 sec . the first stage of preheating is completed within about a hundred of these units , i.e. , @xmath0 sec . the rise of @xmath13 and gradual saturation takes roughly the same time . \ii ) _ second , the dependence of @xmath18 on the coupling @xmath37 for resonant preheating is a non - monotonic function of @xmath37 . _ this is to say that the time during which preheating comes to an end is very weakly ( logarithmically ) dependent on the coupling . as seen from figure [ fig : w1 ] the curves @xmath18 begin to shift to the left towards an earlier end of preheating , as we vary @xmath37 by @xmath88 . however , at some point the curves stop moving to the left and instead begin to return toward the right . as we change @xmath37 by about @xmath89 , the cycle repeats . as we vary @xmath37 , the function @xmath13 not only shifts , but it also varies its detailed shape . still , to characterize these variations , we pick up the moment where @xmath13 is equal to the value @xmath90 ( just for convenience of calculation ) , @xmath91 . this allows us to plot the transition moment @xmath92 as a function of @xmath37 , see figure [ fig : tran ] . = we see that the transition time varies between @xmath93 and @xmath94 . this non - monotonic behavior of the duration of preheating is explained in the theory of broad paremetric resonance @xcite ( see sections 6 and 9 there ) . the @xmath37 dependence of the eos is the critical issue for the theory of modulated cosmological perturbations , which we will discuss in section [ sec : implementations ] . \iii ) _ the third point is that @xmath13 does not necessarily immediately go to the radiation dominated value @xmath15 . _ this is partly because immediately after preheating the light field still has a significant induced effective mass due to the interaction , and partly due to the significant residual contribution from the homogeneous inflaton @xcite . unfortunately , limitations on running longer simulations preclude us from seeing further details of the time evolution of @xmath13 . however , we have a strong theoretical argument to advance the discussion further . in a model with a massive inflaton and light scalar @xmath21 even the radiation dominated stage is transient . indeed , sooner or later the massive inflaton particles , even if significantly under - abundant at the end of preheating , will become the dominant component , and the universe will again be matter - dominated . the occupation numbers @xmath95 and @xmath96 are among the most interesting variables to understand the micro - physics in our system of two interacting fields . first , we shall determine when the definition of @xmath66 is meaningful . to do so , we consider the composition of total energy density @xmath62 of the system of coupled @xmath1 and @xmath21 fields . the total energy density @xmath62 can be decomposed into partial contributions from the kinetic energy of both fields , their gradient energy , their potential energy ( @xmath1 only in this model ) , and finally the interaction energy [ decomp ] = ^2+^2+()^2 + ( ) ^2 + m^2 ^2 + g^2 ^2 ^2 . figure [ fig : comp ] shows the relative contribution to the total energy from each of the components for @xmath97 . = we note two features of this plot . first , the interaction term is comparable to the other terms in the time interval @xmath98 . in this short period , the formula ( [ energy ] ) for the energy of the particles is not a good approximation and the occupation number @xmath66 is not well defined . outside this time interval , @xmath66 is a meaningful quantity . secondly , the contribution from the background homogeneous inflaton is dominant even after preheating , up to approximately @xmath99 . a similar point was made for the @xmath100 model in @xcite . let us now turn to the occupation numbers @xmath66 . we find it more instructive to output not the occupation numbers @xmath66 _ per se _ ( as it is commonly done in the literature ) but the combination of @xmath66 with the energy per mode @xmath101 . this combination can be immediately compared with the rayleigh - jeans spectrum , [ thermal ] n_k , which corresponds to the equipartition spectrum of classical waves ( we introduce the chemical potential @xmath102 for generality ) . the comparison allows to determine how close the distribution is to the thermal one . = = the combination @xmath103 for the two fields is shown at some characteristic times in the two figures [ fig : phinkwk ] and [ fig : chinkwk ] . there are three distinct stages which characterize the evolution covered by our simulation . the first stage , characterized by linear dynamics , is the one of preheating and early rescattering . the first modes to be populated are the ir ones . preheating of @xmath21 particles occurs in the resonant band at comoving momentum @xmath104 @xcite . then , quanta @xmath105 are generated by rescattering . the annihilation @xmath106 amplifies quanta of the inflaton at @xmath107 . even more effective is the rescattering of the @xmath21 quanta against the inflaton zero mode , @xmath108 , which produces quanta of both @xmath1 and @xmath21 at momentum @xmath109 . the second stage is a violent stage of highly nonlinear dynamics . starting at @xmath110 , the higher band at @xmath111 both increases in its amplitude and broadens towards higher momenta ; quite interestingly , the peak location shifts from @xmath112 to @xmath113 during this quick and explosive stage . as we remarked , the particle occupation number is ill defined at this stage , and a nonlinear wave description is more adequate . the explosive stage of rescattering ends at about @xmath114 . in the next stage , characterized by perturbative dynamics , the distributions smooth out and start evolving ( at a much slower rate ) towards higher comoving momenta . the spectra in the ir approach a saturated power - law state , which then slowly propagates towards the uv . although one can observe a greater tendency towards thermalization for the ir modes ( where the rescaled spectra are closer to flat ) , the overall distributions are still typical of the turbulent regime , and they are far from thermal . if we contrast this with the macroscopic behavior we have described above , particularly the evolution of the eos , we see that the system can be considered in a pre thermalized state ( see also @xcite ) , but that thermalization is still far from being complete . it is also instructive to consider the product of the occupation number @xmath66 with the phase space sphere area @xmath115 and energy per mode @xmath116 . this combination represents the energy density of the quanta at momentum @xmath117 ( since the product @xmath118 is the energy density of the quanta whose momentum has a magnitude between @xmath117 and @xmath119 ) . = = in figures [ fig : phispl ] and [ fig : chispl ] , we plots the distributions at different times separated by intervals @xmath120 . this allows us to visualize the growth of the distributions , and to monitor the cascade of energy in the phase space @xmath117 . when the ultraviolet part of the distribution hits the highest momentum of the simulation ( defined by the grid size ) , the energy is artificially reflected back to the ir modes and the simulation is no longer reliable . due to the scale chosen ( natural rather than log scale ) , only the distributions at times greater than about @xmath121 can be appreciated in the two figures [ fig : phispl ] and [ fig : chispl ] . moreover , the double peak structure that can be observed for @xmath21 at late times is due to the rescaling chosen . the plot for the occupation number , @xmath122 ( not shown ) has a high peak at @xmath123 , followed by a plateau up to @xmath124 . the combination @xmath125 shown has a peak at the momenta corresponding to this plateau , showing that these momenta dominate the energy density in the @xmath21 distribution . the saturated spectra can be clearly seen in the green ( thick ) curves on these plots . = = the combination @xmath125 is also given in figures [ fig : phisps ] and [ fig : chisps ] , where we however show only few times , and we focus on the ir part of the distributions . these figures show the rapid broadening ( towards the uv ) of the distributions in the violent rescattering stage , and the tendency to saturation at later times . the evolution of the scalar fields can be strongly affected by the presence of dynamical effective masses for the two fields . a high mass for some of the fields can disfavour or block some of the interactions , with a consequent delay of thermalization . for instance , it is well known that a condensate which is strongly coupled to some field @xmath126 can have a very long lifetime , since the same coupling which would allow the decay also produces a high effective mass for @xmath126 , which can prevent the decay from occuring . in the case we are discussing , the effective masses of the fields acquire significant ( loop ) contributions from the high variances produced at preheating / rescattering , according to eqs . ( [ mass1])-([mass2 ] ) . = figure [ fig : fluc ] shows the evolution of the background inflaton oscillations and fluctuations of @xmath127 and @xmath128 . all values are plotted in `` comoving scales '' , i.e. multiplied by @xmath129 . the corresponding physical quantities can be obtained by the value of the scale factor , which is shown in figure [ fig : sf ] . = figure [ fig : masses ] shows the effective masses for @xmath1 and @xmath130 , multiplied by the scale factor @xmath131 . this combination enters in the comoving dispersion relations for the quanta of the two fields , given by eq . ( [ comom ] ) . therefore , the product @xmath132 is the correct quantity to compare with the comoving momentum @xmath133 . = this comparison is shown in figure [ fig : frel ] , where we show the fraction of relativistic quanta ( that is , with @xmath134 ) for the two fields . we see that the majority of the particles are non - relativistic , so that the effective masses indeed play a significant role in the evolution of the spectra . in particular , we observe that during rescattering very few particles are relativistic ; this confirms the fact that the distributions are very peaked in the ir at this stage . the fraction of relativistic @xmath21 particles rapidly increases during the violent thermalization stage . the fraction of relativistic @xmath135 remains instead always smaller ( confirming the fact that @xmath136 ) , and it slowly increases during the thermalization stage , when the distributions evolve towards the uv . = up to @xmath137 , the coherent mode of the inflaton dominates the effective mass for @xmath130 . ( the rapid variation of @xmath138 due to the coherent oscillations of @xmath1 is the cause of preheating . ) at later times , the variance of @xmath1 gives instead a greater contribution . in the effective mass for @xmath1 the `` bare '' term @xmath86 dominates up to @xmath139 , while the variance of @xmath21 provides the dominant contribution at later times . ( [ fluc3 ] ) provides a useful expression for the variance ; during the turbulence stage , the occupation numbers vary only adiabatically , so one may expect that the growth of @xmath140 results in a decrease of the variance , and vice - versa . this relation is visible in the figures shown . the `` late time '' decrease of @xmath141 is accompanied by an increase of @xmath142 . this in turns cause an increase of @xmath143 , and , consequently , a decrease of @xmath144 . in the simple model discussed so far , with the interaction term @xmath145 , the decay of the inflaton @xmath1 is not complete . indeed , such a term mediates scatterings between inflaton quanta , rather than single particle decay . once @xmath1 is diluted by the expansion of the universe , the scatterings become inefficient , and the number of inflaton quanta remain practically constant . this poses a problem because in order to reach the stage of radiation domination we need to have complete decay of the inflaton . the simplest way to have a complete decay is to consider three legs interactions . for instance , we can replace ( [ int ] ) by @xmath146 where @xmath36 is a mass dimension parameter , which breaks the @xmath147 symmetry . for @xmath148 , the trilinear term produced by this interaction is irrelevant and the inflaton decay occurs as described above . however , when @xmath1 is decreased by the expansion of the universe , the trilinear interaction becomes dominant , leading to a complete inflaton decay . we can avoid discussing the introduction of the scale @xmath36 by considering a yukawa interaction of the inflaton with some fermions @xmath149 . by itself , this interaction has interesting consequences for preheating , as discussed e.g. in @xcite . when both fermionic and bosonic interactions are present , preheating into bosons is typically much more efficient , since the one into fermions is reduced by pauli blocking . however , once the interactions become perturbative , the trilinear @xmath150 decay will eventually dominate over the @xmath151 scattering . three - legs interactions arise very naturally in susy theories . indeed , consider the simple superpotential @xmath152 the corresponding scalar field potential contains three - legs as well as four - legs plus self - interaction terms @xmath153 ( neglecting the imaginary parts of the fields ) . the four - legs interaction dominates over the three - legs one as long as @xmath154 . for reasonable values of @xmath36 , this happens during the first stages of preheating . eventually , the amplitude of @xmath1 decreases due to the expansion of the universe , and the trinlinear interaction dominates , resulting in a complete decay of the massive inflaton . hence , considering a supersymmetric theory automatically introduces trilinear vertices among the scalars . it also allows for stronger couplings between the inflaton and other fields , without spoiling the flatness of the inflaton potential , as we have discussed in sec . [ sec : model ] . the precise history of the expansion of the universe @xmath7 is important for connecting the physical wavelength @xmath155 of the cosmological perturbations we observe to the number of efoldings @xmath9 at which the perturbation was generated ( left the horizon ) during inflation . according to the general lore @xcite , one has [ efold ] n(k ) = 62 - ln + , where @xmath156 is the inverse size of the present cosmological horizon and @xmath157 is defined by the physics after inflation : = -ln + - , [ deltadef ] where @xmath158 is the energy density at the end of reheating , @xmath159 is the value of the inflaton potential at the moment when the mode with the comoving wave number @xmath117 exits the horizon at inflation and finally @xmath160 is the value at the end of inflation . the last term in the right hand side of eq . ( [ deltadef ] ) is related to reheating , specifically that it is not an instantaneous process . it is typically computed under the assumption that the universe is matter dominated ( by the coherent oscillations of the inflaton ) all throughout reheating , and that it ( instantaneously ) becomes radiation dominated at the end of the reheating stage . as we have seen , this is not what happens . indeed , a better definition of the last term in @xmath157 is @xmath161 where @xmath162 is the scale factor at the moment after reheating at which the eos reaches its value at radiation dominance . as we have been discussing , this moment can occur while the distributions of particles are very far from thermal , shortly after the end of preheating . this amounts to taking @xmath163 , rather than @xmath164 , where @xmath165 is the value of the scale factor at the end of reheating under the common assumption that perturbative reheating occurs entirely within a regime of matter domination , with @xmath166 . in this situation one has @xmath167 at the end of matter domination , and the corresponding value of the scale factor @xmath168 can be estimated as @xmath169 . hence , for a preheating scenario with a very wide range of values of the coupling @xmath170 , the first term of ( [ deltadef ] ) can be fixed to @xmath171 this result holds for a wide class of inflationary models , and for many different types of preheating . one of the most important parameters in physics of the early universe is the highest temperature of the hot plasma after the inflaton field decays and transfer its energy into radiation . , defined at the moment when the thermal bath starts dominating . ] this is traditionally called the reheating temperature @xmath6 . as we have seen in many previous papers and above in this paper , ultimate thermal equilibrium is preceeded by several stages . let us distinguish four of them : * preheating , with the duration @xmath172 , * a short violent stage at the end of preheating when non - linearity and non - perturbative effects are dominant ; chaos is onset , erasing the details of the initial conditions , on a timescale @xmath173 , * the stage of turbulence of classical fields , with the saturated spectrum cascading both towards uv and ir modes with duration @xmath174 , longer than the two previous stages , * the last stage of proper qft thermalization , with particle fusion / offshell processes ; it is characterized by the timescale @xmath175 , much longer than the previous ones . traditionally the beginning of the thermal history is related to the moment @xmath175 , which may be very long ; as a result of redshift , the reheating temperature can be rather low . let us reconsider this attitude . while the question of when the ultimate thermal equilibrium will be established and to which value of @xmath6 it corresponds is still very interesting , we will argue that important physics which constrains the model of inflation and interactions takes place long before full thermal equilibrium is completed . indeed , figures [ fig : phinkwk][fig : phispl ] show that the modes at physically interesting scales , up to @xmath176 tens of @xmath177 gev , are quickly excited towards a saturated distribution . this raises several subtle and interesting questions . usually , the relativistic particles embedded in the thermal bath are brought into thermal equilibrium after the relaxation time @xmath178 , where @xmath36 is the cross section of the processes relevant for thermalization , and @xmath179 is the number density of particles . this estimate works in the diluted gas approximation . indeed , consider the vertex @xmath22 , leading to inflaton decay into quanta of @xmath21 . the particle number density enters in the collisional integral in the combination @xmath180 ( @xmath181 is the momentum of the inflaton quantum , while @xmath182 are the momenta of the two quanta of @xmath21 entering in the process ) . the above estimate for the thermalization timescale assumes that the inverse process @xmath183 ( described by the second term in ( [ collfactor ] ) ) is irrelevant , and that all the occupation numbers are much smaller than one , so that @xmath184 . in particular , this last assumption is not valid until the very end of the thermalization stage . therefore , for large @xmath66 we expect that particles are dragged into effective thermal equilibrium faster , due to the stimulated interaction . we have observed this fact also in previous numerical lattice simulations @xcite . another relevant effect , already discussed above , is that large fluctuations of fields can contribute to the effective masses of particles , which can become effectively heavy . this leads to the blocking of some processes which may lead to a faster thermalization , until fields fluctuations are diluted by the expansion . finally , a very relevant question is the one of particle production from the nonthermal but highly excited distributions . in particular , one should verify that dangerous relics are not overproduced at this stage . to see this , let us discuss gravitino production from the decay products of the inflaton . in models where supersymmetry is broken gravitationally , only a very small number of gravitinos can be tolerated , @xmath185 ( the exact value being dependent on the gravitino mass ) @xcite . the typical approach is to compute the production of gravitinos only after the thermal bath has formed . the dominant processes are @xmath186 scatterings with only one gravitino as outgoing particle ( and , hence , only one gravitationally suppressed vertex , while the other vertex is typically a gauge interaction ) . the rate of gravitino production follows from the boltzman equation @xmath187 where @xmath188 is the abundance of the ( mssm ) degrees of freedom in the thermal bath , and where the average cross section is of the order @xmath189 , where @xmath190 is a gauge structure constant . the right hand side decreases very rapidly with time , and the production at the highest possible temperature ( that is , at @xmath6 ) dominates . hence , only the production in the first hubble time is relevant , and the final result can be estimated as @xmath191 where @xmath192 is the hubble parameter at @xmath193 . this leads to the gravitino abundance @xmath194 and to the well known bound @xmath195 on the reheating temperature . this standard computation of gravitinos generated from the thermal bath neglects all the production which may have taken place during the thermalization stage . the underlying idea is that , during reheating , most of the energy is in the inflaton condensate , where the quanta ( loosely speaking ) do not have momentum , and hence can not scatter to produce gravitinos . while this may be true for a slow ( perturbative ) inflaton decay , this is certainly not the case for preheating / rescattering , where the distributions of the decay products form right after the end of inflation . these distibutions may be responsible for significant gravitino production , even before complete thermalization has taken place @xcite . the computation is now more difficult than in the thermal case @xcite . first of all , unlike the thermal computation , the result is expected to be model dependent . second , turbulent nonlinear processes are dominant right after preheating , so that it is possible that processes including more vertices give a sizeable contribution to the amount of gravitinos produced . however , we can obtain a conservative estimate by considering only @xmath186 tree level processes also in this case . for definiteness , let us take the model ( [ pot ] ) , ( [ int ] ) , with @xmath196 ( the value that we have studied above ) , and assume that the @xmath21 field produced at rescattering has a trilinear vertex with its femionic superpartner @xmath197 and another fermion @xmath198 . for instance , this could be a gaugino , if @xmath21 has gauge interactions . however , let us consider here a generic vertex with coupling constant @xmath199 and the interaction @xmath200 , mediated by the fermionic partner of @xmath21 . ( as a technical point , we consider production of transversal gravitinos , since the nature of the longitudinal component changes with time ) . we are interested in evaluating the production at the time @xmath201 , when the ( physical ) number density of quanta of @xmath21 is maximal . we first discuss whether the process we are considering is kinematically allowed . at the time @xmath202 , the non - gravitational vertex provides an effective mass for fermions of the order ( see fig . [ fig : fluc ] ) . @xmath203 the distributions of @xmath21 obtained from the lattice have a typical comoving momenum of the order of @xmath204 , which , at the time @xmath205 , corresponds to the physical momentum @xmath206 . hence , if @xmath207 , most of the quanta of @xmath21 will be able to produce gravitinos through the process we are considering . for higher values ( for example , if the second vertex is a gauge interaction ) , only the few quanta of @xmath21 with very high momenta can contribute to the production , so that this process is significantly suppressed . however , as the universe expands , @xmath208 decreases as @xmath209 , while the physical momenta @xmath210 decrease less . therefore , more and more quanta will have a sufficiently high momentum to produce a @xmath211 pair . indeed , since thermalization proceeds through particle fusion it has the effect of both increasing the typical momenta of the distributions and of decreasing the effective masses . ] for simplicity , @xmath212 is assumed in the following discussion ; we do not expect that the following conclusions will significantly change if a stronger coupling is considered . by proceeding as in the thermal case , we can estimate the number of gravitinos produced in the first hubble time after @xmath213 . ( again , the numerical values are taken from the lattice results ; @xmath214 denotes the physical number density of the @xmath21 quanta at the time @xmath213 . ) @xmath215 for sufficiently high @xmath216 , this value is actually incorrect , since it exceeds pauli blocking . assuming that the whole pauli sphere up to physical momentum @xmath217 is filled gives @xmath218 the number density of produced gravitinos is then the smaller between eqs . ( [ n321 ] ) and ( [ n322 ] ) , according to the value of @xmath219 . in the following , we assume @xmath220 , so that pauli blocking is reached and the number density of gravitinos is given by eq . ( [ n322 ] ) . the gravitino abundance is obtained by dividing the gravitino number density by the entropy . at the time @xmath202 , the distributions of the decay products are far from thermal ( being concentrated in the ir ) , so that their entropy is smaller than that of a thermal bath with the same energy . the entropy density can be computed from the lattice simulations to be approximately @xmath221 at this stage @xcite . however , the ( comoving ) entropy is expected to increase due to thermalization . if we convert the actual particle distribution instanteneously into the thermal distribution , the entropy would reach the maximal value @xmath222 allowed by energy conservation . the difference between the actual entropy density and @xmath223 gives rise to the dilution of the gravitino abundance during thermalization . ( further sources of dilution require extensions of the minimal model . ) therefore , the conservative estimate for the gravitino abundance is obtained by dividing the gravitino number density by @xmath224 where @xmath225 is the ( physical ) background energy density at the time @xmath202 . hence , the ratio @xmath226 is about @xmath227 orders of magnitude higher than the allowed limit . this is a serious concern , which has to be addressed in the model building of inflation . it can be resolved by introduction of extra radical assumptions . for instance , this gravitino abundance can be decreased by an entropy dilution . this happens if a massive species @xmath126 is produced at reheating , and it dominates the energy density of the universe for some time . when it decays , it leads to a significant amount of entropy , which dilutes the particle pieces produced directly at the inflaton decay . concretely , the entropy dilution leads to the decrease of the gravitino abundance @xmath228 . supposing that the universe is matter dominated between the times @xmath229 and @xmath230 , the ratio @xmath231 is decreasing as @xmath232 ( @xmath233 being the scale factor at the time @xmath234 ) . this entropy dilution could also occur through a secondary stage of inflation @xcite . in the specific toy model we have considered , a decrease of six orders of magnitude does not seem to be realistic , so that the production of gravitinos in this model appears to be a serious concern . however , it is important to remember that the result ( [ granont ] ) is very sensitive to the specific model considered , and that a case by case calculation is necessary . one of the motivations for studying the eos after preheating is related to the theory of modulated cosmological fluctuations . we briefly review the idea as it was presented in @xcite , and discuss the implications of our findings on the evolution of the eos for this mechanism . we leave the explicit calculations for subsequent publication . suppose that the coupling @xmath37 depends on some modulus field @xmath235 , ( as is typical in string theory ) @xmath236 during inflation the modulus field @xmath235 can be light , with a mass smaller than the hubble parameter during inflation @xmath237 gev . then , large scale fluctuations @xmath238 with an almost scale free spectrum are inevitably generated from inflation . the wavelength of fluctuations exceeds the size of the causal patch @xmath239 . as we have discussed above , the eos varies very quickly already at the beginning of the thermalization stage . the variation is a non monotonic function of the coupling @xmath240 , see for instance figs . [ fig : w1 ] and [ fig : tran ] . due to the large scale spatial fluctuations of the coupling @xmath241 , the change in the eos proceeds with a slight time shift in different hubble patches . this results in the generation of ( almost ) scale free scalar metric cosmological perturbations . to perform accurate calculations of the amplitude of modulated fluctuations , one needs to know the exact character of the transition from inflaton field to radiation . in the simplified picture of perturbative reheating , when the inflaton energy decays exponentially as @xmath242 , calculations of modulated fluctuations @xcite are based on linearization of the inflaton decay rate @xmath243 , which is linearly dependent on the coupling @xmath37 . however , as we have remarked , the decay of the inflaton field is typically more complicated , and the transition towards rd occurs very quickly ( well before thermalization completes ) . in addition , we stress that the time of the transition is a non monotonic function of @xmath37 . finally , although we focused only on the direct coupling of the inflaton to matter , the details of thermalization ( and of the evolution of the eos ) depend on the other interactions among the decay products ; the strength of these interactions are likely to be dependent on some modulus field as well , and they can also modulate cosmological perturbations . all these features have to be taken into account for precision computations of modulated fluctuations . in this paper we studied the out - of - equilibrium nonlinear dynamics of the fields during and after preheating . this continues earlier works in @xcite . we use lattice numerical simulations of fully non - linear dynamics of interacting classical fields . in this investigation , we computed the evolution of the equation of state during preheating and the early thermalization stage . immediately after inflation , the eos is the one corresponding to inflaton domination . for a massive inflaton , we have matter domination , characterized by @xmath244 . after a very short time , the inflaton transfers its energy into inhomogeneous modes . at the first stage ( preheating ) inhomogeneous field fluctuations are copiously produced in the regime of parametric resonance . then the modes rescatter to redistribute the energy by cascading in the phase space . this process of classical wave turbulence can be rather long . this leads to a final stage , where the occupation numbers are small and the classical approximation breaks down . thermalization finishes when the quantum fields reach the ultimate thermal equilibrium . if all the participating fields are light , the final thermalized state corresponds to an eos of radiation , with @xmath245 . the transition between the matter and radiation dominated eos is hence typically supposed to occur on a ( relatively ) very long timescale . what we found in our numerical simulations is very different from this naive expectation . in fact , the eos jumps from matter domination to ( almost ) radiation domination immediately after the first stage of preheating , i.e. in a couple dozen inflaton oscillations ( @xmath87 sec each ) , long before thermal equilibrium has been established . in other words , the macroscopic eos is close to the one of radiation domination , while the microscopic state is far from thermal . this result is similar to the conclusion of @xcite regarding @xmath10 theory in flat spacetime . the sharp change of the eos we observed can be very conveniently related to the moment of energy transfer from the inflaton to the inhomogeneous radiation . as we mentioned , it is a very fast process , as is well - known from the theory of preheating , which takes place in only about @xmath0 sec . moreover , the time at which the transition occurs depends only weakly on the coupling @xmath37 between the inflaton and the other fields . the exact dependance is non - monotonic , and the transition time oscillates around @xmath246 sec ( as set by preheating ) for a very wide range of couplings . this result is drastically different from perturbative inflaton decay , where the transition timing is proportional to the coupling . \i ) the first is related to the generation of modulated cosmological fluctuations from preheating . it is conceivable @xcite that the strength of the inflaton interactions with its decay products is a function of some moduli fields as expected in string theory . fluctuations of the light moduli ( with mass smaller than @xmath247 gev ) are inevitably generated during inflation . this results in spatial variations of the couplings at large scales , well beyond the size of the causal ( hubble ) patch at the end of inflation . hence , the transition from inflation to radiation domination occurs at slightly diffferent times in different hubble patches , giving rise to scalar metric fluctuations . to calculate metric fluctuations from modulated perturbations , one needs to know the exact evolution of the eos , from the matter domination ( inflaton ) to the radiation domination ( light decay products ) stage . we found that the transition of the eos occurs sharply in a step - like manner . therefore one can introduce a time - like ( but spatially rippled ) hypersurface which divides two cosmological regimes , with @xmath244 and @xmath248 . metric fluctuations can be calculated using gr matching conditions across this hypersurface , similar to how it was done to calculate modulated fluctuations in hybrid inflation @xcite . \ii ) the second application of the eos history is related to the formula which links the number of efoldings during inflation @xmath9 and the present day physical wavelengths of fluctuations in terms of the logs of their momenta @xmath249 . indeed , the cosmological evolution of the scale factor @xmath250 is ultimately defined by the eos . refining the @xmath19 relation is mandatory to put precise observational constraints on inflationary models . in this paper we notice that a ( practically ) immediate transition from inflation to radiation domination is a common feature of inflationary models with preheating , either parametric resonance or tachyonic ones . equation ( [ deltadef ] ) , with our result ( [ finalrat ] ) , refines the @xmath19 relation . \iii ) the next note is related to the model building of inflaton interactions . in the majority of papers discussing inflaton decay , interactions with bosons are considered in the form of four - legs vertices of the type @xmath14 . these interactions are the dominant ones at the early stages of preheating , when the amplitude of the inflaton is large . ( we do not consider here nonrenormalizable interactions , which if present could be more important . ) however , they do not lead to a complete inflaton decay . a complete decay requires vertices with only one inflaton , as the yukawa interaction @xmath251 wih fermions , or the three - legs bosonic interaction @xmath252 ( where @xmath36 is a mass scale ) . as we have argued , these trilinear interactions are very natural in supersymmetric models . supersymmetry has the additional advantage of protecting the flatness of the inflaton potential from too large radiative corrections , which arise from the coupling of the inflaton to the light degrees of freedom . \iv ) finally , the early evolution of the light degrees of freedom from the decay of the inflaton can have very important consequences for the generation of particles far from thermal equilibrium . a particularly relevant application is the possible ( over)production of gravitinos from the decay products of the inflaton . although this is a perturbative generation , it can be very effective , since it occurs at very early times . traditionally , the limit @xmath253 is due to gravitinos produced only after the @xmath21 quanta have thermalized . however , a nonthermal distribution can also be responsible for gravitino overproduction @xcite . these inflaton decay products arise very rapidly , with an energy density much higher than @xmath254 ; although their initial distribution is far from thermal , we have seen that they can be responsible for a significant gravitino production , which in some cases exceeds by several orders magnitude the allowed bound . clearly , this is a less universal bound than the standard one , since it depends on the details of how the thermal bath is produced . ( for instance , the gravitino abundance can be diluted by entropy release during thermalization . ) in addition , the gravitino problem is avoided altogether if the gravitino mass is significantly smaller ( as in gauge mediated supersymmetry breaking ) or larger ( as in the simplest versions of the kklt model @xcite ) than the electroweak scale . still , this possible production must be addressed in many inflationary models . * acknowledgements*. we thank j. berges , j.f . dufaux , a. linde , a. mazumdar , and i. tkachev for useful discussions . gf and mp thank cita for its warm hospitality during different stages of this work . lk was supported by ciar and nserc . mp was supported in part by the doe grant de - fg02 - 94er40823 . r. allahverdi , phys . d * 62 * , 063509 ( 2000 ) [ arxiv : hep - ph/0004035 ] ; s. davidson and s. sarkar , jhep * 0011 * , 012 ( 2000 ) [ arxiv : hep - ph/0009078 ] ; r. allahverdi and m. drees , phys . d * 66 * , 063513 ( 2002 ) [ arxiv : hep - ph/0205246 ] . g. n. felder , j. garcia - bellido , p. b. greene , l. kofman , a. d. linde and i. tkachev , phys . lett . * 87 * , 011601 ( 2001 ) [ arxiv : hep - ph/0012142 ] ; g. n. felder , l. kofman and a. d. linde , phys . d * 64 * , 123517 ( 2001 ) [ arxiv : hep - th/0106179 ] . r. kallosh , l. kofman , a. d. linde and a. van proeyen , phys . d * 61 * ( 2000 ) 103503 [ arxiv : hep - th/9907124 ] , class . * 17 * ( 2000 ) 4269 [ arxiv : hep - th/0006179 ] ; g. f. giudice , a. riotto , and i. tkachev jhep * 9908 * ( 1999 ) 009 [ arxiv : hep - ph/9907510 ] , jhep * 9911 * ( 1999 ) 036 [ arxiv : hep - ph/9911302 ] ; h. p. nilles , m. peloso and l. sorbo , phys . rev . lett . * 87 * ( 2001 ) 051302 [ arxiv : hep - ph/0102264 ] , jhep * 0104 * ( 2001 ) 004 [ arxiv : hep - th/0103202 ] . j. ellis , a. linde and d. nanopoulos , phys . lett . * 118b * , 59 ( 1982 ) ; l.m . krauss , nucl . b227 * , 556 ( 1983 ) ; d. nanopoulos , k. olive and m. srednicki , phys . lett . * 127b * , 30 ( 1983 ) ; m. yu . khlopov and a. linde , phys . 138b * , 265 ( 1984 ) ; j. ellis , j. kim and d. nanopoulos , phys . lett . * 145b * , 181 ( 1984 ) ; j. ellis , g.b . gelmini , j.l . lopez , d.v . nanopoulos , and s. sarkar , nucl . b373 * , 399 ( 1992 ) ; m. kawasaki and t. moroi , progr . theor . phys . * 93 * , 879 ( 1995 ) . j. baacke , k. heitmann and c. patzold , phys . d * 58 * , 125013 ( 1998 ) [ arxiv : hep - ph/9806205 ] ; p. b. greene and l. kofman , phys . b * 448 * , 6 ( 1999 ) [ arxiv : hep - ph/9807339 ] ; g. f. giudice , m. peloso , a. riotto and i. tkachev , jhep * 9908 * , 014 ( 1999 ) [ arxiv : hep - ph/9905242 ] . p. b. greene and l. kofman , phys . d * 62 * , 123516 ( 2000 ) [ arxiv : hep - ph/0003018 ] . m. peloso and l. sorbo , jhep * 0005 * , 016 ( 2000 ) [ arxiv : hep - ph/0003045 ] . l. boubekeur , s. davidson , m. peloso and l. sorbo , phys . d * 67 * , 043515 ( 2003 ) [ arxiv : hep - ph/0209256 ] . for the entropy density we used the formula @xmath255 $ ] . see for example r. h. brandenberger , t. prokopec and v. mukhanov , phys . d * 48 * , 2443 ( 1993 ) [ arxiv : gr - qc/9208009 ] . d. h. lyth and e. d. stewart , phys . lett . * 75 * , 201 ( 1995 ) [ arxiv : hep - ph/9502417 ] ; l. kofman , a. d. linde and a. a. starobinsky , phys . lett . * 76 * , 1011 ( 1996 ) [ arxiv : hep - th/9510119 ] ; g. n. felder , l. kofman , a. d. linde and i. tkachev , jhep * 0008 * , 010 ( 2000 ) [ arxiv : hep - ph/0004024 ] .

we propose a new algorithm for calculation of vibrational spectra of molecules using tensor train decomposition . under the assumption that eigenfunctions lie on a low - parametric manifold of low - rank tensors we suggest using well - known iterative methods that utilize matrix inversion ( lobpcg , inverse iteration ) and solve corresponding linear systems inexactly along this manifold . as an application , we accurately compute vibrational spectra ( 84 states ) of acetonitrile molecule chcn on a laptop in one hour using only mb of memory to represent all computed eigenfunctions .

1605.08422

in this work we consider time - independent schrdinger equation to calculate vibrational spectra of molecules . the goal is to find smallest eigenvalues and corresponding eigenfunctions of the hamiltonian operator . the key assumption that we use is that potential energy surface ( pes ) can be approximated by a small number of sum - of - product functions . this holds , e.g. if pes is a polynomial . we discretize the problem using the discrete variable representation ( dvr ) scheme @xcite . the problem is that the storage required for each eigenvector grows exponentially with dimension @xmath2 as @xmath3 , where @xmath4 is number of grid points in each dimension . even for the dvr scheme where @xmath5 grid points in each dimension is often sufficient to provide very accurate results we would get @xmath6 pb of storage for a 12 dimensional problem . this issue is often referred to as the _ curse of dimensionality_. to avoid exponential growth of storage we use _ tensor train _ ( tt ) decomposition @xcite to approximate the operator and the eigenfunctions in the dvr representation , which is known to have exponential convergence rate . it is important to note that the tt - format is algebraically equivalent to the matrix product state format ( mps ) , which has been used for a long time in quantum information theory and solid state physics to approximate certain wavefunctions @xcite ( dmrg method ) , see the review @xcite for more details . prior research @xcite has shown that the eigenfunctions often can be well approximated in the tt - format , i.e. they lie on a certain low - dimensional non - linear manifold . the key question is how to utilize this a priori knowledge in computations . we propose to use well - established iterative methods that utilize matrix inversion and solve corresponding linear systems inexactly along the manifold to accelerate the convergence . our main contributions are : * we propose a concept of a _ manifold preconditioner _ that explicitly utilizes information about the size of the tt - representation . we use the manifold preconditioner for a tensor version of _ locally optimal block preconditioned conjugate gradient _ method ( lobpcg ) @xcite . we will refer to this approach as _ manifold - preconditioned lobpcg _ ( mp lobpcg ) . the approach is first illustrated on computation of a single eigenvector ( section [ sec : mp1 ] ) and then extended to the block case ( section [ sec : mpb ] ) . * we propose tensor version of simultaneous inverse iteration ( also known as block inverse power method ) , which significantly improves accuracy of the proposed mp lobpcg . similarly to the manifold preconditioner the inversion is done using the a priori information that the solution belongs to a certain low - parametric manifold . we will refer to this method as _ manifold - projected simultaneous inverse iteration _ ( mp sii ) . the approach is first illustrated on computation of a single eigenvector ( section [ sec : iii1 ] ) and then extended to the block case ( section [ sec : iib ] ) . * we calculate vibrational spectra of acetonitrile molecule ch@xmath0cn using the proposed approach ( section [ sec:12 ] ) . the results are more accurate than those of the smolyak grid approach @xcite but with much less storage requirements , and more accurate than the recent memory - efficient h - rrbpm method @xcite , which is also based on tensor decompositions . we note that the smolyak grid approach does not require pes to be approximated by a small number of sum - of - product functions . we follow @xcite and consider schrdinger equation with omitted @xmath7 cross terms and the potential - like term in the normal coordinate kinetic energy operator . the hamiltonian in this case looks as @xmath8 where @xmath9 denotes potential energy surface ( pes ) . we discretize the problem using the discrete variable representation ( dvr ) scheme on the tensor product of hermite meshes @xcite such that each unknown eigenfunction is represented as @xmath10 where @xmath11 denotes one - dimensional dvr basis function . we call arising multidimensional arrays _ tensors_. the hamiltonian consists of two terms : the laplacian - like part and the pes . it is well - known that the laplacian - like term can be written in the kronecker product form @xmath12 where @xmath13 is the one dimensional discretization of the @xmath14-th mode . the dvr discretization of the pes is represented as a tensor @xmath15 . the operator corresponding to the multiplication by @xmath15 is diagonal . finally the hamiltonian is written as @xmath16 for our purposes it is convenient to treat @xmath17 not as a 2d @xmath18 matrix , but as a multidimensional operator @xmath19 . in this case the discretized schrdinger equation has the form @xmath20 hereinafter we use notation @xmath21 implying matrix - by - vector product from . using this notation can be equivalently written as @xmath22 in this section we discuss how to solve the schrdinger equation numerically and present our approach for finding a single eigenvector . the case of multiple eigenvalues is discussed in section [ sec : block ] . the standard way to find required number of smallest eigenvalues is to use iterative methods . the simplest iterative method of finding one smallest eigenvalue is the shifted power iteration @xmath23 where the shift @xmath24 is an approximation to the largest eigenvalue of @xmath25 . the matrix - by - vector product is the bottleneck operation in this procedure . this method was successfully applied to the calculation of vibrational spectra in @xcite . the eigenvectors in this work are represented as sum - of - products , which allows for fast matrix - by - vector multiplications . despite the ease of implementation and efficiency of each iteration , the convergence of this method requires thousands of iterations . instead of power iteration we use inverse iteration @xcite @xmath26 which is known to have much faster convergence if a good approximation @xmath24 to the required eigenvalue @xmath27 is known . question related the solution of linear systems in the tt - format will be discussed in section [ sec : iii1 ] . convergence of the inverse iteration is determined by ratio @xmath28 where @xmath29 is the next closest to @xmath27 eigenvalue . thus , @xmath24 has to be closer to @xmath27 than to @xmath29 for the method to converge . however , the closer @xmath24 to @xmath27 is , the more difficult to solve the linear system with matrix @xmath30 it is . therefore , typically this system is solved inexactly @xcite . parameter @xmath24 can also depend on the iteration number ( rayleigh quotient iteration ) , however in our experiments ( section [ sec : num ] ) constant choice of @xmath24 yields convergence in 5 iterations . as it follows from for the inverse iteration to converge fast a good initial approximation has to be found . to get initial approximation we propose to use locally optimal block preconditioned conjugate gradient ( lobpcg ) method as it is relatively easy to implement in tensor formats , and a preconditioner can be explicitly utilized . the problem with straightforward usage of the iterative processes above is that we need to store an eigenvector @xmath31 . the storage of this vector is @xmath32 , which is prohibitive even for @xmath33 and @xmath34 . therefore we need a compact representation of an eigenfunction which allows to do inversion and basic linear algebra operations in a fast way . for this purpose we use the tensor train ( tt , mps ) format @xcite . tensor @xmath35 is said to be in the tt - format if it is represented as @xmath36 where @xmath37 , @xmath38 , @xmath39 . matrices @xmath40 are called tt - cores and @xmath41 are called tt - ranks . for simplicity we assume that @xmath42 , @xmath43 and call @xmath44 the tt - rank . in numerical experiments we use different mode sizes @xmath45 , @xmath46 , but for simplicity we will use notation @xmath47 in complexity estimates . compared to @xmath32 parameters of the whole tensor , tt decomposition contains @xmath48 parameters as each @xmath49 , @xmath50 has size @xmath51 . the definition of the tt - format of an operator is similar to the tt representation of tensors @xmath52 where tt - cores @xmath53 , @xmath54 . if @xmath55 , @xmath43 then this representation contains @xmath56 degrees of freedom . tt - format can be considered as a multidimensional generalization of svd . other alternatives to generalize the svd to many dimensions are the canonical , tucker and hierarchical tucker formats . we refer the reader to @xcite for detailed survey . the important point why we use the tt decomposition is that it can be computed in a stable way and it does not suffer from the `` curse of dimensionality '' . moreover , there exists well - developed software packages to work with the tt - decomposition @xcite . for the inverse iteration we need to find tt - representation of @xmath57 by approximately solving a linear system @xmath58 assume that both the exact eigenvector @xmath35 and the current approximation @xmath31 belong to the manifold @xmath59 of tensors with tt - rank @xmath44 . the solution of may have ranks larger than @xmath44 and therefore be out of @xmath59 . in the present work we suggest exploiting the information that @xmath35 belongs to @xmath59 and retract the solution of back to the manifold . we refer to this concept as a _ manifold - projected inverse iteration _ ( mp ii ) . in this work we pose the following optimization problem with a rank constraint @xmath60 problem is hard to solve as operator @xmath61 is close to singular . similarly to the inexact inverse iteration framework we are not searching for the solution that finds global minima of , but utilize several sweeps of the alternating least squares ( als ) method with the initial guess @xmath31 @xcite . the als procedure alternately fixes all but one tt - core and solves minimization problem with respect to this tt - core . for instance , an update of a core @xmath62 when all other cores are fixed is @xmath63 found from @xmath64 the minimization over a single core ( see appendix [ sec : als ] ) is a standard linear least squares problem with the unknown vector that has the size @xmath65 size of the corresponding tt - core and is very cheap . moreover these systems can be also solved iteratively . the described minimization over all cores in the tt - representation is referred to as one sweep of the als . the total computational cost of one sweep is then @xmath66 , where @xmath67 is maximum tt - rank of the operator @xmath17 , see appendix [ sec : als ] . _ according to the proposed concept we start from @xmath68 and use only a few sweeps ( one or two ) of the als method rather than running the method till convergence . moreover , we found that one can solve local linear systems inexactly either with fixed number of iterations or fixed low accuracy . _ such low requirements for solution of local linear systems and number of als sweeps results in a very efficient method . however , for this method to converge , a good initial approximation to both eigenvector and eigenvalue has to be found . to get initial approximation we use lobpcg method . the lobpcg algorithm for one eigenvalue looks as follows @xmath69 where @xmath70 denotes preconditioner and vector of coefficients @xmath71 is chosen from minimization of the rayleigh quotient @xmath72 finding @xmath73 is equivalent to solving the following @xmath74 eigenvalue problem @xmath75\ , \alpha = \lambda \begin{bmatrix } { \mathcal{x}}_{k } \\ { \mathcal{r}}_k \\ { \mathcal{p}}_k \end{bmatrix } [ { \mathcal{x}}_{k } , { \mathcal{r}}_k , { \mathcal{p}}_k]\ , \alpha.\ ] ] let us discuss tt version of the lobpcg . operations required to implement the lobpcg are presented below : [ [ preconditioning . ] ] preconditioning . + + + + + + + + + + + + + + + + the key part of the lobpcg iteration is multiplication by a preconditioner @xmath70 . in this work we use @xmath76 as a preconditioner . this preconditioner works well if the density of states is low , see @xcite . to make a preconditioner more efficient one can project it to the orthogonal complement of current approximation of the solution , see jacobi - davidson method @xcite . instead of forming @xmath77 we calculate matrix - by - vector multiplication @xmath78 . hence , similarly to the inverse iteration ( section [ sec : iii1 ] ) we propose solving a minimization problem @xmath79 we also use only several sweeps of als for this problem . we refer to this construction of preconditioner as a _ manifold preconditioner _ as it retracts the residual on a manifold of tensors with fixed rank @xmath44 . note that if @xmath30 is known to be positive definite , then minimization of energy functional @xmath80 can be used instead of minimization of the residual . [ [ summation - of - two - tensors . ] ] summation of two tensors . + + + + + + + + + + + + + + + + + + + + + + + + + given two tensors @xmath35 and @xmath81 with ranks @xmath44 in the tt - format @xmath82 the cores of the sum @xmath83 are defined as @xcite @xmath84 and @xmath85 thus , tensor @xmath86 is explicitly represented with ranks @xmath87 . [ [ inner - product - and - norm . ] ] inner product and norm . + + + + + + + + + + + + + + + + + + + + + + + to find inner product of two tensors @xmath35 and @xmath81 in the tt - format we first need to calculate the hadamard product , which can calculated as @xmath88 therefore , @xmath89 where @xmath90 using special structure of matrices @xmath91 the inner product can be calculated in @xmath92 complexity @xcite . the norm can be computed using inner product as @xmath93 . [ [ reducing - rank - rounding . ] ] reducing rank ( rounding ) . + + + + + + + + + + + + + + + + + + + + + + + + + as we have seen , after summation ranks grow . to avoid rank growth there exists special _ rounding _ operation . it suboptimally reduces rank of a tensor with a given accuracy . in 2d the rounding procedure of @xmath94 looks as follows . first we calculate qr decompositions of matrices @xmath95 and @xmath9 : @xmath96 hence @xmath97 finally , to reduce the rank we calculate the svd decomposition of @xmath98 and truncate singular values up to required accuracy . in @xcite this idea is generalized to the tt case . the complexity is @xmath92 . * @xmath99 : calculates @xmath100 via the cross approximation algorithm ( section [ sec : iib ] ) . * @xmath101 : block multiplication of a vector of tt tensors @xmath102 by real - valued matrix @xmath103 using @xmath104 function , see . * @xmath105 : orthogonalizes tt tensors @xmath106 : @xmath107 via cholesky ( section [ sec : iib ] ) or modified gram - schmidt procedure . matrix - by - vector multiplications are done using using @xmath108 . * @xmath109 : solves @xmath110 , @xmath111`length`@xmath112 using @xmath113 sweeps of als method to minimize @xmath114 with a rank constraint . * @xmath115 : orthogonalizes tt tensors @xmath106 : @xmath107 with respect to @xmath116 : @xmath117 . to avoid rank growth we use rounding if ` length`@xmath118 is small and via ` multifuncrs ` if ` length`@xmath118 is large . * @xmath119 : truncates each tensor @xmath106 : @xmath107 with rank @xmath44 using rounding procedure . [ [ matrix - by - vector - multiplication . ] ] matrix - by - vector multiplication . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to calculate matrix - by - vector operation @xmath120 in it is convenient to have hamiltonian @xmath25 represented in the tt - format . in this case there exists explicit representation of matrix - by - vector multiplication @xmath121 when both @xmath17 and @xmath35 are represented in the tt - format @xcite @xmath122 which gives representation with tt - rank @xmath123 . to reduce the rank one can either use the rounding procedure or use als minimization of the following optimization problem @xmath124 which is faster than rounding for large ranks in the previous section we discussed the algorithm for a single eigenvector . in this section we extend the algorithm to a block case . the difference is that we need to make additional block operations such as othogonalization and block multiplication . in this section we present manifold - projected version of the simultaneous inverse iteration , which is also known as block inverse power method . assume we are given a good initial approximation to eigenvalues and eigenvectors of linear operator @xmath17 given by its tt - representation . in this case inverse iteration yields fast convergence rate with low memory requirements . [ alg : lobpcg ] group close eigenvalues in clusters @xmath125 $ ] , @xmath126 $ ] compute shift @xmath127 average eigenvalue in cluster @xmath128 compute @xmath129 corresponding subvector of @xmath130 @xmath131 ) @xmath132 $ ] , @xmath133 $ ] @xmath134 given initial approximation we first split eigenvalues into clusters of eigenvalues . proximity of eigenvalues is defined using a threshold parameter . if a cluster consists of one eigenvalue , then we run a version described in section [ sec : iii1 ] . otherwise we need additional orthogonalization on each iteration step . the orthogonalization is done using the qr decomposition via cholesky factorization . let us consider orthogonalization procedure for vector @xmath135 $ ] in more details . first we calculate gram matrix @xmath136 where we used notation @xmath137 the calculation of gram matrix can be done in @xmath138 operations as calculation of a single inner product requires @xmath139 operations . then we calculate the cholesky factorization of the @xmath140 gram matrix @xmath141 . we note that @xmath142 consists of numbers , so the standard cholesky factorization of matrices is used . the final and the most time consuming step is to find block matrix - by - vector operation denoted by ` block_matvec`@xmath143 . for a general matrix @xmath144 function ` block_matvec`@xmath145 produces block vector @xmath146 $ ] such that @xmath147 if @xmath148 is small , say @xmath149 , then the summation can be done using summation and rounding . the typical value of @xmath148 in numerical experiments is @xmath150 or @xmath151 , so to reduce the constant in complexity we use the so - called _ cross approximation method _ , which is able to calculate the tt - decomposition of a tensor using only few of its elements . namely , the cross approximation is able to build tt - representation of a tensor of exact rank @xmath44 using only @xmath152 elements ( number of parameters in the tt - format ) using interpolation formula @xcite with @xmath153 complexity . if tensor is approximately of low rank then tensor can be approximated with given accuracy using the same approach and quasioptimality estimates exist @xcite . to find tt - representation of @xmath154 we calculate @xmath152 elements of tensor @xmath154 by explicitly calculating elements in @xmath155 and summing them up with coefficients @xmath156 . this approach allows to calculate general type of functions of a block vector @xmath157 and is referred to as ` multifuncrs ` . it is typically used when either the number of input tensors is large or some nonlinear function of input tensors must be computed . the similar approach was used in @xcite for a fast computation of convolution . the result of solving the block system @xmath158 using @xmath113 of optimization procedure is denoted as @xmath159 , where @xmath160.\ ] ] the overall algorithm is summarized in algorithm [ alg : inverse ] . all discussed auxiliary functions are presented in algorithm [ alg : functions ] . if the cluster size is much smaller than the number of required eigenvalues , the complexity of finding each cluster is fully defined by the inversion operation which costs @xmath161 . thus , the overall complexity of the inverse iteration scales linearly with @xmath162 : @xmath163 . we note that each cluster can be processed independently in parallel . we use the lobpcg method to get initial approximation for the manifold - preconditioned simultaneous inverse iteration . the problem is that each iteration of the lobpcg is much slower compared to the inverse iteration for large number of eigenvalues . we hence only run lobpcg with small ranks and then correct its solution with larger ranks using the inverse iteration . [ alg : lobpcg ] [ alg : lobpcg ] @xmath164 @xmath165 @xmath166 @xmath167 @xmath168 augment @xmath116 with corresponding tensors from @xmath169 restart the algorithm with new @xmath130 ( optionally ) increase truncation rank @xmath170 the lobpcg algorithm is summarized in algorithm [ alg : lobpcg ] . auxiliary functions used in the algorithm are presented in algorithm [ alg : functions ] . we also use matlab like notation for submatrices , e.g. @xmath171 is the corresponding submatrix in matrix @xmath172 . block matrix - by - vector multiplication arises when multiplying by @xmath173 matrix , where @xmath162 is the number of eigenvalues to be found . when @xmath174 is a large number and we use ` multifuncrs ` for block matrix - by - vector product instead of straightforward truncation . this is the most time consuming step in the algorithm and it costs @xmath175 . another time - consuming part is matrix - by - vector multiplication which costs @xmath176 . thus , the overall complexity of each iteration is @xmath177 . in order to accelerate the method , we consider a version of lobpcg with deflation . deflation technique is that we stop iterating converged eigenvectors . in this case the residual must be orthogonalized with respect to the converged eigenvectors . this procedure is denoted as ` ortho ` and is described in more details in algorithm [ alg : functions ] . it might be also useful to increase rank of vectors that did not converged after previous deflation step . the prototype is implemented in python using the ` ttpy ` library https://github.com/oseledets/ttpy . the code of the proposed algorithm can be found at https://bitbucket.org/rakhuba/ttvibr . for the basic linear algebra tasks the mkl library is used . python and mkl are from the anaconda python distribution https://www.continuum.io . python version is 2.7.11 . mkl version is 11.1 - 1 . tests were performed on a single intel core i7 2.6 ghz processor with 8 gb of ram . however , only 1 thread was used . first of all , we test our approach on a model hamiltonian when the solution is known analytically . following @xcite we choose bilinearly coupled 64 dimensional hamiltonian @xmath178 with @xmath179 and @xmath180 . the tt - rank of this hamiltonian is @xmath181 for all inner modes independently of @xmath2 or @xmath4 . to solve this problem we first use manifold - preconditioned lobpcg method ( section [ sec : mpb ] ) with rank @xmath182 and then correct it with the mp inverse iteration . the thresholding parameter @xmath183 for separation of energy levels into clusters in the inverse iteration is @xmath184 ( @xmath185 and @xmath186 are in the same cluster if @xmath187 ) . the mode size @xmath33 is constant for each mode . as it follows from figure [ fig : berror ] the mp inverse iteration significantly improves the accuracy of the solution . we used 10 iterations of the lobpcg . the mp lobpcg computations took approximately 3 hours of cpu time and the mp inverse iteration took additionally 30 minutes . we also tested tensor version of the preconditioned inverse iteration ( pinvit ) @xcite that in case of a single eigenvector looks as @xmath188 where @xmath189 is selected to minimize the rayleigh quotient . figure [ fig : conv_sd ] illustrates convergence behavior of last 10 eigenvalues for different methods . the pinvit which also allows for explicit preconditioner converged to wrong eigenvalues . lobpcg method without preconditioner is unstable due to rank thresholding . we note that all these iterations converged to correct eigenvalues if the number of eigenvalues to be found was less than @xmath190 . in this part we present calculations of vibrational spectra of acetonitrile ( ch@xmath0cn ) molecule . the hamiltonian used is described in @xcite and looks as follows @xmath191 it contains @xmath192 terms : @xmath193 kinetic energy terms , @xmath193 quadratic , @xmath194 cubic , and @xmath195 quartic potential terms . we chose the same basis size that was used in @xcite , namely the mode sizes were @xmath196 corresponding to the order described in that work . we found that ranks of the hamiltonian for this particular molecule do not strongly depend on the permutation of indices , namely the largest rank we observed among random permutations was @xmath197 , while the maximum rank of the best permutation was @xmath198 . in computations we permuted indices such that array of @xmath199 is sorted in a decaying order . table [ tab : hamranks ] contains ranks of the hamiltonian in the tt - representation . we note that total ranks are ranks of a sum of potentials after rounding , and hence they are not equal to the sum of ranks of potentials in table [ tab : hamranks ] . to assemble the potential @xmath9 one needs to add rank-1 terms @xmath200 and rank-1 terms @xmath201 . each rank-1 term can be expressed analytically in the tt - format . as was mentioned in section [ sec : mp1 ] after each summation the rank grows , so the rounding procedure is used . recall that the rounding procedure requires @xmath202 operations . thus , the complexity of assembling the hamiltonian is @xmath203 where ` nnz ` stands for number of nonzeros , @xmath4 is the maximum mode size and @xmath67 is the maximum rank of @xmath25 . the total time of assembling the hamiltonian was less than @xmath204 second . .tt - ranks of the parts of the hamiltonian with @xmath205 threshold . + [ cols="<,^,^,^,^,^,^,^,^,^,^,^,^",options="header " , ] we ran the lobpcg method with tt - rank equal to 12 and used manifold preconditioner . initial guess is chosen from the solution of the harmonic part of the hamiltonian . eigenvectors of multidimensional quantum harmonic oscillator are tensor product of 1d oscillator eigenvectors and therefore can be represented analytically in the tt - format with rank @xmath204 . shift for the preconditioner is set to be the lowest energy of the harmonic part . the convergence of each eigenvalue is presented on figure [ fig : conv_ch3cn ] . the obtained eigenfunctions were used as an initial approximation to the inverse iterations with ranks equal to @xmath206 . shifts were chosen to be lobpcg energies . the thresholding parameter @xmath183 for separation of energy levels into clusters in the inverse iteration is @xmath207 these results were further corrected with the inverse iteration with rank @xmath208 . as it follows from table [ tab : energies ] and figure [ fig : ch3cn_error ] both corrections are more accurate than the h - rrbpm method . the latter correction with rank @xmath208 yields energy levels lower than those of the smolyak quadrature method @xcite which means that our energy levels are more accurate . we note that storage of a solution with @xmath208 is less than the storage of the smolyak method ( 180 mb compared with 1.5 gb ) . timings and storage of the h - rrbpm method were taken from @xcite . on this example the state - of - the - art method ` eigb ` @xcite method converges within approximately several days . the problem is that all basis functions are considered in one basis , which leads to large ranks . nevertheless , the ` eigb ` becomes very efficient and accurate when small number of eigenpairs are required . of eigenvalues of the acetonitrile molecule with respect to the eigenvalue number . are energies obtained by smolyak quadratures @xcite . negative value of error stand for more accurate than smolyak quadratures energies . black line denotes zero value of the error . mp sii stands for manifold - projected inverse iteration.,width=340 ] the simplest basis set for representing unknown eigenfunctions is the direct product ( dp ) of one - dimensional basis functions . if a fast matrix - by - vector operation is given , then krylov iterative methods are available and the only problem is the exponential storage requirements @xmath3 . alternatively one can prune a direct product basis @xcite or use a basis that is a product of functions with more than one variable @xcite . in this work , we focus on dp basis and further reduce @xmath3 storage by approximating unknown eigenvectors in the tt - format . we refer the reader to @xcite for detailed surveys on tensor decompositions . canonical tensor decomposition ( also called cp decomposition or candecomp / parafac model ) of the eigenvectors of vibrational problems was considered in work by leclerc and carrington @xcite . the authors used rank - reduced block power method ( rrbpm ) . each iteration of this method involves matrix - by - vector product , which can be efficiently implemented in tensor formats . the problem is that this method has poor convergence . moreover , canonical decomposition is known to have stability issues @xcite . the hierarchical rrbpm ( h - rrbpm ) proposed in @xcite by thomas and carrington is a significant improvement of the rrbpm . this method also utilizes sum - of - products representation , but treats strongly coupled coordinates together . coupled coordinates are further decomposed hierarchically . the multi configuration time dependent hartree ( mctdh ) approach @xcite also uses tensor format , namely the tucker decomposition . this approach reduces complexity , but suffers from the curse of dimensionality . this problem was solved in the multilayer mctdh @xcite which is similar to the hierarchical tucker representation @xcite . we would also like to discuss tensor algorithms for solving eigenvalue problems developed in mathematical community . there are two natural directions of solving eigenvalue problems in tensor formats . one direction is straightforward generalization of iterative methods to tensor arithmetics with rank reduction on each iteration . for the canonical decomposition power method with shifts was generalized in @xcite and used in the rrbpm method . the preconditioned inverse iteration ( pinvit ) for tensor formats was considered in @xcite . the inverse iteration used in this paper differs from the pinvit , which is basically preconditioned steepest descent . tensor version of the inverse iteration based on iterative solution of arising linear systems was considered in @xcite . the pinvit iteration can explicitly utilize a preconditioner . the construction of preconditioners in tensor formats for eigenvalue problems was considered in @xcite . the approach for a general operator @xcite uses newton - schulz iteration in order to find a good preconditioner . however , due to high amount of matrix multiplications , this approach becomes time - consuming . in order to construct a preconditioner one can use approximate inverse matrix or approximate solution of linear systems . see @xcite for solvers in tensor formats . the more advanced lobpcg method was for the first time generalized in @xcite . we utilize this method and construct preconditioner based on optimization procedure . the pinvit method with the proposed preconditioner and the lobpcg with and without preconditioning were tested in section [ sec : num ] . although rank - truncated pinvit method is fast and is able to accurately calculate small amount of eigenvalues , it fails to converge when a lot of eigenvalues are required . alternatively to iterative methods , one can pose an optimization problem minimization of the rayleigh quotient with the constraint on rank . this approach was recently considered in the matrix product state ( mps ) community @xcite and independently in @xcite . the only disadvantages is that all eigenfunctions are treated in one basis , which leads to large ranks and the method becomes slow ( calculation of the acetonitrile took several days ) . nevertheless , this approach becomes very efficient and accurate when small number of eigenpairs are required . one of the most interesting missing bits is the theory of the proposed approach . first , why the eigenfunctions can be well - approximated in the tt - format and what are the restrictions on the pes . second , the convergence properties of the manifold preconditioner have to be established . these questions will be addressed in future works . from practical point of view , the applicability of the proposed approach for general molecules has to be studied . currently , it requires the explicit knowledge of sum - of - product representation . obtaining such a representation is a separate issue , which can be done by using existing methods : potfit @xcite or more general tt - cross approximation approach @xcite . if pes has large ranks , coordinate transformation can be helpful @xcite . we would like to thank prof . tucker carrington jr . and his research group for providing data for the numerical experiment section . the authors also would like to thank alexander novikov for his help in improving the manuscript . we also thank anonymous referees for their comments and constructive suggestions . this work has been supported by russian science foundation grant 14 - 11 - 00659 . let us discuss technical details of solving the problem . to illustrate the idea let us start from the skeleton decomposition in 2d . in this case @xmath210 , where @xmath211 . this representation is equivalent to the tt - format in 2d . indeed , by defining cores @xmath212 and @xmath213 we get . the als procedure starts with fixing one core . let us fix @xmath9 with orthonormal columns ( can always be done by qr decomposition ) and find the updated @xmath95 from the minimization problem @xmath214 thanks to well - known property of kronecker products @xmath215 the problem can be equivalently represented as a minimization problem on unknown vector @xmath216 @xmath217 where @xmath218 denotes vectorization of @xmath219 by reshaping it into a column vector . finally we get small linear system @xmath220 where matrix @xmath221 is of size @xmath222 , while the initial matrix @xmath223 is of size @xmath224 . to avoid squared condition number one can formally use the projectional approach @xmath225 which corresponds to the zero gradient of the energy functional @xmath226 the problem is that @xmath223 is not positive definite unless the smallest eigenvalue is required . nevertheless , we found that this approach also works if @xmath227 is close enough to the eigenvector corresponding to the eigenvalue closest to @xmath24 . this assumption holds as before running mp ii we find a good initial guess for both @xmath24 and eigenvector , see section [ sec : mp1 ] . note that operator @xmath228 is also given in the low - rank matrix format ( also corresponds to the tt - matrix format in 2d ) : @xmath229 where @xmath230 . hence , the matrix of a local system can be represented as @xmath231 and one can do matrix - vector products with @xmath232 complexity . indeed , matrices @xmath233 are of size @xmath234 and @xmath235 are @xmath236 . hence multiplication of @xmath237 by a vector requires @xmath238 operations . similarly to the 2d case in arbitrary dimension we get the following linear system on the vectorized @xmath239-th core @xmath240 @xmath241 where @xmath242 matrix @xmath243 is a multiplication of first @xmath244 cores reshaped into @xmath245 matrix : @xmath246 matrix @xmath247 is defined by analogy . typically additional orthogonalization of @xmath243 and @xmath247 is done @xcite . since operator @xmath223 is given by its tt - representation , matrix - vector multiplication requires @xmath248 operations , where @xmath44 is tt - rank of the tensor @xmath35 and @xmath67 is the maximum rank of the operator @xmath223 . compared with @xmath249 of the 2d case tt - cores between the first and last dimensions are of size @xmath250 for tt - tensor and @xmath251 for tt - matrix , so squared ranks appear . we note that if dimension and/or mode size are large : @xmath252 , then the total complexity is @xmath253 . + & & & & & reference + & & & & & & & & smolyak @xcite + & & & & & & & & @xmath254 gb + & & & & & & & & + level & sym . & @xmath255 & @xmath256 & @xmath255 & @xmath256 & @xmath255 & @xmath256 & @xmath255 & @xmath256 & @xmath255 & @xmath256 & @xmath256 & @xmath255 & @xmath209 + zpve & & 9837.893 & 0.485 & 9837.525 & 0.118 & 9837.429 & 0.022 & 9837.463 & 0.056 & 9837.408 & 0.001 & & @xmath257 & 9837.4073 + @xmath258 & @xmath255 & 361.24 & 0.25 & 361.04 & 0.04 & 361.00 & 0.01 & 361.016 & 0.03 & 360.991 & 0.000 & -0.001 & 360.990 & 360.991 + & & 361.24 & 0.25 & 361.06 & 0.07 & 361.00 & 0.01 & 361.029 & 0.04 & 360.991 & 0.000 & -0.001 & 360.990 & 360.991 + @xmath259 & @xmath255 & 723.60 & 0.42 & 723.41 & 0.23 & 723.22 & 0.04 & 723.274 & 0.09 & 723.180 & -0.001 & -0.001 & 723.180 & 723.181 + & & 723.60 & 0.42 & 723.41 & 0.23 & 723.22 & 0.04 & 723.276 & 0.09 & 723.180 & -0.001 & -0.001 & 723.180 & 723.181 + @xmath259 & @xmath260 & 724.27 & 0.44 & 724.05 & 0.23 & 723.87 & 0.04 & 723.919 & 0.09 & 723.827 & 0.000 & -0.001 & 723.826 & 723.827 + @xmath261 & @xmath260 & 901.25 & 0.59 & 900.83 & 0.16 & 900.71 & 0.05 & 900.722 & 0.06 & 900.660 & -0.002 & -0.004 & 900.658 & 900.662 + @xmath262 & @xmath255 & 1035.33 & 1.21 & 1034.23 & 0.10 & 1034.19 & 0.07 & 1034.211 & 0.08 & 1034.127 & 0.001 & -0.002 & 1034.124 & 1034.126 + & & 1035.34 & 1.21 & 1034.23 & 0.10 & 1034.20 & 0.07 & 1034.241 & 0.11 & 1034.127 & 0.001 & -0.002 & 1034.124 & 1034.126 + @xmath263 & @xmath264 & 1087.23 & 0.67 & 1086.86 & 0.31 & 1086.66 & 0.11 & 1086.720 & 0.17 & 1086.554 & 0.000 & -0.002 & 1086.552 & 1086.554 + @xmath263 & @xmath260 & 1087.23 & 0.67 & 1086.86 & 0.31 & 1086.66 & 0.11 & 1086.734 & 0.18 & 1086.554 & 0.000 & -0.001 & 1086.553 & 1086.554 + @xmath263 & @xmath255 & 1088.52 & 0.74 & 1088.09 & 0.31 & 1087.88 & 0.11 & 1087.960 & 0.18 & 1087.776 & 0.000 & -0.001 & 1087.775 & 1087.776 + & & 1088.52 & 0.75 & 1088.09 & 0.31 & 1087.88 & 0.11 & 1088.005 & 0.23 & 1087.776 & 0.000 & -0.001 & 1087.775 & 1087.776 + @xmath265 & @xmath255 & 1260.70 & 0.82 & 1260.04 & 0.16 & 1259.90 & 0.02 & 1259.947 & 0.07 & 1259.811 & -0.071 & -0.073 & 1259.809 & 1259.882 + & & 1260.70 & 0.82 & 1260.14 & 0.26 & 1259.90 & 0.02 & 1260.035 & 0.15 & 1259.811 & -0.071 & -0.073 & 1259.809 & 1259.882 + @xmath266 & @xmath260 & 1390.32 & 1.34 & 1389.44 & 0.47 & 1389.16 & 0.18 & 1390.003 & 1.03 & 1389.007 & 0.034 & -0.002 & 1388.971 & 1388.973 + @xmath267 & @xmath255 & 1396.40 & 1.71 & 1394.90 & 0.21 & 1394.79 & 0.10 & 1395.349 & 0.66 & 1394.701 & 0.012 & -0.007 & 1394.682 & 1394.689 + & & 1396.40 & 1.72 & 1394.90 & 0.21 & 1394.80 & 0.11 & 1395.519 & 0.83 & 1394.709 & 0.020 & -0.007 & 1394.682 & 1394.689 + @xmath267 & @xmath264 & 1396.56 & 1.65 & 1395.10 & 0.19 & 1395.01 & 0.10 & 1395.745 & 0.84 & 1394.916 & 0.009 & -0.007 & 1394.900 & 1394.907 + @xmath267 & @xmath260 & 1399.39 & 1.70 & 1398.07 & 0.38 & 1397.85 & 0.16 & 1398.570 & 0.88 & 1397.727 & 0.040 & -0.003 & 1397.684 & 1397.687 + @xmath268 & @xmath255 & 1452.27 & 1.17 & 1451.65 & 0.55 & 1451.35 & 0.25 & 1451.409 & 0.31 & 1451.095 & -0.006 & -0.008 & 1451.093 & 1451.101 + & & 1452.27 & 1.17 & 1451.65 & 0.55 & 1451.35 & 0.25 & 1451.501 & 0.40 & 1451.095 & -0.006 & -0.008 & 1451.093 & 1451.101 + @xmath268 & @xmath255 & 1454.09 & 1.26 & 1453.37 & 0.55 & 1453.08 & 0.25 & 1453.179 & 0.35 & 1452.821 & -0.006 & -0.008 & 1452.819 & 1452.827 + & & 1454.24 & 1.41 & 1453.37 & 0.55 & 1453.08 & 0.25 & 1453.231 & 0.40 & 1452.821 & -0.006 & -0.008 & 1452.819 & 1452.827 + @xmath268 & @xmath260 & 1454.85 & 1.45 & 1453.95 & 0.55 & 1453.65 & 0.25 & 1453.739 & 0.34 & 1453.397 & -0.006 & -0.008 & 1453.395 & 1453.403 + @xmath269 & @xmath255 & 1484.53 & 1.30 & 1483.33 & 0.10 & 1483.30 & 0.07 & 1483.455 & 0.23 & 1483.225 & -0.004 & -0.009 & 1483.220 & 1483.229 + & & 1484.54 & 1.31 & 1483.34 & 0.11 & 1483.30 & 0.07 & 1483.545 & 0.32 & 1483.226 & -0.003 & -0.008 & 1483.221 & 1483.229 + @xmath270 & @xmath255 & 1621.95 & 1.73 & 1620.86 & 0.64 & 1620.45 & 0.22 & 1620.540 & 0.32 & 1620.202 & -0.020 & -0.024 & 1620.198 & 1620.222 + & & 1621.96 & 1.74 & 1620.86 & 0.64 & 1620.45 & 0.22 & 1620.729 & 0.51 & 1620.202 & -0.020 & -0.024 & 1620.198 & 1620.222 + @xmath270 & @xmath260 & 1622.56 & 1.79 & 1621.40 & 0.64 & 1620.99 & 0.22 & 1621.281 & 0.51 & 1620.748 & -0.019 & -0.024 & 1620.743 & 1620.767 + @xmath271 & @xmath255 & 1751.15 & 1.62 & 1750.18 & 0.65 & 1749.76 & 0.23 & 1750.918 & 1.39 & 1749.591 & 0.061 & -0.005 & 1749.525 & 1749.53 + & & 1751.15 & 1.62 & 1750.43 & 0.90 & 1749.76 & 0.23 & 1751.229 & 1.70 & 1749.612 & 0.082 & -0.003 & 1749.527 & 1749.53 + @xmath272 & @xmath260 & 1758.64 & 2.22 & 1756.85 & 0.42 & 1756.60 & 0.18 & 1757.736 & 1.31 & 1756.475 & 0.049 & -0.007 & 1756.419 & 1756.426 + @xmath272 & @xmath264 & 1758.65 & 2.22 & 1756.85 & 0.42 & 1756.60 & 0.18 & 1757.756 & 1.33 & 1756.476 & 0.050 & -0.007 & 1756.419 & 1756.426 + @xmath272 & @xmath255 & 1759.26 & 2.13 & 1757.53 & 0.39 & 1757.30 & 0.17 & 1758.023 & 0.89 & 1757.148 & 0.015 & -0.010 & 1757.123 & 1757.133 + & & 1759.26 & 2.13 & 1757.53 & 0.40 & 1757.31 & 0.18 & 1758.263 & 1.13 & 1757.154 & 0.021 & -0.009 & 1757.124 & 1757.133 + @xmath272 & @xmath255 & 1762.20 & 2.43 & 1760.37 & 0.59 & 1759.98 & 0.21 & 1761.112 & 1.34 & 1759.848 & 0.076 & -0.004 & 1759.768 & 1759.772 + & & 1762.20 & 2.43 & 1760.56 & 0.78 & 1759.98 & 0.21 & 1761.381 & 1.61 & 1759.859 & 0.087 & -0.002 & 1759.770 & 1759.772 + @xmath273 & @xmath260 & 1787.24 & 2.04 & 1785.52 & 0.31 & 1785.37 & 0.16 & 1785.338 & 0.13 & 1785.128 & -0.079 & -0.087 & 1785.120 & 1785.207 + @xmath274 & @xmath255 & 1818.50 & 1.70 & 1817.82 & 1.02 & 1817.07 & 0.27 & 1817.207 & 0.41 & 1816.789 & -0.010 & -0.012 & 1816.787 & 1816.799 + & & 1818.50 & 1.70 & 1817.82 & 1.02 & 1817.07 & 0.27 & 1817.219 & 0.42 & 1816.789 & -0.010 & -0.012 & 1816.787 & 1816.799 + @xmath274 & @xmath260 & 1820.78 & 1.83 & 1819.98 & 1.03 & 1819.22 & 0.27 & 1819.635 & 0.68 & 1818.943 & -0.009 & -0.012 & 1818.940 & 1818.952 + @xmath274 & @xmath264 & 1820.78 & 1.83 & 1819.98 & 1.03 & 1819.22 & 0.27 & 1819.644 & 0.69 & 1818.944 & -0.008 & -0.011 & 1818.941 & 1818.952 + @xmath274 & @xmath255 & 1821.92 & 1.89 & 1821.07 & 1.04 & 1820.30 & 0.27 & 1820.482 & 0.45 & 1820.021 & -0.010 & -0.014 & 1820.017 & 1820.031 + & & 1821.92 & 1.89 & 1821.07 & 1.04 & 1820.30 & 0.27 & 1820.575 & 0.54 & 1820.021 & -0.010 & -0.014 & 1820.017 & 1820.031 + @xmath275 & @xmath264 & 1845.94 & 1.68 & 1844.45 & 0.19 & 1844.36 & 0.10 & 1845.160 & 0.90 & 1844.312 & 0.054 & -0.008 & 1844.250 & 1844.258 + @xmath275 & @xmath255 & 1846.03 & 1.70 & 1844.53 & 0.20 & 1844.43 & 0.10 & 1845.309 & 0.98 & 1844.387 & 0.057 & -0.008 & 1844.322 & 1844.33 + & & 1846.04 & 1.71 & 1844.53 & 0.20 & 1844.43 & 0.10 & 1845.610 & 1.28 & 1844.389 & 0.059 & -0.008 & 1844.322 & 1844.33 + @xmath275 & @xmath260 & 1846.41 & 1.72 & 1844.89 & 0.20 & 1844.79 & 0.10 & 1845.985 & 1.30 & 1844.745 & 0.055 & -0.009 & 1844.681 & 1844.69 + @xmath276 & @xmath255 & 1934.56 & 3.01 & 1931.84 & 0.29 & 1931.70 & 0.15 & 1931.849 & 0.30 & 1931.523 & -0.024 & -0.033 & 1931.514 & 1931.547 + & & 1934.58 & 3.03 & 1931.84 & 0.29 & 1931.70 & 0.16 & 1931.897 & 0.35 & 1931.523 & -0.024 & -0.032 & 1931.515 & 1931.547 + @xmath277 & @xmath260 & 1984.05 & 2.20 & 1983.51 & 1.66 & 1982.14 & 0.29 & 1982.577 & 0.73 & 1981.821 & -0.028 & -0.034 & 1981.815 & 1981.849 + @xmath277 & @xmath264 & 1984.05 & 2.20 & 1983.51 & 1.66 & 1982.14 & 0.29 & 1982.783 & 0.93 & 1981.821 & -0.029 & -0.035 & 1981.815 & 1981.85 + @xmath277 & @xmath255 & 1985.33 & 2.48 & 1984.51 & 1.65 & 1983.14 & 0.28 & 1983.483 & 0.63 & 1982.823 & -0.034 & -0.041 & 1982.816 & 1982.857 + & & 1985.33 & 2.48 & 1984.51 & 1.66 & 1983.14 & 0.29 & 1983.575 & 0.72 & 1982.823 & -0.034 & -0.041 & 1982.816 & 1982.857 + @xmath278 & @xmath260 & 2062.65 & 5.58 & 2058.66 & 1.59 & 2057.17 & 0.10 & 2058.000 & 0.93 & 2057.071 & 0.003 & -0.020 & 2057.048 & 2057.068 + @xmath278 & @xmath255 & 2069.96 & 4.67 & 2066.78 & 1.49 & 2065.38 & 0.10 & 2065.811 & 0.53 & 2065.272 & -0.014 & -0.019 & 2065.267 & 2065.286 + & & 2070.21 & 4.92 & 2066.80 & 1.51 & 2065.38 & 0.10 & 2066.355 & 1.07 & 2065.297 & 0.011 & -0.018 & 2065.268 & 2065.286 + @xmath279 & @xmath255 & 2118.65 & 7.27 & 2112.50 & 1.12 & 2111.75 & 0.37 & 2113.788 & 2.41 & 2111.511 & 0.131 & -0.003 & 2111.377 & 2111.38 + & & 2118.67 & 7.29 & 2112.50 & 1.12 & 2111.75 & 0.37 & 2113.928 & 2.55 & 2111.565 & 0.185 & -0.001 & 2111.379 & 2111.38 + @xmath279 & @xmath260 & 2120.01 & 7.71 & 2113.35 & 1.05 & 2112.67 & 0.38 & 2114.145 & 1.85 & 2112.467 & 0.170 & -0.003 & 2112.294 & 2112.297 + @xmath280 & @xmath255 & 2120.94 & 1.61 & 2119.96 & 0.63 & 2119.66 & 0.33 & 2121.397 & 2.07 & 2119.396 & 0.069 & -0.010 & 2119.317 & 2119.327 + & & 2121.14 & 1.81 & 2119.96 & 0.63 & 2119.66 & 0.33 & 2122.058 & 2.73 & 2119.422 & 0.095 & -0.010 & 2119.317 & 2119.327 + @xmath280 & @xmath255 & 2121.54 & 1.00 & 2121.10 & 0.56 & 2120.84 & 0.30 & 2122.132 & 1.59 & 2120.583 & 0.042 & -0.012 & 2120.529 & 2120.541 + & & 2122.07 & 1.53 & 2121.10 & 0.56 & 2120.85 & 0.31 & 2122.481 & 1.94 & 2120.586 & 0.045 & -0.011 & 2120.530 & 2120.541 + @xmath280 & @xmath264 & 2122.07 & 1.16 & 2121.46 & 0.55 & 2121.21 & 0.30 & 2123.442 & 2.53 & 2120.943 & 0.033 & -0.013 & 2120.897 & 2120.91 + @xmath280 & @xmath255 & 2123.23 & 0.40 & 2124.17 & 1.34 & 2123.16 & 0.32 & 2124.548 & 1.71 & 2122.845 & 0.011 & 0.001 & 2122.835 & 2122.834 + & & 2123.32 & 0.49 & 2124.18 & 1.34 & 2123.16 & 0.33 & 2125.200 & 2.37 & 2122.997 & 0.163 & 0.004 & 2122.838 & 2122.834 + @xmath280 & @xmath260 & 2123.59 & 0.29 & 2124.72 & 1.42 & 2123.63 & 0.33 & 2126.877 & 3.58 & 2123.419 & 0.118 & -0.001 & 2123.300 & 2123.301 + @xmath281 & @xmath255 & 2145.31 & 2.70 & 2144.46 & 1.85 & 2142.68 & 0.06 & 2142.585 & -0.03 & 2142.390 & -0.224 & -0.235 & 2142.379 & 2142.614 + & & 2145.31 & 2.70 & 2144.60 & 1.99 & 2142.68 & 0.06 & 2143.580 & 0.97 & 2142.390 & -0.224 & -0.235 & 2142.379 & 2142.614 + @xmath282 & @xmath255 & 2186.73 & 3.09 & 2185.33 & 1.70 & 2184.00 & 0.37 & 2184.164 & 0.53 & 2183.622 & -0.013 & -0.016 & 2183.619 & 2183.635 + & & 2186.73 & 3.09 & 2185.33 & 1.70 & 2184.00 & 0.37 & 2184.254 & 0.62 & 2183.622 & -0.013 & -0.016 & 2183.619 & 2183.635 + @xmath282 & @xmath255 & 2189.99 & 3.86 & 2187.88 & 1.74 & 2186.50 & 0.36 & 2187.113 & 0.98 & 2186.124 & -0.014 & -0.019 & 2186.119 & 2186.138 + & & 2190.00 & 3.86 & 2187.88 & 1.74 & 2186.50 & 0.36 & 2187.230 & 1.09 & 2186.124 & -0.014 & -0.019 & 2186.119 & 2186.138 + @xmath282 & @xmath255 & 2191.88 & 4.24 & 2189.41 & 1.77 & 2188.00 & 0.36 & 2188.710 & 1.07 & 2187.626 & -0.016 & -0.021 & 2187.621 & 2187.642 + & & 2192.02 & 4.38 & 2189.42 & 1.78 & 2188.00 & 0.36 & 2188.820 & 1.18 & 2187.626 & -0.016 & -0.021 & 2187.621 & 2187.642 + @xmath282 & @xmath260 & 2192.61 & 4.46 & 2189.93 & 1.78 & 2188.50 & 0.36 & 2189.111 & 0.97 & 2188.127 & -0.017 & -0.022 & 2188.122 & 2188.144 + @xmath283 & @xmath260 & 2208.60 & 1.98 & 2207.08 & 0.45 & 2206.79 & 0.17 & 2208.281 & 1.65 & 2206.740 & 0.114 & -0.011 & 2206.615 & 2206.626 + @xmath283 & @xmath264 & 2208.61 & 1.98 & 2207.09 & 0.45 & 2206.80 & 0.17 & 2208.938 & 2.31 & 2206.772 & 0.139 & -0.009 & 2206.624 & 2206.633 + @xmath283 & @xmath255 & 2208.73 & 1.97 & 2207.23 & 0.46 & 2206.95 & 0.18 & 2208.948 & 2.18 & 2206.810 & 0.044 & 0.001 & 2206.767 & 2206.766 + & & 2208.73 & 1.97 & 2207.23 & 0.46 & 2206.95 & 0.18 & 2209.447 & 2.68 & 2206.876 & 0.110 & 0.002 & 2206.768 & 2206.766 + @xmath283 & @xmath255 & 2209.57 & 2.01 & 2208.01 & 0.45 & 2207.72 & 0.16 & 2209.486 & 1.93 & 2207.604 & 0.045 & -0.010 & 2207.549 & 2207.559 + & & 2209.59 & 2.03 & 2208.02 & 0.46 & 2207.72 & 0.16 & 2209.739 & 2.18 & 2207.615 & 0.056 & -0.009 & 2207.550 & 2207.559 + bill poirier and tucker carrington jr . accelerating the calculation of energy levels and wave functions using an efficient preconditioner with the inexact spectral transform method . , 114(21):92549264 , 2001 . gustavo avila and tucker carrington jr . using a pruned basis , a non - product quadrature grid , and the exact watson normal - coordinate kinetic energy operator to solve the vibrational schrdinger equation for c2h4 . 135(6):064101 , 2011 . richard dawes and tucker carrington jr . how to choose one - dimensional basis functions so that a very efficient multidimensional basis may be extracted from a direct product of the one - dimensional functions : energy levels of coupled systems with as many as 16 coordinates . , 122(13):134101 , 2005 .

models of universal extra dimensions ( ued ) at the tev scale lead to the presence of kaluza klein ( kk ) excitations of the ordinary fermions and bosons of the standard model that may be observed at hadron and lepton colliders . a conserved discrete symmetry , kk - parity , ensures the stability of the lightest kk particle ( lkp ) , which , if neutral , becomes a good dark matter particle . it has been recently shown that for a certain range of masses of the lkp a relic density consistent with the experimentally observed one may be obtained . these works , however , ignore the impact of kk graviton production at early times . whether the is the lkp or not , the tower thus produced can decay to the lkp , and depending on the reheating temperature , may lead to a modification of the relic density . in this article , we show that this effect may lead to a relevant modification of the range of kk masses consistent with the observed relic density . additionally , if evidence for ued is observed experimentally , we find a stringent upper limit on the reheating temperature depending on the mass of the lkp observed .

hep-ph0608140

theoretical models constructed to understand physics beyond the standard model ( sm ) ( most notably string theory ) frequently imply the existence of extra dimensions . it has been a particular challenge in phenomenology to understand how these extra dimensions would be realized and manifested in our observable @xmath2 dimensional world . the number , shape and size of these dimensions , as well as the particles allowed to propagate in them give rise to several different models , all having different phenomenological implications . we will be considering universal extra dimensions ( ued ) where the sm fields propagate in all the extra dimensions . for toroidal compactification , this would imply a tower of kaluza - klein ( kk ) particles for every sm particle , each carrying kk number . momentum conservation in the extra dimensions implies kk number conservation . however , the requirement of obtaining the proper sm chiral modes at low energies leads to constraints on the possible compactification geometries . an example for @xmath3 is the orbifold @xmath4 where the @xmath5 projects out half of the zero modes , leaving only sm fields . additionally , this breaks translation invariance along the extra dimensions , so kk number is no longer conserved . a residual symmetry , kk - parity , is still present and it is sufficient to ensure the stability of the lightest kk particle ( lkp ) . it also ensures that kk particles are always produced in pairs , allowing for good agreement between theory and experiment for small values of the compactification scale , of the order of a few hundred gev @xcite . the stability of the lkp allows for an interesting candidate for dark matter ( dm ) @xcite . the lkp is expected to be weakly interacting and electrically neutral if it is to be considered a candidate for dark matter . the usual candidate is @xmath6 , the kk partner of the hypercharge gauge boson . for @xmath7 tev , @xmath6 gives excellent agreement with the observed relic density . there have been many further analyses of the relic density in ued , including a more proper treatment of coannihilation effects , and the impact of the inclusion of second kk level resonances @xcite . however , most of these studies ignore the gravitational sector . we shall mostly work within the context of one extra dimension . the gravitons couple extremely weakly and therefore are generally considered to be unimportant for collider studies . cosmologically , however , they can have a significant effect , and they can also be a candidate for the lkp @xcite . we are interested in investigating the inclusion of the graviton tower in the scenario with @xmath6 as the lkp . the gravitons do not evolve thermally , and have very long lifetimes , so if they are present , we expect them to decay to the lkp sometime after big bang nucleosynthesis ( bbn ) . the preservation of the light element abundances sets a bound on the amount of energy that can be released in a decay @xcite . secondly , since the gravitons decay ultimately into the lkp , we expect the relic density to increase . therefore , the mass of the lkp consistent with the observed relic density is lowered . additionally , if any of the gravitons decay after matter domination , we have to consider the effects of non - thermalized photons released in the decays on the spectrum of the diffuse photon flux ( dpf ) @xcite . this article is organized as follows . in section 2 , we review the standard calculation for relic density . section 3 reiterates the calculation for the density of gravitons presented in ref . @xcite , emphasizing the points relevant for our analysis . in section 4 , we analyze the lifetimes for the decay of the graviton tower and study its implication on the diffuse photon spectrum . section 5 goes over the constraints on the energy released in the decay of the kk gravitons to the lkp . in section 6 , we combine the above results with the standard relic density calculation . our goal is to analyze the effect of the gravitons on the predicted mass of the lkp consistent with the known dark matter density . we will show that , in fact , almost any value of @xmath8 lower than the one obtained in the absence of gravitons would be allowed , provided the reheating temperature , @xmath9 , is large enough . for the calculation of the relic density of @xmath6 , we ignore complicating factors such as coannihilation and second kk level resonance effects @xcite . since these effects can be parameterized in the effective cross - section , we present the effect the graviton tower will have on the mass of the lkp without the gravitons , @xmath10 , and note that as @xmath11 increases , the contribution due to the graviton tower becomes large , and should be included in any precise calculation of @xmath8 . we also find that all the gravitons , except for @xmath0 , decay right after bbn . at these early times , the electromagnetic bbn constraints are very weak . however it has been shown that at small lifetimes , the hadronic constraints are very stringent @xcite . therefore , including constraints from both hadronic and electromagnetic decays , requiring consistency with bbn light element predictions , we find a stringent limit on the mass difference between @xmath12 and the lkp mass , @xmath13 . since the @xmath0 is long - lived , we also derive another constraint on the mass difference such that the observed spectrum of the dpf is not destroyed . comparing the two constraints , we find that there exists a region of parameter space where both constraints are satisfied and which gives mass differences of the same order of magnitude as that obtained by radiative loop corrections to @xmath14 @xcite . additionally , we observe that if experimental evidence for ued is found , and the relic density induced by standard interactions of the kk modes is found to be lower than the one observed experimentally , the reheating temperature may be determined by assuming that the graviton kk modes provide the contribution necessary to achieve consistency between theory and experiment . alternatively , including the possibility of other unobserved exotic particles contributing to the relic density , an upper bound on the reheating temperature is obtained . let us start by recalling the computation of the density of thermal relics . for any particle @xmath15 , the evolution of the number density @xmath16 is governed by the boltzmann equation : @xmath17 where @xmath18 is the thermally averaged cross - section multiplied by the relative velocity , and eq . ( [ eq : neq ] ) is valid in the non - relativistic approximation , with @xmath19 and @xmath20 being the mass and number of degrees of freedom associated with the particle @xmath15 . the temperature at which the particle decouples from the thermal bath is denoted by @xmath21 ( _ freeze - out temperature _ ) and roughly corresponds to when @xmath22 is of the same order as @xmath23 . changing variables from @xmath24 to @xmath25 ( we will drop the subscript @xmath15 from now on ) , where the entropy density is given by @xmath26 , and using the fact that @xmath27 remains constant , we obtain , @xmath28 introducing the variable @xmath29 , in the radiation dominated era : @xmath30 then , eq . ( [ eq : y ] ) may be rewritten as @xmath31 we are interested in obtaining the relic density , or the equivalent @xmath32 , at late times . the solution of eq . ( [ eq : yx ] ) enables the determination of the relic density and also the value of @xmath33 as a function of @xmath34 . the ratio of the mass to the freeze - out temperature , @xmath35 is approximately given by @xmath36 and can be solved iteratively , with @xmath37 and @xmath38 tev . as calculated in @xcite , for @xmath6 as the lkp , without including co - annihilation , the non - relativistic approximation is : @xmath39 where @xmath40 and @xmath41 denote a summation over the fourth power of the hypercharges of fermions and bosons produced in the annihilation of @xmath6 respectively . moreover , ignoring the splitting between the different kk states of a given tower , one obtains , approximately @xmath42.\ ] ] in terms of this , @xmath32 is found to be : @xmath43 where we have assumed masses of @xmath44 tev , and used @xmath45 for @xmath46 gev and @xmath47 . for more than one extra dimension this analysis is more complicated but a rough estimate may be obtained by assuming independent towers of kk modes for each extra dimension , @xmath48 . the current relic energy density then is simply given by @xmath49 . we will however restrict ourselves to one extra dimension in most of our numerical work . as emphasized in the introduction , the above derivation for the density of @xmath6 ignores the contribution of the kk modes of the gravitons , which may become relevant for sufficiently large values of the reheating temperature @xmath9 . after reheating , except for gravitons , all other particles are initially in equilibrium and follow the pattern described in the previous section . after dilution of their density due to inflation , the gravitons are produced mainly in collisions involving gauge bosons , either in the initial or the final state , with cross - sections proportional to @xmath50 , where @xmath51 is the relevant gauge coupling . since their density is so small , they are never in equilibrium and further we can ignore graviton annihilation processes to first order . because the decay times for gravitons are of the order of @xmath52 s , the gravitons produced in the early universe are still present when the lkp freezes out . as the universe cools down further , the gravitons start to decay . including higher order mass corrections @xcite , we will have decays of the form : @xmath53 where @xmath54 denotes sm particles . here we are considering only kk number preserving decays since the kk number violating decays would only take place at the orbifold fixed points and so would only have a sizeable contribution in the case when the kk number preserving decays are suppressed due to phase space considerations . hence , after decaying , the kk gravitons contribute to the lkp relic density and a proper computation of the dark matter density demands the addition of this effect to the relic density calculated in the previous section . in order to obtain a quantitative estimate of this effect , we follow the derivation for the number density of gravitons presented in ref . the number density of the kk gravitons at each level @xmath24 is again determined by the boltzmann equation : @xmath55 where @xmath56 is the collision operator and can be parameterized as follows : @xmath57 ^ 2,\ ] ] where @xmath58 here @xmath59 is the strong coupling constant and @xmath60 can be understood as the fraction of all possible collisions which will interact strongly to produce gravitons . the exact calculation of the total production cross section is quite complicated , but can be estimated by the method presented in appendix [ appendixa ] . the authors of ref . @xcite used an analogy with the calculation of gravitino abundances @xcite , and estimated @xmath61 . however the sample calculation of the cross - section for the production of relativistic gravitons presented in appendix [ appendixa ] leads us instead to expect @xmath62 . for completeness we will consider a range of values , @xmath63 , in our numerical work . for ued theories , @xmath64 , where @xmath65 is the effective number of degrees of freedom per kk level . @xmath66 can be approximated by counting all modes with masses below @xmath67 : @xmath68^d\ ] ] where @xmath69 is the @xmath48 dimensional volume of a unit sphere , and the factor @xmath70 in eq . ( [ eq : d_d ] ) accounts for the restriction to non - negative @xmath24 . assuming entropy conservation , @xmath71 : @xmath72 as in the previous section , making the substitution @xmath25 and with the help of eq . ( [ eq : ent_cons ] ) the boltzmann equation becomes : @xmath73 ^ 2.\ ] ] @xmath74 changes until @xmath1 production stops at temperatures @xmath75 and then remains constant until gravitons begin to decay . after bbn and before gravitons decay : @xmath76 & \mbox{for } n<\displaystyle\frac{t_r}{m_{\rm kk}};\nonumber\\ \\ & = & 0 & \mbox{for } n>\displaystyle\frac{t_r}{m_{\rm kk}}.\nonumber\\ \\ \end{array}\ ] ] to find the @xmath77 contribution to the lkp , assuming only kk number preserving decays , @xmath78^{2+\frac{3d}{2}}\nonumber\\ & = & \alpha(d)\;c\;\frac{m_{\rm kk}}{\mbox{tev}}\;\left[\frac{t_r}{m_{\rm kk}}\right]^{2+\frac{3d}{2}}\\ a_d&=&\frac{2\pi^{d/2}}{\gamma(d/2)}=2,2\pi , ... \;\;\ ; \mbox{for d}= 1,2, ... \ ; , \end{aligned}\ ] ] where @xmath79 and @xmath80 . the decay widths for the decay of the gravitons are calculated in appendix [ appendixb ] . as shown there , the decay of @xmath1 for @xmath81 is primarily into gauge bosons of kk number @xmath82 ( @xmath83 for @xmath24 odd ) , assuming that all fermions are heavier . the lifetime for these is suppressed by powers of @xmath24 , and is given by : @xmath84^{5/2}\left[\frac{\mbox{tev}}{\delta_n}\right]^{1/2}\nonumber \\ & \sim & 5.56\times10 ^ 6 \mbox { s}\frac{1}{n^3}\left[\frac{\mbox{tev}}{m_{\rm kk}}\right]^3\label{eq : gn_life}\end{aligned}\ ] ] where @xmath85 and in the last line we have approximated @xmath86 with @xmath87 , which is of the order of the mass corrections induced at one loop @xcite , and , as we will show below , is also of the order of the mass differences required to satisfy the phenomenological constraints coming from bbn and diffuse gamma ray constraints . the only long - lived graviton is @xmath0 , with a lifetime given by : @xmath88 ^ 3,\nonumber\\ & \sim & 2.33\times 10^{13}\mbox { s}\left[\frac{\mbox{tev}}{m_{\rm kk}}\right]^3,\label{eq : g1_life}\end{aligned}\ ] ] where we have again assumed in the last line that @xmath89 . the different dependence on the mass difference in eq . ( [ eq : gn_life ] ) and ( [ eq : g1_life ] ) is due to the fact that one of the decay products for @xmath0 is massless . [ 0.8 ] the diffuse photon flux is only sensitive to photons released after matter domination @xcite since the late produced photons do nt have time to thermalize . therefore we just need to ensure that the energy released in the decay of the @xmath0 does not destroy the spectrum . the diffuse photon flux is given by @xcite : @xmath90\left[\frac{e_\gamma}{\epsilon_\gamma}\right]^{1/2}e^{-\left[\left(\frac{e_\gamma}{\epsilon_\gamma}\right)^{3/2}\frac{t_0}{\tau_{g^1}}\right]}\theta(\epsilon_\gamma - e_\gamma),\ ] ] where @xmath91 is initial energy released in the decay , @xmath92 is energy of photons observed now , @xmath93 and @xmath94 are the current time and volume of the universe , and @xmath95 is the total number of @xmath0s initially . using @xmath96 calculated in the previous section : @xmath97\nonumber\\ & & \left[\frac{\delta_1}{\mbox{mev}}\right]^{3/2}\left[\frac{m_{\rm kk}}{\mbox{tev}}\right]\left[\frac{e_\gamma}{\mbox{mev}}\right]^{1/2}e^{-\left[\left(\frac{e_\gamma\delta_1}{\mbox{\tiny mev}^2}\right)^{3/2}1.85\times10^{-5}\right]}.\label{eq : dphi / de2}\end{aligned}\ ] ] this has maximal value at : @xmath98.\ ] ] observe that , although @xmath91 is proportional to @xmath99 , the inverse dependence of @xmath100 on @xmath99 comes from the energy redshift induced by the expansion of the universe ( @xmath101 ) . this , in turn , leads to a maximum value for the flux , @xmath102\left[\frac{\delta_1}{\mbox{gev}}\right]\left[\frac{m_{\rm kk}}{\mbox{tev}}\right].\ ] ] if , for example , we assume a reheating temperature @xmath103 , we obtain @xmath104\left[\frac{m_{\rm kk}}{\mbox{tev}}\right].\ ] ] this is two orders of magnitude less than was obtained in ref . @xcite for a similar mass range , and is marginally consistent with the observed diffuse photon flux @xcite . the source of the difference is not computational , but based on the different assumptions made in both works . in ref . @xcite , the nlkp was assumed to be @xmath6 and had a density comparable to the dark matter relic density , while in our framework we are concerned with the decay of the @xmath0s , which have a density much smaller than that associated with the dark matter relic density . in order to understand the constraints imposed by the photon flux on the @xmath0 decay , we can parameterize the experimentally observed flux in terms of @xmath105 as follows , @xmath106 ^ 3 . \label{eq : obsflux}\ ] ] this is plotted in fig . [ fig : dpf ] . for comparison we have also plotted the maximum flux according to eq . ( [ eq : maxflux ] ) for a range of values of @xmath60 and @xmath13 , assuming a @xmath9 such that @xmath107 . the mass difference allowed by the diffuse photon flux can then be quantified by requiring that the maximum differential flux calculated from eq . ( [ eq : maxflux ] ) is less than the observed flux eq . ( [ eq : obsflux ] ) . it should also be noted that here we assumed that all the photons produced in the decay of the gravitons did not thermalize and are available to distort the spectrum now . this is certainly an overestimate . the constraints on @xmath99 will be discussed in more detail in section 6.6 . the electromagnetic showers produced by the decay of @xmath108 can create and destroy light elements . as discussed in @xcite , this sets constraints on the lifetime @xmath109 of the unstable @xmath1 as well as on @xmath110 , the energy released per background photon in the decay . the dependence on @xmath109 can be understood by looking at the characteristic energy scales in the initially produced photon spectrum . the primary photon created in the decay interacts with the background and creates an em cascade . the fastest interactions are pair production and inverse compton scattering . these processes rapidly redistribute the energy and the non - thermal photons reach a quasi - static equilibrium ( qse ) . the zeroth order qse photon spectrum depends inversely on the temperature of the background plasma . if we make the _ uniform decay _ approximation , i.e. all particles decay at @xmath111 , this corresponds to a cut - off energy of @xmath112 @xcite . therefore higher energy photo - erosion processes occur for longer lifetime values . we can understand the @xmath110 dependance of photodestruction and secondary production in a similar way . in the limit of small @xmath110 , the decaying particle does not influence the light element abundances . beyond this trivial case , we can again use the uniform decay approximation to gain some insights . as long as a reaction can take place ( @xmath113 , the threshold energy for a reaction , @xmath114 ) , a typical shower photon has energy : @xmath115 therefore , the number of such photons per decay is @xmath116 , where @xmath117 is the energy released in each decay . thus , the non - thermal photon density is : @xmath118 these photons then further thermalize as well as cause photodestruction of particles to yield other species of particles . therefore the dependence of the fractional change in the abundance of a particle is proportional to @xmath110 and inversely proportional to @xmath119 and therefore to @xmath109 . detailed numerical work gives a complicated but weak constraint on @xmath110 for getting the correct light element abundances for small lifetimes . however after about @xmath120 s , the dependance on @xmath110 steadies out to @xmath121 tev @xcite . comparing eq . ( [ eq : gn_life ] ) and ( [ eq : g1_life ] ) , we see that the only long - lived graviton is the @xmath0 with @xmath122 ^ 3 $ ] , much larger than @xmath123 s. hence , electromagnetic energy release bounds only constrain the decay of the @xmath0 to the lkp . similarly , hadrons produced in the decays can scatter off photons , electrons and , more importantly , background nuclei . scattering off photons and electrons just causes them to lose energy , but the nuclei become more energetic and cause `` hadronic showers '' . additionally if inelastic scattering occurs , the background nuclei get dissociated and the light element abundances are changed . due to the multitude of possible interactions , these are a lot more complicated to analyze than the purely electromagnetic case . previously it had always been assumed that since the branching ratio into hadrons is small , the effect would also be negligible . however detailed numerical work @xcite shows that in fact at early times , the constraints due to hadronic processes are much stronger then those due to electromagnetic processes . as can be seen from eq . ( [ eq : gn_life ] ) the lifetime of the @xmath1s are suppressed by powers of @xmath24 : @xmath124 ^ 3 . $ ] therefore all the @xmath1s except for @xmath0 decay right after bbn . the bounds at these relatively early decay times depend strongly on the branching ratio of the decay of the gravitons into hadrons . if all the visible energy released is in the form of hadrons , the bounds are quite severe : @xmath125 tev . if most of the decay is into leptons , then the bounds on @xmath126 may be significantly weaker . a preferential decay into leptons could happen if , for any given tower , the kk leptons are very close in mass to the kk hypercharge gauge bosons , while the kk quarks are heavier ( see appendix [ appendixb ] for calculation of the graviton decay widths into fermions ) . this would be the result one would obtain if including only the radiative corrections to the kk particle masses computed in ref . @xcite . since this is very model dependent , we will again use the most constrained case of @xmath127 tev . hence , both for early and late time decays we will use a conservative bound for the energy released per background photon , @xmath128 tev , in our numerical work . observe that while strong violations of this bound will certainly induce strong effects on the light element abundances , and therefore are ruled out , small violations of this bound are still possible due to the fact that we used very conservative limits on the energy released . one loop effects introduce corrections of two forms , constant ( independent of @xmath24 ) and proportional to @xmath24 ( bulk and boundary correction terms)@xcite . therefore , the mass of a particle at kk level @xmath24 can be written as : @xmath129 where the corrections @xmath130 and @xmath131 for each particle type is the same at each level . the graviton kk modes , of course , do not receive corrections of this kind at the one loop level , and we shall assume , for simplicity , that their masses remain unperturbed : @xmath132 . to analyze the effects of the decays , we are going to treat the two cases ( @xmath130 , @xmath133 ) separately . note that this is only for ease of understanding . in fact , the true effect would just be the sum of the two terms . in the decay @xmath134 assuming that @xmath54 is massless to first order , @xmath135 , and small momenta , the energy released in each decay is given by : @xmath136 for this energy release to be consistent with the light element abundances @xcite we need to have : @xmath137 here @xmath138 as given in eq . ( [ eq : ygn ] ) , and @xmath139 is the electromagnetic / hadronic branching ratio . this fraction depends on whether the lkp is a lepton or a gauge boson and the mass splittings at each kk level @xcite . throughout this article we choose @xmath140 as the most stringent constraint possible . in this section , we will discuss the determination of the lightest kk particle mass consistent with the observed dark matter relic density . since the energy density of lkp is determined , in part , by the primordial graviton kk density , the lightest lkp mass @xmath8 will depend on the reheating temperature and the parameter @xmath60 governing graviton production at early times . [ 0.8 ] including the decay of the gravitons in the density for @xmath6 , we have : [ 0.8 ] @xmath141 where @xmath142 , and @xmath32 is the abundance of the lkp without the inclusion of the gravitons , eq . ( [ eq : y_infty ] ) , and @xmath77 is the abundance of the gravitons , eq . ( [ eq : yg ] ) . we have not included either coannihilation or second kk resonance effects in the calculation of @xmath32 . requiring that @xmath143 , we get : @xmath144^{\frac{2}{4 + 3d}}\ ] ] in principle , the above expression shows that the relic density constraint can be satisfied for any value of the lightest kk mode mass , @xmath13 , provided the value of @xmath9 is adjusted according to eq . ( [ eq : tr_const ] ) . in practice , [ fig : trd5 ] shows that in order to get a consistent dark matter density for low values of @xmath13 , one would need large values of the ratio of the reheating temperature to @xmath13 , but as is discussed in the next section , these are disfavored due to strong coupling constraints . similar results are presented in fig . [ fig : trd6 ] for the case of @xmath145 . the mass of the lkp that would be obtained without the graviton tower contribution , @xmath10 , is shown in the figure as the point where the bound on @xmath9 becomes very strong , @xmath146 tev for @xmath3 and @xmath147 tev for @xmath145 , since in this case the reheating temperature must become smaller than the lightest kk mass so that the relic density is not effected in any significant way by the gravitons . it is important to stress that for @xmath3 , values of the kk masses smaller than about 500 gev , are also restricted in the minimal ued case by precision electroweak constraints @xcite . somewhat lower masses may be obtained for large values of the higgs mass . however even then , within this minimal framework , values of @xmath13 smaller than 350 gev are strongly disfavored . from fig . [ fig : trd5 ] we see that for @xmath3 , @xmath148 and values of the reheating temperature @xmath149 ( consistent with the strong coupling bounds discussed in the next section ) , a consistent relic density may be obtained for any value of @xmath8 larger than 580 gev . this range of values of the lkp mass are also consistent with precision electroweak constraints . even for @xmath150 and for the same range of values for the reheating temperature there could be modifications of @xmath13 up to ten percent of the value obtained without the inclusion of gravitons . larger ( smaller ) modifications are possible for larger ( smaller ) values of @xmath9 . as shown in fig . [ fig : trd6 ] , for @xmath145 one could obtain larger effects for smaller values of the ratio of @xmath151 . however , as will be shown in the next section , the bounds obtained from the requirement of perturbative consistency of the theory become much more stringent in this case . therefore , large departures from the @xmath13 values obtained in the absence of gravitons seem to be disfavored for @xmath145 within this minimal framework . large values of the reheating temperature compared to @xmath13 immediately raise the question of the ultraviolet cut - off for our effective 4d theory . in ref . @xcite , the authors assumed the limit on the kk masses to be the order of @xmath152 , based on the quantum corrections to the strong gauge coupling . more stringent bounds have been evaluated in @xcite . here we show that , if one computes the running of the zero mode hypercharge gauge coupling , and require it to remain weak in the ultraviolet regime , one obtains a bound similar to the one obtained in ref . @xcite . let us then analyze the running of the zero mode gauge couplings for one extra dimension : @xmath153 - \ln[n_{kk}!])\label{eq : kkcoupling}\\ & & \nonumber\\ \mbox{where } & & \nonumber\\ \tilde{b}_j&=&\left(\frac{81}{10 } , \frac{(-44 + 48 + 1 + 2d)}{6 } , \frac{(-22 + 16+d)}{2}\right ) , \mbox { each kk level}\nonumber\\ b_j&=&\left(\frac{41}{10},-\frac{19}{6},-7\right)\mbox { sm}\label{eq : beta}\end{aligned}\ ] ] for @xmath3 , the most dangerous one for the kk scenario was the u(1 ) case and we found that @xmath154 develops a landau pole at scales @xmath155 @xcite . our conclusion is then that above energy scales of about @xmath156 our effective theory breaks down . similarly , for @xmath145 , a landau pole would develop at scales larger than about @xmath157 . however , these bounds may be avoided by assuming that this theory is just the low energy manifestation of a theory with more degrees of freedom and based on a higher , asymptotically free , gauge group . this would not only prevent @xmath158 from developing a landau pole , but additionally it would increase the value of @xmath60 . we have conservatively kept values up to @xmath159 for @xmath3 and @xmath160 for @xmath145 . the impact of these bounds on the possible values of the lightest kk mass are displayed in figs . [ fig : trd5 ] and [ fig : trd6 ] . as we mentioned before , we see that if we restrict ourselves to the bounds implied by perturbative consistency , the mass of the lkp can be as low as @xmath161 tev for both @xmath3 and @xmath145 , but lower values , as low as those implied by consistency with precision electroweak measurements , would be only obtained by the possible relaxation of these bounds by new physics . on the other hand , if the most stringent bounds obtained in @xcite were imposed , very small modifications to the mass of the lkp would be obtained . [ 0.8 ] [ 0.8 ] [ 0.8 ] until now , we have only considered the effects induced by the annihilation of @xmath6s on the total relic density . however , several effects can change the density of the @xmath6s . for example , if we assume that the mass difference between the lkp and the nlkp is less than about 10% , the nlkp is still thermally accessible to the lkp , and the effective cross - section , including coannihilations must be used in eq . ( [ eq : boltz ] ) . also , possible resonant effects induced by the second kk - level modes must be included @xcite . note that even though including these effects changes the mass of the lkp consistent with the dark matter density without the inclusion of the gravitons , @xmath10 , and will certainly change any numerical results we obtain , the qualitative picture we have presented here will remain unchanged . the corrections to the cross - section can be quantified by considering @xmath162 and @xmath163 in eq . ( [ eq : y_infty ] ) to encode all the information about the processes that affect @xmath32 . due to the weak , logarithmic dependence of @xmath33 on @xmath13 and the interaction cross section , we can approximately parameterize @xmath32 as being proportional to @xmath13 , @xmath164 then @xmath165 , such that @xmath10 ( including coannihilation , second kk resonances etc . , but without the graviton tower ) is consistent with the experimentally observed dark matter density , @xmath166 , is given by : @xmath167 we can now replace @xmath32 using eq . ( [ eq : y_par ] ) and ( [ eq : y ] ) in eq . ( [ eq : yb1 ] ) . using this , we get the change in @xmath10 necessary to reproduce the observed dark matter density for a given reheating temperature @xmath9 : @xmath168^{\frac{4 + 3d}{2}}}\right)^{1/2}\ ] ] this is plotted in fig . [ fig : mwg20 ] , [ fig : mwg40 ] and [ fig : mwg100 ] for @xmath169 and @xmath170 and a range of @xmath10 predicted in refs . we see that even for @xmath171 with @xmath172 , the mass consistent with the observed relic density undergoes a relevant modification due to the inclusion of the gravitons . this effect becomes increasingly important with increasing @xmath10 . therefore , depending on the reheating temperature , any precise calculation of @xmath13 must include the contribution from the graviton tower to be accurate . it is interesting to observe that for @xmath173 and @xmath174 , the necessary values of @xmath13 are below the bounds imposed by precision electroweak constraints . conversely , this shows that for this value of @xmath60 and for values of the kk masses consistent with precision electroweak constraints , such large values of the reheating temperature will lead to an excess of dark matter density and are therefore disfavored . on the contrary , for values of @xmath175 , the obtained values of @xmath8 are in good agreement with the scale set by precision electroweak data this shows an interesting correlation between the dark matter density bounds and those coming from requiring the perturbative consistency of the theory up to scales of the order of @xmath9 . in the discussion above , we have ignored the bounds on the energy released in the graviton decays . as we have stated in section 4 , the amount of energy released will depend on the type of mass correction the graviton and @xmath6 receive ( see eq . ( [ eq : erel ] ) and eq . ( [ eq : xi_bound ] ) ) . for the case of @xmath133 , the energy released per background photon is given by [ 0.8 ] @xmath176 assuming , as stated before , that there are no corrections to the kk graviton masses @xmath177 , then @xmath178 . in addition we can identify @xmath13 with @xmath14 . the corrections to @xmath179 are then equivalent to corrections to @xmath180 for small @xmath130 . if we now assume that the reheating temperature is such that one reproduces the correct dark matter relic density , @xmath107 , as obtained in the previous section , then we get a constraint on the mass difference between @xmath0 and @xmath6 : @xmath181 this is plotted in fig . [ fig : delta1 ] for @xmath3 . this constraint is independent of @xmath60 . the mass difference between @xmath6 and @xmath0 is very tightly constrained to be less than a gev for most of the mass range considered . as @xmath13 approaches the value of the mass consistent with a proper dark matter relic density in the absence of gravitons , the bounds become very weak . this is due to the fact that in this case , as mentioned before , @xmath9 may be of the order of ( or smaller than ) @xmath13 , and hence very few gravitons would be produced , implying that the mass difference can be large without inducing any dramatic effects . for most of the parameter space , we have derived here a very stringent constraint on the mass difference . therefore , we need to consider whether such small mass differences would require too much fine tuning to be considered natural . as is discussed in section 6.7 , we find that in fact the mass differences obtained here are the order of magnitude of the one loop corrections induced to the @xmath6 mass . [ 0.8 ] here we depart from the previous assumption and assume that all graviton kk modes receive a constant , positive mass correction with respect to @xmath183 . the energy released for this kind of mass correction tends to be much smaller than in the case of @xmath184 analyzed above ( see eq . ( [ eq : erel1 ] ) ) , and is given by : @xmath185^{1+\frac{3d}{2}}\nonumber\\ & = & \beta(d)\;c\;\frac{m_{\rm kk}}{\mbox{tev}}\;\left[\frac{t_r}{m_{\rm kk}}\right]^{1+\frac{3d}{2}}\label{eq : yg'}.\end{aligned}\ ] ] for this case , since the dependance of @xmath186 on @xmath9 is different from that of @xmath187 , we get a more complicated expression for @xmath131 , depending explicitly on the number of extra dimensions and @xmath60 : @xmath188^{-\frac{2 + 3d}{4 + 3d}}\label{eq : delta'}\ ] ] this is plotted in fig . [ fig : delta2 ] . as anticipated , the bounds on @xmath131 are much weaker than on @xmath117 . however , even though the constraint is much weaker now than before , only small mass differences , smaller than about 10 gev , are allowed in most of the parameter space . since the @xmath0 decays late to the @xmath6 we have to consider its effect on the diffuse photon flux . as long as the @xmath0 decays after matter domination , it can have an effect on the diffuse photon flux . since the observed flux is smooth to a high degree , we can derive a constraint on the mass difference @xmath99 between @xmath0 and @xmath6 by requiring that eq . ( [ eq : maxflux ] ) be less than eq . ( [ eq : obsflux ] ) . using this inequality , we are led to : [ 0.8 ] @xmath190^{3/2}-1\right)\left[\frac{m_{\rm kk}}{\mbox{tev}}\right]\right)^{1/2}\;\mbox{gev},\label{eq : delta1}\ ] ] where @xmath9 is given by eq . ( [ eq : tr_const ] ) . this is plotted in fig . [ fig : deldpf ] . we see again that as @xmath13 approaches the value consistent with the dark matter density without gravitons , the constraint on @xmath99 becomes very weak . at these masses , even if the @xmath0 were decaying right now , their density is so small that the decays are basically invisible and do nt affect the photon spectrum . [ 0.8 ] [ 0.8 ] [ 0.8 ] comparing the constraints on the mass difference between @xmath6 and @xmath0 , plotted in fig . [ fig : delsuper ] , we see that there is a region of parameter space where both conditions are satisfied . the allowed region increases for larger values of @xmath60 , and for @xmath191 , extends up to values of order 600 gev , which is about the value of the bound placed by precision electroweak tests for a light higgs boson . the range of allowed values of @xmath99 shrinks for decreasing values of @xmath192 . for instance , in the case represented in fig . [ fig : delsuper ] for @xmath172 , values of @xmath193 tev may be obtained for a range of values of @xmath194 gev , while values as low as @xmath195 tev are only allowed for @xmath196 gev . interestingly enough the allowed values of @xmath13 agree with the ones consistent with the observed dark matter relic density for @xmath197 depicted in fig . [ fig : trd5 ] . the effects on the allowed values of @xmath99 as @xmath10 changes are illustrated in figs . [ fig : delsuper15 ] and [ fig : delsuper2 ] . as shown in these figures , the qualitative behavior for the different @xmath10 is the same as the one found in fig . [ fig : delsuper ] . we want to compare the bounds obtained on @xmath99 due to bbn and the diffuse photon flux found in the previous section , with the one - loop corrections induced in the @xmath6 mass , assuming that @xmath198 . radiative corrections induced by boundary terms at the orbifold fixed points affect the spectrum of the standard kk particles . since radiative corrections are divergent they may be regularized by counterterms localized at these boundaries . the possibility of including localized counterterms also implies that one could have started with different values of the localized boundary terms than the ones assumed in the minimal scenario , and this would lead to modifications of the spectrum different from the one discussed below @xcite . we include , however , a discussion of this minimal scenario in order to show that , interestingly enough , the magnitude of the one - loop induced corrections is indeed of the order of the one necessary to satisfy the bbn and diffuse gamma ray bounds . in order to compute the lkp mass , apart from the one - loop corrections , one must also take into account the @xmath199 correction to the gauge boson mass matrix due to the vev of the higgs field @xcite . in the @xmath200 , @xmath201 basis : @xmath202 since the mixing is very small compared to the mass difference between the @xmath200 and the @xmath201 induced at the one - loop level , the neutral gauge bosons approximately become pure @xmath200 and @xmath201 . therefore , the mass correction to @xmath6 is very well approximated by : @xmath203\left(\frac{1}{r}\right)^2+\frac{g_1 ^ 2v^2}{4 } , \;\;\ ; \normalsize{{\mbox{d=5}}}\ ] ] this correction to @xmath6 becomes positive for masses below about @xmath204 gev , and the absolute value is plotted in fig . [ fig:1loop ] . therefore , as noted in @xcite , below this mass , assuming only one - loop corrections , the graviton is the lkp . this case has been extensively studied and the constraints are derived in ref . the @xmath6s , with a density comparable to @xmath205 , would be decaying late to @xmath0 and based on an analysis similar to the one performed in section 4 , one concludes that , apart from the finely tuned case in which the mass difference approximately vanishes , small mass differences between the @xmath6 and the @xmath0 would lead to a large impact on the diffuse photon flux . therefore , unless some evidence is seen in the diffuse photon flux for new physics for values of @xmath206 mev ( see fig . [ fig : dpf ] ) , this minimal scenario would be ruled out for @xmath207 gev . [ 0.8 ] we see that in the allowed region demarcated in figs . [ fig : delsuper ] , [ fig : delsuper15 ] and [ fig : delsuper2 ] the values of @xmath99 are , as emphasized before , of the same order of magnitude as the range predicted by the one loop corrections shown in fig . [ fig:1loop ] . the precise quantitative constraints on the possible values of @xmath208 depend on @xmath10 . while for @xmath209 , 2 tev , only corrections of the order of ten percent would be allowed , for @xmath210 tev , the one - loop corrections are consistent with those necessary to satisfy both the bbn and diffuse photon flux constraints tev would be disfavored since a charged higgs would become the lkp . ] . in evaluating these constraints we have to stress again that we have used a very conservative estimate for the hadronic branching ratio of @xmath211 . the actual energy released into hadrons is approximately proportional to @xmath212 and would be somewhat lower since the kk quarks are much heavier then the kk right handed leptons . additionally there are large errors in the detection of the diffuse photon flux . a proper computation of the hadronic branching ratio of the decay of gravitons , as well as a more accurate spectrum for the diffuse photon flux , would be necessary in order to determine the compatibility of this minimal scenario with the energy release constraints . [ 0.8 ] let us finish this section by commenting on the possible experimental probes of this scenario . due to their small couplings to matter , provided @xmath0 is not the lkp , the kk gravitons will not be produced at laboratory experiments . therefore , their presence may only be probed by indirect effects , like their impact on cosmology as we have discussed in this article . in order to be able to evaluate these effects , one would have to find conclusive evidence of ued at , for instance , collider experiments and , in addition , measure the properties of the relevant first and second kk modes that contribute to the @xmath6 annihilation cross section . then , under the assumption that no other exotic particle contributes to the dark matter density one could estimate the relic density associated with the lkp for the specific value of @xmath13 measured . if the obtained relic density disagrees with the experimentally measured one , then one would get information about the possible kk graviton contribution . knowledge of the graviton contribution and the kk mass @xmath13 will , in turn , allow us to determine the reheating temperature . by analogy with eq . ( [ eq : tr_const ] ) , the reheating temperature for one extra dimension would be given by , @xmath213 where @xmath214 is the difference between the measured value of @xmath187 and the theoretically estimated one from the measured value of @xmath13 and the corresponding annihilation cross section . this is plotted for @xmath172 in fig . [ fig : trom ] for several different values of @xmath215 for reference . even including the possibility of exotic particles contributing to the dark matter density , eq . ( [ eq : trom ] ) gives an upper limit on the reheating temperature . in this article , we analyzed the effects of including the kk graviton tower on the determination of the relic density associated with the lightest kk particle in the scenario of universal extra dimensions . gravitons may be copiously produced in the early universe and their subsequent decay into the lkp may lead to a modification of the observed lkp relic density . graviton production is governed by a parameter @xmath216 and by the reheating temperature , which is bounded from above by the requirement of maintaining the perturbative consistency of the theory , @xmath217 ( @xmath218 ) for @xmath219 ( @xmath145 ) . in the case that universal extra dimensions are observed in laboratory experiments , we show that an upper limit on the reheating temperature can be deduced from the requirement that only the lkp contributes to the observed dark matter density . throughout this work we have assumed kk parity conservation , leading to the stability of the lkp particle . we found that including the graviton effectively lowers the lightest kk particle mass consistent with the observed dark matter relic density from values of about 0.92 tev to values as low as @xmath220 tev for @xmath172 , @xmath3 and @xmath221 . additionally , including effects which change the lkp density ( coannihilation , second kk - level mode resonant contributions , etc . ) , we show that the graviton tower has a large impact on the predicted mass of the lkp with increasing mass . it should be stressed here that these results are independent of which kk particle is the lkp ( or the nlkp ) . additionally , there are bounds on the mass difference of the lkp with the graviton kk modes induced by the requirement that the energy released in the graviton decay does not lead to a disturbance of the light element abundance or the diffuse photon flux . under the assumption that the graviton spectrum is @xmath222 , we have obtained a bound on the mass difference of @xmath0 and the lkp ( @xmath6 ) mass , that is consistent with the minimal one - loop corrections obtained in ref . @xcite for a large range of values of @xmath13 . + + + we would like to thank jonathan feng , csaba balazs and tim tait for useful discussions and comments . work at anl is supported in part by the us doe , div . of hep , contract w-31 - 109-eng-38 . this work was also supported in part by the u.s . department of energy through grant no . de - fg02 - 90er40560 . 99 i. antoniadis , phys . b * 246 * , 377 ( 1990 ) . t. appelquist , h. c. cheng and b. a. dobrescu , phys . d * 64 * , 035002 ( 2001 ) [ arxiv : hep - ph/0012100 ] . t. appelquist and h. u. yee , phys . d * 67 * , 055002 ( 2003 ) [ arxiv : hep - ph/0211023 ] . g. servant and t. m. p. tait , nucl . b * 650 * , 391 ( 2003 ) [ arxiv : hep - ph/0206071 ] . k. griest and d. seckel , phys . rev . d * 43 * , 3191 ( 1991 ) . m. kakizaki , s. matsumoto , y. sato and m. senami , phys . d * 71 * , 123522 ( 2005 ) [ arxiv : hep - ph/0502059 ] . m. kakizaki , s. matsumoto , y. sato and m. senami , nucl . b * 735 * , 84 ( 2006 ) [ arxiv : hep - ph/0508283 ] . k. kong and k. t. matchev , jhep * 0601 * , 038 ( 2006 ) [ arxiv : hep - ph/0509119 ] . f. burnell and g. d. kribs , phys . d * 73 * , 015001 ( 2006 ) [ arxiv : hep - ph/0509118 ] . m. kakizaki , s. matsumoto and m. senami , phys . rev . d * 74 * , 023504 ( 2006 ) [ arxiv : hep - ph/0605280 ] . e. w. kolb , g. servant and t. m. p. tait , jcap * 0307 * , 008 ( 2003 ) [ arxiv : hep - ph/0306159 ] . j. l. feng , a. rajaraman and f. takayama , phys . d * 68 * , 085018 ( 2003 ) [ arxiv : hep - ph/0307375 ] . j. r. ellis , k. a. olive and e. vangioni , phys . b * 619 * , 30 ( 2005 ) [ arxiv : astro - ph/0503023 ] . m. kawasaki , k. kohri and t. moroi , phys . d * 71 * , 083502 ( 2005 ) [ arxiv : astro - ph/0408426 ] . r. h. cyburt , j. r. ellis , b. d. fields and k. a. olive , phys . d * 67 * , 103521 ( 2003 ) [ arxiv : astro - ph/0211258 ] . p. sreekumar , f. w. stecker and s. c. kappadath , aip conf . * 510 * , 459 ( 2004 ) [ arxiv : astro - ph/9709258 ] . m. kawasaki , k. kohri and t. moroi , phys . b * 625 * , 7 ( 2005 ) [ arxiv : astro - ph/0402490 ] . k. jedamzik , phys . rev . d * 70 * , 063524 ( 2004 ) [ arxiv : astro - ph/0402344 ] . h. c. cheng , k. t. matchev and m. schmaltz , phys . d * 66 * , 036005 ( 2002 ) [ arxiv : hep - ph/0204342 ] . m. bolz , w. buchmuller and m. plumacher , phys . b * 443 * , 209 ( 1998 ) [ arxiv : hep - ph/9809381 ] . m. bolz , a. brandenburg and w. buchmuller , nucl . b * 606 * , 518 ( 2001 ) [ arxiv : hep - ph/0012052 ] . j. l. feng , a. rajaraman and f. takayama , phys . rev . lett . * 91 * , 011302 ( 2003 ) [ arxiv : hep - ph/0302215 ] . j. l. feng , a. rajaraman and f. takayama , phys . d * 68 * , 063504 ( 2003 ) [ arxiv : hep - ph/0306024 ] . t. flacke , d. hooper and j. march - russell , phys . d * 73 * , 095002 ( 2006 ) [ erratum - ibid . d * 74 * , 019902 ( 2006 ) ] [ arxiv : hep - ph/0509352 ] . i. gogoladze and c. macesanu , arxiv : hep - ph/0605207 . r. s. chivukula , d. a. dicus , h. j. he and s. nandi , phys . b * 562 * , 109 ( 2003 ) [ arxiv : hep - ph/0302263 ] . r. s. chivukula , d. a. dicus , h. j. he and s. nandi , arxiv : hep - ph/0402222 . m. masip , phys . d * 62 * , 105012 ( 2000 ) [ arxiv : hep - ph/0007048 ] . m. e. peskin and d. v. schroeder , `` an introduction to quantum field theory '' , addison wesley , 1995 . p. ramond , `` journeys beyond the standard model '' , perseus books , 1999 . d. r. t. jones , phys . rev . d * 25 * , 581 ( 1982 ) . m. carena , t. m. p. tait and c. e. m. wagner , acta phys . b * 33 * , 2355 ( 2002 ) [ arxiv : hep - ph/0207056 ] . c. macesanu , a. mitov and s. nandi , phys . d * 68 * , 084008 ( 2003 ) [ arxiv : hep - ph/0305029 ] . s. matsumoto , j. sato , m. senami and m. yamanaka , arxiv : hep - ph/0607331 . djmp / tensorialpage.html we will calculate explicitly the graviton production cross - section for @xmath223 . the feynman rules we will need are @xcite : @xmath224{\includegraphics[bb=2.9 cm 0 cm 20.1 cm 16.1 cm , clip = false]{amp1.eps } } & \scalebox{0.3}[0.35]{\includegraphics[bb=0.9 cm 2.9 cm 18.1 cm 16.1 cm , clip = false]{amp2.eps } } \\ \mathcal{m}^1 & \mathcal{m}^2 \end{array}$ ] @xmath224{\includegraphics[bb=2.9 cm 0 cm 20.1 cm 16.1 cm , clip = false]{amp3.eps } } & \scalebox{0.3}[0.3]{\includegraphics[bb=0.9 cm 0.9 cm 18.1 cm 16.1 cm , clip = false]{amp4.eps } } \\ & \\ \mathcal{m}^3 & \mathcal{m}^4 \end{array}$ ] @xmath225\label{eq:56}\\ & & \nonumber\\ & & \nonumber\\ g^l_\alpha(k_1)&\to & g^m_\beta(k_2)+g^n_{\mu\nu}:\nonumber\\ & & \nonumber\\ x^{\mu\nu\alpha\beta}&=&-\frac{i}{2m_4}\left[(-m_lm_m+k_1.k_2)\left(-\eta^{\mu\nu}\eta^{\alpha\beta}+\eta^{\mu\alpha}\eta^{\nu\beta}+\eta^{\mu\beta}\eta^{\nu\alpha}\right)+\eta^{\mu\nu}k_1^\beta k_2^\alpha\right.\nonumber\\ & & \nonumber\\ & & \left.-\left(\eta^{\mu\beta}k_1^\nu k_2^\alpha+\eta^{\mu\alpha}k_1^\beta k_2^\nu - \eta^{\alpha\beta}k_1^\mu k_2^\nu+(\nu\leftrightarrow\mu)\right)\right]\delta_{n,|l - m|}\nonumber\\ & & \\ & & \nonumber\\ & & \nonumber\\ q+\bar{q}&\to&g:\nonumber\\ & & \nonumber\\ z^\mu & = & ig\gamma^\mu t^a\label{eq:58}\\ & & \nonumber\\ & & \nonumber\\ q^l+\bar{q}^m&\to & g^{k\rho}+g^{n\mu\nu}:\nonumber\\ & & \nonumber\\ a^{\mu\nu\rho}&=&\frac{igt^a}{4m_4}\left(-2\eta^{\mu\nu}\gamma^\rho+\eta^{\mu\rho}\gamma^\nu+\eta^{\nu\rho}\gamma^\mu\right)\gamma^5\delta_{l+m , k+n}\label{eq:59}\\ & & \nonumber\end{aligned}\ ] ] four diagrams contribute to this process ( fig . [ fig:7 ] , [ fig:8 ] ) . the amplitudes are given by : @xmath226\gamma^5 u(k_1)\label{eq:60}\\ & & \nonumber\\ \mathcal{m}^2&:&\nonumber\\ & & -\frac{igt^a}{4m_4}\bar{\nu}(k_2)\gamma^\rho\frac{\not{k_1}-\not{p_2}+m_{l - n}}{2(k_1.p_2-m_lm_n)}\left(\eta^{\mu\nu}(\not{p_2}-m_n)+\gamma^\mu k_1^\nu+\gamma^\nu k_1^\mu\right)u(k_1)\epsilon^*_\rho e^*_{\mu\nu}\nonumber\\ & & \label{eq:61}\\ & & \nonumber\\ & & \nonumber\\ \mathcal{m}^3&:&\nonumber\\ & & \frac{igt^a}{4m_4}\bar{\nu}(k_2)\frac{\eta^{\mu\nu}(\not{k_1}-\not{p_1}+m_{l - k})-\gamma^\mu k_2^\nu-\gamma^\nu k_2^\mu}{2(k_1.p_1-m_lm_k)}\gamma^5\left(2k_1^\rho-(\not{p_1}+m_k)\gamma^\rho\right)u(k_1)\epsilon^*_\rho e^*_{\mu\nu}\nonumber\\ & & \label{eq:62}\\ & & \nonumber\\ \mathcal{m}^4&:&\nonumber\\ & & \frac{igt^a}{2m_4}\frac{\bar{\nu}(k_2)}{(k_1+k_2)^2-m_{l+m}^2}\left[(m_km_{k - l - m}+p_1.p_2)\left(\gamma^\rho\eta^{\mu\nu}-\gamma^\mu\eta^{\nu\rho}-\gamma^\nu\eta^{\mu\rho}\right)\right.\nonumber\\ & & \nonumber\\ & & -\left(\eta^{\mu\nu}p_2^\rho-\eta^{\mu\rho}p_1^\nu-\eta^{\nu\rho}p_1^\mu\right)\not{p_1}-2\gamma^\rho p_1^\mu p_1^\nu+(\gamma^\mu p_1^\nu + \gamma^\nu p_1^\mu)p_2^\rho{\huge{]}}u(k_1)\epsilon^*_\rho e^*_{\mu\nu}\nonumber\\ & & \label{eq:63}\end{aligned}\ ] ] in the center - of - mass frame , the cross - section is given by @xcite : @xmath227 here , @xmath228 and @xmath229 label the initial particle energies and velocities , and @xmath230 is the three - momentum of the graviton . since we are interested in the relativistic limit , we can use the more general form for all four particles with @xmath231 : @xmath232 averaging over initial spins and colors and summing over final polarizations , @xmath233 was calculated using the * tensorial 3.0 * package for _ mathematica _ @xcite . the cross - section is then given by : @xmath234 the fraction of strongly interacting degrees of freedom per kk level is computed by counting the different allowed interactions . four strong interactions can produce gravitons : @xmath235 there are @xmath236 degrees of freedom for the quarks and @xmath237 for the gluons . the fraction can then be calculated : @xmath238+\left(\frac{24}{197.5}\right)^2\sim 0.19.\label{eq:47}\ ] ] assuming that all the other graviton production processes have a cross - section similar to ( [ eq:46 ] ) , @xmath60 used to parameterize the graviton production cross - section is given by : @xmath239 the graviton polarization sum is given by @xcite : @xmath247^{5/2}\left[1-\frac{\left(m_l+m_m\right)^2}{m_n^2}\right]^{1/2}\nonumber\\ & & \times\left[\frac{13}{6}+\frac{7}{3}\frac{\left(m_l+m_m\right)^2}{m_n^2}+\frac{1}{2}\frac{\left(m_l+m_m\right)^4}{m_n^4}\right].\nonumber\\ & & \label{eq:68}\end{aligned}\ ] ] we see again that the decay into kk modes other than @xmath82 is suppressed . therefore for all @xmath81 , assuming now that the main decay is in fact into @xmath248 with an extra factor of @xmath249 for identical particles in the final state :

we consider magnetotransport in a disordered two - dimensional electron gas in the presence of a periodic modulation in one direction . existing quasiclassical and quantum approaches to this problem account for weiss oscillations in the resistivity tensor at moderate magnetic fields , as well as a strong modulation - induced modification of the shubnikov de haas oscillations at higher magnetic fields . they do not account , however , for the operation at even higher magnetic fields of the integer quantum hall effect , for which quantum interference processes are responsible . we introduce a field theory approach , based on a nonlinear model , which encompasses naturally both the quasiclassical and quantum mechanical approaches , as well as providing a consistent means of extending them to include quantum interference corrections . a perturbative renormalization - group analysis of the field theory shows how weak localization corrections to the conductivity tensor may be described by a modification of the usual one - parameter scaling , such as to accommodate the anisotropy of the bare conductivity tensor . we also show how the two - parameter scaling , conjectured as a model for the quantum hall effect in unmodulated systems , may be generalized similarly for the modulated system . within this model we illustrate the operation of the quantum hall effect in modulated systems for parameters that are realistic for current experiments .

cond-mat0210211

the problem of electron motion in a disordered conductor in a periodic potential and strong magnetic field displays a rich combination of phenomena of both classical and quantum origin . the complexity arises from the appearance of two independent types of periodicities , of the potential and of the cyclotron orbits , which may interplay in complicated ways . perhaps the most striking effect is known as weiss oscillations@xcite , whereby very large oscillations in the resistivity are induced by even a weak periodic potential at moderate magnetic fields . at different magnetic fields or temperatures , other types of contributions to the resistivity become significant . for example , at very small fields , a positive magnetoresistance results from a classical mechanism of channeling of orbits@xcite . at larger magnetic fields , the resistivity develops shubnikov de haas oscillations , which originate from the onset of landau quantization and may still be affected by the periodic potential@xcite . at even higher fields , the integer quantum hall effect@xcite ( iqhe ) becomes operative due to the contribution of quantum interference processes . much of the above phenomenology has been thoroughly analyzed theoretically , by a variety of techniques including quantum mechanical approaches@xcite involving diagrammatics@xcite or solution of a quantum boltzmann equation@xcite , and quasiclassical approaches@xcite . the experimental realization of these systems is also well advanced@xcite with precise confirmation of the theoretical predictions already possible . so far however a theory for the influence of quantum interference processes in a periodic potential and high magnetic field has been missing . such processes lead to weak localization corrections to the conductivity which become significant at low temperatures ( see e.g. refs.@xcite for reviews ) . at high magnetic fields , they are also responsible for the operation of the iqhe@xcite , whereby the hall conductivity becomes quantized at low temperatures . at the same time , the observation of the iqhe in systems with a weak potential modulation is well within current experimental capability@xcite . this paper aims to fill this gap by showing how the standard theory for weak localization and the iqhe may be generalized so as to incorporate the presence of a periodic potential . we employ a field - theory approach based on a nonlinear @xmath0 model@xcite which is well established in the study of mesoscopic disordered conductors . in deriving the appropriate field theory , we are able to show how previous theoretical calculations of the conductivity , within both quasiclassical@xcite and quantum mechanical approaches@xcite , may be recovered in a natural way within the field theory formalism . this enables us to extend previous theoretical results in a consistent way so as to include the influence of quantum interference processes . the experiments of weiss _ et al_.@xcite employed weakly modulated two - dimensional ( 2d ) electron systems of high mobility with a well - known period , @xmath1 , much less than the mean free path , @xmath2 . such samples were engineered using a holographic modulation technique , based on the persistent photoconductivity effect in gaas / al@xmath3ga@xmath4as heterostructures at low temperatures @xmath5 . the weiss oscillations appear in only one component of the resistivity tensor , @xmath6 , when the modulation is in the @xmath7 direction . furthermore they appear at magnetic fields @xmath8 such that the cyclotron radius @xmath9 ( where @xmath10 is the cyclotron frequency , @xmath11 is the electron charge and @xmath12 is the fermi velocity ) satisfies the commensurability condition @xmath13 , for integer @xmath14 . hence the oscillations are periodic in @xmath15 . in addition they are relatively stable with respect to temperature , suggesting a quasiclassical origin . at even higher magnetic fields , such that the cyclotron radius is much less than the period of modulation ( @xmath16 ) , the quasiclassical theory predicts that the @xmath6 component shows a large , nonoscillatory increase proportional to @xmath17 , leading to a strong positive magnetoresistance . this result has been confirmed in experiments@xcite for which the temperature was kept deliberately high so as to avoid the intervention of the iqhe . a limitation of the quasiclassical approach is that it fails to account for the renormalization of single - particle properties , such as the density of states and scattering lifetime , by the strong magnetic field according to landau quantization . even in unmodulated samples , quantization leads to oscillations in the density of states with respect to magnetic fields , and hence to shubnikov de haas oscillations in the resistivity . in samples with the periodic potential , the shubnikov de haas oscillations start to appear at higher magnetic fields than the weiss oscillations ( for @xmath18 an integer multiple of the fermi wavelength @xmath19 ) . a quantum - mechanical approach that does allow for such quantization effects has been provided by zhang and gerhardts@xcite ( see also peeters and vasilopoulos@xcite ) ; it is a diagrammatic treatment that generalizes the approach of ando and co workers@xcite to modulated samples . the quantum - mechanical approach is then capable of describing both weiss oscillations and the shubnikov de haas oscillations ( also affected by the modulation ) at higher magnetic fields . in principle , the calculation of weak localization corrections in a strong magnetic field ( unitary ensemble ) is possible even in the presence of a periodic potential , by a generalization of the diagrammatic approach of zhang and gerhardts@xcite , but the procedure would be complicated ( although the simpler orthogonal case has been examined diagrammatically for a periodic magnetic field@xcite ) . instead , the calculation of high - order diagrams is more convenient in the field - theory formalism and , furthermore , with this method the possibility exists of calculating contributions of diagrams to all orders by the renormalization - group technique . we show how the field - theory takes the form of a nonlinear @xmath0 model with a topological term @xcite . the effective lagrangian is slightly nonstandard since it contains an anisotropy in the coefficients , corresponding to the difference between the longitudinal conductivities in the @xmath7 and @xmath20 directions , due to the periodic potential . the effect of weak localization corrections to the conductivity for unmodulated samples is accounted for by a scaling ( one - parameter scaling ) of the conductivity with the system size . as a first step we derive the analog of the one - parameter scaling for the conductivity@xcite in the modulated system by means of a perturbative rg analysis of the effective lagrangian . we then turn to the study of the iqhe , implementing a generalization of a two - parameter scaling , which has been conjectured@xcite as a model for the iqhe in unmodulated systems . we examine how the resistivity tensor should be affected at low temperatures by the iqhe in modulated samples for parameters that are realizable in actual experiments . we see , for example , how the hall conductivity becomes quantized under scaling at low temperatures , while in the regions between the plateaus the longitudinal conductivities develop peaks of differing heights according to the anisotropy in the @xmath7 and @xmath20 directions . _ in the following , we employ the hamiltonian for the disordered conductor in a magnetic field and periodic potential in two dimensions : @xmath21 here @xmath22 is the vector potential , so that @xmath23 , where @xmath8 is the perpendicular , uniform magnetic field . also @xmath24 is the periodic potential which is taken to be weak ( @xmath25 , where @xmath26 is the fermi energy ) and with modulation period @xmath27 . @xmath28 is the disorder potential , which we assume to be @xmath29-correlated in space , with the associated scattering time @xmath30 . we assume that the mean free path @xmath31 greatly exceeds the modulation period @xmath32 . the plan for the remainder of the paper is as follows . in sec . [ sec : sigma ] we describe the field theory approach and derive the effective lagrangian for the system . in sec . [ sec : scaling ] we show how the field theory provides the scaling of the conductivity tensor under changes of length scale due to the contribution of quantum interference processes , and hence a description of the iqhe in these samples . section [ sec : summary ] concludes with a summary and discussion . in this section we describe a field - theoretical approach to the description of a disordered conductor in the presence of a periodic potential and a strong magnetic field . the approach is based on a diffusive nonlinear @xmath0 model , which will serve as a tool for the analysis of quantum interference effects in the remainder of this paper . in section [ mainfindings ] we discuss this model and the relation to previous semiclassical@xcite and quantum mechanical@xcite approaches . technical details of the derivation are presented in section [ derivation ] . the field - theory apparatus that we employ is by now well established in the study of spectral and wave - function statistics of disordered conductors . it is based on a functional integral expression of electron green functions in the presence of disorder , and in the diffusive regime takes the form of a nonlinear @xmath0 model in terms of a @xmath33 matrix@xcite . in two dimensions , the field - theory technique provides a convenient means of calculating weak localization corrections to the conductivity , a task that may become very cumbersome by the more conventional technique of diagrammatics@xcite . the field - theory may also provide a resummation of diagrams to all orders by a renormalization - group procedure@xcite . we show below that the effective lagrangian for the disordered conductor in the presence of a periodic potential and strong magnetic field takes the form @xmath34=\frac{1}{16 } \int dr\ , { \rm str } \left\{\sigma^0_{xx}(\nabla_x q)^2+\sigma^0_{yy}(\nabla_y q)^2-\sigma^0_{xy}\tau_3q [ \nabla_x q,\nabla_y q]\right\},\end{aligned}\ ] ] where @xmath35 is the classical conductivity tensor of the system , in units of @xmath36 . the 8@xmath378 supermatrix field @xmath38 satisfies the nonlinear constraint @xmath39 . the supertrace operation is defined in ref . @xcite . in the absence of a strong magnetic field or modulation , a renormalization - group analysis of the first two terms in eq . ( [ diffaction ] ) leads to the well - known one - parameter scaling@xcite of the conductivity with system size due to weak localization . the final term in eq . ( [ diffaction ] ) is known as a topological term and appears in the theory of the iqhe proposed by pruisken and collaborators@xcite . the influence of the extra term does not appear within perturbation theory , but becomes evident only through a nonperturbative analysis . such an analysis has been conjectured@xcite in the form of a two - parameter renormalization - group procedure . the two parameters are the longitudinal and hall conductivities that follow coupled scaling equations with respect to changes of length scale . while the validity of the two - parameter scaling has been vigorously debated ( see , e.g. , refs . @xcite ) , it has remained a valuable guide to experimental@xcite and numerical@xcite data for a number of years . in the following , we take a pragmatic approach by not contesting the validity of the two - parameter scaling and its relation to the proposed field theory . instead , we assume its validity and explore how it may be generalized to take account of the periodic potential . the main difference of the lagrangian ( [ diffaction ] ) from the unmodulated case is the anisotropy in the diffusion coefficients for the @xmath7 and @xmath20 directions . the lagrangian , therefore , contains three , rather than two parameters , which scale together under changes of length scale . a simple scale transformation , however , maps the lagrangian to an isotropic version for which the usual two - parameter scaling may be applied . this lagrangian applies in the diffusive limit , that is , to configurations of the @xmath38 field that vary on scales much longer than the mean free path . this allows the calculation of the contribution of low - momentum relaxational modes to weak localization corrections to the conductivity . although momentum relaxation is diffusive on large length scales , electron motion on the scale of the periodic potential is ballistic , since the modulation period is much less than the mean free path . to arrive at the above lagrangian it is therefore necessary to integrate over degrees of freedom corresponding to length much smaller than the mean free path . one way to do so is to start from a description of ballistic electron motion on length scales much less than the mean free path . such a description is provided by the `` ballistic @xmath0 model''@xcite which , as the name suggests , generalizes the diffusive @xmath0 model to the ballistic regime . starting from this model we show in sec . [ derivation ] how the contribution from short length scales may be integrated out . the result is the lagrangian of a diffusive @xmath0 model that describes the interaction of diffusion modes on scales longer than the mean free path . when derived in this way , the effective lagrangian takes the form of eq . ( [ diffaction ] ) , but with the @xmath40 coefficients replaced by @xmath41 . here @xmath42 correspond precisely to the components of the quasiclassical conductivity tensor as derived from the boltzmann equation for the modulated system @xcite . one finds that , in the quasiclassical approach , only the xx component of the quasiclassical resistivity tensor is affected by the modulation , all other components being the same as in the unmodulated case , so that @xmath43 and @xmath44 ( here @xmath45 is the diffusion coefficient in dimension @xmath46 ) . for details of the solution of the boltzmann equation , we refer the reader to beenakker@xcite . here we only remark that for moderate magnetic fields , weak enough that the cyclotron radius is much larger than the modulation period ( @xmath47 ) , but strong enough that @xmath48 , the quasiclassical result simplifies to @xmath49 equation . ( [ been2 ] ) demonstrates the oscillatory dependence of the resistivity known as weiss oscillations . for strong magnetic fields , such that @xmath50 , a very large oscillatory increase in @xmath6 proportional to @xmath17 is predicted and has been observed experimentally in , e.g. , geim et al . @xcite . however , for high magnetic fields ( @xmath51 ) , the quasiclassical results are reliable only at sufficiently high temperatures ( @xmath52 ) , while at lower temperatures the effects of quantization on single - particle properties such as the density of states must be taken into account by a quantum - mechanical approach . these effects are not taken into account in the derivation via the ballistic @xmath0 model , since here the magnetic field is treated only as a weak perturbation . indeed , once temperatures are low enough for weak localization corrections to be significant ( @xmath53 ) , then quantization is already well established ( as @xmath54 , since @xmath48 ) . hence this method of deriving the lagrangian is unfortunately of no use in describing the effects of quantum interference processes at high magnetic fields . an improvement on this situation may be found by proceeding instead along a second route to derive the effective lagrangian , following more closely the original lines of pruisken@xcite . he showed how to derive a @xmath0 model in the presence of a strong magnetic field , including the effects of quantization . the lagrangian is applicable at high values of the landau - level index @xmath14 , such that the fluctuations of the density of states may be neglected . we show below in sec . [ derivation ] how this method may be adapted to describe a disordered conductor in a strong magnetic field and a periodic potential . this route bypasses the intermediate step of a ballistic @xmath0 model , instead computing more directly the final diffusive lagrangian . while the resulting lagrangian takes again the form displayed in eq . ( [ diffaction ] ) , now the @xmath40 coefficients correspond precisely to @xmath55 , where @xmath56 are the components of the conductivity tensor calculated in the fully quantum - mechanical approach of ref . @xcite . since the effects of quantization are included this approach correctly describes certain features that are observed in experiment but are beyond the quasiclassical approach . some typical results for the resisitivity components ( @xmath57 ) calculated within the quantum - mechanical approach are shown in fig . [ fig : respv ] . as well as the appearance of weiss oscillations in @xmath58 at moderate magnetic fields ( up to @xmath59 in fig . [ fig : respv ] ) , we see that both @xmath60 and @xmath61 show strong , in - phase shubnikov de haas oscillations at higher magnetic fields . these latter oscillations reflect those of the density of states . at weaker fields , for which the weiss oscillations in @xmath58 are visible , weaker , out - of - phase oscillations in @xmath61 also appear . again , the latter oscillations in @xmath61 are beyond the quasiclassical approach . the quasiclassical result ( [ been2 ] ) for the change in @xmath62 ( for @xmath48 and @xmath63 ) may be reproduced by the quantum mechanical approach , as long as the temperature is high enough for the thermal broadening to be greater than the separation of the landau levels , @xmath64 ( see , e.g. , refs.@xcite ) . as the temperature is lowered , however , the region of applicability of the quasiclassical formula ( [ been2 ] ) shrinks to progressively lower magnetic fields : the sinusoidal weiss oscillations start to interfere at low temperatures with the shubnikov de haas oscillations . indeed , at vanishingly low temperatures , the quasiclassical result becomes unjustified for all magnetic fields . the lagrangian ( [ diffaction ] ) , supplied with the values of the conductivities determined by the quantum - mechanical approach , may then be used for a reliable description of quantum interference processes at high magnetic fields , and hence the operation of the iqhe . in this section we describe in more detail the derivation of the effective lagrangian ( [ diffaction ] ) . in the field - theory approach , it is necessary to average the functional integral expression for products of green functions over the disorder configurations . to perform the disorder average , several possible techniques exist , including the use of either supersymmetry or the replica trick . although we may use either of these two approaches , we choose here the supersymmetry approach , since it is free of certain technical problems regarding nonperturbative calculations that appear for the replica approach . the starting point for both routes is a functional integral representation of the partition function , @xmath65 , in terms of `` superfields '' @xmath66 : @xmath67 p[v ] d[\psi,\bar{\psi } ] \exp { -\cal l}(\psi,\bar{\psi},v ) , \\ { \cal l } & = & i \int dr \bar{\psi}({\bf r } ) ( e - h - i\delta\lambda)\psi({\bf r } ) . \label{lag1}\end{aligned}\ ] ] here the superfields @xmath66 contain eight components , corresponding to fermion / boson , retarded / advanced and time - reversal sectors @xcite . also @xmath68 in advanced - retarded space space , @xmath29 is a positive infinitesimal and @xmath69 is the hamiltonian ( [ hamiltonian ] ) up to the substitution @xmath70 , where @xmath71diag@xmath72 in time - reversal space . the short - range random potential @xmath73 is taken to be gaussian distributed , according to @xmath74 = \frac{1}{{\cal z } } \exp\left\ { -\frac{\pi\nu\tau}{\hbar}\int dr v({\bf r})^2\right\},\end{aligned}\ ] ] where @xmath75 provides the normalization . _ ballistic @xmath0 model . _ the first route is to derive a ballistic @xmath0 model from the lagrangian in eq . ( [ lag1 ] ) . et al_. @xcite have suggested a derivation of the ballistic @xmath0 model by means of an energy averaging procedure , leading to a lagrangian of the same form as that originally proposed by muzykantskii and khmelnitskii@xcite . following the lines of andreev _ et al_.@xcite , one treats the disorder field @xmath28 as a perturbation in eq . ( [ lag1 ] ) . performing an energy averaging on the clean hamiltonian leads to a term in the lagrangian that is nonlocal and quartic in the @xmath76 , @xmath77 fields . the quartic term is then decoupled by a hubbard - stratonovich field , @xmath78 , and the @xmath76 , @xmath77 fields are integrated out . the resulting lagrangian is then simplified by subjecting it to a saddle point analysis , and performing a gradient expansion around the saddle point . note that , since at this stage we are neglecting the effects of quantization on single - particle properties , we do not include the renormalization of this saddle point solution by the magnetic field , instead we treat the magnetic field as a perturbation . following a wigner transform , and after applying a semiclassical@xcite approximation , the @xmath33 matrix becomes a function of position , @xmath79 , and the direction of momentum , @xmath80 , where @xmath80 is a unit vector . we do not repeat these steps here but refer to refs . @xcite for further details . the resulting lagrangian ( ballistic @xmath0 model ) corresponding to the hamiltonian ( [ hamiltonian ] ) is @xmath81=\frac{\pi\nu \hbar}{4 } \int dr { \rm str}\left\{-\frac{1}{2\tau}\int \frac{dn}{2\pi } \frac{dn'}{2\pi}q_{\bf n}({\bf r})q_{{\bf n}'}({\bf r})-2\int \frac{dn}{2\pi}\lambda\tau_3 t_{\bf n}^{-1}{\mathcal l}^0({\bf n},{\bf r})t_{\bf n}\right\ } , \label{freeb}\end{aligned}\ ] ] in this equation @xmath82 is the louiville operator defined by @xmath83 where @xmath84 is the polar angle of @xmath80 , @xmath85 is the density of states per spin direction at the fermi level , and @xmath86 is the electron velocity at the fermi surface . the supermatrix @xmath87 satisfies the nonlinear constraint @xmath88 , as a result of which it may be parametrized by @xmath89 due to the presence of the magnetic field , the @xmath90 field satisfies unitary symmetry , that is , @xmath91=0 $ ] . the @xmath87 field also satisfies the further standard symmetries @xmath92 , where @xmath93 . the matrices @xmath94 and @xmath95 , as well as supertrace operation , are defined in ref . @xcite . the lagrangian ( [ freeb ] ) is in its current form too complicated for our purposes : it describes fluctuations of @xmath87 on length scales smaller than the mean free path ( and longer than the fermi wavelength ) and for any dependence on @xmath80 . at the same time , electron motion on length scales much longer than the mean free path is diffusive . the strategy therefore is to integrate out the modes corresponding to fluctuations on length scales smaller than the mean free path , to produce a lagrangian that describes the interaction of diffusion modes on larger length scales . to perform the integration , we isolate the the fluctuations of lowest mass ( termed massless ) , corresponding to matrices @xmath96 that are independent of @xmath80 and vary more slowly than the mean free path , @xmath97 . we then integrate out all other fluctuations that preserve @xmath98 . a similar procedure has been carried out by wlfle and bhatt@xcite to derive the diffusive lagrangian for a disordered conductor with anisotropic masses in the @xmath7 and @xmath20 directions ( see also refs.@xcite for further examples ) . to perform integration over the weakly massive modes , we write @xmath99 here the @xmath100 matrix , obeying the symmetries @xmath101 and @xmath102 , represents the massless fluctuations , while @xmath103 contains the weakly massive fluctuations . the matrix @xmath87 may , in turn , be parametrized by its generator @xmath104 , for example , by@xcite @xmath105 here @xmath106 is off - diagonal in retarded - advanced space and satisfies the symmetry @xmath107 . since the fluctuations represented by the @xmath90 matrix are weakly massive , it is sufficient to treat them within a gaussian approximation : @xmath108 . inserting into the lagrangian ( [ freeb ] ) , we find @xmath109 where @xmath110 , @xmath111\lambda$ ] . the next step is to integrate over matrices @xmath106 . this amounts to a set of gaussian integrals with a linear term in @xmath106 , and hence may be performed by means of a shift of @xmath106 . the resulting lagrangian has the form @xmath112=\frac{h}{16e^2 } \int dr\ , { \rm str } \left\{\sigma^{\rm qc}_{xx}(\nabla_x q)^2+\sigma^{\rm qc}_{yy}(\nabla_y q)^2-\sigma^{\rm qc}_{xy}\tau_3q [ \nabla_x q,\nabla_y q]\right\},\end{aligned}\ ] ] it is written in terms of a matrix @xmath113 that depends only on position and varies on length scales much longer than the mean free path ( and hence the modulation period , @xmath32 ) . the @xmath38 field also satisfies the nonlinear constraint @xmath39 , as well as the symmetries @xmath114 . we see that the lagrangian takes the usual form of a diffusive @xmath0 model with a topological term@xcite , the only non - standard feature being the anisotropy of @xmath115 . the coefficients @xmath116 are defined by @xmath117 where the averaging in eq . ( [ conductivities ] ) is over one period of the modulation @xmath32 ( this averaging arises naturally in the lagrangian ( [ diffaction ] ) since the @xmath118 matrix varies slowly on this scale ) . also , @xmath119 is defined by the boltzmann - like equation @xmath120 here , the collision operator is given as @xmath121 as we will show now the @xmath116 coefficients in the lagrangian ( [ diffaction2 ] ) are the components of the quasiclassical conductivity tensor as derived by beenakker @xcite . in his approach , the boltzmann equation is expressed @xcite ( see also refs.@xcite ) in terms of the distribution function f(*r*,*n * ) @xmath122 and the current @xmath123 is given by @xmath124 where the averaging is over both the velocity direction @xmath80 and one period of the modulation , @xmath32 . the relation @xmath125 defines the quasiclassical conductivity tensor @xmath116 , whose inversion gives the quasiclassical resistivity tensor @xmath126 . comparing eqs . ( [ boltz2 ] ) and ( [ boltzmann ] ) , we see that @xmath127 is related to the distribution function @xmath128 by @xmath129 comparing also eqs . ( [ conductivities ] ) and ( [ jdef ] ) , we see that the @xmath116 coefficients in the lagrangian ( [ diffaction2 ] ) calculated by this method coincide precisely with the components of the quasiclassical conductivity as derived@xcite from the boltzmann equation ( [ boltzmann ] ) . in this way we see that we have rederived the quasiclassical results within the field - theory formalism . as discussed above , the inadequacy of the lagrangian ( [ diffaction2 ] ) is that it neglects the renormalization of the single - particle properties by the strong magnetic field . at high magnetic fields ( @xmath48 ) , once the temperature is low enough for the weak localization corrections to be significant ( @xmath53 ) , the quasiclassical results for the conductivity@xcite have already become unreliable due to their neglect of quantization . in order to include these effects , and hence account for weak localization effects reliably , we need to follow a different route to derive the lagrangian . such a route for unmodulated systems has been provided by pruisken@xcite , whose method may be adapted ( as we show here ) to include a periodic potential . this method includes the renormalization of the saddle point equation for the @xmath33 matrix by the strong magnetic field . it does not require a derivation of a ballistic @xmath0-model as an intermediate step , but provides more directly the final form of the lagrangian . while the first route contains certain parallels with the quasiclassical approach @xcite , the second route contains closer parallels with the quantum - mechanical approach@xcite . _ generalization of pruisken derivation . _ we now present the second route to the derivation of the effective lagrangian ( [ diffaction ] ) , starting again from form ( [ lag1 ] ) for the lagrangian . the approach now is to average over the short - range disorder . this produces a term in the lagrangian that is quartic in the @xmath76 , @xmath77 fields . the quartic term may then be decoupled by a ( now local ) hubbard - stratonovich field @xmath38 . after integrating out the @xmath76 , @xmath77 fields , one finds @xmath130 = \int dr\left\{\frac{\hbar\pi\nu}{8\tau}{\rm str}\,{q}^2 -\frac{1}{2}{\rm str}\,{\rm ln } \left[-i\left(\pi_{\mu}\pi^{\mu}-e+u_0\cos(qx)\right ) + \delta\lambda + \frac{\hbar q}{2\tau}\right]\right\ } \label{trln}\end{aligned}\ ] ] where @xmath131 . here q satisfies the symmetry @xmath132 . the lagrangian ( [ trln ] ) , in principle , provides an exact description of the system , although in its current form it is too general to be useful . instead , one proceeds by finding the saddle point value of @xmath33 that minimizes the lagrangian , and by performing a gradient expansion about this minimum . the saddle point equation may be written as @xmath133 in order to find the matrix inverse of the operator in eq . ( [ sad1 ] ) , we make use of the eigenvalues and eigenfunctions of the operator @xmath134 . using the landau gauge @xmath135 , one may write the eigenfunctions of @xmath136 in the form @xmath137 , where @xmath138 is a normalization length . the center coordinate @xmath139 remains a good quantum number despite modulation , where @xmath140 is the magnetic length . the @xmath141 are eigenfunctions of the hamiltonian @xmath142 in the absence of modulation ( @xmath143 ) , eq . ( [ h0a ] ) represents the hamiltonian of a harmonic oscillator . we proceed by a first - order perturbative expansion modulation @xmath144 . this turns out to be a very good approximation for typical parameters that we consider , such as a weak periodic potential . it is very difficult to improve on this approximation analytically , although exact numerical diagonalizations have been performed for the density of states@xcite . within the perturbative expansion , the eigenvalues @xmath145 of the hamiltonian ( [ h0a ] ) are given by @xmath146 where @xmath147 are the unperturbed landau energies , and @xmath148 , where @xmath149 and @xmath150 are the laguerre polynomials@xcite . we see that the modulation lifts the degeneracy of the landau levels and discrete levels are broadened into bands , whose width depends on the band index @xmath14 in an oscillatory manner ( due to the behavior of the laguerre polynomials at large @xmath14 ) . it is this oscillatory dependence that leads to the weiss oscillations in the resistivity . the validity of the perturbation theory depends only on the smallness of the @xmath151 parameter , which is assured for large values of the landau - level index @xmath14 . we refer the reader to , e.g. , refs.@xcite for further details of this perturbative expansion . we also make the assumption that in this basis , the saddle point value of @xmath33 is independent of the landau indices and @xmath152 : this approximation is analogous to the c - number approximation ( cna ) introduced by zhang and gerhardts @xcite for the self - energy matrix in the presence of modulation , and is valid as long as the magnetic field is not too high . in order to go beyond the cna , one would need to generalize the saddle point @xmath33 to include a matrix structure in the space of landau indices and a dependence on @xmath152 : such a task is of interest for future work but beyond the scope of this paper . the saddle point equation ( [ sad1 ] ) may now be written @xmath153 where @xmath154 is the width of the landau level in the absence of modulation and @xmath155 the saddle point equation ( [ sad3 ] ) coincides with the self - energy equation derived in the self - consistent born approximation ( scba ) by zhang and gerhardts@xcite , under the replacement @xmath156 , where indices refer to advanced / retarded space and @xmath157 denotes the self - energy . the solution for @xmath33 is then of the form @xmath158 , where @xmath159 is proportional to the density of states . in the absence of modulation , the density of states reduces to that determined by ando and co - workers@xcite within the scba ( as shown by pruisken@xcite ) . having identified the saddle point value @xmath33 , we proceed by performing a gradient expansion of the lagrangian ( [ trln ] ) around the saddle point . we follow very closely the calculation of pruisken@xcite , although we use the supersymmetric rather than the replica formulation . we use the representation @xmath160 where the @xmath161 fields are diagonal in retarded / advanced space and represent the massive modes . the procedure is to integrate over @xmath161 and @xmath162 separately . to do so , we split the fields of integration , @xmath163),\end{aligned}\ ] ] in the process acquiring the associated jacobian @xmath164 $ ] ( as discussed in ref .@xcite ) . @xmath165 becomes @xmath166 & = & \int dr \left\ { -{\rm str}\,{\rm ln}[i[p ] ] + \frac{\hbar \pi\nu}{8\tau } { \rm str}p^2 \right . \\ & & \left . -\frac{1}{2}{\rm str}\,{\rm ln}\left [ i\left ( e - t^{-1 } \pi_{\mu}\pi^{\mu } t -u_0\cos(qx)-\frac{i \hbar p}{2\tau}\right ) \right]\right\}.\end{aligned}\ ] ] to integrate over the @xmath161 modes , wsplit @xmath165 into two parts : @xmath167 & = & \int dr\left\ { -{\rm str}\,{\rm ln}[i[p ] ] + \frac{\hbar \pi\nu}{8\tau } { \rm str}p^2 -\frac{1}{2}{\rm str}\,{\rm ln}\left[e-\pi_{\mu}\pi^{\mu } -u_0\cos(qx)-\frac{i \hbar p}{2\tau}\right ] \right\ } , \\ \delta { \cal l}[p , t ] & = & { \cal l}[p , t]-{\cal l}_0[p].\end{aligned}\ ] ] integration over the @xmath161 fields then proceeds by cumulant - expanding @xmath168 $ ] with respect to @xmath169 $ ] . in turn , @xmath168 $ ] is computed by a gradient expansion of the str ln term up to second order in the combination @xmath170 . the propagators for the latter expansion are of the form @xmath171 weighted with respect to @xmath169 $ ] . a typical second - order term is @xmath172 [ d_{\nu}({\bf r } ' ) ] \langle g({\bf r}',{\bf r})\pi^\mu g({\bf r},{\bf r } ' ) \pi^\nu\rangle,\end{aligned}\ ] ] where the averaging is over @xmath161 with respect to @xmath173 . at this point , we now exploit the assumption that the @xmath174 matrices vary in space more slowly than the modulation . thus the averages of the products of the green functions are short ranged with respect to the @xmath162 matrices , and are translationally invariant after averaging over one cycle of the modulation . this allows us to perform a gradient expansion in the propagator averages : @xmath175 here the averaging includes that over a cycle of the modulation . a series of ward identities may now be used as in ref . @xcite to simplify the resulting expressions up to second order in @xmath176 . the final lagrangian is then of the form displayed in eq . ( [ diffaction ] ) , where @xmath177 . again , we have the standard form of a nonlinear @xmath0 model with a topological term , although with the anisotropy @xmath178 . again , @xmath33 satisfies @xmath179 and the symmetries @xmath180 . also the bare coefficients , @xmath35 , are the components of the conductivity tensor at zero temperature in units of @xmath36 , according to the formulas @xmath181 here the averaging is over @xmath161 as well as over one cycle of the modulation with respect to @xmath79 . the propagator averages may now be evaluated by their saddle point values with respect to @xmath161 . the bare coefficients @xmath35 are the components of the conductivity tensor , in units of @xmath36 , calculated in the scba at zero temperature . in the absence of modulation , the expressions ( [ condtensor2 ] ) for the bare conductivity then reduce to those computed in the scba by ando and co - workers @xcite . in the presence of modulation , they reduce to the conductivity tensor computed in the scba by zhang and gerhardts @xcite : @xmath182 ( the equivalence of may be seen by following the working of streda@xcite ) . therefore , in contrast to those derived via the ballistic @xmath0 model , the bare conductivities in the lagrangian ( [ diffaction ] ) represent the values calculated by a fully quantum - mechanical approach . having derived the final form of the lagrangian ( [ diffaction ] ) , we are now in a position to use it to calculate weak localization corrections to the conductivity tensor . the effect of the corrections may be expressed as a scaling of the conductivity tensor under changes of length scale . a perturbative treatment of the lagrangian ( [ diffaction ] ) is valid in the limit of large longitudinal conductivity ( @xmath183 ) . within perturbation theory , however , the hall conductivity is not renormalized by weak localization corrections . instead , the effect of the topological term in the lagrangian may only be made apparent through a nonperturbative analysis . such an analysis has been conjectured@xcite in the form of a two - parameter renormalization - group procedure . the two parameters are the longitudinal and hall conductivities which follow coupled scaling equations with respect to changes of length scale . the derivation of the scaling equations presented in refs . @xcite is based on approximating configurations of the field - theory parameter as a gas of instantons . while the validity of this derivation is being vigorously debated ( see , e.g. , refs . @xcite ) , it has remained a valuable guide to experimental@xcite and numerical@xcite data for a number of years . in the following , we take a pragmatic approach by not contesting the validity of the two - parameter scaling and its relation to the proposed field theory . instead , we assume its validity and explore how it may be generalized to take account of the periodic potential . first we subject the lagrangian to a perturbative rg analysis . the perturbation theory is valid in the limit of @xmath184 . the topological term in the lagrangian does not contribute at any perturbative order and hence @xmath185 is unrenormalized . for an unmodulated system , @xmath186 and the perturbative rg procedure has been explained in detail , e.g. , in ref . @xcite . in the presence of modulation , the @xmath0 model is slightly nonstandard due to the anisotropy of the longitudinal conductivities , @xmath187 . however , quantum interference processes in a diffusive , anisotropic conductor have been considered before , for example by wlfle and bhatt@xcite , where the origin of the anisotropy was envisaged as due to a difference in effective electron masses in the two directions . in the latter paper , a diffusive @xmath0 model was derived with the same form as eq . ( [ diffaction ] ) , although without the topological term . when @xmath187 , the two parameters , @xmath188 and @xmath189 , follow coupled flow equations under changes of length scale . while it is straightforward to derive and solve the two flow equations , it is also instructive to follow a different strategy whereby we perform a scale transformation after which the lagrangian ( [ diffaction ] ) maps to an isotropic form : we scale @xmath190 under which the lagrangian ( [ diffaction ] ) transforms to an isotropic @xmath0 model : @xmath130 = \frac{1}{16 } \int dr ' { \rm str}\left\ { \tilde{\sigma}^0 ( \nabla q)^2 -\sigma^0_{xy } \tau_3 q[\nabla_{x ' } q,\nabla_{y ' } q]\right\ } , \label{pruisken2}\end{aligned}\ ] ] where @xmath191 . the perturbative scaling for @xmath192 may now be derived in the standard way@xcite : the flow equation is ( to leading order ) @xmath193 by integrating the flow equation from microscopic to macroscopic length scales ( of the order of the system size @xmath194 ) , we find the conductivity @xmath195 using the fact that the ratios @xmath196 are invariant under the scaling , we may recover @xmath197 from @xmath198 as follows : @xmath199 we remark that the perturbative scaling derived from a field theory as above in the presence of anistropy in the conductivities in the @xmath7 and @xmath20 directions ( due to an anisotropy in the electron mass ) has been confirmed by direct diagrammatics ( in the orthogonal ensemble ) by wlfle and bhatt@xcite . equation ( [ pert ] ) shows how weak localization corrections affect the longitudinal conductivity within perturbation theory . we can see how the conventional scaling applies to @xmath192 in the presence of the periodic potential . by inputting the scba values for @xmath200 and @xmath201 weak localization corrections are obtained , according to eq . ( [ pert ] ) they depend on _ both _ bare values @xmath200 and @xmath201 . corrections to @xmath188 , for example , are influenced by the strong oscillatory behavior of @xmath202 . upon inverting the conductivity tensor to find the resistivity tensor , the weak localization corrections will therefore give rise to additional oscillatory corrections to the resistivity tensor as a function of the magnetic field . as mentioned earlier , strong weiss oscillations in @xmath6 are accompanied by weak out - of - phase oscillations in @xmath203 as long as quantum interference effects may be neglected . it is interesting to notice that weak localization corrections to @xmath203 oscillate in phase with the ( dominant ) oscillations in @xmath6 . in practice@xcite , rather than changing the system size @xmath194 , one varies the temperature to change the effective system size , which is given by the diffusion length , @xmath204 . the perturbative result ( [ pert ] ) is only valid when the corrections are much smaller than the bare conductivities , and hence may be difficult to verify experimentally . a potentially more promising strategy is to illustrate the effect of weak localization corrections when their contribution is relatively large such as is the case for the iqhe . we now generalize from the perturbative analysis to discuss the nonperturbative scaling that would follow from the lagrangian ( [ diffaction ] ) . again , we are able to make the scale transformation ( [ scale ] ) as above , which maps the lagrangian to the isotropic version , eq . ( [ pruisken2 ] ) . the coupled flow of the two coupling constants , @xmath205 and @xmath185 , may written in a general form , @xmath206 the above beta functions , @xmath207 and @xmath208 , would take precisely the same form as the beta - functions that describe the flow of @xmath188 and @xmath185 in the equivalent , unmodulated system . the starting point of the flow is determined by the bare conductivity tensor @xmath209 . in general , the coupled flow equations ( [ genscale ] ) for @xmath192 and @xmath185 need to be integrated up to length scales of the system size . the values of @xmath210 and @xmath211 may then be recovered from @xmath198 by eqs . ( [ sigtil ] ) . as a consequence of eq . ( [ sigtil ] ) , we see that the ratio @xmath212 remains constant under the scaling . this conclusion is furthermore independent of the form of the beta - functions , @xmath207 and @xmath208 , in eq . ( [ genscale ] ) . in the absence of modulation , the scba values@xcite of the bare conductivities correspond to an approximate semicircular dependence of @xmath200 on @xmath213 . in the presence of modulation , the dependance of @xmath214 on @xmath213 is modified from the semicircle in a complicated way ; a typical dependence , calculated using the scheme proposed in ref . @xcite , is shown in fig . [ fig : scaling ] . the corresponding dependency of the bare conductivity tensor on magnetic field for the same parameters is part of fig . [ fig : condscaling ] below ( thick curves ) . we remark that for the peaks centered around @xmath8=1.5 t one may already expect vertex corrections to become effective , although they are not included in the cna approach that we have employed . such corrections have the general tendency@xcite to enhance @xmath201 in comparison to @xmath200 . in order to provide an illustration of the iqhe in the modulated sample , it is necessary to assume a particular form of the scaling equations ( [ genscale ] ) . the following scaling equations@xcite were derived ( originally in the replica formulation ) within a dilute gas approximation of the instantonic configurations of the @xmath33 matrix from the lagrangian ( [ diffaction ] ) : @xmath215 the dimensionless constant @xmath216 is of order unity and is related to the density of instantons . it may be seen immediately from the form of eq . ( [ twopar ] ) , that along the lines @xmath217 , where @xmath14 is an integer , the hall conductance is unrenormalized . it may also been seen that the points @xmath218 , for integer @xmath14 , are ( attractive ) fixed points . upon scaling the system from microscopic to macroscopic length scales , the coordinates @xmath219 scale from the bare nonuniversal values towards the quantized values @xmath220 , for integer @xmath14 . this tendency reflects the quantization of the hall conductivity under scaling at low temperatures . as mentioned before , the validity of the approximations underlying eq . ( [ twopar ] ) has long been the subject of debate@xcite . keeping this in mind we use these equations to provide an illustration of the operation of the iqhe in the modulated system in figs . [ fig : condscaling ] and [ fig : focuscondscaling ] . fig . [ fig : condscaling ] shows the evolution of the conductivity for the modulated system under the two - parameter scaling . we see how the hall conductivity becomes quantized under scaling in the modulated system . between the plateaus in the hall conductivity , the longitudinal conductivities develop peaks under scaling of differing heights , due to the anisotropy in the system . [ fig : focuscondscaling ] shows in more detail how these peaks in the longitudinal conductivities develop , while the ratio between the two conductivities remains constant under scaling . we remark that the behavior described above should be within current experimental capabilities : for example , geim _ et al_.@xcite have studied the high magnetic field ( @xmath16 ) regime , although they kept the temperature high to avoid the effect of quantum interference processes . et al_.@xcite have also studied the iqhe in modulated samples ; in their work however they have not focused on the regime of a weak periodic potential or effects that are independent of the device boundaries . in this paper we have considered transport properties of a disordered conductor in a periodic potential and strong magnetic field . we focused on the contribution of quantum interference processes , whose influence at high fields is missing from previous approaches despite being responsible for a whole class of phenomena . for example , they lead to weak localization corrections to the conductivity and the operation of the iqhe at high magnetic fields . to this end , we introduced a field - theory approach , which is well established in the study of unmodulated disordered conductors . the effective lagrangian of the field theory takes the form of a nonlinear @xmath0 model , which describes the interaction of diffusion modes on large length scales . the presence of the strong magnetic field leads to an extra , topological term in the lagrangian . the form of the lagrangian is the same as for unmodulated systems , except for an anisotropy in the coefficients corresponding to the bare longitudinal conductivities in the @xmath7 and @xmath20 directions . we provided two different routes to deriving the lagrangian . the first route was via a so - called `` ballistic '' @xmath0 model , and demonstrated how the results of the quasiclassical approach@xcite may be recovered within the field - theory formalism . the drawback of this route , in common with the quasiclassical approximation , is that it neglects the renormalization of single - particle properies by landau quantization . consequently the resulting lagrangian was too approximate to be useful in determining weak localization properties at low temperatures . the second route improved on this situation by including the effects of landau quantization . it bypasses a derivation of a ballistic @xmath0 model and instead follows more closely the original derivation of pruisken@xcite for unmodulated systems . indeed , while the first route contained more parallels with the quasiclassical approach , the second route contained more parallels with the quantum - mechanical approach@xcite for the bare conductivity tensor . having derived the effective lagrangian , we showed how it leads to the scaling of the conductivity tensor under changes of length scale ( and hence temperature ) . a perturbative renormalization - group analysis of the free energy leads to a generalization , to modulated systems , of one - parameter scaling for the longitudinal conductivities . perturbatively , the hall conductivity is unrenormalized . in the regime of weiss oscillations@xcite , weak localization corrections give rise to an additional oscillatory dependence of the longitudinal conductivities as a function of the magnetic field . due to their smallness , these corrections may , however , be hard to detect experimentally . in order to describe the iqhe , whereby the hall conductivity becomes quantized at low temperatures , a nonperturbative analysis of the lagrangian is necessary . this is provided by the conjecture of the two - parameter scaling for unmodulated systems . assuming the validity of this conjecture and the underlying instanton gas approximation , we have shown how the two - parameter scaling may be generalized to the case of the modulated system . this has allowed us to illustrate the evolution of the resistivity tensor under the iqhe for parameters that are realistic for experiments . we find , for example , that the ratio of the two longitudinal components of the conductivity remains constant under scaling ( an observation that does not itself depend on the assumptions used to derive the flow equations ) . while the hall conductivity still becomes quantized under scaling in the modulated system , between the plateaus the longitudinal conductivities develop peaks of differing heights due to the anisotropy in the system . there are several directions in which our analysis may be generalized . a simple generalization is to consider a periodic magnetic field , rather than potential , a situation which leads to a similar phenomenology@xcite . the derivation of the field theory for this case follows very similar lines to the case of a periodic potential , with similar results . another direction in which the analysis may be extended straightforwardly is to the study of the low magnetic field regime , in which a positive magnetoresistance has been observed@xcite and explained within a quasiclassical approach@xcite . as long as @xmath221 , one may derive the field theory for this regime according to the first route of sec . [ sec : sigma ] , neglecting quantization ; the coefficients of the field theory then coincide with the components of the conductivity tensor derived within the boltzmann approach . the weak localization to the conductivity ( or related quantities such as the dephasing time@xcite ) may then be calculated using the lagrangian for either the orthogonal or the unitary ensemble , or in the crossover between the pure symmetry classes . a less straightforward generalization is to improve on the cna approximation of the quantum - mechanical approach and its analog in the field - theory formalism . according to this approximation , the self - energy matrix ( and the saddle point solution of the @xmath33 matrix ) are assumed to have a trivial structure in the space of landau indices and coordinate @xmath152 . while this simplifies the analysis considerably , we have seen already how this approximation breaks down for very high magnetic fields , such that the cyclotron radius is much less than the period of modulation , @xmath16 . improvement on the cna is necessary , not only for such high fields , but also to study models of disorder with long - range correlations , for which vertex corrections are not negligible . such models have been analyzed in the quasiclassical approach@xcite and shown to represent experimental data more closely in certain respects . very recently it has been shown how to improve on the cna@xcite within the quantum - mechanical approach , so as to analyze models at high magnetic fields or with long - range disorder correlations . in the field theory formalism , improvement on the cna approximation would require enlarging the space of the @xmath33 matrix even further to include the additional matrix structure in the space of landau indices , as well as a dependence on @xmath152 . inclusion of such a structure within the @xmath33 field is to our knowledge a novel direction to pursue , although this task is left as a future project . a further area to explore is the case of a strong periodic potential , for which the first - order perturbative expansion in the potential that we use is no longer valid . while an analytical approach would be very difficult , a numerical treatment would be better suited to this regime . for a strong enough potential that @xmath144 is of the order of @xmath222 , the smearing of the landau levels is so great that they can no longer be individually resolved even at low temperatures . this leads to a quenching of the shubnikov de haas oscillations ( as noted by beton _ et al_.@xcite ) and hence one would expect the quantum hall effect to be destroyed . we would like to thank b. huckestein and m. langenbuch for helpful discussions during the course of this work . the authors gratefully acknowledge the financial support of trinity college , cambridge ( d. t .- s . ) , the _ graduiertenkolleg _ 384 ( g.s . ) , the _ schwerpunktprogramm `` quanten hall systeme '' _ and the _ sonderforschungsbereich _ 491 . d. weiss , k. v. klitzing , k. ploog and g. weimann , europhys . * 8 * , 179 ( 1989 ) ; also in _ high magnetic fields in semiconductor physics ii _ , vol . 87 of _ springer series in solid - state sciences _ , ed . g. landwehr , springer - verlag : berlin ( 1989 ) . p. h. beton , e. s. alves , p. c. main , l. eaves , m. w. dellow , m. henini , o. h. hughes , s. p. beaumont and c. d. w. wilkinson , phys . b * 42 * , 9229 ( 1990 ) . p. streda and a. h. macdonald , phys . b * 41 * , 11892 ( 1990 ) ; p. streda , j. kucera and j. van de konijnenberg , physica scripta * t39 * , 162 ( 1991 ) ; j. kucera , p. streda and r. r. gerhardts , phys . rev . b * 55 * , 14439 ( 1997 ) . r. menne and r. r. gerhardts , phys . b * 57 * , 1707 ( 1998 ) . r.r . gerhardts , d. weiss and k. v. klitzing , phys . 62 * , 1173 ( 1989 ) . r. w. winkler , j. p. kotthaus and k. ploog , phys . * 62 * , 1177 ( 1989 ) . c. zhang and r.r . gerhardts , phys . b * 41 * , 12850 ( 1990 ) . p. vasilopoulos and f. m. peeters , phys . lett . * 63 * , 2120 ( 1989 ) ; f. m. peeters and p. vasilopoulos , phys . b * 46 * , 4667 ( 1992 ) . k. v. klitzing , g. dorda and m. pepper , phys . * 45 * , 494 ( 1980 ) . a. manolescu , r.r . gerhardts , m. suhrke and u. rssler , phys . b * 63 * , 115322 ( 2001 ) . j. gross and r. r. gerhardts , physica b * 298 * , 83 ( 2001 ) . c. w. j. beenakker , phys . 62 , 2020 ( 1989 ) . a. d. mirlin and p. wlfle , phys . b * 58 * , 12986 ( 1998 ) . a. k. geim , r. taboryski , a. kristensen , s. v. dubonos and p. e. lindelof , phys . b * 46 * , 4324 ( 1992 ) . p. f. bagwell , s. l. park , a. yen , d. a. antoniadis , h. i. smith , t. p. orlando and m. a. kastner , phys . b * 45 * , 9214 ( 1992 ) . b. milton , c. j. emeleus , k. lister , j. h. davies and a. r. long , physica e * 6 * , 555 ( 2000 ) . f. wegner , z. phys . b * 35 * , 207 ( 1979 ) . k. b. efetov , a. i. larkin and d. e. khmelnitskii , zh . fiz . * 79 * , 1120 ( 1980 ) [ jetp * 52 * , 568 ( 1980 ) ] . k. b. efetov , adv . phys . * 32 * , 53 ( 1983 ) ; k. b. efetov , _ supersymmetry in disorder and chaos _ , cambridge university press , new york ( 1997 ) . a. m. m. pruisken , nucl . b * 235 * , 277 ( 1984 ) h. levine , s. b. libby and a. m. m. pruisken , nucl . phys . b240 [ fs12 ] , 30 ; 49 ; 71 ( 1984 ) . t. ando and y. uemura , j. phys . . jpn . * 36 * , 359 ( 1974 ) ; t. ando , j. phys . * 36 * , 1521 ( 1974 ) ; * 37 * , 622 ( 1974 ) ; * 37 * , 1233 ( 1974 ) ; t. ando , a. fowler and f. stern , rev . * 45 * , 437 ( 1982 ) . b. l. altshuler , a. g. aronov , d. e. khmelnitskii and a. i. larkin , in _ quantum theory of solids _ , ed . i. m. lifshitz ( mir publishers : moscow ) ( 1983 ) . g. bergmann , phys . * 101 * , 1 ( 1984 ) . s. chakravarty and a. schmid , phys . rep . * 140 * , 193 ( 1986 ) . d. belitz and t. r. kirkpatrick , rev . phys . * 66 * , 261 ( 1994 ) . l. p. gorkov , a. i. larkin and d. e. khmelnitskii , pisma zh . . fiz . * 30 * , 248 ( 1979 ) [ jetp lett . * 30 * , 228 ( 1979 ) ] . s. hikami , a. i. larkin and y. nagaoka , prog . phys . * 63 * , 707 ( 1980 ) ; s. hikami , phys . rev . b * 24 * , 2671 ( 1981 ) . j. rammer and a. l. shelankov , phys . b * 36 * , 3135 ( 1987 ) . wang , phys . b * 65 * , 115303 ( 2002 ) . d. e. khmelnitskii , pisma zh . . fiz . * 38 * , 454 ( 1983 ) [ jetp lett . * 38 * , 552 ( 1983 ) ] . a. m. m. pruisken , phys . b * 32 * , 2636 ( 1985 ) . a. m. m. pruisken , in _ the quantum hall effect _ , ed . r. e. prange and s. m. girvin , springer : berlin ( 1987 ) . h. a. weidenmuller and m. r. zirnbauer , nucl . b * 305 [ fs25 ] * , 339 ( 1988 ) . m. r. zirnbauer , ann . ( leipzig ) * 3 * , 513 ( 1994 ) . b. huckestein , rev . phys . * 67 * , 357 ( 1995 ) . h. p. wei , d. c. tsui and a. m. m. pruisken , phys . b * 33 * , 1488 ( 1985 ) . m. abramowitz and i. a. stegun , _ handbook of mathematical functions _ , dover publications , : new york ( 1972 ) . b. a. muzykantskii and d. e. khmelnitskii , pisma zh . fiz . * 62 * , 68 ( 1995 ) [ jetp lett . * 62 * , 76 ( 1995 ) ] . a. v. andreev , b. d. simons , o. agam and b. l. altshuler , nucl . b * 482 * , 536 ( 1996 ) . p. wlfle and r.n . bhatt , phys . b * 30 * , 3542 ( 1984 ) . a. g. aronov , a. d. mirlin and p. wlfle , phys . b * 49 * , 16 609 ( 1994 ) . d. taras - semchuk and k. b. efetov , phys . b * 64 * , 115301 ( 2001 ) . the semiclassical approximation referred to here differs from the quasiclassical approximation associated with the boltzmann equation approach of ref . d. p. xue and g. xiao , phys . b * 45 * , 5986 ( 1992 ) . r. r. gerhardts , phys . b * 53 * , 11064 ( 1996 ) . p. streda , j. phys . c * 15 * , l717 ( 1982 ) .

in the present work we show a comprehensive comparison of our new stellar population synthesis miuscat models with photometric data of globular clusters and early - type galaxies . the models compare remarkably well with the colours of milky way globular clusters in the optical range . likewise , the colours of m31 globular clusters can also be explained by the models by assuming younger ages then their galactic counterparts . when compared with quiescent galaxies we reproduce the colour evolution at intermediate redshift . on the other hand we find that the colour relations of nearby early - type galaxies are still a challenge for present - day stellar population synthesis models . we investigate a number of possible explanations and establish the importance ofenhanced models to bring down the discrepancy with observations . [ firstpage ] galaxies : elliptical and lenticular , cd galaxies : stellar content globular clusters : general galaxies : star clusters : individual : m31

1205.5497

stellar population synthesis ( sps ) models are commonly used for solving a great variety of studies , from age / metallicity estimates in globular clusters to the reconstruction of the star formation history ( sfh ) in galaxies . the modeling of the spectral energy distribution ( sed ) emitted by evolving stellar populations requires at least three main ingredients : stellar evolutionary tracks , a spectral or photometric library with well - established atmospheric parameters for each library star , and an initial mass function ( imf ) . a variety of models exists in the literature that differ in their treatment of each of these ingredients , e.g. , @xcite . in particular , the adoption of the spectral library used to convert physical quantities in observed flux turned out to be critical in the comparison with observations . although theoretical libraries ( e.g. , @xcite ) are widely used among the most popular sps models @xcite because they are easier to handle , given the wide stellar parameter and spectral coverages and the arbitrary wavelength resolution , empirical - based libraries represent an advantage as the stars are real . the adoption of empirical spectra allows to overcome the uncertainties in the underlying model atmosphere calculations , giving more reliable predictions when compared with both spectroscopic @xcite and photometric @xcite data . in vazdekis et al ( 2012 , hereafter paper i ) a library of single - burst stellar population model seds covering the spectral range 3500 - 9500 at moderately high resolution is presented . for building - up these models we combine three extensive empirical stellar spectral libraries , namely miles @xcite , indo - us @xcite and cat ( @xcite , @xcite , @xcite ) . here we focus on calibrating these models by comparing these new predictions to a variety of photometric data ranging from globular clusters ( gc ) to quiescent galaxies . broadband colours are among the simplest predictions of population synthesis models , thus comparing the model predictions to observed colours is the natural zeroth - order test of compatibility between models and observations . since gc are almost coeval and mono - metallic population of stars that closely resemble a single stellar population ( ssp ) , they are the ideal benchmark to test the accuracy of the models . on the other hand , quiescent galaxies can reach larger metallicities than the metallicity distribution of most globular cluster systems and allow us , with some caveats , to calibrate the models even in the high metallicity regime . although they can be modelled as an ssp as a first approximation , pieces of evidence of recent star formation in early - type galaxies have been found by different authors ( e.g. , @xcite ) , which indicate the need of more complex stellar populations to obtain better fits . moreover , galaxies may contain a certain amount of internal reddening that can further complicate the analysis . the paper is organized as follows . in section 2 we present a brief summary of the stellar population synthesis models presented in paper i. in section 3 we describe the webtool developed to download and handle these models . in sections 4 and 5 we present the comparison of these models with globular cluster and galaxy data , respectively . section 7 summarizes our results . throughout the paper , we adopt the concordance cosmology : @xmath1=70 , @xmath2 and @xmath3 . the stellar population models we use throughout the paper have been presented in paper i , here we briefly summarize them . these models represent an extension of the @xcite models , based on the cat and miles empirical stellar spectral libraries . miles includes 985 stars covering the range @xmath4 3525 - 7500 @xcite with a spectral resolution of 2.5fwhm , ( falcn - barroso et al . the cat library @xcite consists of 706 stars in the range @xmath4 8350 - 9020 at resolution 1.5 . the indo - u.s . stellar library @xcite has also been added to fill - in the spectral range around 8000 that is not covered by the miles and cat libraries and to extend blueward and redward the wavelength coverage of the models . the indo - u.s . includes @xmath5 1200 stars with spectra that typically cover from 3465 to 9469 at resolution @xmath61.36 fwhm ( falcn - barroso et al . 2011 ) . although all the stars of the indo - u.s . library have been used to obtain an homogenized set of stellar parameters matching those of miles and cat libraries , only a subsample composed of 432 stars have been included to build - up the final miuscat catalogue to feed the models . only stars without gaps in the stellar spectra or significant telluric absorption residuals , among other reasons , were considered ( see paper i for details ) . as the flux - calibration quality of the indo - u.s . is not as accurate as in the other two libraries , a procedure was applied to correct the spectrum shape in the region @xmath4 @xmath5 7400 - 8350 to match the miles and cat stellar spectra . finally the indo - u.s . spectrum was used to extend the spectral range blueward the miles and redward the cat spectral ranges without applying any correction to the indo - u.s . spectrum . finally , the cat and indo - u.s . spectra were smoothed to match the resolution of miles , i.e. 2.5(fwhm ) . the miuscat stellar spectra were implemented in the models as described in @xcite and updated in falcn - barroso et al . finally the relevant spectral ranges of the resulting ssp seds were extracted and plugged to the original model seds computed on the basis of the miles and cat libraries . for the latter step a further correction , similar to that described above for the individual stellar spectra , was applied to the model seds to obtain a unique spectrum , which is identical to the miles and cat based models in the corresponding spectral ranges , respectively . unlike miles ssp spectra , the new miuscat models allow us to measure from b to i broad - band filters and related colours . it is also possible to derive the u filter , which starts at @xmath5 3050 @xcite , by correcting for the missing flux blueward 3464.9 . for this purpose we make use of the fractions provided in paper i , which are obtained with the aid of the library of stellar spectra of @xcite ( @xmath4 1150 - 25000 ) . for each ssp the fractions obtained for the different stellar spectral types are integrated along the isochrone to obtain the corresponding missing flux for the ssp sed . finally the total u magnitude is derived by measuring the flux of the miuscat ssp spectrum through the u filter redward 3464.9 and correcting by the missing flux . to compare our model predictions to the observational data discussed in this paper we follow the same method to compute the sdss _ u _ filter at two redshift values ( i.e. 0 and 0.04 ; see section 6 ) . the model predictions used in this work have been obtained with the aid of the webtool facilities that we provide in the miles website : http://miles.iac.es . the miles and cat stellar libraries and the stellar population synthesis models can be retrieved and handled according to the user requirements . the webtools include the transformation of the spectra to match the instrumental set - up of the observations ( spectral resolution and sampling ) , measurements of line - strength indices and synthetic magnitudes derived from the spectrum . as in this work we apply the miuscat models to a large sample of galaxy data we describe in more detail another webtool facility that will be extensively used in section 6 . the tool returns the model spectrum corresponding to a given star formation history ( sfh ) , which can be either user - defined or follow some parametric description . the first case is devoted to handle with complex sfhs , with an arbitrary number of bursts . for instance , this case will be particularly useful for the sfhs retrieved from cosmological simulations . given the mass fraction and the metallicity at different ages , the tool computes the corresponding sed according to : @xmath7 where @xmath8 is the number of input bursts , @xmath9 is the fraction of mass formed in the i - th bin and @xmath10 is the i - th ssp spectrum for a chosen imf . in the case that @xmath11 and @xmath12 are not included in the age / metallicity grid of the models , the closest ssp model to the desired parameters is selected . thus no interpolation is performed on the ssp spectra . for the parametric sfhs four different cases are allowed : * multiple bursts : up to five bursts occurring at different ages and with different metallicities are summed up * truncated : the star formation rate ( sfr ) is assumed constant between the time of formation , @xmath13 , and the truncation time , @xmath14 ( all the times are referred to the present time ) * exponential : the sfh is modelled as a power - law rising term at early times plus an exponentially declining sfr : @xmath15 where @xmath13 is the time when star formation starts , @xmath16 is the e - folding time and @xmath17 parametrizes the power - law rising term at early times . * exponential plus bursts : up to five bursts can be added on top on the exponential sfr as we will show in the following sections , the sfh webtool is particularly useful to construct galaxy seds , for which the use of complex stellar populations provide better fits than the ssp approach . in section 6 several applications of the parametric sfh will be explored . here we use the catalogue of @xcite ( 2010 edition ) as a reference for the milky way ( mw ) globular clusters . the catalogue is a compilation of 150 clusters including the johnson - cousins ubvri photometry , [ fe / h ] metallicity and reddening e(b@xmath18v ) among other data . colours , except b@xmath18v , have been de - reddened using the extinction law of @xcite . figure [ gc_bvri ] shows the comparison of the optical broad - band colours of the gcs with the ones derived from the models . the model metallicity indicates the total abundance ratio [ z / h ] . a fit to the data points with a second order polynomial is indicated to guide the eye . in order to avoid clusters with uncertain photometry due to high extinction , we excluded the most reddened clusters with e(b@xmath18v ) @xmath190.5 ( open circles ) . the miuscat models are shown for ages of 10.0 and 12.6gyr for a kroupa universal imf . the use of a salpeter imf would leave the colours almost unaffected ( with differences smaller than 0.01 mag ) . to derive the synthetic colours we adopt the @xcite filter responses and vega magnitudes . the zero point has been set by using the @xcite spectrum of vega with a flux of @xmath20erg @xmath21s@xmath22@xmath22 at 5556(see falcon - barroso et al . 2011 for further details on the computation of synthetic magnitudes ) . the colours of vega in the johnson - cousins filters are assumed to be u@xmath18b = u@xmath18v = b@xmath18v = v@xmath18r=0 mag , v@xmath18i=-0.04 mag . the black dotted line represents the photometric ssp model predictions computed by adopting temperature - gravity - metallicity relations from extensive photometric stellar libraries . the colour predictions resulting from these two approaches give very similar results ( with differences of the order of 0.01 - 0.02 mag ) showing the reliability of the flux - calibration of the model seds ( see paper i for more details ) . figure [ gc_bvri ] also shows that the models reproduce the observed colours fairly well . note that the v@xmath18i colour is particularly relevant to test the reliability of our models . this colour is heavily sensitive to the details of the method used to join the three libraries ( miles , indo - u.s . and cat ) . indeed , the v magnitude is entirely measured within the miles range , whereas the i filter response falls within the indo - u.s./cat range . the good match to the observed colours shows the robustness of the method and the flux - calibration of the miuscat seds . it is also remarkable the ability of our models to reproduce the b@xmath18v colour . this has been for a long time a problem for stellar population synthesis models . as quoted in previous works ( e.g. @xcite ) , ssp model seds tend to provide systematically redder b@xmath18v colour than the observed values ( @xmath50.05 mag ) . one of the reasons of these discrepancies may be ascribed to difficulties of the theoretically - based models in modeling the spectral energy distribution around @xmath23 5000 - 6000 , where the theoretical spectra predict a flux excess . the use of empirical libraries @xcite has proven to be effective at improving the match with colours of luminous red galaxies @xcite . likely , the use of empirical libraries with good flux - calibration quality and good parameter coverage , including the metal - poor regime , has allowed us to improve the matching of the gc colours in comparison to previous works . the improved match of the b@xmath18v vs [ z / h ] relation when using miles - based models instead of the stelib - kurucz library has been also shown in @xcite . in figure [ gc_ubv ] we show the results obtained for the u@xmath18v and u@xmath18b colours . we found that colours involving the u band are significantly more difficult to reproduce . indeed , our models predict bluer colours ( @xmath50.1 mag in the high metallicity regime ) . as explained in sect . [ sect_u ] , the u magnitude is not fully derived from the miuscat seds , which do not extend to the blue end of the u filter response . the missing flux is accounted for with the aid of the @xcite stellar library . however we rule out this calibration as the a main source for the obtained discrepancy , as the obtained magnitudes following two different approaches ( the second one employing ssp model seds from other authors ) do agree within @xmath50.02 mag . moreover , even the photometric predictions show a similar level of disagreement with the data . a possible explanation may be linked to the various u filter response curves that can be adopted to reproduce the observations . figure [ gc_ubv ] shows the obtained results for various u filter responses , which differ from our reference filter ( @xcite ) . whereas for other johnson broad - band filters the adoption of different references for a given filter usually implies minor changes , in the case of the u filter , the exact definition of the curve makes a significant impact on the obtained colours . for instance , if we use the johnson ( 1965 ) response curve , we obtain a much better match . since it is difficult to trace back the compilation of data to recover which filter has been used for every gc , the u@xmath18v and u@xmath18b colours are subject to this source of uncertainty . + + in this section we compare our models with the globular clusters of m31 . for this purpose we use different photometric catalogs and metallicity estimates . we consider the spectroscopic metallicity determinations of @xcite , which is a compilation of various sources @xcite , the revised bologna catalog ( rbc ) , which are determined in a homogeneous manner @xcite and the spectroscopic catalog of @xcite , hereafter c11 . integrated colours in the johnson - cousins system have been taken from the rbc , while photometry in the sdss bands is from @xcite . the extinction in each band has been derived from the reddening values given in the corresponding catalog and the @xcite extinction law . given the higher uncertainty in the m31 measurements compared to the mw gcs , we only consider clusters with low extinction values ( e(b@xmath18v ) @xmath240.5 ) to avoid unreliable colour estimates . only globular clusters classified as old in the catalogs are considered . figure [ m31_j ] shows the johnson - cousins colours of the m31 gcs as a function of metallicity . given the high dispersion of the points we found that the polynomial fit is not a good description of the data . to guide the eye , we overplotted the median of the colours computed in intervals of metallicity and compare it to the fit to the mw gcs . we also show the model lines corresponding to the ssps with ages ranging from 6 to 12.6gyr ( kroupa universal imf ) . the vega system used to compute the synthetic magnitudes is the same as adopted for the mw data . the sdss colours in the ab system are shown in figure [ m31_sdss ] . note that the scatter obtained for the m31 data in these two figures is considerably larger than that observed in the mw clusters . moreover , as shown by the arrows , the m31 clusters display a considerable amount of reddening . although the absolute colours of m31 can be subject to several uncertainties , a robust result inferred from figures 3 is that the slope of the colour - metallicity relation is flatter in the m31 clusters . given that the mw slope is consistent with that of the old models , we argue that in m31 a significant age variations with metallicity is needed to account for the different slope . a flatter relation than that indicated from the models is also seen in figure 4 with sdss colours . old stellar populations in the lower metallicity regime and young ages for the metal rich clusters could explain such a trend . we obtain the worse fits to the data with the old models for the v@xmath18r and _ g@xmath18r _ colours . indeed , the m31 gc colours are in general bluer than their galactic counterparts , mainly in the high metallicity regime , as it can be inferred from comparing to the mw fit . the reason for this discrepancy at nearly solar metallicity may be linked to the fact that m31 clusters show higher metallicities than their mw counterparts ( e.g. , @xcite ) . in general the sdss colours , even for intermediate metallicities , are matched better with younger models . however , it is worth noticing that the choice of the catalog may change our conclusion . in fact young models are required to fit the trend in the _ g@xmath18r _ colour of @xcite dataset , whereas models of 6 - 8gyr provide a fair match for all the colours of the rbc and c11 catalogs . from the comparison with our model lines the best match to the data is obtained combining old models at lower metallicities and younger models at the high metallicity regime . in @xcite the synthetic _ g@xmath18r _ colour derived from the models of @xcite has been compared to the sdss colours of m31 clusters in a similar way as it is done here . since the _ g@xmath18r _ colour falls in the miles range , it coincides with the predictions of the miuscat models presented here . in @xcite paper the @xcite colours turned out to be offset by @xmath50.1 mag at solar metallicity with respect to m31 gc colours . although for the oldest model considered here we find similar difference in colour , by using different colours and different photometric systems we conclude that the data can be matched by considering younger models than those in @xcite . finally , as we obtain with the old models very good fits to the mw clusters , which have more reliable data , and since the m31 cluster colours in the same photometric bands differ from the ones of the mw , we can conclude that we require younger models , at least in the high metallicity regime , to fit the data . + + our results agree with evidences supporting that metal - rich gcs tend to be younger than the more metal - poor clusters , at least in the mw ( e.g. , @xcite ) . moreover , it has been claimed that the gcs of m31 tend to be on average younger than their galactic counterparts @xcite , though this result is yet to be conclusive . spectroscopic ages for the high - metallicity m31 clusters have been derived in c11 . in the low - metallicity regime the authors could not get accurate age determination , as these clusters are contaminated by blue hb stars , hence we can not test the age variation with metallicity . for the high - metallicity clusters in our sample ( [ z / h]@xmath25 ) the c11 determinations give a mean age of 11.3 gyr , that apparently contradicts the young nature of the m31 clusters . note however that these authors employed the balmer absorption line indices as age indicators and it is well known that this method can give very uncertain absolute ages , even larger than the age of the universe @xcite . moreover , the models of @xcite used in c11 give broad - band colours in excellent agreement with our values . for instance , a solar metallicity model with age of 11.2 gyr has a b@xmath18v colours of 0.978 mag , very similar to b@[email protected] mag that we obtain for the miles models . both photometric predictions are much larger than the m31 colours ( figure [ m31_j ] ) , confirming the discrepancy between the spectroscopic and photometric age measurements . furthermore , given the high uncertainties involved in the extinction , even our results should be taken with caution . indeed , the parameter that varies mostly among the three catalogs is the reddening . the c11 estimates tend to be biased towards larger values than the other two catalogs , leading to bluer de - reddened clusters ( see for instance the g@xmath18r colour in figure [ m31_sdss ] ) . in c11 the quoted uncertainties for the @xmath27 is 0.1 mag , hence of the order of colour variations due to different ages . therefore , it is not possible to draw robust conclusions . in this section we consider the luminous red galaxy ( lrg ) sample of @xcite , for which typically the definition of early - type galaxy ( etg ) also applies . the lrg sample has been extracted from the sdss spectroscopic galaxy survey ( data release 7 , @xcite ) , by selecting galaxies with the galaxy_red flag in the sdss database . lrgs constitute a uniform sample of objects with the reddest rest - frame colours , selected with cuts in the ( _ g@xmath18r , r@xmath18i , r _ ) colour - colour - magnitude cube ( see @xcite ) . to obtain the same galaxy population at different redshifts , the lrg sample has been selected using two different selection criteria below and above z=0.4 , as the balmer break moves from the _ g _ band to the _ r _ band at this redshift . to increase the redshift range , galaxies from the 2df sdss lrg and quasar ( 2slaq ) survey @xcite have been also included ( @xmath515000 objects ) and they contribute to the high redshift tail of the redshift distribution ( see figure [ zdistr_lrg ] ) . by restricting the sample to galaxies with reliable redshift determination ( zwarning@xmath260 ) we end up with @xmath5170000 galaxies . the sdss database provides a variety of measured magnitudes for each detected object . throughout this paper , we use dereddened model magnitudes because they provide an unbiased colour estimate in the absence of colour gradients @xcite . previous works @xcite have highlighted the difficulties of stellar population models in reproducing the colour evolution in observed frame of lrg at intermediate redshift ( @xmath28 ) , with the prediction of too red colours in the _ g@xmath18r _ and too blue in the _ r@xmath18i _ for the lowest redshift lrgs , while the opposite effect is observed for the highest redshift lrgs . similar disagreements have been reported for red sequence colours in galaxy clusters at similar redshifts @xcite . + + in figure [ col1_z ] we compare the predictions from the miuscat models to the colour evolution of the lrg sample . according to the fiducial model for the formation of lrg , in the simplest case we assume that they formed in a uniform burst at high redshift ( z@xmath265 ) and evolve passively thereon . solar metallicity and kroupa universal imf are assumed . the observed colours are reproduced by shifting the model spectra to the corresponding redshift and taking into account the ageing effect . synthetic colours in the sdss filters have been computed in the ab system @xcite . for consistency , the observed colours from the sdss database have been transformed to ab system by applying the relative shifts : @xmath29 mag ; @xmath30 mag . the model predictions are shown for the redshift interval where the miuscat spectra cover the _ r@xmath18i _ and _ i@xmath18z _ colours . filter falls outside the measurable range at [email protected] , thus colours containing _ u _ and _ g _ are not shown . we see that the simplest model is able to reproduce the observed distribution at low redshift in both colours . the model starts to deviate from the median colours at [email protected] , producing too red colours . in figure [ col1_z ] it is represented also a second model , where a small fraction of the mass ( 5% ) is contributed by a metal - poor population ( [ z / h]@xmath26 - 2.32 ) . it is expected @xcite that galaxies exhibit some spread in the metallicity distribution with a contribution also from metal - poor stars . the effect of including stars with such a low metallicity is to bluen the colours , with the result of loosing the match with the observed distribution in the lowest redshift range , but improving it at high redshift . it is worth to say that the differences between this model and the median of observed colour is only @xmath5 0.02 mag . the uncertainties derived from the zero - point calibration of the sdss system ( 0.01 mag ) and from the choice of the imf ( 0.01 mag of difference between salpeter and kroupa universal ) could also account for this difference . hence , it turns out to be difficult to discriminate among these two possibilities . nevertheless , in both cases the miuscat models perform rather well , confirming the results of @xcite , according to which models constructed with empirical stellar libraries provide a good fit to the lrg colour distribution . although the lrg sample has been selected to follow a passive evolution model ( see @xcite ) , this does not guarantee that all the galaxies accomplishing the selection criteria follow this model . in @xcite it has been shown that a non negligible fraction of lrgs does not evolve passively in a dynamical sense . they can undergo mergers and the sample can be contaminated with galaxies that belong to the lrg sample for a short period of time . however , the brightest tail of the sample turns out to be coeval ( i.e. with almost no merging ) for galaxies in the redshift range : 0.22 - 0.45 . for this reason , in the bottom panels of figure [ col1_z ] we compare our passive evolution models with these bright galaxies ( @xmath31 -23 mag ) . for this subsample , the high - redshift galaxies are truly the progenitors of the low - redshift ones . in the redshift range where the galaxies appear coeval , our passive model performs remarkably well . 0.04 , which includes 370 objects . ] a further insight into the capabilities of the models can be obtained by comparing them to a sample of nearby galaxies , for which we can use three different colours as constraints . for this purpose we select a homogeneous sample in the local universe with a large number of sdss galaxies with visual morphological classification @xcite . we select a subsample of early - type galaxies ( morphological type @xmath32 ) with reliable redshift measurements . we further restrict our sample to a narrow redshift slice , @xmath33 , to end up with 370 targets . this choice of redshift is motivated by two main reasons . first , at this redshift the number of galaxies peaks ( see figure [ zdistr ] ) . second , given the limited spectral coverage of the miuscat models in the blue end , the correction factors we need to apply to measure the sdss _ u _ magnitude become too large for redshifts above this value , as the wavelengths at which the _ u _ filter response peaks start to fall outside the spectral range of the models ( see paper i ) . before comparing the colours derived from the miuscat seds , once redshifted to [email protected] , we have tested that the selected redshift interval is small enough that colour differences due only to variation in redshift within such interval are negligible ( @xmath340.01 mag ) . as for the photometric estimates , we rely on the model magnitudes from the dr7 catalog . it is worth noticing that all the colours involved in this study belong to the optical spectral range , which are known to be heavily affected by the age / metallicity degeneracy , irrespective of the colour - colour diagnostic diagram in use . therefore the results obtained in this section are not intended to provide a well - constrained picture for the stellar populations of these galaxies , which might require a deeper analysis involving colours in other wavelengths and/or spectra . in figure [ etg_ssp ] we show the colour distribution of the etg sample compared to the miuscat ssp models with ages ranging from 3 to 14gyr for three metallicity values : [ z / h]@xmath26 - 0.4 , 0 and 0.22 . we also include an arrow indicating the effect of internal dust extinction on the observed colours . we assume @xmath35 mag , in agreement with the typical value derived for low redshift lrg @xcite . the de - reddening in the various sdss bands has been computed with the extinction law : @xmath36 @xcite . we see that none of the ssp models is able to match the observed colour distribution . although the fiducial model for the formation of etgs predicts that they form in a uniform burst at high redshift with high metallicity , the old models tend to produce too red colours in the _ g@xmath18r _ vs _ u@xmath18 g _ plot . on the contrary , the _ r@xmath18i _ colours turn out to be too blue with respect to the real galaxies . hence , this simple picture does not seems to appropriate to describe the stellar populations of this galaxy sample . the miuscat models adopt the stellar isochrones of @xcite ( i.e. padova 2000 ) , as described in paper i. it has been claimed that these tracks provide blue colours for old elliptical galaxies . @xcite claimed that models computed with the isochrones of @xcite ( i.e. padova 1994 ) are preferred because redder v - k colours are predicted , in better agreement with the observations . for the optical spectral range the net effect of adopting these two sets of stellar tracks is almost negligible . using the padova 1994 tracks , @xcite obtained redder colours than using padova 2000 : _ u@xmath18 g _ by @xmath50.02 mag and _ g@xmath18r _ and _ r@xmath18i _ by @xmath50.01 mag . therefore these results tend to worsen the obtained mismatch for the bluer colours . + to understand better the origin of this mismatch and to avoid possible sample selection biases we have divided the etg sample in intervals according to their masses . in figure [ etg_sigma ] we show the colour distributions corresponding to galaxies with velocity dispersions above and below 180 . this value allows us to separate galaxies with different properties , while keeping almost equally populated these two galaxy mass ranges . etgs with high velocity dispersion are generally found to exhibit older ages and higher metallicities than their low velocity dispersion counterparts @xcite , resulting in redder colours . we see that none of the ssp models plotted in this figure is able to match the distribution of massive etgs . galaxies with lower velocity dispersion tend to be matched with models of lower metallicity and age @xmath238gyr , even though the predicted _ u@xmath18 g _ remain too blue ( @xmath230.1 mag ) . although a better agreement is obtained , the models do not provide fully satisfactory fits . the colour predictions computed by adopting ssp models from different authors are shown in figure [ etg_mod ] . we considered the bc03 and @xcite models , the two based on the basel theoretical stellar library @xcite and the @xcite models that use the @xcite empirical library . the bc03 , @xcite and @xcite adopt a salpeter imf , whereas the miuscat models are shown for both salpeter and kroupa imf , showing that the choice among these commonly used imfs has little effect on these colours . although none of the models can fairly reproduce the observed distribution , it is remarkable the advantage of using models based on empirical - based libraries . these models predict _ g@xmath18r _ bluer by 0.06 mag and _ r@xmath18i _ redder by 0.05 mag compared to the models based on theoretical libraries . the @xcite models predict slightly bluer _ g@xmath18r _ and redder _ r@xmath18i _ colours than the miuscat seds by @xmath50.01 mag , but also redder _ u@xmath18 g _ colour . here we explore how these colours vary when adopting different imf shapes and slopes . we consider the power - law unimodal imf defined in vazdekis et al . ( 1996 ) for different slopes for the solar metallicity ssp models with the same ages considered in figure [ etg_ssp ] . figure [ etg_imf ] shows a very modest variation of the bluer bands . however varying the imf turns out to have a significant effect on the _ r@xmath18i _ colour . indeed , by assuming a very steep imf slope ( @xmath37 ) the colour distribution in the _ g@xmath18r _ vs _ r@xmath18i _ diagram can be matched , though for relatively young ages ( 3 - 4gyr ) . on the other hand the _ g@xmath18r _ vs _ u@xmath18 g _ colour distribution is not reproduced for none of the represented models . we relax here the assumption that etgs are ssps and use stellar populations with two main bursts to model these galaxies . the first model tested is the same used for the description of the lrg colour evolution shown in section 5 . we include a small contribution of 5% to the total mass , with a primordial chemical composition ( [ z / h]=-2.32 ) and very old age ( 14gyr ) . this should resemble the first population of stars formed before being enriched by the subsequent star formation . as shown also by @xcite , the contribution of metal - poor stars is effective in bluening the _ g@xmath18r _ , without perturbing too much the _ r@xmath18i _ colour . the resulting miuscat two - burst colours are shown in figure [ etg_mp ] . as expected , all the colours become bluer , improving the match with the observed colour - colour distribution in the _ g@xmath18r _ vs _ u@xmath18 g _ diagram . the resulting _ r@xmath18i _ colour becomes slightly bluer , but since the effect of the metal - poor component is stronger on the bluer bands the colour distribution in _ g@xmath18r _ vs _ r@xmath18i _ improves as well , in agreement with the finding of @xcite . however the inclusion of such metal - poor component alone is not sufficient to fairly reproduce the observed colour - colour distributions . a similar effect is seen when a young component is included . in figure [ etg_yp ] the plotted model includes a small contribution ( 3% of the total mass ) of a young stellar population of 1gyr . in this case the effects on all the colours is more pronounced than in section [ sect_mp ] . this model can successfully explain the colour - colour distribution in the _ g@xmath18r _ vs _ u@xmath18 g _ diagram , but not the _ r@xmath18i _ colour , since the predicted values are far too blue ( @xmath50.06 mag ) . therefore , these models are not able to match the two colour - colour diagrams simultaneously . here we switch for the first time to the combination of more than two stellar populations . in sections [ sect_mp ] and [ sect_age ] we have seen that the combination of an old metal - rich stellar population with a small fraction of either a metal - poor or a young component has significant effects on the resulting colour - colour plots . furthermore in section [ sect_imf ] we tested the effects on the colours from non standard imfs . in fact , very recently @xcite showed spectroscopic evidences pointing towards steeper imfs in massive galaxies , in agreement with previous studies ( e.g. , @xcite ) . therefore it is worth to test the colours that result from combining all these stellar populations . an example of such scenario is the one derived in @xcite from the best fits to both broad - band colours ( from @xmath38 to @xmath39 ) and the lick line - strength indices , of a number of prototype massive galaxies , employing a full chemo - evolutionary version of our models @xcite . in this scenario the imf varies with time , being skewed towards high - mass stars during an initial short period of time ( @xmath340.5gyr ) that quickly enriches the interstellar medium and , afterwards , the imf becomes even steeper than the salpeter one . this is reproduced with a multiple bursts model by adding on top of the main episode of star formation , that forms 87% of the mass , two populations . in the first population 10 % of the mass formed at 14gyr with a top - heavy imf ( slope @xmath40=0.8 ) . half of the mass formed in this burst is associated with a metal - poor component , [ z / h]=-2.32 dex ( as in section [ sect_mp ] ) , and the other half has subsolar composition ( [ z / h]=-0.4 ) . another 3% of the mass is formed at the age of 1gyr ( as in [ sect_age ] ) . the stars formed in the main episode and at late times are skewed towards low - mass ( @xmath40=2.3 ) ( as in section [ sect_imf ] ) . a bimodal imf is assumed for all these models . the adoption of the imf with slope 2.3 in the main burst is consistent with recent evidences for a steep imf in massive galaxies @xcite . . the solid black line shows the relation found in @xcite , whereas the dotted line is the linear fit to the points , obtained with a least squares fit in the x direction . ] this model is shown in fig . 14 . for solar metallicity it is quite successful at matching the bulk of the galaxy distribution , although some problems persist in the _ r@xmath18i _ colour , which remains too blue . although quite unlikely , part of this discrepancy can also be attributed to internal extinction , leading too red colours that we can not fit with dust - free models . in order to explore all the possible combinations of sfhs that can match the two colour - colour distributions we have run a set of montecarlo simulations , where we allow all the stellar population parameters to vary : age , metallicity and imf slope . we only consider seven representative imf slopes : 0.3 , 0.8 , 1.3 , 1.8 , 2.3 , 2.8 and 3.3 . to keep the case as general as possible we use three bursts , where the fraction of mass involved in each of the burst is also a free parameter . figure [ etg_mc ] shows the results for 100000 simulations . on one hand we see that the cloud of models does not properly match the observed galaxy distribution in the _ g@xmath18r _ vs _ u@xmath18 g _ plot . on the other hand , the cloud of models matches the region enclosed by data in the _ g@xmath18r _ vs _ r@xmath18i _ plot . to understand which kind of ssp mixtures most likely resemble the data we selected the closest 1700 points to the median of the observed colours in each of the distributions . among these combinations we only choose those that are common to these two distributions to end up with 26 models , whose sfhs are shown in the appendix . we have ordered all these models following an increasing contribution of young stellar populations , i.e. , with ages smaller than 5gyr . we see that a large number of models require very important contributions from ssps with old ages , high metallicity ( solar or higher ) and imf slopes steeper than salpeter ( i.e. @xmath41 ) . all these models require smaller , but not negligible , contributions from young or / and metal - poor stellar populations . finally ssps with slopes lower than the salpeter value seem to be required for a number of models . a mean solution for these models can therefore be summarized as mainly composed of metal - rich and dwarf - dominated stellar populations with much smaller contributions from young or metal - poor components . interestingly the scenario discussed in section [ sect_chemo ] would be a good representative of this mean trend of solutions . in fact we nearly recover it in the models shown in the first and second rows of panels of figure [ sfh ] . whereas many studies support the presence of old ages and metal - rich stellar populations in massive galaxies ( see for example the review of @xcite ) , the requirement of non standard imfs is a matter of debate as recently pointed out by @xcite . among our solutions , the steeper imf slopes are recovered for the highest metallicity objects . in figure [ mu_met ] we show the mass weighted mean imf slope as a function of the mass weighted mean metallicity for the 26 solutions . the existence of a imf slope - metallicity relation has already pointed out by @xcite . interestingly , the linear fit to the simulated points lies close to the relation derived by the above authors by means of near - ir imf sensitive spectral features . it is worth to note that , using completely independent diagnostic techniques , it is inferred a similar imf - metallicity relationship in the sense that larger metallicities correspond to larger effective imf slopes . there are a number of models that depart significantly from the mean trend depicted above ( see last rows of figure [ sfh ] ) . these models require important contributions from stellar populations with ages smaller than @xmath65gyr . however we also see that most of these models also require steeper imf slopes than the models that are clearly dominated by old stellar populations . this can be explained by the fact that the colours redden with increasing age , metallicity or imf slope as shown in paper i. such degeneracy between these three parameters for the optical colours does not allow us to provide a unique solution . to constrain the solutions it is required to use spectroscopic data or / and colours in other spectral ranges ( e.g. , @xcite ) . however it is worth recalling that none of the sfhs explored in the simulations provide satisfactory fits to these two observed colour - colour distributions . indeed , the departure of the models from the median of the observed colours is non negligible , with the @xmath42 colour showing the largest residuals . it is beyond the scope of this paper to recover the sfhs of the sdss early - type galaxy sample , but the fact that in these complex models we allow to vary all , the age , metallicity and imf slope of the ssps , without reaching a fully satisfactory solution , supports the need for a fourth parameter to be taken into account in the models . we consider such a possibility in the next section . as explained in paper i the miuscat ssp seds employed here are computed on the basis of scaled - solar isochrones and stellar spectra with nearly scaled - solar abundance ratios for the high metallicity regime that is required to fit massive galaxies . however , it is well known that the spectra of these galaxies show an enhancement in [ mg / fe ] . to assess the effects of @xmath43-enhanced element partitions on the ssp colours we correct them , in a differential way , with the aid of the models of @xcite . these fully theoretical models are constructed on the basis of a high resolution library of synthetic spectra @xcite , where all the @xmath43 elements , including magnesium , are enhanced with respect to iron with the same proportion , i.e. , [ @xmath43/fe]=0.4 . we correct our miuscat ssp colours with the difference in colour obtained from the scaled - solar and @xmath43-enhanced models of @xcite , keeping constant the iron abundance . the corrected ssp colours are shown for different metallicities in figure [ etg_alpha ] . interestingly , the effect of @xmath43-enhancement tends to bluen the _ u@xmath18 g _ and _ g@xmath18r _ colours and to redden _ r@xmath18i _ , as required by the data . note that although this effect is relatively modest it certainly goes in the right direction for improving our fits to the most massive galaxies of our sample . preliminary tests performed with our own version of the models constructed with @xmath43-enhanced mixtures based on the miles library , making use of the [ mg / fe ] determinations of @xcite , lead to even larger differences with respect to the scaled - solar predictions in the same directions as with the models of @xcite . in the present work we have explored the constraints on the miuscat models provided by photometric data of globular clusters and quiescent galaxies . in paper i we have presented the new models and described the improvements . here , we have focused on the comparison of the model predictions with broad - band photometry . the integrated colours of mw globular clusters can be remarkably reproduced by the miuscat models in all the visible bands . only colours involving the u filter can not be used to constrain the models , as the uncertainty relative to the filter response used to reproduce the u magnitude is too high . the ability of our models for matching the colours of the mw stellar clusters in the other bands is remarkable . in particular , the b@xmath18v colour has always challenged models based on theoretical stellar libraries , which fail at matching the range covered by the @xmath44 filter . here we have shown that the problem can be solved by the use of empirical stellar libraries ( see also @xcite ) . the comparison with m31 globular clusters is less straightforward but it allows us to test the models even in the region of solar metallicity . the comparison with our models suggests that the m31 globular clusters are on average younger than usually assumed for the mw globular clusters . although an old model ( 12.6 gyr ) can fit the metal - poor clusters , we need ages in the range 6 - 10gyr to match the colours of the clusters with high metallicities . the spectral coverage of the miuscat library allows us to test here the models in several bands and , by relaxing the assumption that all the m31 gc are old , we can obtain a good fit to the data if we assume that metal - rich m31 clusters are younger than metal - poor ones . the fact that in mw clusters we do not need young models may be ascribed to the lack of metal - rich clusters . therefore our results disagree with previous works @xcite claiming that the _ g@xmath18r _ colours of m31 globular clusters could not be matched accurately with our models . nevertheless , the large amount of extinction in these clusters makes their colours very uncertain and not suitable for calibration purposes . when the models are compared to the colour evolution of lrgs , under the assumption of a passive evolution model , the observed colours are fairly reproduced . both , a simple model where stars formed in a single burst at high redshift , and a model that includes a small component of metal - poor stars , can be used to fit the observed distribution with increasing redshift . however , in the high redshift interval ( z @xmath45 0.4 ) where the observed colours probe the rest - frame _ g@xmath18r _ , the models predict redder colours than the median distribution . although the colour evolution of the lrgs can be fairly described within the fiducial model of purely passive stellar evolution , the combination of three colours , _ u@xmath18 g _ , _ g@xmath18r _ , _ r@xmath18i _ , poses a challenge . we find that the colour - colour diagrams based on these colours allow us to discriminate among different models . given the wavelength coverage of the miuscat models , we restrict our study to a limited redshift range , at [email protected] , where we can measure all these three colours . our study shows that single - burst models do not provide a good description for the local population of quiescent galaxies . irrespective of their metallicity , the old ssps provide too red _ u@xmath18 g _ and _ g@xmath18r _ colours and too blue _ r@xmath18i _ colour . we also find that the ssp model predictions from different authors do not match these colour - colour distributions . such disagreement has also been shown by @xcite . we note that the use of only two colours , _ u@xmath18 g _ and _ g@xmath18r _ does not allow to highlight the problem in the _ r@xmath18i _ colour , which goes in the opposite direction . nonetheless , an important result drawn from such a comparison is that models based on empirical stellar libraries ( miuscat ; @xcite ) predict colours much closer to the observed distribution than models based on theoretical spectra @xcite . we rule out that the inability of the miuscat ssp models for providing fully satisfactory fits to the colours of lrgs is due to flux - calibration issues . as shown in paper i , the miuscat spectra provide colours with photometric uncertainties @xmath46 0.02 mag . furthermore , the colours of our models are in good agreement with globular cluster data . we have tested the hypothesis that quiescent galaxies formed their stars in a strong burst at high redshift . our study based on the _ u@xmath18 g _ vs. _ g@xmath18r _ and _ r@xmath18i _ vs. _ g@xmath18r _ colour - colour diagrams suggest that the constituent stellar populations of local etgs are not necessarily fully coeval and old , and might require small contributions from either young or / and metal - poor stellar populations . we have explored several models with such complex stellar populations . we find that these smaller contributions are very effective at bluening the colours and matching the distribution in the _ u@xmath18 g _ and _ g@xmath18r _ colours . note however that the resulting _ r@xmath18i _ colour is too blue . the chemo - evolutionary model shown in @xcite provides one of the best fits to these two colour - colour diagrams . apart from the inclusion of a metal - poor population at early - times and a small burst at 1gyr , this model adopts a variable imf scenario . in this model the first generation of stars is skewed toward high - mass ( top - heavy imf ) , while at later times stars formed with a steeper imf slope . the adoption of such a steep imf is the main reason for the improvement also in the _ r@xmath18i _ colour . although this model is just one among several possible solutions , it clearly shows the need for a more complex sfh than the single burst model to explain the formation of etgs . moreover , the finding that quiescent galaxies have experienced a small amount of recent star formation is not new and has been revealed by several observational evidences @xcite and predicted by hierarchical models of galaxy formation @xcite . however , it is worth noting that although the colours obtained from these complex models are approaching the observed values , the agreement is not satisfactory . an experiment done by simulating three - burst montecarlo sfhs confirmed the impossibility of the three model parameters ( age , metallicity and imf slope ) to fairly match the color plane _ g@xmath18r _ vs _ u@xmath18g_. we are aware of the degeneracies affecting the optical colours , which do not allow us to properly constrain the solutions . it is beyond the scope of this work to fully constrain the sfhs , for which spectroscopic data or colors in other spectral range would be required , but the inability of these simulations to reproduce the observed colours suggests the need of another parameter to model the observations . we have identified that the inclusion of @xmath43-enhancement improves the fits . a limitation of the models based on empirical libraries is that they rely on stars from the solar neighborhood , which , most likely , has experienced a chemical enrichment that differs from that of massive elliptical galaxies . indeed , it is now well established that massive early - type galaxies are enhanced in magnesium with respect to iron , showing @xmath47 \simeq 0.2 $ ] @xcite . we have estimated the effect of @xmath43-enhanced models on these colours by applying differential corrections derived from the models of @xcite predictions on the miuscat colours . the overall effect is in the direction of improving the fits to the observations , as the colours in the bluer bands become bluer whereas the _ r@xmath18i _ tends to redden . therefore , the combination of both effects , a more complex sfh and the inclusion in the models of abundance ratios in agreement with the observations should provide better fits . we conclude that the impact of @xmath43-enhancement on the colours of massive galaxies can not be neglected to properly reproduce the observations . our forthcoming models taking into account the @xmath43-enhanced mixture of miles stars will clarify this issue . we are very grateful to paula coelho for providing us with her stellar population models . we thank the referee for helpful suggestions that improved this paper . ajc and jfb are _ ramn y cajal _ fellows of the spanish ministry of science and innovation . this work has been supported by the programa nacional de astronoma y astrofsica of the spanish ministry of science and innovation under grants aya2010 - 21322-c03 - 01 and aya2010 - 21322-c03 - 02 and by the generalitat valenciana under grant prometeo-2009 - 103 . 99 abazajian k. n. , et al . , 2009 , apjs , 182 , 543 barmby p. , huchra j. p. , 2000 , apj , 531 , l29 barmby p. , huchra j. p. , brodie j. p. , forbes d. a. , schroder l. l. , grillmair c. j. , 2000 , aj , 119 , 727 beasley m. a. , brodie j. p. , strader j. , forbes d. a. , proctor r. n. , barmby p. , huchra j. p. , 2005 , aj , 129 , 1412 bertelli g. , bressan a. , chiosi c. , fagotto f. , nasi e. , 1994 , a&as , 106 , 275 bessell s. , 1990 , pub . a.s.p . , 102 , 1181 bressan a. , et al . , 2006 , apj , 639 , l55 bruzual g. , charlot s. , 2003 , mnras , 344 , 1000 buser r. , kurucz r. l. , 1978 , a&a , 70 , 555 caldwell n. , schiavon r. , morrison h. , rose j. a. , harding p. , 2011 , aj , 141 , 61 cannon r. , et al . , 2006 , mnras , 372 , 425 cardelli j. a. , clayton g. c. , mathis j. s. , 1989 , apj , 345 , 245 carter d. , visvanathan n. , pickles a. j. , 1986 , apj , 311 , 637 cenarro a. j. , cardiel n. , gorgas j. , peletier r. f. , vazdekis a. , prada f. , 2001a , mnras , 326 , 959 cenarro a. j. , gorgas j. , cardiel n. , pedraz s. , peletier r. f. , vazdekis a. , 2001b , mnras , 326 , 981 cenarro a. j. , gorgas j. , vazdekis a. , cardiel n. , peletier r. f. , 2003 , mnras , 339 , l12 cenarro a. j. , et al . , 2007 , mnras , 374 , 664 charlot s. , fall s. m. , 2000 , apj , 539 , 718 coelho p. , barbuy b. , melndez j. , schiavon r. p. , castilho b. v. , 2005 , a&a , 443 , 735 coelho p. , bruzual g. , charlot s. , weiss a. , barbuy b. , ferguson j. w. , 2007 , mnras , 382 , 498 conroy c. , gunn j. e. , 2010 , apj , 712 , 833 de angeli f. , piotto g. , cassisi s. , busso g. , recio - blanco a. , salaris m. , aparicio a. , rosenberg a. , 2005 , aj , 130 , 116 de lucia g. , springel v. , white s. d. m. , croton d. , kauffmann g. , 2006 , mnras , 366 , 499 eisenstein d. j. , et al . , 2001 , aj , 122 , 2267 fan z. , ma j. , de grijs r. , zhou x. , 2008 , mnras , 385 , 1973 faber s. m. , french . h. , 1980 , apj , 235 , 405 falcn - barroso j. , snchez - blzquez p. , vazdekis a. , ricciardelli e. , cardiel n. , cenarro a. j. , gorgas j. , peletier r. f. , 2011 , a&a , 532 , a95 fernie j. d. , 1974 , pub . . , 86 , 837 galleti s. , bellazzini m. , buzzoni a. , federici l. , fusi pecci f. , 2009 , a&a , 508 , 1285 girardi l. , bressan a. , bertelli g. , chiosi c. , 2000 , a&as , 141 , 371 hayes , d. s. , 1985 , in d. s. hayes , l. e. pasinetti , & a. g. d. philip , eds . proc . iau symposium 111 , calibration of fundamental stellar quantities . dordrecht , reidel , p. 225 . harris w. e. , 1996 , aj , 112 , 1487 huchra j. p. , brodie j. p. , kent s. m. , 1991 , apj , 370 , 495 johnson h. l. , 1965 , apj , 141 , 923 kaviraj s. , et al . , 2007 , apjs , 173 , 619 lejeune t. , cuisinier f. , buser r. , 1998 , a&as , 130 , 65 maraston c. , 2005 , mnras , 362 , 799 maraston c. , stromback g. , portsmouth i .- o. , kingdom u. , 2011 , arxiv , arxiv:1109.0543 maraston c. , strmbck g. , thomas d. , wake d. a. , nichol r. c. , 2009 , mnras , 394 , l107 marn - franch a. , et al . , 2009 , apj , 694 , 1498 mendel j. t. , proctor r. n. , forbes d. a. , 2007 , mnras , 379 , 1618 milone a. d. c. , sansom a. e. , snchez - blzquez p. , 2011 , mnras , 414 , 1227 nair p. b. , abraham r. g. , 2010 , apjs , 186 , 427 oke j. b. , gunn j. e. , 1983 , apj , 266 , 713 pagel b. e. j. , 1997 , nceg.book , peacock m. b. , zepf s. e. , maccarone t. j. , kundu a. , 2011 , arxiv , arxiv:1105.3365 peacock m. b. , maccarone t. j. , knigge c. , kundu a. , waters c. z. , zepf s. e. , zurek d. r. , 2010 , mnras , 402 , 803 perrett k. m. , bridges t. j. , hanes d. a. , irwin m. j. , brodie j. p. , carter d. , huchra j. p. , watson f. g. , 2002 , aj , 123 , 2490 poole v. , worthey g. , lee h .- c . , serven j. , 2010 , aj , 139 , 809 pickles a. j. , 1998 , pasp , 110 , 863 renzini a. , 2006 , ara&a , 44 , 141 ricciardelli e. , franceschini a. , 2010 , a&a , 518 , a14 rosenberg a. , saviane i. , piotto g. , aparicio a. , 1999 , aj , 118 , 2306 snchez - blzquez p. , et al . , 2006 , mnras , 371 , 703 snchez - blzquez p. , gorgas j. , cardiel n. , gonzlez j. j. , 2006 , a&a , 457 , 809 sarzi m. , et al . , 2008 , aspc , 390 , 218 schiavon r. p. , 2007 , apjs , 171 , 146 schiavon r. p. , faber s. m. , rose j. a. , castilho b. v. , 2002 , apj , 580 , 873 stoughton c. , et al . , 2002 , aj , 123 , 485 tojeiro r. , percival w. j. , heavens a. f. , jimenez r. , 2011a , mnras , 413 , 434 tojeiro r. , percival w. j. , 2011b , mnras , 417 , 1114 trager s. c. , faber s. m. , worthey g. , gonzlez j. j. , 2000 , aj , 120 , 165 valdes f. , gupta r. , rose j. a. , singh h. p. , bell d. j. , 2004 , apjs , 152 , 251 van dokkum p. g. , conroy c. , 2010 , nature , 468 , 940 vazdekis a. , snchez - blzquez p. , falcn - barroso j. , cenarro a. j. , beasley m. a. , cardiel n. , gorgas j. , peletier r. f. , 2010 , mnras , 404 , 1639 vazdekis a. , cenarro a. j. , gorgas j. , cardiel n. , peletier r. f. , 2003 , mnras , 340 , 1317 vazdekis a. , 1999 , apj , 513 , 224 vazdekis a. , peletier r. f. , beckman j. e. , casuso e. , 1997 , apjs , 111 , 203 vazdekis a. , casuso e. , peletier r. f. , beckman j. e. , 1996 , apjs , 106 , 307 vazdekis a. , salaris m. , arimoto n. , rose j. a. , 2001 , apj , 549 , 274 vazdekis a. , ricciardelli e. , cenarro a.j . , rivero - gonzalez j.g . , diaz - garcia l.a . and falcn - barroso j. , 2012 , mnras , in press wake d. a. , collins c. a. , nichol r. c. , jones l. r. , burke d. j. , 2005 , apj , 627 , 186 wake d. a. , et al . , 2006 , mnras , 372 , 537 worthey g. , 1994 , apjs , 95 , 107 worthey g. , espaa a. , macarthur l. a. , courteau s. , 2005 , apj , 631 , 820 worthey g. , faber s. m. , gonzalez j. j. , 1992 , apj , 398 , 69 yamada y. , arimoto n. , vazdekis a. , peletier r. f. , 2006 , apj , 637 , 200 in this section we show the sfhs of the best solutions found in the montecarlo simulations described in section [ sect_mc ] . we selected the 1700 closest points to the median of the observed colours in each of the colour - colour plots . then we only considered the solutions that satisfy both distributions , ending up with 26 models , whose sfhs are shown in figure [ sfh ] . the sfhs are ordered according to an increasing fraction of mass in bursts younger than 5gyr ( the first panels show the solutions with the smaller fractions ) . the resulting sfhs can be divided into two broad categories . in the first group we can identify sfhs dominated by an old , metal - rich population , enriched with dwarf stars ( high imf slopes ) . on top of the old populations , we find small contributions from young populations . in some cases we do not see any significant young component . in these cases , g@xmath18r _ and _ u@xmath18 g _ colours result from the contribution of a metal - poor population ( see also section [ sect_mp ] ) . the other group of solutions includes sfhs with an important component of young populations ( @xmath24 5 gyr ) . in some cases , the fraction of mass formed in the youngest burst can be as large as 80% ( see sfh in the last row ) . as already shown in section [ sect_age ] , the age is the dominant parameter at decreasing both the _ g@xmath18r _ and _ u@xmath18 g _ colours and , thus , it provides a good match of the _ g@xmath18r _ vs _ u@xmath18 g _ plot . the sfhs with the largest component of young stellar populations also show the steepest imf slopes , which are needed in order to redden the _ r@xmath18i _ colour . this confirms the results found in sections 6.1 , 6.2 and 6.3 , where we have illustrated how the imf is mainly affecting the _ r@xmath18i _ while age and metallicity variations are needed to match the _ g@xmath18r _ vs _ u@xmath18 g _ colour distribution . given the age / metallicity / imf degeneracies affecting the optical colours , the two kind of sfhs produce similar colours and , hence , lead to the same level of agreement with the observed colour - colour distributions . there are many observational evidences , from spectroscopic and photometric studies favouring the first group of solutions for the formation of massive galaxies ( e.g. @xcite ) . indeed , galaxies that formed a considerable amount of stars very recently , as in the second group , would show strong @xmath48 absorption in the spectra that are not observed in the local population of massive ellipticals ( e.g. @xcite ) . interestingly , within the first group of models we can identify a number of solutions that resemble the ones found in section [ sect_chemo ] . a small burst at early times occurring with a flat imf and metal - poor stars is followed by a strong burst at nearly solar metallicity and steeper imf and then by a small burst at later times ( see first and second rows ) .

we present mid - infrared spectra of thirty two high redshift ultraluminous infrared galaxies , selected via the stellar photospheric feature at rest - frame 1.6 m , and an observed - frame 24 m flux ofjy . nearly all the sample reside in a redshift range of , and have rest - frame 1 - 1000 m luminosities of 10 - 10l . most of the spectra exhibit prominent polycyclic aromatic hydrocarbon emission features , and weak silicate absorption , consistent with a starburst origin for the ir emission . our selection method appears to be a straightforward and efficient way of finding distant , ir - luminous , star - forming galaxies in narrow redshift ranges . there is however evidence that the mid - ir spectra of our sample differ systematically from those of local ulirgs ; our sample have comparable pah equivalent widths but weaker apparent silicate absorption , and ( possibly ) enhanced pah 6.2m/7.7 m and 6.2m/11.2 m flux ratios . furthermore , the composite mid - ir spectrum of our sample is almost identical to that of local starbursts with ir luminosities ofl rather than that of local ulirgs . these differences are consistent with a reduced dust column , which can plausibly be obtained via some combination of ( 1 ) star formation that is extended over spatial scales of 1 - 4kpc , and ( 2 ) star formation in unusually gas - rich regions .

0801.1842

first discovered in the 1970s @xcite , ultraluminous infrared galaxies ( ulirgs , those objects with 1 - 1000@xmath0 m luminosities in excess of @xmath7l@xmath5 ) are among the brightest objects in the low redshift universe , with luminosities rivalling the bolometric output of quasars . their immense luminosities , implying astrophysical processes operating on extreme scales , have made ulirgs the object of intense study over the last thirty years , and in the last decade or so a nascent consensus has developed on their nature . local ulirgs are probably ` composite ' objects ; most are powered mainly by a starburst , but a significant fraction ( @xmath8 ) also contain an ir - luminous agn . this is suggested from several lines of evidence , including optical / uv spectroscopy @xcite , mid - infrared spectroscopy @xcite , and x - ray observations @xcite . local ulirgs are also associated almost exclusively with galaxy mergers @xcite , and may be involved to some degree in triggering qsos @xcite . reviews of the properties of ulirgs can be found in @xcite , and more recently in @xcite . initially , studies of ulirgs were motivated mostly by their luminosities , as they are rare at low redshifts ( with less than 50 examples known over the whole sky at @xmath9 ) , and so were not thought to play a fundamental role in the broader picture of galaxy evolution . this perception changed abruptly however , when it was realized that ulirgs are vastly more numerous at high redshift . first hinted at by spectroscopic followup of _ infrared astronomical satellite _ ( iras ) surveys @xcite , which showed strong evolution in the ulirg luminosity function with redshift , this was confirmed by surveys with the _ infrared space observatory _ ( iso @xcite ) and in the sub - mm @xcite , which showed that there were several hundred ulirgs _ per square degree _ at @xmath10 . later surveys with the _ spitzer _ space telescope @xcite showed that ulirgs contributed an increasing fraction of the comoving bolometric energy density with increasing redshift , reaching a level comparable to the contribution from lower luminosity systems at @xmath11 ( @xcite , their figure 14 ) . distant ulirgs seem to be similar to their low redshift counterparts , in that they appear to be powered by both starburst and agn activity @xcite , and are probably mostly mergers @xcite . the much higher space density of ulirgs in the high redshift universe compared to locally implies that they play an increasingly important role in galaxy evolution with increasing redshift ; for example , their high star formation rates make distant ulirgs excellent candidates for being the rapid growth phases of massive ( @xmath12 ) elliptical galaxies @xcite . the sheer number of ulirgs at high redshift however also raises some serious problems , notably for theories for the formation of large - scale structures . these theories usually invoke ` biased ' hierarchical buildup , in which overdense halos in the dark matter distribution undergo successive mergers to build haloes of increasing mass , with galaxies forming from the baryonic matter in these haloes @xcite . in this scenario , ulirgs , being the likely antecedents of massive systems , should be found in overdense peaks in the underlying dm distribution , and observations seem to bear this out @xcite . despite this , the prevalence of ulirgs at high redshift , and their very high inferred star formation rates , cause major problems for models as they find it difficult to concentrate enough baryons in sufficiently small volumes , and a diverse variety of solutions have been proposed to overcome this @xcite . there is therefore a pressing need for both an efficient census of ulirgs at @xmath11 , and a thorough understanding of how these distant ulirgs form stars . _ spitzer _ is ideal for undertaking such studies ; the infrared array camera ( irac , @xcite ) and multiband imaging photometer for spitzer ( mips , @xcite ) imaging instruments , and the infrared spectrograph ( irs , @xcite ) all offer dramatic improvements in sensitivity and resolution compared to previous generation facilities . in this paper , we use the irs to study 32 ulirgs at @xmath10 , selected using irac and mips photometry . we quantify the efficiency of the selection method and use mid - ir spectral features to set constraints on the nature of their star formation . an overview of the sample selection and a description of the observations is given in [ sectobs ] . results are summarized in [ takeaguess ] with discussion and conclusions given in [ discuss ] and [ conc ] respectively . we assume a spatially flat cosmology with @xmath13 km s@xmath14 mpc@xmath14 , @xmath15 , and @xmath16 . unless otherwise stated , the term ` infrared ( ir ) luminosity ' refers to the total luminosity integrated over 1 - 1000@xmath0 m in the rest - frame . the basis of our selection is an apparent emission feature at rest - frame 1.6@xmath0 m , produced by thermal emission from late - type stars and enhanced by an apparent emission feature caused by h@xmath17 ions in the atmospheres of giant stars is a major source of opacity in the atmospheres of such stars , but shows a minimum in opacity at around 1.6@xmath0 m @xcite , leading to a broad apparent emission feature at this wavelength ] . this 1.6@xmath0 m ` bump ' is almost ubiquitous in the integrated spectra of galaxies . previous authors @xcite have investigated the use of this feature in deriving photometric redshifts for galaxies using the spitzer irac channels , and conclude that it is likely to be effective . a 1.6@xmath0 m bump also implies that any agn continuum is not significant in the rest - frame near - ir , although it does not rule out the possibility of a low luminosity and/or very obscured agn . therefore , combining the presence of the 1.6@xmath0 m feature with a prior on ir luminosity should select high redshift ulirgs that harbor active star formation . a complete description of the selection method can be found in lonsdale et al ( 2008 , in preparation ) . angular clustering measurements for large samples of ` bump ' selected ulirgs can be found in @xcite ( see also @xcite ) , and millimeter - wave fluxes and analysis can be found in @xcite . other studies that include small numbers of ` bump ' sources include irs spectroscopy @xcite , and optical spectroscopy @xcite . other studies of the irac colors of swire galaxies can be found in @xcite . we select our sources from the swire @xcite lockman field , which spans 10.2 square degrees , and has been imaged in all seven irac and mips bands to 5@xmath18 depths of 4.2@xmath0jy , 7.5@xmath0jy , 46.1@xmath0jy , 47.3@xmath0jy , 210@xmath0jy , 18mjy and 108mjy respectively . we first excluded all sources brighter than @xmath19 to remove local and low redshift galaxies , and all sources fainter than 500@xmath0jy at 24@xmath0 m to ensure that spectra could be obtained with the irs in reasonable integration times . we then selected all sources that exhibited excess emission in the 4.5@xmath0 m channel , i.e. where @xmath20 and @xmath21 ( or ` bump 2 ' sources ) . this resulted in a total of 350 sources . we then randomly selected 31 objects for observation with irs . we also included one further object in the swire elais n1 field that satisfies all of our selection criteria , and was observed as part of a complementary program . the complete sample of 32 objects is listed in table [ sample ] . the irac and mips fluxes in this table are from the latest swire data release ( the ` swire3 ' catalogs ) . the sources were however originally selected using the previous generation , ` swire2 ' catalog . the changes to the irac pipeline between these two catalog versions resulted in flux differences between the catalogs of only a few percent , but our original selection did not demand that the bump was detected with any statistical robustness , and in some cases the band 2 excess was not significant within the errors . as a result , some of our sample no longer display an excess in the 4.5@xmath0 m channel using the swire3 catalogs . the irac fluxes of all of the sample are however consistent within the errors with an excess in the 4.5@xmath0 m channel , and for simplicity we treat all of the sources in table [ sample ] as ` bump 2 ' sources for the remainder of this paper . each source was observed with the long - low module onboard the irs ( @xmath22 , 5.1 pix@xmath14 , r@xmath23 ) in both first ( ll1 , 19.5@xmath0 m - 38.0@xmath0 m ) and second ( ll2 , 14.0@xmath0 m - 21.3@xmath0 m ) order , as part of spitzer program 30364 and ( for source 32 ) program 15 . the targets were placed in the center of each slit using the blue peak - up array . each target was observed for eight ramp cycles , with an individual ramp time of 120 seconds within the _ leopard _ software , available from the spitzer science center ] . the data were processed through the _ spitzer _ science center s pipeline reduction software ( version 15.3 ) , which performs standard reduction tasks such as ramp fitting and dark current subtraction , and produces basic calibrated data ( bcd ) frames . starting with these frames , we flagged rogue and otherwise ` bad ' pixels using the _ tool , which uses a pixel mask for each campaign to first flag and then replace rogue pixels . the individual frames at each nod position were then combined into a single image using the smart software package @xcite . sky background was removed from each image by subtracting the image for the same object taken with the adjacent order ( i.e. ` order - order ' sky subtraction ) . one - dimensional spectra were then extracted from the images using the spice software package using ` optimal ' extraction and default parameters for the long - low modules . this results in separate spectra for each nod and for each order , all of which must be combined to give the final spectrum . the spectra for each nod were inspected ; features present in only one nod were treated as artifacts and removed manually . the two nod positions were then combined . in most cases the two orders differed in flux by @xmath24 ; we first scaled ll2 to ll1 , and then scaled the resulting spectrum to the 24@xmath0 m flux . finally , the first and last 4 pixels on the edge of each order , corresponding to regions of decreased sensitivity on the array , were removed to give the final spectrum for each object . the spectra are presented in figures [ spectraa ] and [ spectrab ] . redshift data are presented in table [ measurements ] , and measurements of spectral features are given in table [ pahsils ] . we derive spectroscopic redshifts from broad emission features at 6.2@xmath0 m , 7.7@xmath0 m , 8.6@xmath0 m , 11.2@xmath0 m and 12.7@xmath0 m , attributed to bending and stretching modes in neutral and ionized pah molecules ( the 12.7@xmath0 m feature likely also contains a contribution from the [ neii]@xmath2512.814 fine - structure line ) , and/or a broad absorption feature at 9.7@xmath0 m arising from large silicate dust grains . in most cases ( 29/32 ) an unambiguous redshift can be determined from the pah features ; the associated uncertainty on these redshifts is governed by the variations in pah peak wavelengths seen in local galaxies , and is of order @xmath26 . in three cases however ( sources 11 , 15 and 27 ) the spectral features can not be reliably identified . for these three sources we derive tentative redshifts based on what are plausibly pah or silicate features . in these cases the error on the redshift is larger , @xmath27 , and the redshift itself should be treated with caution . a histogram of the redshifts is plotted in the left panel of figure [ zphotzspec ] . all of the sample lie in the redshift range @xmath28 , with all but three lying within @xmath29 . the redshift distribution is well fitted by a gaussian centered at @xmath30=1.71 and with a full width at half maximum of @[email protected] . an accurate measure of the ir luminosities is difficult due to the paucity of data at far - ir wavelengths . therefore , we initially simply estimate the range in ir luminosity within which our sample likely resides , and defer estimates of ir luminosities based on the irs spectra to [ nature ] . to do this , we normalize a suite of radiative transfer models for starbursts @xcite and agn @xcite to the 24@xmath0 m fluxes , and reject those models that exceed the irac fluxes , or 70@xmath0 m or 160@xmath0 m upper limits . the ir luminosities derived from the remaining models then constitute the probable range in ir luminosity of our sample . the resulting ir luminosity range is @xmath33l@xmath5 , placing our sample at the upper end of the ir luminosity range seen for low - redshift ulirgs the sample is faint at 24@xmath0 m , hence our spectra are of low signal - to - noise , and detailed spectral diagnostics are not possible . most of the spectra however show one or more pah features at 6.2@xmath0 m 7.7@xmath0 m 8.6@xmath0 m , 11.2@xmath0 m and ( depending on redshift ) 12.7@xmath0 m . the pah fluxes and equivalent widths ( ews ) were computed by integrating the flux above a spline interpolated local continuum fit ( for a description of the method see @xcite and @xcite ) . due to the low s / n and restricted wavelength range of the spectra , we can not correct for water ice and/or aliphatic hydrocarbon absorption , however the effect of this lack of correction is likely to be insignificant . the errors on the ews are large because the method used to measure them requires a spline fit to the underlying continuum , and the continua of our sample are in most cases only weakly detected . this method does however offer the advantage that our ews can be compared directly to those of local ir - luminous galaxies as measured by spitzer @xcite . we derive star formation rates from the combined luminosity of the 6.2@xmath0 m and 11.2@xmath0 m pah features , via the formula from @xcite : @xmath34 where @xmath35 is the star formation rate in solar masses per year , and @xmath36 is the pah 6.2@xmath0 m + 11.2@xmath0 m luminosity in watts . the resulting star formation rates are listed in table [ pahsils ] . as described in @xcite , the errors on these star formation rates are of order @xmath37 . we measured the strengths of the silicate features via : @xmath38 where @xmath39 is the observed flux density at rest - frame 9.7@xmath0 m , and @xmath40 is the underlying continuum flux at rest - frame 9.7@xmath0 m deduced from a spline fit to the continuum on either side . a complete description of the method used to measure @xmath41 can be found in @xcite and @xcite . the silicate feature is not detected ( at @xmath42 ) in any object individually . we also see no evidence for molecular hydrogen lines or any fine - structure emission lines in the individual spectra . the average spectrum for all 32 objects is shown in figure [ composite_all ] . all of the pah features between 6.2@xmath0 m and 12.7@xmath0 m are prominent . there is also evidence for two fine - structure lines ; the 12.7@xmath0 m pah feature appears asymmetric , likely due to the presence of [ neii]@xmath2512.814 , and there is a weak detection of what may be [ arii]@xmath256.99 . both these lines are commonly seen in local ulirgs @xcite . conversely , the average spectrum shows no evidence for significant apparent silicate absorption . we also see no signs of the h@xmath43,s(3)@xmath259.662 and h@xmath43,s(2)@xmath2512.275 lines , which are both common , though weak , in local ulirgs , or the [ siv]@xmath2510.511 line . these results do not change appreciably if we selectively coadd the spectra , examples of which are also shown in figure [ composite_all ] . the average of the ten spectra with the largest pah 6.2@xmath0 m ews shows no apparent silicate absorption , no molecular hydrogen features , but strong pah features with an asymmetric 12.7@xmath0 m pah feature . the average of the ten spectra with the ` strongest ' silicate absorption features is similar ; there is a hint of absorption , but it is not detected with any statistical significance . the redshifts of our sample have a remarkably tight distribution . all of the sample lie in the range @xmath28 , with all but three lying within @xmath29 . if we exclude these three outliers , then the mean redshift for our sample is @xmath2 ( assuming fwhm@xmath44 ) . this is significantly tighter than the redshift distribution of sub - mm selected galaxies , which have @xmath45 @xcite , and the redshift distribution for lyman break galaxies , which have @xmath46 @xcite , though a larger sample would be needed to confrm this . our redshift distribution is much tighter than that of sources found by selecting on 24@xmath0m/0.7@xmath0 m flux ratio @xcite , which span @xmath47 . combined with the constraint that the ir luminosities of our sample must ( for any reasonable sed shape ) be in excess of 10@xmath48l@xmath5 , we conclude that our selection method is an excellent way of finding distant ulirgs in specific , narrow redshift intervals . this narrow redshift range likely arises due to a combination of two factors ; the demand of an excess flux at observed frame 4.5@xmath0 m favors sources with a rest - frame 1.6@xmath0 m opacity feature at @xmath49 , and the demand that sources be bright at observed - frame 24@xmath0 m biases towards sources whose redshifts place the 7.7@xmath0 m and 8.6@xmath0 m pah features within the mips 24@xmath0 m band . this method offers many advantages over other methods for finding high redshift ulirgs . for example , it avoids all of the observational difficulties in assigning optical counterparts to sub - mm selected ulirgs ( see for example @xcite ) . remarkably , we have arrived at this redshift distribution without demanding that the ` bump ' be detected with any statistical robustness , and it is notable that none of the seven sources that ` lost ' their bump in the 4.5@xmath0 m channel due to reprocessing ( table [ measurements ] ) are unusual in redshift or spectral shape . given the difficulties in obtaining accurate redshifts for distant ulirgs @xcite , we examine the accuracy of photometric redshifts for our sample obtained by @xcite and listed in table [ measurements ] . given the high redshifts and limited optical photometry , the photometric redshifts are surprisingly good . ` reliable ' solutions ( those with @xmath50 and detected in at least three of the @xmath51 bands ) are obtained for 14/32 objects ; these are plotted in the right panel of figure [ zphotzspec ] . of those 14 , four are classified as ` catastrophic ' failures , i.e. those where the photometric redshift deviates from the spectroscopic redshift by more than @xmath52 in log(1+z ) ( although two of these four are sources 11 and 15 , which have uncertain spectroscopic redshifts ) , while ten are classified as successes . we conclude that the @xcite photometric redshift method remains effective even for faint , high redshift sources . finally , we note that , of the seven objects in table [ measurements ] with ambiguous bump classifications , six have photometric and spectroscopic redshifts that agree very well . it seems therefore that , for the sources with an ambiguous bump classification , most are indeed ` bump 2 ' sources . the irs spectra provide the best diagnostics of the power source behind the ir emission . the presence of prominent pah features in 29/32 objects is strong evidence that @xmath53 of our sample harbor powerful starbursts . the star formation rates computed using equation [ pahstarconc ] and listed in table [ pahsils ] are all extremely high . if we extrapolate these star formation rates to ir luminosities via the relation given by @xcitem luminosity rather than 1 - 1000@xmath0 m luminosity , but the difference in wavelength range is unlikely to be significant ] , then these luminosities lie in the same range as those determined in [ takeaguess ] . we infer that dust heated by star formation can account for all of the ir emission from the sources that show strong pahs , and therefore that most of our sample are consistent with being pure starbursts . interestingly , the strong pahs and high degree of homogeneity of our sample make them similar to sub - mm selected ulirgs @xcite , suggesting that we may be drawing from the same parent population . we can not , however , rule out the presence of an agn in any of our sample . it is unlikely that an agn contributes significantly in the rest - frame near - ir , as we see no power law continua in the irac bands ( see e.g. @xcite ) , but it is possible that some of our sample contain mid - ir luminous agn , even in those objects with prominent pahs . an example of such an object at low redshift is ngc 6240 , which shows strong pahs accompanied by an agn continuum and the [ nev]@xmath2514.322 fine - structure [email protected] has @xmath54 ev , and models indicate that it is unlikely to be strong in galaxies without an agn @xcite . observationally , [ nev]@xmath2514.32 is occasionally seen in planetary nebulae and young supernova remnants @xcite , but is not seen in the integrated spectra of starburst galaxies . it is however strong in the spectra of agn @xcite ] @xcite . furthermore , previous authors have noted correlations between shallow ( @xmath55 ) or deep ( @xmath56 ) silicate features , and the presence of agn ; @xcite show that ulirgs with shallow silicate features usually also show the [ nev]@xmath2514.322 line , @xcite find that qsos and seyfert 1 galaxies often show the silicate feature in emission , and @xcite find that many high redshift ulirgs with deep silicate absorption features also show powerful radio jets . the shallow silicate features of our sample are therefore consistent with ( though not supportive of ) the presence of an agn . finally , we examine the irac colors as a diagnostic of the ir emission . in figure [ colcol ] we plot 5.8@xmath0m/3.6@xmath0 m color against 8.0@xmath0m/4.5@xmath0 m color , together with the selection ` wedge ' for agn from @xcite , defined empirically using the spitzer colors of sdss qsos . this diagnostic is not , on its own , particularly useful , as our 4.5@xmath0 m ` bump ' selection already excludes much of the agn selection region , but it serves as a useful check . nearly all of our sample lie outside the agn selection region , and are consistent with the positions of the bulk of the galaxies in the main spitzer first look survey ( fls ) dataset . our sample lies some distance from the sdss qsos and sy1 galaxies in figure 1 of @xcite . based solely on the irac colors , the rest - frame near - ir emission from our sample is consistent with arising from starlight . the prevalence of starbursts in our sample means we can investigate the nature of star formation in high redshift ulirgs . we start by considering diagnostics based solely on the pah features . in figure [ pahlumratio ] we plot pah luminosities against pah flux ratios . the 6.2@xmath0 m and 7.7@xmath0 m pah features are extraordinarily luminous , lending further weight to the conclusion that the starbursts in our sample are more intense than seen in any local ( @xmath9 ) ulirg . the dispersion in pah luminosity is perhaps slightly greater for our sample , but not significantly so given the sizes of the error bars . we see a more interesting picture if we consider pah color - color plots ( figure [ pahfluxratio ] ) . considering first only the low redshift sources ; local ir - luminous galaxies show a distinct ` environmental segregation ' with ir luminosity , corresponding to elevated 6.2@xmath0m/11.2@xmath0 m and 7.7@xmath0m/11.2@xmath0 m pah ratios in ulirgs compared to lower luminosity starbursts and agn . such a segregation is also seen in hii regions and photo - dissociation regions , and the reasons for it are not clear @xcite . the two most likely reasons are pah ionization and pah molecular structure . a higher fraction of ionized pahs gives rise to stronger pah features at shorter wavelengths , as does a higher fraction of open , uneven pah molecules compared to smooth , compact pah molecules ( see figure 9 of @xcite for diagrams of these pah features ) . the drivers behind pah molecular structure are unknown , but pah ionization fractions can be affected in two ways ; ( 1 ) a harder interstellar radiation field ( isrf ) would ionize more pahsm range ] , and ( 2 ) decreasing the electron density would decrease the recombination rate , leading to a higher fraction of ionized pahs . it is therefore likely that local ulirgs have either a harder isrf , a lower electron density , or a higher fraction of open , uneven pah molecules than do lower luminosity systems . turning now to the bump sources ; our sample appears to lie preferentially toward one ` edge ' of the local ulirg distribution , corresponding to even further elevated 6.2@xmath0 m pah emission compared to the 7.7@xmath0 m and 11.2@xmath0 m features , though the error bars are large even for the average bump point . these two diagrams alone suggest that the star formation in our sample _ may _ differ in some way from that seen in local ulirgs , although it is difficult to say what the physical origin of those differences could be . one possibility is a lower extinction level in our sample compared to the mean for low redshift ulirgs . we move on to consider diagnostics based on both the pah emission features and the 9.7@xmath0 m silicate absorption feature . in figure [ fork ] we plot our sample on the ` fork ' diagram of @xcite . our sources occupy a small region on the lower right of this plot , corresponding almost entirely to class 1c , with a small number in classes 2c and 2b , though the large error bars on the pah ews mean our sources are in many cases also consistent with class 1a/1b . these classes label our sources as pah dominated , with relatively weak apparent silicate absorption . to check whether this is just a chance similarity between these specific spectral features , we plot in figure [ comp_fork ] the averaged spectrum from figure [ composite_all ] against the averaged spectra for the classes in the fork diagram . clearly , our averaged spectrum closely resembles the average spectrum for the class 1c sources . figure [ fork ] raises interesting questions on the nature of starburst activity in our sample . as described in [ select ] , our selection method biases towards ulirgs harboring ongoing star formation , and selects against those ulirgs containing a rest - frame near - ir luminous agn . our sample should therefore resemble the majority of local ulirgs , as most low redshift ulirg are powered mainly by star formation ( e.g. @xcite ) . furthermore , our demand for an excess flux in one of the irac channels should not remove many ` local type ' ulirgs from the sample , as many local ulirgs have a rest - frame sed that either has a near - ir excess , or is not obviously a power law . moreover , previous studies of high redshift , l@xmath57l@xmath5 ulirgs selected using _ spitzer _ data have shown that they seem to resemble local ulirgs ; for example , @xcite find that @xmath10 ulirgs selected on the basis of red 24@xmath0m/8@xmath0 m and 24@xmath0m/0.7@xmath0 m color often show pahs accompanied by an agn continuum and in some ( @xmath58 ) cases strong silicate absorption , and their figure 13 appears to place the bulk of their sample in the upper branch of figure [ fork ] . we might therefore expect that our sources should be distributed mainly along the upper branch in figure [ fork ] , along with the majority of local ulirgs , and previous small samples of spitzer selected high redshift ulirgs . this however is not the case ; our sources are located in a specific region , classifying them as starbursts with ( relatively ) weak apparent silicate absorption , and almost completely avoid the upper branch in figure [ fork ] , where most local ulirgs reside . there is no _ a priori _ reason why our relatively crude selection method should select against ulirgs with strong silicate absorption , and the almost complete absence of such absorption in our sample is peculiar , for two reasons . first , many local ulirgs have a near - ir excess coupled with strong silicate absorption @xcite . second , the 24@xmath0 m flux cut ( in combination with the demand for a rest - frame 1.6@xmath0 m ` bump ' ) selects in favor of sources with strong pahs , but among local ulirgs , systems with strong pahs but weak silicate absorption are rare - e.g. @xcite examine ulirg local spectra medianed by optical spectra type , and find strong pahs and prominent silicate absorption in the hii and liner composites , and weak pahs and weak silicate absorption in the seyfert composite . if our sample do not resemble local ulirgs , then what do they look like ? we compare in figure [ comp_brandl ] our average spectrum with three different starburst templates ; arp 220 @xcite , the nucleus of m82 @xcite , and the average spectrum of the local starburst galaxies from @xcite ( which have ir luminosities between 10@xmath59l@xmath5 and 10@xmath60l@xmath5 ) . clearly , the brandl et al spectrum is the closest match to our spectrum , even though the parent samples differ in ir luminosity by over two orders of magnitude . both show strong pahs with an asymmetric 12.7@xmath0 m pah profile , and negligible silicate absorption . interestingly , another class of high redshift galaxy that tends to show strong pahs but weak silicate absorption are the sub - mm galaxies @xcite . the m82 nuclear spectrum is also a good match , except for slightly stronger silicate absorption . the arp 220 spectrum on the other hand is a poor match , with much stronger silicate absorption . the similarity of our composite spectrum to that of local starburst galaxies allows us to estimate the ir luminosities for our sample using the relation between ir luminosity and 6.2@xmath0 m pah luminosity in @xcite : @xmath61 where @xmath62 is the flux of the pah 6.2@xmath0 m feature in units of @xmath63 w @xmath64 , @xmath65 is the luminosity distance in kpc , and @xmath66 is in units of solar luminosities . the ir luminosities for our sample computed using equation [ pahlirs ] are presented in table [ pahsils ] , and are in good agreement with the ir luminosities estimated in [ takeaguess ] . we explore the robustness of the conclusion that our sample have systematically weaker apparent silicate absorption than is seen in local ulirgs using a different approach ; fitting their broad band optical through ir seds ( i.e. the @xmath51 , irac and mips data ) with the code of @xcite , with the redshifts fixed at the spectroscopic values , and comparing the results from these fits with the irs spectra . we consider only those objects ( 4 , 12 , 13 , 15 , 17 , 21 , 29 , 31 ) with detections in at least six optical through mips bands , and specifically do _ not _ fit to the irs spectra , as we wish to test the conclusions independently . the limited available photometry ( less than nine points in all cases ) renders this approach crude compared to the analysis of the irs spectra , but it serves as a useful check . the results from this fitting are shown in figure [ mrrdiag ] . in four cases ( 4 , 13 , 21 29 ) , an m82 template gives an excellent fit to the photometry , and is well matched to the irs spectrum . in two cases ( 12 and 17 ) an m82 template gives a good fit to the photometry , but predicts a higher flux at rest - frame 6@xmath0 m than is seen in the irs spectrum , which may indicate a steeper mid - ir continuum . in the final two cases ( 15 and 31 ) there is marginal evidence for strong silicate absorption , consistent with an arp 220 template , though the low s / n of the irs spectra means we can not rule out an m82 template either . overall , the results from the template fitting agree with the results from the individual irs spectra , and from the comparison between our average spectrum and arp 220 , m82 , and the @xcite composite in figure [ comp_brandl ] . all of the evidence points towards an unexpected result ; despite the fact that our sample are ultraluminous in the ir , their mid - ir spectra have systematically lower silicate absorption than low - redshift ulirgs , and instead resemble the mid - ir spectra of local starbursts that are over two orders of magnitude less luminous in the ir . what makes this even more puzzling is that high redshift ulirgs appear to be physically similar to local ulirgs in many ways ; they show evidence for being mergers ( @xcite , though see also @xcite ) between massive , gas rich systems @xcite . we can think of six possible reasons to explain this difference : 1 - _ ir luminosity dependent silicate strength : _ our sample is significantly more ir - luminous than most local ulirgs , so we would see weak apparent silicate absorption if compact , dusty starbursts and agn exhibit weaker silicate absorption as they increase in ir luminosity , but remain otherwise identical . while we can not rule this possibility out , we can think of no plausible physical reason why this should be the case . 2 - _ a specific evolutionary stage : _ we may be selecting high redshift ulirgs at a specific evolutionary stage , when a compact but relatively old starburst ( @xmath67myr if we assume an exponentially decaying burst ) becomes less obscured as supernova winds clear dust from the star - forming regions . while this is in principle plausible , our selection method is relatively crude compared to others cited as selecting distant ulirgs at specific evolutionary stages , which rely on the presence of specific fine - structure absorption lines @xcite . it seems unlikely that our method should find all 32 of our sample at this specific stage . 3 - _ silicate poor dust : _ weak apparent silicate absorption could arise from dust that is unusually poor in silicates . if the dust is made mainly in asymptotic giant branch ( agb ) stars @xcite , then silicate poor dust could be made in two ways ; a low metallicity ( z@xmath68 ) would suppress the formation of silicate dust , and a steep or ` top - heavy ' stellar imf may shift more agb stars into the carbon dust producing mass range , thereby reducing the production of silicate dust @xcite . the metallicities of distant ulirgs are not well constrained @xcite , but it seems unlikely that the metallicities of our sample could be lowered to the extent required , as low metallicities also suppress pah features @xcite . a top heavy imf is more plausible , and is consistent with the demands from structure formation models ( e.g. @xcite ) . interestingly , if the imf were top heavy , it would result in a harsher radiation field , which ( as described in [ pahfluxsect ] ) may result in enhanced 6.2@xmath0 m pah emission compared to the longer wavelength pah features , a result hinted at in figures [ pahfluxratio ] and [ comp_brandl ] . there is however no direct evidence for a top - heavy imf in distant ulirgs . moreover , some authors have suggested there is no evidence for evolution in dust properties with redshift , at least up to @xmath69 ( e.g. @xcite ) . 4 - _ an intrinsically weak underlying continuum : _ if the underlying near / mid - ir continua in our sample are weak , then we would see three effects ; the rest - frame 1.6@xmath0 m feature would be prominent , the pah ews would be higher , and we would see weak silicate absorption as there would be no background continuum upon which to make such absorption apparent . a weak continuum would arise if our sample harbored lower luminosity agn than those seen in local ulirgs . in this scenario the absence of silicate absorption , and the differences between our sample and that of @xcite , arises due to selection bias , as our selection method favors systems harboring active star formation . it is also possible that the differences between our sample and that of @xcite arise because our sample is fainter at observed - frame 24@xmath0 m ( most of our sample fall below the 24@xmath0 m flux cut of 0.9mjy used by @xcite ) which , if true , would imply a ` transition ' in the mid - ir properties of high redshift ulirgs at @xmath70mjy . this possibility is simple , but does suffer from three drawbacks . first , as described in [ select ] , we do not demand that the 1.6@xmath0 m bump is detected with any statistical significance , and many of our sources have a weak 1.6@xmath0 m bump . our selection method therefore does not demand that the underlying continuum in the rest - frame near - ir is negligible , and therefore by extension does not demand that the rest - frame mid - ir continuum is absent either . second , as can be seen in figures [ spectraa ] and [ spectrab ] , several of our sample have significant mid - ir continua . third , several local ulirgs with strong mid - ir continua , in some cases accompanied by silicate absorption , also show a near - ir ` bump ' ( e.g. arp 220 , ngc6240 , @xcite ) , hence the presence of a bump does not necessarily select against objects in the upper branch of figure [ fork ] . 5 - _ gas - rich mergers : _ a simple way to reduce the apparent silicate strength is to lower the total extinction , which can be achieved by reducing the amount of dust in the starburst regions , or equivalently by increasing the amount of gas so that the same level of star formation can be sustained for a smaller dust mass . this scenario is compatible with the presence of strong pah emission , as changing the dust - to - gas ratio mainly affects the total dust column to the central ionizing source , whereas pah emission is thought to originate from the surface layers of star - forming regions . there is some support for this possibility from millimeter interferometry of sub - mm selected ulirgs at high redshift , which shows that they may be unusually gas - rich compared to local examples @xcite , and also from models for the star formation history of the universe , which suggest a lower mean @xmath71 at @xmath72 compared to locally ( e.g. @xcite , their figure 20 ) . 6 - _ starburst geometry : _ the final possibility is a different star - forming region geometry , such that the total dust column is reduced . this can be achieved in two ways ; either make the star formation diffuse and extended over several kpc instead of the sub - kpc scale starbursts seen in local ulirgs , or by distributing the star formation in multiple compact dusty star forming regions spread over a few kpc , instead of a single sub - kpc nuclear starburst . there is observational evidence for both these ` modes ' of star formation in distant starburst galaxies ; there is evidence for extended star formation from high resolution radio observations @xcite and ly@xmath73 imaging @xcite , while other authors have found evidence for star formation in multiple small regions distributed across a few kpc in forming massive galaxies at high redshift @xcite . furthermore , @xcite have recently suggested that disklike , ` inefficient ' star formation may be prevalent among distant massive galaxies . there is however an observational limit on the spatial extents of these starbursts ; high resolution millimeter imaging of small samples of sub - mm selected ulirgs at high redshift have shown that the starbursts in these systems can not be much more than 4kpc in diameter @xcite . it is difficult to discriminate reliably between these six possibilities , as the irs spectra can not formally rule any of them out . the first two possibilities however do not seem likely ; one would require some as - yet unknown dependence on luminosity for the silicate feature , while the other would require an extraordinary level of serendipity to find nearly all of the sample in the same evolutionary stage . the third possibility is more plausible , but would require a ` top - heavy ' imf , for which there is currently no direct evidence . the fourth possibility is also plausble , and some level of selection bias is undoubtedly present in our sample as we are not selecting solely on the basis of ir luminosity , but selection bias likely plays only a minor role , for the reasons given above . we tentatively propose therefore that some combination of the final two possibilities is the most likely origin for most ( but not all ) of this effect . it is worth noting that these conclusions may also apply to sub - mm selected ulirgs , given the similarity between their respective irs spectra . we have presented mid - infrared spectra , taken using the irs onboard spitzer , of 32 objects selected on ( 1 ) an optical band magnitude fainter than @xmath19 , ( 2 ) excess flux in the irac 4.5@xmath0 m passband compared to the other three irac channels , and ( 3 ) a 24@xmath0 m flux in excess of 0.5mjy . our conclusions are : 1 - all of the sample lie within @xmath28 , with more than 90% lying in a much tighter redshift range , characterized by @xmath2 . the rest - frame ir luminosities are consistent with the range @xmath74l@xmath5 . our selection method is therefore a straightforward and efficient way of finding large samples of distant ulirgs confined to narrow redshift ranges . it avoids all of the observational challenges inherent in finding distant ulirgs via their far - ir emission , and produces samples with redshift distributions akin to those of optically selected samples such as lyman - break galaxies . 2 - nearly all of the sample show two or more strong pah emission features in their mid - ir spectra , and systematically lack strong silicate absorption . the individual spectra closely resemble the irs spectra of high - redshift sub - mm selected ulirgs . the star formation rates computed from the pah luminosities are all extremely high , of order @xmath75m@xmath5yr@xmath14 , sufficient to power all of the ir emission . coupled with the lack of a power law continuum in the irac channels , we infer that star formation is likely to be the dominant power source in most of the sample , though it is possible that some also contain a luminous agn . 3 - our sample occupy a different region on the pah 6.2@xmath0 m ew vs silicate strength plane to most local ulirgs , and there is marginal evidence for enhanced pah 6.2@xmath0m/7.7@xmath0 m and 6.2@xmath0m/11.2@xmath0 m ratios . furthermore , the composite mid - ir spectrum of our sample resembles that of local starburst galaxies with ir luminosities of @xmath76l@xmath5 , rather than that of local ulirgs . though selection bias likely plays a role , we propose that the most likely reasons for this are that the star formation in distant ulirgs is extended over scales of a few kpc rather than the sub - kpc starbursts seen in local ulirgs , and/or occurs in unusually gas - rich environments . we thank the referee for a very helpful report . this work is based on observations made with the spitzer space telescope , which is operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . support for this work by the irs gto team at cornell university was provided by nasa through contract number 1257184 issued by jpl / caltech . support for this work was provided by nasa . this research has made extensive use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with nasa . mp acknowledges financial support from the marie - curie fellowship grant meif - ct-2007 - 042111 ccclcccccccccc 1 & swire3 j103205.16 + 574817.5 & 17420288 & 158.02150 & 57.804861 & 68 & 87 & 74 & 75 & 699 + 2 & swire3 j103707.79 + 591204.5 & 17422080 & 159.28250 & 59.201250 & 42 & 50 & 50 & 37 & 625 + 3 & swire3 j104011.60 + 580542.6 & 17416704 & 160.04837 & 58.095194 & 63 & 76 & 65 & & 751 + 4 & swire3 j104012.86 + 592712.4 & 17414912 & 160.05358 & 59.453472 & 150 & 150 & 116 & 101 & 838 + 5 & swire3 j104034.35 + 582314.5 & 17417472 & 160.14312 & 58.387389 & 42 & 49 & 54 & & 680 + 6 & swire3 j104129.25 + 581712.2 & 17421824 & 160.37192 & 58.286750 & 43 & 62 & 46 & & 561 + 7 & swire3 j104139.78 + 573723.8 & 17421056 & 160.41575 & 57.623306 & 40 & 52 & 48 & & 601 + 8 & swire3 j104232.04 + 575439.4 & 17418240 & 160.63350 & 57.910972 & 37 & 49 & & & 704 + 9 & swire3 j104343.93 + 571322.5 & 17414656 & 160.93304 & 57.222917 & 97 & 116 & 112 & 69 & 1009 + 10 & swire3 j104349.41 + 575438.7 & 17419776 & 160.95592 & 57.910750 & 78 & 102 & 72 & 62 & 700 + 11 & swire3 j104402.55 + 593204.6 & 17421568 & 161.01067 & 59.534639 & 63 & 73 & 77 & & 615 + 12 & swire3 j104427.54 + 593811.7 & 17418752 & 161.11475 & 59.636611 & 62 & 75 & 76 & 57 & 686 + 13 & swire3 j104436.55 + 593252.4 & 17416448 & 161.15233 & 59.547889 & 51 & 64 & 58 & 55 & 644 + 14 & swire3 j104514.38 + 575708.8 & 17415424 & 161.30996 & 57.952472 & 65 & 78 & 70 & & 874 + 15 & swire3 j104551.62 + 594234.8 & 17416192 & 161.46513 & 59.709694 & 49 & 55 & 49 & & 703 + 16 & swire3 j104614.90 + 594134.3 & 17420800 & 161.56208 & 59.692861 & 42 & 52 & & & 715 + 17 & swire3 j104627.82 + 592843.4 & 17420544 & 161.61596 & 59.478722 & 56 & 67 & 69 & & 607 + 18 & swire3 j104632.93 + 563530.2 & 17416960 & 161.63721 & 56.591750 & 76 & 91 & 68 & & 761 + 19 & swire3 j104643.29 + 575851.0 & 17415936 & 161.68037 & 57.980833 & 82 & 98 & 66 & 64 & 772 + 20 & swire3 j104653.07 + 592652.1 & 17421312 & 161.72117 & 59.447833 & 86 & 100 & 87 & & 637 + 21 & swire3 j104656.23 + 594008.0 & 17419264 & 161.73433 & 59.668889 & 50 & 67 & 57 & & 632 + 22 & swire3 j104754.66 + 583905.9 & 17418496 & 161.97775 & 58.651667 & 48 & 60 & 38 & & 688 + 23 & swire3 j104843.21 + 584537.8 & 17417728 & 162.18008 & 58.760500 & 48 & 62 & 48 & & 687 + 24 & swire3 j104845.19 + 561055.8 & 17419008 & 162.18833 & 56.182194 & 69 & 75 & & & 674 + 25 & swire3 j104922.64 + 564032.5 & 17417984 & 162.34438 & 56.675694 & 93 & 110 & 103 & 71 & 645 + 26 & swire3 j105056.08 + 562823.0 & 17419520 & 162.73371 & 56.473056 & 69 & 80 & 67 & 52 & 621 + 27 & swire3 j105152.73 + 564719.2 & 17417216 & 162.96975 & 56.788667 & 65 & 76 & 59 & & 590 + 28 & swire3 j105308.24 + 591447.5 & 17415680 & 163.28433 & 59.246528 & 75 & 88 & 79 & & 793 + 29 & swire3 j105334.59 + 574242.3 & 17414400 & 163.39417 & 57.711750 & 71 & 128 & 91 & 69 & 1108 + 30 & swire3 j105539.92 + 571711.8 & 17415168 & 163.91637 & 57.286639 & 74 & 100 & 85 & 66 & 863 + 31 & swire3 j105908.46 + 574511.4 & 17420032 & 164.78525 & 57.753167 & 60 & 74 & 71 & 56 & 688 + 32 & swire2 j161744.64 + 540031.4 & 16161792 & 244.43583 & 54.008722 & 62 & 88 & 68 & 69 & 1074 + clcc 1 & 2 & & 1.63 + 2 & 2/3 & 1.88 ? & 1.84 + 3 & 2 & & 1.83 + 4 & 1/2 & 1.18 & 1.43 + 5 & 2/3 & 1.73 & 1.89 + 6 & 2 & & 1.61 + 7 & 2 & & 1.69 + 8 & 2 & & 1.93 + 9 & 2 & & 1.71 + 10 & 2 & & 1.77 + 11 & 2/3 & 1.00 ? & 1.74 ? + 12 & 2/3 & 2.03 & 1.83 + 13 & 2 & 1.61 & 1.49 + 14 & 2 & & 1.78 + 15 & 2 & 1.30 & 0.76 ? + 16 & 2 & 0.58 & 2.11 + 17 & 2/3 & 1.63 & 1.80 + 18 & 2 & & 1.76 + 19 & 2 & 2.36 ? & 1.60 + 20 & 2 & & 1.69 + 21 & 2 & 1.61 & 1.66 + 22 & 2 & & 1.63 + 23 & 2 & & 2.23 + 24 & 2 & & 1.25 + 25 & 2 & & 1.70 + 26 & 2 & & 1.54 + 27 & 2 & & 1.88 ? + 28 & 2 & & 1.47 + 29 & 2 & 1.55 & 1.65 + 30 & 2 & & 1.68 + 31 & 2/3 & 1.42 & 1.65 + 32 & 2 & 1.81 ? & 1.80 + cccccccccc 1 & 3.88 @xmath78 1.13 & 1.01 @xmath78 1.14 & 5.67 @xmath78 1.53 & 0.53 @xmath78 0.26 & 2.68 @xmath78 0.58 & 0.63 @xmath78 0.20 & 0.00 @xmath78 0.45 & 1352 & 13.28 + 2 & @xmath794.13 & @xmath790.70 & @xmath795.82 & 0.43 @xmath78 0.31 & 2.53 @xmath78 0.70 & 0.48 @xmath78 0.19 & 0.35 @xmath78 0.53 & @xmath791850 & @xmath7913.46 + 3 & @xmath794.49 & @xmath795.39 & 8.62 @xmath78 1.97 & 0.77 @xmath78 0.28 & 3.99 @xmath78 0.95 & 1.44 @xmath78 1.01 & 1.07 @xmath78 1.00 & 2324 & 13.49 + 4 & 8.82 @xmath78 0.99 & 0.64 @xmath78 0.21 & 17.79 @xmath78 1.40 & 0.61 @xmath78 0.10 & 5.03 @xmath78 0.45 & 0.51 @xmath78 0.07 & 0.17 @xmath78 0.15 & 2067 & 13.53 + 5 & 2.47 @xmath78 0.95 & 2.66 @xmath78 14.15 & 5.27 @xmath78 1.35 & 0.58 @xmath78 0.23 & 1.55 @xmath78 0.77 & 0.37 @xmath78 0.32 & 0.24 @xmath78 0.62 & 1194 & 13.24 + 6 & 3.15 @xmath78 1.12 & 0.52 @xmath78 0.43 & 9.22 @xmath78 1.82 & 0.80 @xmath78 0.47 & 2.39 @xmath78 0.44 & 0.50 @xmath78 0.16 & 0.68 @xmath78 0.37 & 1108 & 13.16 + 7 & @xmath794.40 & @xmath793.16 & 5.96 @xmath78 1.46 & 0.69 @xmath78 0.36 & 1.72 @xmath78 0.46 & 0.39 @xmath78 0.14 & 0.06 @xmath78 0.43 & @xmath791378 & @xmath7913.39 + 8 & 4.18 @xmath78 1.36 & 0.91 @xmath78 1.04 & 3.98 @xmath78 0.79 & 0.34 @xmath78 0.09 & 2.65 @xmath78 0.70 & 0.52 @xmath78 0.20 & 0.87 @xmath78 0.39 & 2133 & 13.52 + 9 & 7.49 @xmath78 0.95 & 1.88 @xmath78 2.14 & 11.82 @xmath78 1.71 & 0.79 @xmath78 0.19 & 4.86 @xmath78 0.63 & 0.62 @xmath78 0.13 & 0.55 @xmath78 0.25 & 1160 & 13.27 + 10 & @xmath794.35 & @xmath790.64 & 7.52 @xmath78 1.85 & 0.77 @xmath78 0.32 & 2.95 @xmath78 0.61 & 1.38 @xmath78 1.24 & 1.19 @xmath78 0.80 & @xmath791844 & @xmath7913.44 + 11 & @xmath794.03 & @xmath790.83 & @xmath796.0 & @xmath794.00 & @xmath791.79 & @xmath790.34 & 0.02 @xmath78 0.40 & @xmath791410 & @xmath7913.38 + 12 & 3.46 @xmath78 1.05 & 1.02 @xmath78 1.39 & 4.48 @xmath78 0.84 & 0.38 @xmath78 0.10 & @xmath791.86 & @xmath790.41 & 0.67 @xmath78 0.43 & @xmath791458 & 13.37 + 13 & 4.72 @xmath78 1.38 & 0.63 @xmath78 0.59 & 4.62 @xmath78 1.53 & 0.22 @xmath78 0.08 & 2.30 @xmath78 0.42 & 0.25 @xmath78 0.06 & 0.20 @xmath78 0.26 & 1160 & 13.27 + 14 & 4.91 @xmath78 0.96 & 1.26 @xmath78 1.52 & 6.84 @xmath78 1.51 & 0.50 @xmath78 0.17 & 2.87 @xmath78 0.62 & 0.66 @xmath78 0.29 & 0.67 @xmath78 0.62 & 1991 & 13.50 + 15 & & & & & @xmath794.37 & @xmath791.49 & 1.89 @xmath78 0.43 & & + 16 & @xmath795.07 & @xmath791.23 & 2.80 @xmath78 0.93 & 0.22 @xmath78 0.08 & & & & & @xmath7913.72 + 17 & 1.53 @xmath78 0.54 & 0.63 @xmath78 1.09 & 3.94 @xmath78 1.84 & 0.57 @xmath78 0.33 & 2.33 @xmath78 0.00 & 0.59 @xmath78 0.00 & 0.01 @xmath78 0.44 & 1014 & 12.94 + 18 & 3.96 @xmath78 1.49 & 0.98 @xmath78 1.01 & 4.25 @xmath78 1.36 & 0.26 @xmath78 0.09 & 2.89 @xmath78 0.69 & 0.53 @xmath78 0.19 & 0.45 @xmath78 0.43 & 1706 & 13.38 + 19 & 4.94 @xmath78 1.06 & 0.95 @xmath78 0.75 & 9.65 @xmath78 1.98 & 0.59 @xmath78 0.18 & 2.99 @xmath78 0.65 & 0.44 @xmath78 0.14 & 0.27 @xmath78 0.51 & 1562 & 13.38 + 20 & @xmath794.16 & @xmath790.81 & @xmath796.00 & @xmath791.50 & 4.23 @xmath78 0.52 & 0.93 @xmath78 0.24 & 0.40 @xmath78 0.45 & @xmath793008 & @xmath7913.74 + 21 & 4.35 @xmath78 0.98 & 0.56 @xmath78 0.35 & 7.64 @xmath78 1.42 & 0.54 @xmath78 0.18 & 1.45 @xmath78 0.38 & 0.24 @xmath78 0.08 & 0.58 @xmath78 0.31 & 1423 & 13.42 + 22 & 5.85 @xmath78 1.10 & 7.34 @xmath78 40.91 & 8.28 @xmath78 1.59 & 0.58 @xmath78 0.21 & 3.89 @xmath78 0.66 & 0.71 @xmath78 0.23 & 0.54 @xmath78 0.52 & 2009 & 13.48 + 23 & @xmath794.86 & @xmath790.49 & 4.68 @xmath78 0.76 & 0.51 @xmath78 0.15 & & & & & @xmath7913.77 + 24 & & & 8.16 @xmath78 1.43 & 0.31 @xmath78 0.07 & 8.83 @xmath78 1.29 & 3.32 @xmath78 3.05 & 0.83 @xmath78 0.64 & & + 25 & @xmath793.10 & @xmath790.28 & @xmath795.00 & 0.25 @xmath78 0.11 & 2.39 @xmath78 0.68 & 1.06 @xmath78 0.66 & 1.34 @xmath78 1.34 & @xmath791253 & @xmath7913.22 + 26 & 8.39 @xmath78 1.21 & 0.65 @xmath78 0.22 & 7.22 @xmath78 1.47 & 0.28 @xmath78 0.07 & 5.13 @xmath78 0.75 & 0.66 @xmath78 0.15 & 0.51 @xmath78 0.35 & 2424 & 13.59 + 27 & @xmath793.31 & @xmath790.65 & @xmath794.80 & 0.41 @xmath78 0.22 & @xmath791.30 & @xmath790.26 & 0.01 @xmath78 0.43 & @xmath791281 & @xmath7913.35 + 28 & 3.49 @xmath78 1.19 & 0.47 @xmath78 0.45 & 5.97 @xmath78 1.23 & 0.31 @xmath78 0.09 & 2.96 @xmath78 0.40 & 0.35 @xmath78 0.06 & 0.00 @xmath78 0.24 & 1030 & 13.10 + 29 & 6.22 @xmath78 1.10 & 0.59 @xmath78 0.25 & 10.56 @xmath78 1.57 & 0.45 @xmath78 0.09 & 2.86 @xmath78 0.55 & 0.24 @xmath78 0.05 & 0.73 @xmath78 0.27 & 1929 & 13.53 + 30 & 4.50 @xmath78 1.12 & 1.08 @xmath78 1.06 & 5.77 @xmath78 1.62 & 0.33 @xmath78 0.11 & 2.40 @xmath78 0.50 & 0.33 @xmath78 0.09 & 0.37 @xmath78 0.32 & 1533 & 13.39 + 31 & 4.81 @xmath78 0.93 & 2.62 @xmath78 5.95 & 9.36 @xmath78 1.18 & 0.77 @xmath78 0.31 & 1.64 @xmath78 0.49 & 0.20 @xmath78 0.07 & 0.85 @xmath78 0.47 & 1370 & 13.40 + 32 & 5.80 @xmath78 0.90 & 0.36 @xmath78 0.10 & 4.40 @xmath78 0.96 & 0.26 @xmath78 0.07 & @xmath791.85 & @xmath790.19 & 0.02 @xmath78 0.50 & @xmath792017 & 13.60 + . the number in the top left of each panel corresponds to the i d number in the tables . the vertical dotted lines mark the wavelengths of the 6.2@xmath0 m , 7.7@xmath0 m , 8.8@xmath0 m , 11.2@xmath0 m and 12.7@xmath0 m pah features . [ spectraa],width=642 ] . _ right panel : _ comparison between the irs redshifts , and the photometric redshifts from @xcite , listed in table [ measurements ] . the dotted lines denote a deviation of 0.06 in log(1+z ) , the boundary defined as a ` catastrophic failure ' by @xcite . sources plotted as triangles have an uncertain irs redshift . sources plotted in grey have a photometric redshift derived from three bands . [ zphotzspec],title="fig:",width=321 ] . _ right panel : _ comparison between the irs redshifts , and the photometric redshifts from @xcite , listed in table [ measurements ] . the dotted lines denote a deviation of 0.06 in log(1+z ) , the boundary defined as a ` catastrophic failure ' by @xcite . sources plotted as triangles have an uncertain irs redshift . sources plotted in grey have a photometric redshift derived from three bands . [ zphotzspec],title="fig:",width=321 ] dispersion . the average of the ten objects with the largest pah 6.2@xmath0 m ews is plotted in red , and the average of the ten objects with the largest silicate strengths is plotted in blue . a ` p ' denotes the wavelength of a pah feature . [ composite_all],width=642 ] l@xmath5 - 10@xmath60l@xmath5 are plotted in red and blue , respectively . the large orange point is the mean value for our sample . [ pahfluxratio],title="fig:",width=321 ] l@xmath5 - 10@xmath60l@xmath5 are plotted in red and blue , respectively . the large orange point is the mean value for our sample . [ pahfluxratio],title="fig:",width=321 ] l@xmath5 are plotted in blue and red respectively . the large orange point is the average for all the bump sources . our sources occupy a small region , mostly the 1c class . this contrasts with local ulirgs , which are found in nearly all classes . [ fork],width=642 ] errors denoted by the grey region ) and averaged spectra for classes in the ` fork ' diagram of @xcite , normalized at 14@xmath0 m ; class 1a ( green ) , class 1b ( cyan ) , class 1c ( blue ) , class 2a ( yellow ) , class 2b ( magenta ) , class 2c ( purple ) , class 3a(red ) , and class 3b ( orange ) . [ comp_fork],width=642 ] 10@xmath3 - 10@xmath4l@xmath5 ) and three different starburst galaxy templates , all normalized at 6@xmath0 m ; the averaged spectrum of the starburst galaxies in @xcite ( blue , l@xmath8010@xmath59 - 10@xmath60l@xmath5 ) , arp 220 ( @xcite , green , l@xmath8110@xmath82l@xmath5 ) , and m82 ( @xcite , orange , l@xmath8110@xmath59l@xmath5 , this spectrum was taken with the iso - sws and hence is of higher spectral resolution than the other spectra ) . despite being approximately two orders of magnitude less luminous , the brandl et al spectrum is an excellent match to our average spectrum . m82 is also a reasonable match , except for slightly stronger apparent silicate absorption . arp 220 on the other hand is a very poor match , with much stronger absorption . [ comp_brandl],width=642 ]

the basic elements of the relativistic positioning systems in a two - dimensional space - time have been introduced in a previous work [ phys . rev . d * 73 * , 084017 ( 2006 ) ] where geodesic positioning systems , constituted by two geodesic emitters , have been considered in a flat space - time . here , we want to show in what precise senses positioning systems allow to make _ relativistic gravimetry_. for this purpose , we consider stationary positioning systems , constituted by two uniformly accelerated emitters separated by a constant distance , in two different situations : absence of gravitational field ( minkowski plane ) and presence of a gravitational mass ( schwarzschild plane ) . the physical coordinate system constituted by the electromagnetic signals broadcasting the proper time of the emitters are the so called _ emission coordinates _ , and we show that , in such emission coordinates , the trajectories of the emitters in both situations , absence and presence of a gravitational field , are identical . the interesting point is that , in spite of this fact , particular additional information on the system or on the user allows not only to distinguish both space - times , but also to complete the dynamical description of emitters and user and even to measure the mass of the gravitational field . the precise information under which these dynamical and gravimetric results may be obtained is carefully pointed out .

gr-qc0607037

a _ relativistic positioning system _ is defined by four clocks @xmath0 ( _ emitters _ ) in arbitrary motion broadcasting their proper times @xmath1 in some region of a ( four - dimensional ) space - time @xcite . the future light cones of the points @xmath2 constitute coordinate hypersurfaces @xmath3 of a coordinate system for that region . indeed , at every event of the domain , four of these cones broadcasting the times @xmath1 intersect , endowing thus the event with the _ emission coordinates _ @xmath4 the four proper time signals received by any observer at the event from the four clocks . as a _ location system _ ( i.e. as a physical realization of a mathematical coordinate system ) @xcite , the above positioning system presents interesting qualities and , among them , those of being _ generic _ , ( gravity-)_free _ and _ immediate_. this fact has been pointed out elsewhere @xcite and some explicit results have been recently obtained for the generic four - dimensional case @xcite . a full development of the theory for this generic case requires a hard task and a previous training on simple and particular situations . in @xcite we have presented a two - dimensional approach to relativistic positioning systems introducing the basic features that define them . we have obtained the explicit relation between emission coordinates and any given null coordinate system wherein the proper time trajectories of the emitters are known . we have also developed in detail the positioning system defined in flat space - time by geodesic emitters . finally , we have shown that , in arbitrary two - dimensional space - times , the data that a user obtains from the positioning system determine the gravitational field and its gradient along the emitters and user world lines . this two - dimensional approach has the advantage of allowing the use of precise and explicit diagrams which improve the qualitative comprehension of general four - dimensional positioning systems . moreover , two - dimensional scenarios admit simple and explicit analytic results . these advantages encourage us in proposing and solving new two - dimensional problems , many of them already suggested in a natural way by the results presented in @xcite . in the present work we want to detect the possibility of making relativistic gravimetry or , more generally , the possibility of obtaining the dynamics of the emitters and/or of the user , as well as the detection of the absence or presence of a gravitational field and its measure . this possibility is here examined by means of a ( non geodesic ) _ stationary positioning system _ , that is to say a positioning system whose emitters are uniformly accelerated and the radar distance from each one to the other is constant . such a stationary positioning system is constructed in two different scenarios : minkowski and schwarzschild planes . in both scenarios , and for any user , the trajectory of the emitters in the _ grid _ , i.e. in the plane @xmath5 are two parallel straight lines . at first glance , this fact would seem to indicate the impossibility to extract dynamical or gravimetric information from them , but this appearance is deceptive . we shall prove that the simple qualitative information that the positioning system is stationary ( but with no knowledge of the acceleration and mutual radar distance of every emitter ) and that the space - time is created by a given mass ( but with no knowledge of the particular stationary trajectories followed by the emitters ) allows to know the actual accelerations of the emitters , their mutual radar distances and the space - time metric in the region between them in emission coordinates @xmath6 also , the transformations to other standard coordinate systems may be obtained . the important point for gravimetry is that , in the above context , the data of the schwarzschild mass may be substituted by that of the acceleration of one of the emitters . then , besides all the above mentioned results , including the obtaining of the acceleration of the other emitter , the actual the schwarzschild mass of the corresponding space - time may be also calculated . these relatively simple two - dimensional results strongly suggests that relativistic positioning systems can be useful in gravimetry at least when parameterized models for the gravitational field may be proposed . a comment about the presentation of the results . the user is supposed to continuously receive data from the positioning system and have some a priori information about it and/or the space - time . because of their operational importance , the statements involving exclusively consequences of these two types of knowledge , have been emphasized and numbered . the work is organized as follows . in sec . [ section - previous ] we summarize the basic concepts and notation about relativistic positioning systems in a two - dimensional space - time . sections [ section - iii ] and [ section - iv ] are devoted to the analysis of stationary positioning systems respectively in minkowski and in schwarzschild planes . in sections [ section - gravimetry ] and [ section - grav - along ] we analyze the roles that different user s data play in obtaining the gravitational field and in recovering the characteristics of the positioning system . finally , we finish with a short discussion about the present results and comments on prospective work in sec . [ discussion ] . in a two - dimensional space - time , let @xmath7 and @xmath8 be the world lines of two clocks measuring their proper times @xmath9 and @xmath10 respectively . suppose they broadcast them by means of electromagnetic signals , and that the signals from each one of the world lines reach the other . this system of two clocks ( _ emitters _ ) is called a _ relativistic positioning system_. a domain @xmath11 in the region between both emitters constitutes a _ coordinate domain_. indeed , the past light cone of every event in @xmath11 cuts the emitter world lines at @xmath12 and @xmath13 , respectively . then @xmath14 are the coordinates of the event : the two proper time signals received by any observer at the event from the two clocks . we shall refer to them as the _ emission coordinates _ @xmath15 of the system in @xmath11 . also , by continuity , we shall call _ emission coordinates of the emitters _ the pair of proper times formed by the proper time of the clock in question and that received from the other clock . nevertheless , the signals @xmath9 and @xmath10 do not constitute coordinates for the events in the outside region . the plane @xmath16 ( @xmath17 ) in which the different data of the positioning system can be transcribed is called , as already remembered , the grid of the positioning system . in this grid , the trajectories of the two emitters ( which , for the auto - locating positioning systems considered below , can be drawn from the positioning data broadcast by the emitters ) define an interior region and two exterior ones . this interior region in the grid is in one - to - one correspondence with the interior region in the space - time , i.e. with the set of events that can be distinguished by the pair of times @xmath14 that reach them . but the exterior regions in the grid are not in one - to - one correspondence with the corresponding exterior regions in the space - time , because these last regions are foliated by null geodesics whose events have all the same values @xmath14 that the emission coordinates of the emitter that borders the region . so , the pairs @xmath14 of the exterior regions in the grid _ are not _ emission coordinates : they do not correspond to pairs of signals broadcast by the two emitters . consequently , for this emission protocol , they do not correspond to points of the two - dimensional space - time ( nevertheless , as mathematical extensions , they could be of interest in some situations ) . the grid origin depends exclusively on the choice of the instant zero of the emitter clocks . this synchronization is not a necessary prior information for positioning and it can be obtained from the system data . emission coordinates are null coordinates and thus the space - time metric depends on the sole _ metric function _ @xmath18 : @xmath19 an observer @xmath20 traveling throughout an emission coordinate domain @xmath11 and equipped with a receiver allowing the reading of the received proper times @xmath14 at each point of his trajectory , is called a _ user _ of the positioning system . from now on we will consider _ auto - locating positioning systems _ , which are systems in which every emitter clock not only broadcasts its proper time but also the proper time that it receives from the other , i.e. the systems that broadcast the emission coordinates of the emitters . thus , the physical components of an auto - locating positioning system are @xcite a _ spatial segment _ constituted by two emitters @xmath7 , @xmath8 broadcasting their proper times @xmath21 @xmath10 _ and _ the proper times @xmath22 , @xmath23 that they receive each one from the other , and a _ user segment _ constituted by the set of all users traveling in an internal domain @xmath11 and receiving these four broadcast times @xmath24 . it is worth remarking that any user receiving continuously the emitted times @xmath25 knows his trajectory in the grid . indeed , from these _ user positioning data _ @xmath25 the equation @xmath26 of the user trajectory may be extracted : @xmath27 on the other hand , any user receiving continuously the _ emitter positioning data _ @xmath24 may extract from them not only the equation ( [ user - trajectory ] ) of his trajectory , but also the equations of the trajectories of the emitters in the grid : @xmath28 eventually , the positioning system can be endowed with complementary devices . thus , in order to obtain the dynamic properties of the system , the emitters @xmath7 , @xmath8 could carry accelerometers and broadcast their acceleration @xmath29 , @xmath30 , meanwhile the users @xmath31 could be endowed with receivers able to read , in addition to the emitter positioning data , also the broadcast emitter accelerations @xmath32 . these new elements allow any user to know the acceleration scalar of the emitters : @xmath33 in some cases , it can be useful that the users generate their own data , carrying a clock to measure their proper time and/or an accelerometer to measure their proper acceleration @xmath34 . the user s clock allows any user to know his proper time function @xmath35 ( or @xmath36 ) and , consequently by using ( [ user - trajectory ] ) , to obtain the proper time parametrization of his trajectory : @xmath37 the user s accelerometer allows any user to know his proper acceleration scalar : @xmath38 thus , a relativistic positioning system may generate the _ user data _ : @xmath39 the emitter trajectories ( [ emitter - trajectories ] ) and the emitter accelerations ( [ emitter - accelerations ] ) do not depend on the user that receives them . thus , among the user data ( [ user - data ] ) we can distinguish besides the emitter positioning data @xmath40 , the _ emitter dynamical data _ @xmath32 , their union or _ public data _ @xmath41 and the _ user proper data _ @xmath42 . the purpose of the ( relativistic ) theory of positioning systems is to develop the techniques necessary to determine the space - time metric as well as the dynamics of emitters and users by means of physical information carried by light signals . but in order to develop it in known space - times , it is useful to connect our emission coordinates to the usual coordinates in which these space - times have been already studied . we finish this section by summarizing the explicit obtaining of the emission coordinates from an arbitrary null coordinate system @xmath43 in a generic two - dimensional space - time @xcite . let us assume the _ proper time history of two emitters _ to be known in this coordinate system : @xmath44 we can introduce the proper times as coordinates @xmath15 by defining the transformation to the null system @xmath43 given by : @xmath45 thus , relations ( [ coordinatechange0 ] ) define _ emission coordinates _ in the _ emission coordinate domains _ @xmath11 between both emitters . note that outside this region of domains the transformation ( [ coordinatechange0 ] ) also determines null coordinates , but they are not emission coordinates , i.e. they can not be constructed by means of broadcast signals @xcite . in emission coordinates , the emitter trajectories take the expression : @xmath46 where , from ( [ principalemit ] ) and ( [ coordinatechange0 ] ) , the emitter trajectories @xmath47 are given by : @xmath48 we consider here the positioning system defined by two non inertial stationary emitters @xmath7 , @xmath8 in absence of gravitation , that is to say , in minkowski plane ( two - dimensional flat space - time ) . thus , the emitters have a uniformly accelerated motion and they are at rest with respect to each other , i.e. the hyperbolic emitter trajectories have the same asymptotes ( these observers have been largely studied by rindler in @xcite ) . then we can choose inertial null coordinates @xmath43 such that the trajectories of the emitters are [ see fig . [ fig : accelerated-1](a ) ] : @xmath49 where @xmath50 , @xmath51 , is the acceleration parameter of the emitter @xmath0 . it is known that the space - like half - straight lines cutting at the coordinate origin determine the locus of simultaneous events for both emitters . thus , we can define a synchronization of the emitter clocks by giving their proper times @xmath52 and @xmath53 at two simultaneous events . with this synchronization , the proper time history of the emitters is [ see fig . [ fig : accelerated-1](a ) ] : @xmath54 \\[3 mm ] \,\displaystyle \texttt{v } = v_i(\tau^i ) = -\frac{1}{\alpha_i } \exp[-\alpha_i ( \tau^i- \tau^i_0 ) ] \end{cases}\ ] ] where the repeated index does not indicate summation . according to ( [ coordinatechange0 ] ) , the emission coordinates @xmath15 are defined by the transformation to the inertial system @xmath43 : @xmath55 \\[3 mm ] \ ! \ ! \ ! \ ! \displaystyle \texttt{v } = v_2(\tau^2 ) = -\frac{1}{\alpha_2 } \exp[-\alpha_2 ( \tau^2 - \tau^2_0 ) ] \end{array}\ ] ] and the inverse transformation is : @xmath56 \displaystyle \tau^2 = \tau^2_0 -\frac{1}{\alpha_2 } \ln ( -\alpha_2 \texttt{v } ) \end{array}\ ] ] from ( [ principalemitintaus0 ] ) and ( [ phis ] ) , we can obtain the expression of the trajectories in emission coordinates : [ iii - recti - trayec ] in minkowski plane , the trajectories @xmath7 , @xmath8 of the emitters of a stationary positioning system in emission coordinates @xmath15 are parallel straight lines of the form : @xmath57 the _ slope parameter _ @xmath58 , the _ separation parameter _ @xmath59 and the _ synchronization parameter _ @xmath60 are related to the acceleration parameters @xmath50 and the synchronization instants @xmath61 of the emitters by : @xmath62 \sigma&\equiv&\tau^1_0 - \frac{\alpha_2}{\alpha_1 } \ , \tau^2_0 \quad \label{iii - parametres - b}\end{aligned}\ ] ] fig . [ fig : accelerated-1](b ) illustrates this situation for a vanishing synchronization parameter . trajectories of the uniformly accelerated emitters in flat space - time . ( a ) in inertial null coordinates @xmath63 . ( b ) in the grid @xmath15 ; in this picture we have chosen the synchronization parameter @xmath64.,scaledwidth=48.0% ] let us consider a user traveling on the emission domain @xmath11 defined by this positioning system and receiving the emitter positioning data @xmath40 . firstly , these data , under the form of the two sequences @xmath65 and @xmath66 determine the emitter trajectories @xmath67 on the grid which , according to statement [ iii - recti - trayec ] , will be straight lines , from which the user can extract the slope , separation and synchronization parameters @xmath68 , @xmath69 and @xmath60 ( see ( [ iii - principaltaus1 ] ) and ( [ iii - principaltaus2 ] ) ) . in fact , these parameters can be extracted from the emitter positioning data at two different events : [ iii - two - events ] in terms of the emitter positioning data @xmath70 and @xmath71 at two different events @xmath72 and @xmath73 of any user , the slope , separation and synchronization parameter @xmath68 , @xmath69 and @xmath74 characterizing the parallel trajectories of the emitters in the grid @xmath16 , are given by : @xmath75 \displaystyle q = \frac12 [ \tau^1_p + \bar{\tau}^1_p + \frac{\delta\bar{\tau}^1}{\delta\tau^2}(\tau^2_p + \bar{\tau}^2_p ) ] \\[3.5 mm ] \displaystyle \sigma = \frac12 [ \tau^1_p + \bar{\tau}^1_p - \frac{\delta\bar{\tau}^1}{\delta\tau^2}(\tau^2_p + \bar{\tau}^2_p ) ] \end{array}\ ] ] once these parameters are evaluated , expression ( [ iii - parametres - a ] ) implies that @xmath58 and @xmath59 determine one - to - one the acceleration of the emitters and ( [ iii - parametres - b ] ) shows that the parameter @xmath60 is necessary in obtaining the emitter synchronization : [ iii - alfas - taus ] for stationary positioning systems in minkowski plane , the accelerations @xmath29 , @xmath30 of the emitters may be obtained in terms of the slope data parameter @xmath68 and the separation data parameter @xmath76 as : @xmath77 and their synchronization times @xmath52 and @xmath53 indicated by the emitter clocks at two simultaneous events , are related to @xmath58 and the synchronization data parameter @xmath60 by @xmath78 this statement gives the parameters @xmath29 and @xmath30 , as well as the relation between @xmath52 and @xmath79 in terms of the data parameters @xmath80 @xmath59 and @xmath60 , which can be extracted from the data received by the user . apart from their specific physical meaning , as accelerations and synchronization times of the emitters , these parameters @xmath81 @xmath82 @xmath52 and @xmath79 yield an operational definition of the null canonical inertial coordinates @xmath43 associated with the emitters , i.e. those with origin in the intersection of the null asymptotes of the stationary system to which the emitters belong . this operational definition is offered by relations ( [ iii - change - a ] ) . every set @xmath43 of null coordinates so determined by ( [ iii - change - a ] ) is one - to - one related to a standard inertial coordinate system @xmath83 by @xmath84 @xmath85 , and inertial systems of fixed origin are related by a one - parametric lorentz transformation . it is such a parameter ( say @xmath86 here ) that relations ( [ iii - acceleracions ] ) and ( [ iii - sigma ] ) leave free for given data parameters @xmath58 , @xmath59 and @xmath87 now , let us analyze the metric information that any user can obtain from the emitter positioning data . the expression of the metric function in emission coordinates follows from the coordinate transformation ( [ iii - change - a ] ) : @xmath88\ ] ] consequently , by using ( [ iii - acceleracions ] ) , we can write the metric function in terms of the data @xmath25 and the parameters @xmath89 : [ iii - prop - metric ] in minkowski plane , the space - time metric in emission coordinates of a stationary positioning system is given , in terms of the emitter positioning data @xmath90 by : @xmath91 d \tau^1 d \tau^2\ ] ] where the data parameters @xmath58 , @xmath59 , @xmath60 can be obtained by ( [ iii - par - taus ] ) in terms of the values of these data at two events . this statement shows that the sole public quantities @xmath92 received by any user allow him to know the space - time metric everywhere . nevertheless , it is worth remarking that , in order for a user to obtain this result , the user must be informed that _ the emitter positioning data come from a stationary positioning system in absence of gravitational fields_. otherwise , the above deduction based on the existence of the coordinate transformation ( [ iii - change - a ] ) would not be valid . remark also that this required information does not demand the specific value of the accelerations of the emitters , which may be also determined by the user by relations ( [ iii - acceleracions ] ) . the users may know that the system of the two emitter clocks is stationary by an _ a priori information _ , forming part of the dynamical characteristics of the positioning system and its foreseen control . but they may also obtain this _ information in real time _ , if clocks and users are endowed with devices allowing the users to know the emitter accelerations @xmath32 at every instant ( see comments in sections [ section - gravimetry ] and [ discussion ] and ref . @xcite ) in any case , the user information of the dynamics of the pair of clocks by any of these two methods is generically necessary , because _ emitter trajectories that are parallel straight lines in the grid are not necessarily uniformly accelerated trajectories in a flat space - time _ @xcite . in sec . [ section - iv ] we show that in a non flat space - time we can also find in the grid the same picture that fig . [ fig : accelerated-1](b ) . the considerations that follow in this section are qualitatively independent of the synchronization of the emitters . for this reason , we assume for the sake of simplicity the synchronization parameter to be zero , @xmath64 . all the above considerations are valid for any user @xmath31 which , as we have already said , can also obtain from the user positioning data @xmath25 his own trajectory in the grid . then , the expression ( [ iii - metric ] ) of the metric tensor allows him to calculate as usual his proper time lapse and acceleration in terms of the emitter positioning data @xmath93 . suppose now that a user obtains from these data the following trajectory : @xmath94 that is , a straight line parallel to the emitter trajectories [ see fig . [ fig : accelerated-2](a ) , where we plot the case @xmath95 . what is his dynamics and his relation to the emitters ? the answer is given by [ see fig . [ fig : accelerated-2](b ) ] : [ iii - prop - user - estat - acc ] the users which find that their trajectory in the grid is a straight line parallel to the emitter ones [ eq . ( [ iii - user1-a ] ) ] , are the users which are stationary with respect to the emitters . their acceleration is the weighted geometric mean of the emitters accelerations : @xmath96 and their proper time lapse @xmath97 is , for any of the two emitter clocks ( no summation ) @xmath98 to obtain these results , let us construct as above indicated the canonical inertial coordinate systems @xmath43 associated with the emitters . the transformation equations from the emission coordinates @xmath25 to these inertial coordinates are given by ( [ iii - change - a ] ) . in them , using the inverse transformation ( [ iii - change - b ] ) , the trajectory equation ( [ iii - user1-a ] ) becomes : @xmath99 where the constant @xmath34 is given by ( [ iii - user1-b ] ) . this means that this trajectory belongs to the stationary family of trajectories that contains that of the two emitters , and that its constant acceleration is @xmath100 to finish the proof of the statement [ iii - prop - user - estat - acc ] we consider the proper time parametrization of the stationary user ( [ iii - user1-bb ] ) : @xmath101 \\[3 mm ] \,\displaystyle \texttt{v } = v(\tau ) = -\frac{1}{\alpha } \exp[-\alpha ( \tau- \tau_0 ) ] \end{cases}\ ] ] then , ( [ iii - user1-bbb ] ) and the coordinate transformation ( [ iii - change - a ] ) give the expression ( [ iii - user - tau ] ) of the proper time lapse of the user . ( a ) the user positioning data @xmath15 determine the trajectory of the user @xmath31 in the grid : @xmath102 . the parallel straight lines @xmath103 are the locus of simultaneous events for the uniformly accelerated emitters . ( b ) trajectory of the user @xmath31 in inertial null coordinates @xmath63 : he also has an hyperbolic motion.,scaledwidth=48.0% ] in order to better understand the grid associated with the emission coordinates of our two - dimensional stationary positioning systems , let us consider the locus of simultaneous events for both emitters , i.e. the orthogonal lines to the stationary family of trajectories that contains that of the two emitters which , as it is well known , are the space - like geodesics cutting at the common point of the two asymptotes of the emitters . how can they be constructed in the grid ? the answer is : [ iii - prop - simultaneous ] the locus of simultaneous events for the emitters of a stationary positioning system in minkowski plane are parallel straight lines of slope @xmath104 , the same , up to sign , as that of the trajectory of the emitters . to see this , take into account that in inertial null coordinates these lines are @xmath105 . it is then sufficient to change to emission coordinates using ( [ iii - change - a ] ) to obtain : @xmath106 which shows directly the statement [ see fig . [ fig : accelerated-2](a ) ] . to finish this section , let us consider the user positioning data received by an inertial ( geodesic ) user . for simplicity , but with no qualitative differences , we shall choose him to be at rest with respect to the emitters at the initial instant at which @xmath107 ( remember that we have taken @xmath64 ) . for him we have : [ iii - prop - inertial ] the trajectory in emission coordinates of inertial users initially at rest with respect to two synchronized stationary emitters is : @xmath108 \ , , \ ] ] where @xmath109 is a constant . this equation may be obtained as follows . choose the null inertial system @xmath43 associated with the inertial observer having the same initial instant as the two emitters , that is to say , such that @xmath110 because the user is at rest with respect this null inertial system , its trajectory is given by [ see fig . [ fig : accelerated-3](a ) ] : @xmath111 then , making use of the coordinate transformation ( [ iii - change - a ] ) , a straightforward calculation leads to expression ( [ iii - user - geo ] ) , with @xmath112 . note that the coordinate @xmath10 run in @xmath113 -\frac{1}{\alpha_2 } \ln a , + \infty[$ ] when @xmath9 run in @xmath113 - \infty , \frac{1}{\alpha_1 } \ln ( a/\omega)\ [ $ ] . thus the data @xmath25 of a geodesic user lead to a trajectory in the grid with a double asymptotic behavior [ see fig . [ fig : accelerated-3](b ) ] . trajectory of a geodesic user @xmath31 with @xmath114 : ( a ) in inertial null coordinates @xmath63 . ( b ) in the grid @xmath15 defined in flat space - time by two uniformly accelerated emitters.,scaledwidth=48.0% ] in this section we study stationary positioning systems defined by two emitters at rest with respect to the gravitational field in _ schwarzschild plane_. the usual expression for the space - time metric in the exterior region to the schwarzschild radius @xmath115 is : @xmath116 we can introduce the stationary time @xmath117 and the radial coordinate @xmath118 from the horizon , both in units of schwarzschild radius : @xmath119 to obtain a conformally flat expression for the metric , one can change the radial coordinate @xmath118 for the new one @xmath120 : @xmath121 the inverse function that gives the coordinate @xmath118 in terms of @xmath120 may be expressed by using the lambert function @xmath122 , which is defined as the inverse of the function @xmath123 . indeed , from ( [ iv - x - rho ] ) we have : @xmath124 finally , one can define the _ stationary _ null coordinates @xmath63 [ see fig . [ fig : schwarzschild-1](a ) ] : @xmath125 \texttt{v } = t - x \ & \qquad \qquad x = \frac12 ( \texttt{u } - \texttt{v } ) \ \end{array}\ ] ] with all these transformations one obtains : _ in the exterior region , the metric of schwarzschild plane takes the expression : @xmath126 b(\rho ) \equiv \frac{\rho}{\rho + 1 } \ , , & \ & \rho \equiv w\left(\exp\left[\frac{\texttt{{\rm u } } - \texttt{{\rm v}}}{2}\right]\right ) \label{iv - b - rho - x}\end{aligned}\ ] ] where @xmath127 is the lambert function . _ the trajectories of two stationary emitters @xmath0 are defined by the conditions [ see fig . [ fig : schwarzschild-1](a ) ] : @xmath128 the method exposed in sec . [ section - previous ] to introduce emission coordinates is based in obtaining the metric tensor in a null coordinate system , as expressions ( [ iv - metric - uv ] ) and ( [ iv - b - rho - x ] ) do , and in knowing the proper time history of the emitters in this system . in coordinates @xmath63 for the two trajectories ( [ iv - rho - is ] ) one has , by ( [ iv - x - rho ] ) and ( [ eq - uvtx ] ) , @xmath129 and then from ( [ iv - metric - uv ] ) the unit character of their velocity gives @xmath130 . consequently , the two trajectories are given by ( the repeated index does not indicate summation ) : @xmath131 \ , \texttt{v } = v_i(\tau^i ) = \lambda_i ( \tau^i- \tau^i_0 ) - x_i\end{cases}\ ] ] where the constants @xmath132 and @xmath133 depend exclusively on the schwarzschild radius @xmath115 and the radial parameter @xmath134 : @xmath135 and where @xmath61 are the proper times that indicate the emitter clocks at two events with the same schwarzschild time @xmath117 . it is important to remark that _ the radial parameter @xmath134 is a dynamical characteristic of the trajectory , equivalent to the acceleration parameter @xmath50_. indeed , as the emitter acceleration scalars are constant because emitter trajectories are tangent to a killing vector , a standard computation of them gives : @xmath136 is is easy to see that this relation admits a unique real positive solution @xmath137 showing their one - to - one equivalence and , thus , the stated dynamical character of @xmath138 according to ( [ coordinatechange0 ] ) , the emission coordinates @xmath15 are defined by the transformation to the stationary null system @xmath43 : @xmath139 \texttt{v } = v_2(\tau^2 ) = \lambda_2 ( \tau^2 - \tau^2_0 ) - x_2 \end{array}\ ] ] and the inverse transformation is : @xmath140 \displaystyle \tau^2 = \tau^2_0 + \frac{1}{\lambda_2}(\texttt{v } + x_2 ) \end{array}\ ] ] trajectories of stationary emitters in schwarzschild plane . ( a ) in ` stationary ' null coordinates @xmath63 associated with the static time @xmath117 and the coordinate @xmath141 . the stationary observers are defined by @xmath142 . ( b ) in the emission grid @xmath15.,scaledwidth=48.0% ] making use of ( [ phis ] ) , we can obtain the expression of the emitter trajectories in emission coordinates @xmath15 : [ iv - recti - trayec ] in schwarzschild plane , the trajectories @xmath7 , @xmath8 of the emitters of a stationary positioning system in emission coordinates @xmath15 are parallel straight lines of the form : @xmath143 \gamma_2 \equiv \begin{cases } \tau^1 = \omega \tau^2 - q + \sigma \equiv \varphi_2(\tau^2)\\ \tau^2 = \tau^2\end{cases } \quad \label{iv - emitt - taus - b}\end{aligned}\ ] ] the _ slope parameter _ @xmath58 , the _ separation parameter _ @xmath59 and the _ synchronization parameter _ @xmath60 are related to the schwarzschild radius @xmath115 , the radial parameters @xmath134 and the synchronization instants @xmath61 of the emitters by : @xmath144 \displaystyle q = \frac{x_1 - x_2}{\lambda_1 } = r_s q(\rho_1 , \rho_2)\\[3.5 mm ] \sigma = \tau^1_0 - \omega \tau^2_0 \end{array}\ ] ] @xmath145 and @xmath146 being the bi - parametric expressions : @xmath147 \displaystyle q(\rho_1 , \rho_2 ) \equiv \sqrt{\frac{\rho_1}{\rho_1 + 1 } } \left[\rho_1-\rho_2 + \ln \frac{\rho_1}{\rho_2}\right ] \end{array}\ ] ] fig . [ fig : schwarzschild-1](b ) illustrates these trajectories in the grid . as a consequence of the above statement , any user traveling on the emission coordinate domain @xmath11 of this positioning system and receiving the emitter positioning data @xmath93 can find a particular linear relation between the @xmath148 s and the @xmath149 s , leading him to find that the emitter trajectories @xmath67 are parallel straight lines in the grid of the emission coordinates he is receiving . from them , he can extract the parameters @xmath150 , @xmath69 and @xmath60 [ see ( [ iv - emitt - taus - a ] ) and ( [ iv - emitt - taus - b ] ) ] : [ iv - two - events ] the slope , separation and synchronization parameters @xmath68 , @xmath151 and @xmath74 characterizing the parallel trajectories in the grid @xmath16 of the stationary emitters in schwarzschild plane have , in terms of the emitter positioning data at two different events @xmath72 and @xmath73 , the same expression as in minkowski plane , given by ( [ iii - par - taus ] ) . once these parameters are evaluated , expressions ( [ iv - pi - q - sigma ] ) and ( [ iv - pi - q - rhos ] ) tell us that @xmath58 and @xmath59 determine one - to - one the dynamical radial parameters @xmath134 , which characterize the emitter trajectories . indeed , from the first expression in ( [ iv - pi - q - rhos ] ) we can obtain : @xmath152 and , substituting in the second one , we have : @xmath153\right]\ ] ] this expression of @xmath154 is an effective function on @xmath155 . consequently it admits an inverse and this radial parameter may be obtained as a function of @xmath58 and @xmath156 . moreover , the third expression in ( [ iv - pi - q - sigma ] ) shows that the parameter @xmath60 gives the emitter synchronization . all these results can be summarized in the following : [ iv - rhos - taus ] for stationary positioning systems in schwarzschild plane , the radial parameters @xmath155 , @xmath157 of the emitters may be obtained in terms of the slope data parameter @xmath68 and the separation data parameter @xmath76 by ( [ iv - rho2 ] ) and @xmath158 where @xmath159 is the inverse of ( [ iv - q - rho1 ] ) . also , their synchronization times @xmath52 and @xmath53 indicated by the emitter clocks at two simultaneous events , are related to @xmath58 and the synchronization data parameter @xmath60 by @xmath160 this statement gives the parameters @xmath155 and @xmath157 , as well as the relation between @xmath52 between @xmath79 in terms of the data parameters @xmath80 @xmath59 and @xmath60 , which can be extracted from the data received by the user . in analogy to minkowskian case , these parameters yield an operational definition of the null canonical stationary coordinates @xmath43 associated with the emitters and given by ( [ iv - cochange - a ] ) . by ( [ iv - rhos - pi - q ] ) and ( [ iv - sigma ] ) these coordinates depend on a unique parameter ( say @xmath53 ) , which now controls the time - like translations in the direction @xmath161 . what is the metric information that the emitter positioning data offer to the users ? to obtain the metric function in emission coordinates , starting from ( [ iv - metric - uv ] ) and using the coordinate transformation ( [ iv - cochange - a ] ) , one obtains : @xmath162 & = & r_s^2\ , b\left(\rho(x(\tau^1 , \tau^2))\right ) \lambda_1 \lambda_2 \end{array}\ ] ] but from ( [ iv - cochange - a ] ) and ( [ iv - pi - q - sigma ] ) we have : @xmath163 & = & \lambda_1 ( \tau^1 - \omega \tau^2 - \sigma - q ) + 2x(\rho_1 ) \end{array}\ ] ] so that , by using ( [ iv - lambda ] ) and ( [ iv - rhos - pi - q ] ) , we can write the metric function in terms of the data @xmath25 and the parameters @xmath89 . moreover these parameters can be obtained as in ( [ iii - par - taus ] ) from the emitter positioning data @xmath164 and @xmath165 at two events @xmath72 and @xmath73 . thus , we can state : [ iv - prop - metric ] in schwarzschild plane , the space - time metric in emission coordinates of a stationary positioning system is given , in terms of the emitter positioning data @xmath166 by : @xmath167 where @xmath168 @xmath127 being the lambert function and @xmath169 where the parameters @xmath89 are given by ( [ iii - par - taus ] ) in terms of the values of the data at two events , and the constant @xmath155 is defined in ( [ iv - rhos - pi - q ] ) . this statement shows that the sole public quantities @xmath92 received by any user allow him to know the space - time metric everywhere . it is worth remarking that the user must be informed that _ the emitter positioning data come from a stationary positioning system in presence of a mass of the schwarzschild radius @xmath115_. remark also that this information does not require the specific value of the emitter radial parameters @xmath134 , which may be determined by the user from the relations ( [ iv - rho2 ] ) and ( [ iv - rhos - pi - q ] ) . as statements [ iii - recti - trayec ] and [ iv - recti - trayec ] show , the sole emitter positioning data @xmath24 give the same picture in the grid in minkowski and in schwarzschild planes . consequently , the additional quantitative information of the schwarzschild radius @xmath115 is a necessary information in order to obtain the metric and the dynamical quantities . moreover , as we have already commented for minkowski plane , the user s knowledge of the accelerations of the clocks is generically necessary because different dynamics ( including non uniformly accelerated ones ) may produce the same emitter trajectories _ in the grid _ @xcite . on the other hand , we will show in next section that the data of one of the emitter s accelerations allows the user to determine schwarzschild mass . ( a ) user @xmath31 with a trajectory in the grid parallel to the stationary emitters in schwarzschild plane . note that parallel straight lines @xmath103 are the hypersurfaces with schwarzschild time @xmath170 . ( b ) trajectory of the user @xmath31 in ` stationary ' null coordinates @xmath63 : it is an stationary user.,scaledwidth=48.0% ] all the results above are independent of the user motion . let us consider now the specific user @xmath31 which obtains from the public data @xmath25 his own trajectory in the grid as [ see fig . [ fig : schwarzschild-2](a ) ] : @xmath171 that is , a straight line parallel to the emitter trajectories . the dynamics of this user and its relation with the emitters are given by [ see fig . [ fig : schwarzschild-2](b ) ] : [ iv - prop - user - estat - acc ] the users which find that their trajectory in the grid is a straight line parallel to the emitter s ones [ eq . ( [ iii - user1-a ] ) ] are stationary users in schwarzschild plane . their radial parameter @xmath172 is : @xmath173 , \ ] ] their constant acceleration scalar @xmath34 is @xmath174 and their proper time lapse @xmath97 is , for any of the two emitter clocks ( no summation ) @xmath175 observe that the parameters @xmath133 , @xmath132 and @xmath60 , and then the value of the radial parameter @xmath172 of the user depend on the emitter positioning data @xmath93 and the schwarzschild radius @xmath115 , as a consequence of ( [ iv - lambda ] ) , ( [ iv - rho2 ] ) and ( [ iv - rhos - pi - q ] ) . in order to show these relations , let us construct , starting from the emitter positioning data , the stationary null coordinate system @xmath43 given by ( [ iv - cochange - a ] ) . in it , using the inverse transformation ( [ iv - cochange - b ] ) , the trajectory equation ( [ iv - user - a ] ) becomes : @xmath176 where @xmath177 is given in ( [ iv - rho - bar ] ) . this means that this trajectory belongs to the stationary family of trajectories as the statement says . furthermore , the user can also obtain , up to an origin , his own proper time @xmath149 . indeed , in the null coordinate system @xmath63 the proper time user trajectory is : @xmath178 where the constant @xmath179 is given by the second of expressions ( [ iv - user - tau ] ) . then , ( [ iv - user - traj - tau ] ) and the coordinate transformation ( [ iv - cochange - a ] ) imply the first of expressions ( [ iv - user - tau ] ) . the simultaneity loci of the stationary emitters in stationary null coordinates are given by the lines @xmath180 . using the coordinate transformation ( [ iv - cochange - a ] ) one obtains that these loci take in the grid the following expression : @xmath181 thus , as in the flat case , one has : [ iv - prop - simultaneous ] the locus of simultaneous events for the emitters of a stationary positioning system in schwarzschild plane are parallel straight lines of slope @xmath104 , the same , up to sign , as that of the trajectory of the emitters . it is worth remarking that the emitter positioning data received by a user give , in the two physically different two - dimensional minkowski and schwarzschild space - times , the same qualitative information about the emitter trajectories in the grid , as we have seen in this and the precedent section [ see figs . [ fig : accelerated-1](b ) and [ fig : schwarzschild-1](b ) ] . moreover , the trajectory in the grid of any stationary user is also the same in both cases [ see figs . [ fig : accelerated-2](a ) and [ fig : schwarzschild-2](a ) ] . in spite of this fact , we shall see in the next section that complementary data that afford dynamic information of the system allow the user to distinguish between both systems . the relevance of the dynamical data for the determination of the system is manifest in the case of a geodesic user : his trajectory in the grid is different in both cases . is is easy to show this fact because the trajectory ( [ iii - user - geo ] ) of a geodesic user in the grid of the system in the flat case is not geodesic for the schwarzschild metric ( [ iv - metric - uv ] ) , as a straightforward calculation shows . in this last case the geodesics , qualitatively similar to those of the flat case shown in fig . [ fig : accelerated-3](b ) , have nevertheless a different trajectory equation . by comparison of this trajectory with the one constructed by the geodesic user from the received sole data @xmath182 , he is able to distinguish in which space - time he is evolving . the interest of auto - locating positioning systems in gravimetry has been pointed out in @xcite , where we have shown that , if a user has neither a priori information on the gravitational field nor on the positioning system , the user data determine the metric and its first derivatives along the user and emitter trajectories . in the next section we apply these results to obtain the gravitational information that can be extracted from the user data generated by the positioning systems of the precedent sections . but in some cases the user can have some a priori information about the space - time or about the kind of positioning system he is using . the different levels of not null information in these two aspects establishes different kinds of gravity problems . for example , in the last two precedent sections , we have supposed the space - time fully known ( minkowski or schwarzschild planes ) , and also the partial information on the positioning system assuring its stationarity , but without specifying the space - time trajectories of the emitters . in this section we will consider other situations of total or partial information on the space - time itself and/or on the positioning system , and we will analyze the role that the emitter accelerations and the proper user data can play in obtaining the gravitational field and in recovering the complementary characteristics of the positioning system . we consider two levels in the a priori information that the user can have about the gravitational field . he may know generically that : * ( gf-1 ) the space - time belongs to the family of schwarzschild planes expanded to include minkowski plane , or specifically that : * ( gf-2 ) the space - time is a particular schwarzschild plane with a known the schwarzschild radius @xmath115 or it is minkowski plane . we also consider two levels in the a priori user information about the positioning system . he may know generically that : * ( ps-1 ) the positioning system is defined by two stationary emitters , or specifically that : * ( ps-2 ) the positioning system is defined by two particular stationary emitters with known radial coordinates @xmath134 and a fixed synchronization @xmath60 . if the user has the widest information ( gf-2 ) about the gravitational field and the widest information ( ps-2 ) about the positioning system , then , from ( gf-2 ) he can take a convenient coordinate system and express in it the metric components and from ( ps-2 ) he can describe in this coordinate system the trajectory of the emitters and the future light rays that spread from them . then , he may construct the emission coordinates and calculate the coordinate transformation relating them to the initial ones . consequently , he can evaluate the emitter trajectories and the space - time metric in emission coordinates . thus , without receiving any data , the user can know by calculation the pairs @xmath183 and @xmath184 locating the emitters as well as their dynamics . with this full information about the metric and the positioning system , a user receiving the user positioning data @xmath182 may extract his trajectory @xmath185 in the grid . moreover at every event @xmath186 on his trajectory , he may know the metric function @xmath187 and his acceleration . thus , in particular , the user may obtain his own units of time and distance . for example , in the particular case of a stationary user , his proper time is given by ( [ iv - user - tau ] ) . the situation in which a user has the widest information about the gravitational field ( gf-2 ) and partial information about the positioning system ( ps-1 ) is the one that has been analyzed in detail in sec . [ section - iii ] for minkowski plane and in sec . [ section - iv ] for schwarzschild plane . in particular , we have shown that the sole emitter positioning data @xmath40 complete the system information [ see statements [ iii - alfas - taus ] and [ iv - rhos - taus ] ] . moreover we have shown that the space - time metric may be obtained in terms of these data ( see statements [ iii - prop - metric ] and [ iv - prop - metric ] ) . once completed the system information in this way , the user is in the situation of the precedent subsection and , as commented there , he can also obtain his trajectory and his own units of time and distance . suppose now that the user has partial information ( gf-1 ) about the gravitational field and also partial information ( ps-1 ) about the positioning system . the user knows thus that the system is defined by stationary emitters and , consequently , that they have constant acceleration . if the emitters carry accelerometers and broadcast their accelerations , the user can receive the precise value @xmath29 @xmath30 of these accelerations . can this dynamical information determine the schwarzschild mass ? the answer is affirmative . furthermore , apart from the emitter positioning data @xmath40 , _ only one of the two accelerations _ is sufficient to the user to obtain the schwarzschild mass and , consequently , to acquire the information level considered in the precedent subsections . in order to specify exactly this result let us assume firstly that the user knows the full set of the public data @xmath188 . we know that the data parameters @xmath189 depend on the emitter positioning data [ see ( [ iii - par - taus ] ) ] . because the accelerations @xmath50 are also known , the first expression in ( [ iv - pi - q - rhos ] ) and the the two in ( [ iv - accelerations ] ) constitute three relations in @xmath190 that allow to obtain them explicitly in terms of @xmath191 . on the other hand , the resulting expressions have to be compatible with the expression of @xmath59 given in ( [ iv - pi - q - sigma ] ) . in the case of minkowski plane , the user data parameters @xmath192 are submitted not to one but to two compatibility conditions , those given by ( [ iii - parametres - a ] ) . this fact and the considerations in the above paragraph lead to the following [ prop - mass - accelerations ] consider a user of a stationary positioning system in a space - time of which he only knows that it is either a schwarzschild or minkowski plane . let @xmath193 be the public data that he receives , @xmath189 the data parameters that can be extracted from them by means of ( [ iii - par - taus ] ) , and @xmath194 the data parameter given by : @xmath195 then @xmath194 is necessarily such that : @xmath196 moreover : + ( i ) the value @xmath197 is the one that informs the user that his space - time is minkowski plane . in this case , the public data are submitted to the constraints : @xmath198 \(ii ) any value @xmath199 informs the user that his space - time is schwarzschild plane with radius @xmath200 that the radial coordinate of the emitters are : @xmath201 and that the public data are submitted to the constraint : @xmath202 the compatibility condition ( [ restriction ] ) for the data parameters @xmath203 suggests that the emitter positioning data @xmath24 and a sole acceleration , say @xmath29 , are sufficient data to determine the schwarzschild mass . indeed , from ( [ iv - accelerations ] ) we obtain : @xmath204 this expression allows eliminate @xmath115 in the expression ( [ iv - pi - q - sigma ] ) of @xmath59 , which becomes : @xmath205 % \label{theta - rhos}\end{aligned}\ ] ] then , substituting ( [ iv - rho2 ] ) in ( [ theta ] ) , ( [ q - alpha1 ] ) takes the form : @xmath206 this expression of @xmath207 is an effective function on @xmath208 consequently it admits an inverse : @xmath209 and we can state : [ prop - mass - acceleration ] consider a user of a stationary positioning system in a space - time of which he only knows that it is a schwarzschild plane . let @xmath40 be the emitter positioning data that he receives and suppose that he also receives but one of the emitter dynamical data , say @xmath29 . then , the schwarzschild radius @xmath115 is given by ( [ radisch ] ) where @xmath155 may be extracted from ( [ rho1-implicit ] ) as ( [ rho1-rs ] ) , in terms of the data parameters @xmath210 . let us assume again that the user has partial information ( gf-1 ) and ( ps-1 ) respectively about the gravitational field and the positioning system . we show here that if the user is stationary and generates his proper time , he can also obtain the schwarzschild mass . let us suppose that a user receiving the emitter positioning data @xmath40 carries a clock that measures his proper time @xmath149 and that he is stationary with respect to the stationary positioning system . remember that , from statement [ iii - prop - user - estat - acc ] , he knows that he is stationary if his trajectory in the grid is parallel to the trajectories of the stationary emitters and that this property may be extracted from the user positioning data he is receiving . from his trajectory in the grid , he may also extract by ( [ iv - user - a ] ) the value of the specific data parameter @xmath211 . on the other hand , by comparing the user time @xmath149 with one of the received times , say @xmath9 , the user can obtain the data parameter : @xmath212 this parameter relates the radial coordinates of the user and the emitter @xmath7 . indeed , from ( [ iv - lambda ] ) and ( [ iv - user - tau ] ) one obtains : @xmath213 then , from the second expression in ( [ iv - pi - q - sigma ] ) and relations ( [ iv - rho - bar ] ) and ( [ iv - user - tau ] ) we have : @xmath214 and substituting ( [ iv - rho2 ] ) and ( [ rouser - ro1 ] ) one obtains : @xmath215 this expression of @xmath216 is an effective function on @xmath155 . consequently , it admits an inverse : @xmath217 thus , we can state : [ prop - mass - time ] in a stationary positioning system of a space - time of which it is only known that it is a schwarzschild plane , let us consider a stationary user receiving the emitter positioning data @xmath218 and carrying a clock measuring his proper time @xmath149 . then , the schwarzschild radius is : @xmath219 where @xmath155 and then @xmath220 are given in terms of the data parameters @xmath221 respectively by ( [ rho1-pi - p - c ] ) and ( [ iv - q - rho1 ] ) . now we suppose that , in a schwarzschild or minkowski space - time , and in a stationary positioning system , we have a user which is unaware of these two facts , i.e. he has no information about the gravitational field and the positioning system . we know that the emitter positioning data @xmath93 determine the user trajectory @xmath185 as well as the emitter trajectories @xmath222 in the grid , and that this happens whatever be the curvature of the two - dimensional space - time . in @xcite we have studied the information that these data give about the space - time metric and we have shown that _ the emitter positioning data @xmath24 determine the space - time metric function along the emitter trajectories_. if in addition the user knows the emitter dynamical data @xmath32 , and consequently the acceleration scalars @xmath223 , @xmath224 , we have also shown in @xcite that _ the public data @xmath225 determine the gradient of the space - time metric function along the emitter trajectories_. we can apply the general explicit expressions given in @xcite and the results of the sections [ section - iii ] and [ section - iv ] to obtain the metric component and its gradient corresponding to the parallel straight lines emitter trajectories and constant accelerations broadcast by the stationary emitters . the result is : [ prop - along - emitters ] in a schwarzschild or minkowski space - time , and in a stationary positioning system , a user unaware of these two facts can obtain : + ( i ) from the emitter positioning data @xmath93 , the metric function on the emitter trajectories : @xmath226 \(ii ) from the public data @xmath227 , the gradient of the metric function along the emitter trajectories : @xmath228 ( \ln m),_1 ( \tau^2 ) = \omega \alpha_2 \ , \quad ( \ln m),_2 ( \tau^2 ) = - \alpha_2 \end{array}\ ] ] where @xmath58 is obtained from the emitter positioning data by ( [ iii - par - taus ] ) . it is worth remarking that the same public data giving the metric ( [ metric - along - emitters ] ) and its gradient ( [ gradmetric - along - emitters ] ) along the stationary emitter trajectories can be generated by others , not necessarily stationary , positioning systems and/or in space - times different from schwarzschild planes . it follows that the compatibility restrictions ( [ kappa ] ) and ( [ restriction ] ) [ or ( [ restrictions - flat ] ) for minkowski plane ] are necessary but not sufficient conditions to characterize a stationary positioning system in schwarzschild planes . consequently , the user needs to receive complementary data or a priori information in order to obtain the gravitational field everywhere or to get the specific characteristics of the positioning system . alternatively , even if the positioning system is not auto - locating ( broadcasting the sole proper times @xmath25 ) , when the user also knows the proper user data @xmath42 , he may obtain gravimetric information along his trajectory . indeed , from these data the user can obtain his parameterized proper time trajectory ( [ user - proper - time ] ) and his acceleration scalar @xmath229 . we have then shown in @xcite that _ the public - user data @xmath230 determine the space - time metric function and its gradient on the user trajectory_. thus , applying the explicit expressions given in @xcite to the stationary user considered in the sections [ section - iii ] and [ section - iv ] , one has : [ prop - along - user ] in a schwarzschild or minkowski space - time , and in a stationary positioning system , a user unaware of these two facts can obtain , from the user positioning data @xmath25 and the proper user data @xmath231 , the metric function and its gradient along the user trajectory : @xmath232 ( \ln m),_1 ( \tau ) = \alpha \frac{\delta \tau}{\delta \tau^1 } \ , \quad ( \ln m),_2 ( \tau ) = - \alpha \frac{\delta \tau}{\delta \tau^2 } \label{gradmetric - along - user}\end{aligned}\ ] ] the user data that give the metric ( [ metric - along - user ] ) and the metric gradient ( [ gradmetric - along - user ] ) along the user trajectory can be received by a non stationary user of other positioning systems and/or in other different space - times . again , the user needs complementary information in order to obtain the metric everywhere and the characteristics of the positioning system . in this work we have continued the two - dimensional approach to relativistic positioning systems initiated in @xcite . in that work we explained the basic features of these systems , we studied in detail the positioning system defined by two geodesic emitters in flat space - time , and we showed that in an arbitrary space - time the user data determine the gravitational field and its gradient along the emitters and user world lines . here we have considered stationary positioning systems , i.e. positioning systems defined by two uniformly accelerated emitters at rest with respect to each other . in order to understand the behavior of the emission positioning systems in a non necessarily flat gravitational field , we have studied stationary positioning systems in both , minkowski ( sec . [ section - iii ] ) and schwarzschild planes ( sec . [ section - iv ] ) . we have shown that a user that receives the emitter positioning data will find , for the emitter trajectories , parallel straight lines in the grid , no matter the plane be ( statements [ iii - recti - trayec ] and [ iv - recti - trayec ] ) , and we have given the parameters of these straight lines in terms of the emitter positioning data ( statements [ iii - two - events ] and [ iv - two - events ] ) . we have also obtained the dynamics ( including the radial coordinate in the schwarzschild case ) and synchronization of the emitters in terms of data parameters ( statements [ iii - alfas - taus ] and [ iv - rhos - taus ] ) . we have found that , if a user knows the stationary character of the positioning system , the emitter positioning data make him able to know explicitly the space - time metric in emission coordinates ( statements [ iii - prop - metric ] and [ iv - prop - metric ] ) . we have studied the proper data received by a stationary user and the expressions of his acceleration and proper time in terms of the emitter positioning data that he receives ( statements [ iii - prop - user - estat - acc ] and [ iv - prop - user - estat - acc ] ) . we have also obtained in the emission grid the locus of simultaneous events for the emitters ( statements [ iii - prop - simultaneous ] and [ iv - prop - simultaneous ] ) . in minkowski plane , the trajectory of a geodesic user in emission coordinates has been obtained ( statement [ iii - prop - inertial ] ) . the study of these two scenarios brings to light an interesting situation : the emitter positioning data of both systems lead to an identical emission grid . how a user can distinguish both systems ? we have analyzed this question in sec . [ section - gravimetry ] and have shown that the knowledge of complementary user data determines the schwarzschild mass ( statements [ prop - mass - accelerations ] , [ prop - mass - acceleration ] and [ prop - mass - time ] ) . these simple two - dimensional situations suggests that the relativistic positioning systems could be useful in four - dimensional gravimetry for reasonable parameterized models of the gravitational field . the gravimetry cases analyzed here are only particular situations of the general gravimetry problem in relativity where user data are the unique information that a user has . we showed in @xcite that these data allow the user to obtain the metric function and its gradient on emitter and on user trajectories . in sec . [ section - grav - along ] we have applied these results to obtain the gravitational information acquired by the user considered in sections [ section - iii ] and [ section - iv ] ( statements [ prop - along - emitters ] and [ prop - along - user ] ) . finally , for a user that knows the space - time where he is immersed ( flat , schwarzschild , ... ) but he has no information about positioning system , we can ask the following questions . can the user data determine the characteristics of the positioning system ? can he obtain information on his local units of time and distance and his acceleration ? the answer to these questions is still an open problem for a generic space - time . but in a future paper @xcite we undertake this query for minkowski plane and we analyze the minimum set of data that determine all the user and system information . a remarkable result is that the user data are not independent quantities : the accelerations of the emitters and of the user along their trajectories are determined by the sole knowledge of the emitter positioning data and of the acceleration of only one of the emitters and only during the interval between the emission time of a signal by an emitter and its reception time after being reflected by the other emitter .

an approximate analytical and non - linear solution of the einstein field equations is derived for a system of multiple non - rotating black holes . the associated space - time has the same asymptotic structure as the brill - lindquist initial data solution for multiple black holes . the system admits an arnowitt - deser - misner ( adm ) hamiltonian that can particularly evolve the brill - lindquist solution over finite time intervals . the gravitational field of this model may properly be referred to as a _ skeleton _ approximate solution of the einstein field equations . the approximation is based on a conformally flat truncation , which excludes gravitational radiation , as well as a removal of some additional gravitational field energy . after these two simplifications , only source terms proportional to dirac delta distributions remain in the constraint equations . the skeleton hamiltonian is exact in the test - body limit , it leads to the einsteinian dynamics up to the first post - newtonian approximation , and in the time - symmetric limit it gives the energy of the brill - lindquist solution exactly . the skeleton model for binary systems may be regarded as a kind of analytical counterpart to the numerical treatment of orbiting misner - lindquist binary black holes proposed by gourgoulhon , grandclment , and bonazzola , even if they actually treat the corotating case . along circular orbits , the two - black - hole skeleton solution is quasi - stationary and it fulfills the important property of equality of komar and adm masses . explicit calculations for the determination of the last stable circular orbit of the binary system are performed up to the tenth post - newtonian order within the skeleton model .

gr-qc0311018

the description of the motion of binary black holes within the einsteinian theory of gravity is quite a challenging problem and is likely to be finally solvable by numerical means only . however , it seems that the full success will occur in a far future . it has turned out in the past that analytical developments did always influence numerical investigations , hence further progress on the analytical side is certainly desirable . in this paper we shall present an analytical solution of truncated einstein equations for systems of non - spinning point - like objects , which shows many aspects of past numerical models . most of the numerical computations of quasi - stationary initial data for black - hole binaries are based on the assumption that the space metric is conformally flat @xcite . they often assume as well the existence of an approximate helical killing vector which permits to elaborate some manageable formulations ( also see @xcite ) . recently , the time - symmetric initial data of brill - lindquist @xcite were generalized by adding a non - conformally flat contribution that incorporates pieces of information provided by perturbative post - newtonian calculations @xcite . the present paper does not follow this line , but rather keeps the spirit of former numerical simulations . we essentially perform two simplifications : on the one hand , we adopt the often - used assumption of conformally flat space metric ; on the other hand , we attribute the non - linear gravitational field energy to point - like sources , which are modeled as dirac delta distributions . we solve the resulting truncated einstein equations analytically and get what we call , in reference to the point - like character of the source support , the _ skeleton _ solution . the calculations are carried out for @xmath0 bodies in a mathematically sound way by working from the beginning in @xmath1-dimensional space - time and taking the limit @xmath2 in the final expressions , so that the dirac delta distributions modeling the sources are consistently handled @xcite . our space metric and extrinsic curvature may be regarded as some generalization of the multiple black hole brill - lindquist initial value solution @xcite to arbitrary black - hole momenta preserving its original asymptotic structure . moreover , the energy function can be used as generator of the dynamical evolution of the system . we shall refer to it as the skeleton hamiltonian associated to the full arnowitt - deser - misner ( adm ) hamiltonian . it agrees with the exact hamiltonian _ ( i ) _ in the test body limit , _ ( ii ) _ at the first post - newtonian ( 1pn ) approximation , i.e. , at the first order in powers of @xmath3 ( @xmath4 denoting the speed of light ) beyond the newtonian dynamics , and _ ( iii ) _ in the limit of special relativity in absence of gravity . the skeleton gravitational field is given explicitly , as well as the equations of motion . as a first application , we shall investigate the location of the last stable ( circular ) orbit ( lso ) for binary systems . in order to reach a proper accuracy , the last - stable - orbit parameters are worked out up to the tenth post - newtonian ( 10pn ) approximation within the skeleton model . the comparison of second and third post - newtonian results derived from the skeleton model with the corresponding quantities computed in the full einstein theory delivers a quantitative measure for the missing terms in the skeleton approach . we are interested in a system of @xmath0 black holes interacting gravitationally . its description will be achieved within the adm hamiltonian framework using a @xmath1 splitting of space - time . the spatial indices are denoted by latin letters and vary from @xmath5 to @xmath6 , whereas the space - time indices are greek and vary from @xmath7 to @xmath6 . in a given chart , @xmath8 , the motion of the @xmath9th black hole is described by a curve @xmath10 , for @xmath11 . the conjugate momentum of the position @xmath12 is called @xmath13 . the constant mass parameter of the @xmath9th black hole is denoted by @xmath14 . the hamiltonian variables associated to the gravitational field are taken to be ( i ) the space metric @xmath15 induced by the full space - time metric @xmath16 on the hypersurface @xmath17 , and ( ii ) its conjugate momentum @xmath18 where @xmath19 is the newtonian gravitational constant . the other components of @xmath16 are parametrized by the lapse and shift functions : @xmath20 and @xmath21 , with @xmath22 standing for the inverse of @xmath15 . it is also useful to introduce the shift vector @xmath23 . the spatial indices will be raised and lowered with the help of @xmath22 and @xmath15 , respectively . we now assume that the `` center - of - field '' @xmath24 of object @xmath9 can be modeled in the fictitious space - time associated with the flat metric by means of the dirac distribution @xmath25 . this prescription is known to lead to the brill - lindquist hamiltonian when the black hole velocities are set to zero , provided the mass parameters @xmath14 , @xmath11 , are identified with the brill - lindquist masses @xcite . it also leads to the correct 2pn dynamics @xcite for compact binaries , which can itself be recovered by means of the extended - body approach @xcite . moreover , the singularities generated by the distributions @xmath26 have shown to be curable by dimensional regularization at the 3pn order @xcite . similarly , we shall see in the present work that , for an appropriate choice of the space dimension @xmath6 , the @xmath26-type sources do not generate any ambiguity , so that the 3-dimensional quantities can be obtained by dimensional regularization . finally , this point - particle approach allows to bypass delicate investigations about the behavior of the gravitational field near the objects . in an asymptotically flat space - time of dimension @xmath27 , the hamiltonian of @xmath0 point masses provided by the adm formalism reads @xcite @xmath28 where @xmath29 . the super - hamiltonian density @xmath30 appearing under the integral sign is defined by @xmath31 while the supermomentum density @xmath32 is given by @xmath33 in eq . @xmath34 denotes the space curvature of the hypersurface @xmath17 and @xmath35 is the determinant of the matrix @xmath15 . as the lapse and shift functions are lagrange multipliers , the variation of the hamiltonian with respect to them leads to the constraint equations @xmath36 before solving eqs . , we must first fix the coordinate system . it is convenient to work in the so - called adm transverse trace - free ( admtt ) gauge @xcite defined by the two conditions @xmath37 in this gauge , the metric decomposes into its trace @xmath38 , which may be parametrized as @xmath39 ( in order to get a simple expression for the curvature density ) , and a transverse trace - free part @xmath40 . we thus have @xmath41 \pi^{ii } = 0.\end{gathered}\ ] ] the momentum @xmath42 needs itself to be split into a transverse trace - free contribution @xmath43 , and a rest @xmath44 , @xmath45 these two terms are uniquely defined assuming they decay as @xmath46 at spatial infinity and are given as a combination of derivatives of the poisson integral @xmath47 , and the poisson integral of the latter quantity , namely @xmath48 . we can then try to compute @xmath49 and @xmath44 from the two constraint equations supplemented by the boundary conditions at spatial infinity resulting from the asymptotic flatness of space - time ; notably @xmath50 , @xmath51 . it is actually more convenient to solve the hamiltonian constraint for the quantity @xmath52 rather than @xmath49 to get rid off the spurious constant at spatial infinity . we thus have @xmath53 after solving the constraint equations , we insert the ensuing expressions for @xmath54 and @xmath44 into the right - hand side of eq . . we get a _ reduced _ hamiltonian depending on @xmath12 , @xmath13 , @xmath55 , and @xmath43 only : @xmath56 \nonumber\\ & = \frac{c^4}{16 \pi g } \int d^d\mathbf{x}~\partial_i ( \partial_j \gamma_{ij}-\partial_i \gamma_{jj } ) \nonumber\\ & = -\frac{c^4}{16 \pi g } \int d^d\mathbf { x}~\delta \phi.\end{aligned}\ ] ] this hamiltonian contains the full information about the evolution of the matter @xmath57 and field @xmath58 variables @xcite . as the change of variables @xmath59 is not canonical , the poisson brackets of @xmath55 and @xmath43 take a form involving a transverse trace - free projection operator @xcite . the reduced hamiltonian @xmath60 provides the equations of motion of the @xmath0 bodies together with the evolution equations of the gravitational field . as they are strongly coupled it is hopeless to solve them analytically , but they can be solved perturbatively using , notably , the post - newtonian iterative scheme @xcite . the resulting dynamics is highly reliable as long as the typical speeds remain smaller than , say , about @xmath61 . when binary black holes reach their last stable orbits , the reliability of the approximation breaks down since the system enters a strong field regime . the only way to understand precisely this crucial stage is to perform numerical simulations in full general relativity . our purpose here is to search for a simple non - perturbative model applicable to arbitrary strong fields , eventually leaving the einsteinian theory but always staying close enough to it , in order to give an element of comparison with numerical investigations . the space metric obtained in our approach should also be usable as initial condition . we shall adopt here a wilson - mathews - type prescription @xcite , setting the non - conformally flat part of @xmath15 , namely @xmath40 , equal to zero . as @xmath55 is of order @xmath62 , the difference between our @xmath15 and the einsteinian space metric shows up at the 2pn level . for spherically symmetric space - times the difference is zero . in the case where the present model is used to fix initial cauchy data , the condition of a conformally flat space metric needs to hold on only one hypersurface , say @xmath63 . choosing @xmath55 to be zero on @xmath64 is always possible , though it is not physically realistic in general for isolated self - gravitating systems . our attitude in this instance will be to accept the resulting error . in return , we shall be able to determine the field @xmath15 without assuming zero velocities of the bodies at initial time . we thus demand here that @xmath65 and the problem reduces to solve the constraints @xmath66 , @xmath67 under this condition . the space metric @xmath15 is then proportional to @xmath68 , which implies that in the supermomentum @xmath32 the term @xmath69 vanishes [ cf . eq . ] . the field momentum @xmath42 relates to @xmath70 through the identities @xmath71 in particular , @xmath70 is symmetric and trace - free ( stf ) as is @xmath42 . the determinant @xmath35 appearing in @xmath30 is merely @xmath72 finally , the 3-curvature density @xmath73 takes a remarkably simple form @xcite : @xmath74 collecting these relations together with the constraint equations , we find : [ eq : constraint_system ] @xmath75 & -\frac{16\pi g}{c^2 } \sum_{a=1}^\mathcal{n } m_a \delta_a \psi^{-1 } \left ( 1 + \psi^{-\frac{4}{d-2 } } \frac{p_a^2}{m_a^2 c^2 } \right)^{\frac{1}{2 } } \ , , \\ \label{eq : constraint_momentum } \partial_j\pi^j_{\,i } = & -\frac{8\pi g}{c^3 } \sum_{a=1}^\mathcal{n } p_{ai } \delta_a \ , , \end{aligned}\ ] ] with @xmath76 . in the system the second equation decouples from the first one and is solved by searching for a particular solution proportional to the symmetric trace - free part of the derivative of a certain vector potential @xmath77 ; namely , @xmath78 the equation for the divergence @xmath79 is obtained by applying the operator @xmath80 simultaneously on the two sides of , hence @xmath81 . this yields immediately @xmath82 , and from the constraint it follows that : @xmath83 we now notice that the @xmath6-dimensional laplacian of the @xmath84th power of @xmath85^{\frac{1}{2}}$ ] satisfies @xmath86 in the sense of functions , and that , in the sense of distributions , @xmath87 as a result we may choose : @xmath88 which is nothing but the 1pn vector potential @xmath89 of ref . @xcite verifying @xmath90 . our second simplification consists in stating that the homogeneous solution of eq . , @xmath91 , is precisely zero , which implies @xmath92 . when this prescription is imposed on one unique hypersurface , it merely represents a particular choice of initial data . the resulting field momentum @xmath93 is identical to the rescaled extrinsic curvature proposed by bowen and york in their time - asymmetric formulation of the cauchy problem @xcite and adopted e.g. in the puncture method @xcite or various calculations of the last stable circular orbit @xcite . although it is a mathematical solution of the constraint equations for a conformally flat metric , it is not realistic in the sense that it does not result from an evolution process in absence of incoming radiation . it does not coincide either with the wilson - mathews momentum @xmath94 derived from the condition @xmath95 : @xmath96 where the quantity @xmath97 , defined as the @xmath98 part of the solution [ matching the form ( 3.6 ) ] of the equation @xmath99 , is explicitly given in paper @xcite for @xmath100 . the difference @xmath101 is only of order @xmath98 , which corresponds to the 2pn approximation . on the hamiltonian level , it is of 3pn order only . one might then argue that the wilson - mathews condition @xmath102 , valid for all times , i.e. particularly @xmath103 , is in contradiction to the condition @xmath104 . however , it is not the case because the momentum constraint equation is linear in @xmath105 and decouples from any other field equation when the space metric is conformally flat , so that the quantity obtained by adding an arbitrary homogeneous solution to @xmath70 still satisfies the constraint . as our @xmath70 differs from @xmath106 by such a homogeneous solution ( i.e. a transverse - trace free object ) , there is no contradiction between the two approaches . the integration of the hamiltonian constraint is much more involved , principally because of the appearance of @xmath54 in the source and the presence of a non - compact support term . on the contrary , the compact support term is harmless . it is made of a sum of dirac distributions @xmath26 that formally determine the boundary conditions near the black holes . by subtracting from the conformal factor the @xmath107 terms entering @xmath54 , we get an object that satisfies an equation whose source is entirely free of @xmath26 s ( this is the essence of paper @xcite ) . by contrast , we shall keep here the constraint equation under its original form in order to determine the dependence of the coefficient of @xmath107 on the positions @xmath12 and momenta @xmath13 . to solve the hamiltonian constraint iteratively , we have to face the delicate computation of @xmath108 . the expansion of the prefactor @xmath109 in powers of @xmath54 is known to give rise to poles when @xmath2 at the level corresponding to the 3pn approximation @xcite . in full general relativity , these poles cancel in the hamiltonian with similar quantities coming from @xmath55 , but such a cancellation can not occur in our case . in particular , _ the 3pn approximation of the wilson - mathews model does not exist _ for binary black - holes . we must therefore treat the @xmath110 term differently . let us first note that this term enters the space metric at the 3pn order only , since @xmath111 , but contributes to the 1pn hamiltonian [ because of the global factor @xmath112 in eq . ] . to be consistent with the accuracy of the former assumption @xmath102 which amounts to neglect some 2pn corrections in the hamiltonian , we have to keep at least the leading term in the post - newtonian expansion of @xmath113 , modulo a possible total space derivative . now , the tensor density @xmath70 is itself proportional to the symmetric trace - free part of the space derivative @xmath114 , so that @xmath115 as a consequence , the latter expression can be split into a `` skeleton '' term @xmath116 with compact support , and a `` flesh '' term @xmath117 which takes into account the field between the point particles . the flesh term contributes to the hamiltonian through the quantity : @xmath118 we are allowed to regard it as negligible in our scheme and keep the skeleton term alone in the decomposition of @xmath119 . our third simplification consits thus in performing the substitution @xmath120 = \frac{16 \pi g}{c^3 } \psi^\frac{2 - 3d}{d-2 } \sum_{a=1}^\mathcal{n } p_{aj } v_j \delta_a \ , . \end{gathered}\ ] ] after the replacement is made , the right - hand side of eq . takes the form @xmath121 , where the @xmath122 s are some unknown functions . for @xmath6 belonging to an appropriate range of values , @xmath122 is regular at @xmath123 , and @xmath124 . we see that in our model the source of the hamiltonian constraint reduces to some `` skeleton '' made of a linear combination of dirac distributions . for this reason , all quantities computed in the present approximation will be referred to as skeleton quantities . defining the @xmath0 quantities @xmath125 by the relation @xmath126 the function @xmath54 can be written as @xmath127 and the conformal factor reads @xmath128 where we have posed , in order to lighten the notation , @xmath129 it is also useful to introduce the ( regularized ) value of @xmath49 at the particle position @xmath130 : @xmath131 where @xmath132^\frac{1}{2}$ ] . after making the replacement on the right - hand side of eq . , inserting there expressions and , we arrive at @xmath133^{\frac{1}{2 } } \nonumber\\ & + \bigg(1 + a_d \sum_{b \neq a } \frac{\alpha_b}{r_{ab}^{d-2 } } \bigg)^{\frac{2 - 3d}{d-2 } } \frac{p_{ai } v_{ai}}{c } \bigg\ } \delta_a,\end{aligned}\ ] ] with @xmath134 . making use of eq . , we see that @xmath135 because of the linear independence of the distributions @xmath26 , the hamiltonian constraint is finally equivalent to a system of @xmath0 algebraic equations which reads @xmath136^\frac{1}{2 } + \frac{p_{ai } v_{ai}/c}{\left(1 + a_d \sum_{b \neq a } \frac{\alpha_b}{r^{d-2}_{ab } } \right)^\frac{3d-2}{d-2 } } , \quad a=1,\,\ldots,\,\mathcal{n}.\ ] ] solving the above system provides in principle the values of all the @xmath125 s . these quantities represent some effective masses , as suggested by the approximate equality @xmath137 . they are functions of @xmath138 ( for @xmath139 ) and @xmath13 [ defined implicitly by eqs . ] . the skeleton hamiltonian reads @xmath140 which shows that the adm mass @xmath141 is just the sum of the @xmath0 effective masses @xmath125 . the hamiltonian depends only on the matter variables : @xmath142 , so that the system is automatically conservative . the space metric is deduced from eqs . , , and : @xmath143 by construction , @xmath144 and @xmath15 coincide with their counterparts in the einsteinian theory up to the 1pn order . the test - body limit is achieved by taking the mass of one black hole , say of label 1 , much larger than the masses of its companion : @xmath145 , where @xmath146 . to get the test - body hamiltonian exactly , one has to let formally @xmath147 go to infinity ( @xmath148 ) in the hamiltonian while keeping @xmath149 , @xmath146 , constant as well as all linear momenta @xmath13 . it follows that particularly the flesh term @xmath117 does not contribute to this approximation . in absence of gravity ( @xmath150 ) our hamitlonian has the form of the exact special - relativistic hamitlonian for the system of @xmath151 free point particles . when @xmath152 for all @xmath9 , we recover the brill - lindquist solution , for which eqs . reduce to @xmath153 . as a consequence our model also generalizes the brill - lindquist data . let us observe that the particle masses @xmath14 can not be equated to the adm masses computed at the spatial infinities of the @xmath0 black hole sheets when one or several momenta are non - zero . when such an identification is performed , it leads automatically to the same relation between @xmath125 and @xmath14 as in the zero momentum case , whereas the correct dynamics up to the 1pn order is known to be the one that derives from the hamiltonian determined by the set of eqs . with @xmath154 . the actual dependence of the effective masses on the @xmath14 s must take the kinetic energy of the system into account . this indicates a relative boost of the black hole sheets with respect to each others . the issue is complicated by the fact that the bowen - york data for one single black hole do not result from a boosted schwarzschild field @xcite . this point , being still rather unclear , needs to be investigated in detail . we do not intend to discuss it here in the adm formalism , the evolution of the space metric in @xmath1 dimensions is given by [ cf . ( 4.70 ) in ref . @xcite ] . ] @xmath155 where the operator @xmath156 denotes the space covariant derivative . its explicit action on the shift function @xmath157 after symmetrization is @xmath158 for the conformally flat 3-metric @xmath159 , we have @xmath160 , hence @xmath161 after replacing in the above equation @xmath162 and @xmath163 by the expressions following from eq . and dropping all contributions proportional to @xmath164 , we arrive at : @xmath165 the evolution equation for the field momenta @xmath42 is known in @xmath6 dimensions as well [ cf . ( 4.71 ) in ref . it reads @xmath166 + \frac{n}{\sqrt{\gamma } } \bigg [ \pi^{ij } \pi^k_{\ , k } - 2 \pi^i_{\ , k}\pi^{kj } + \frac{1}{2 } \gamma^{ij } \left(\pi^{kl } \pi_{kl } - \frac{1}{2 } ( \pi^k_{\,k})^2 \right ) \bigg ] \nonumber\\[1ex ] & - \left [ \pi^{kj } d_k n^i + \pi^{ki } d_k n^j - d_k ( \pi^{ij } n^k ) \right ] + \frac{8\pi g}{c^4 } n \sum_{a = 1}^\mathcal{n } \frac{p_{ak}p_{al}}{m_a } \gamma^{ik}\gamma^{jl } \left(1+\frac{p_{am}p_{an}}{m_a^2 c^2}\gamma^{mn}\right)^{-\frac{1}{2 } } \delta_a \,.\end{aligned}\ ] ] in the adm gauge , the trace ( with respect to flat metric @xmath68 ) of reduces to @xmath167 + \frac{8\pi g}{c^4 } n \psi^{-\frac{8}{d-2 } } \sum_{a = 1}^\mathcal{n } \frac{p_{a}^2}{m_a } \left(1+\frac{p_a^2}{m_a^2 c^2}\psi^{-\frac{4}{d-2}}\right)^{-\frac{1}{2 } } \delta_a = 0 \ , . \end{gathered}\ ] ] making now use of eqs . , , , and of the two useful relations @xmath168 and @xmath169 , eq . can be rewritten as @xmath170 + \frac{8\pi g}{c^4 } \psi^{-\frac{d+2}{d-2 } } n \sum_{a = 1}^\mathcal{n } \frac{p_a^2}{m_a } \left(1+\frac{p_a^2}{m_a^2 c^2}\psi^{-\frac{4}{d-2}}\right)^{-\frac{1}{2 } } \delta_a = 0 \,.\end{gathered}\ ] ] the form of the latter equation suggests to solve it for the auxiliary function @xmath171 rather than for the lapse @xmath172 @xcite . at this stage , we apply to the right - hand side of eq . the replacement which amounts to deleting a flesh term in the product @xmath173 . since the newtonian potential comes from the @xmath3 part of the lapse , the term @xmath174 is indeed a 2pn quantity and may be neglected consistently with our approximation . finally , we obtain a poisson equation for @xmath175 with the source term being some linear combination of the dirac deltas : @xmath176 + m_a \psi_a^{-2 } \bigg(1+\frac{p_a^2}{m_a^2c^2}\psi_a^{-\frac{4}{d-2}}\bigg)^{-\frac{1}{2 } } \left ( 1 + \frac{d}{d-2}\frac{p_a^2}{m_a^2c^2}\psi_a^{-\frac{4}{d-2 } } \right ) \bigg ) \delta_a \ , . \end{gathered}\ ] ] following the same argument as in the derivation of the conformal factor in sec . [ sec : skeleton ] , we may write @xmath177 so that @xmath178 . we next replace the left - hand side of eq . by virtue of the latter relation , and equate the coefficients of @xmath26 . we are led to @xmath179 & \quad + m_a \psi_a^{-2 } \left(1+\frac{p_a^2}{m_a^2c^2 } \psi_a^{-\frac{4}{d - 2}}\right)^{-\frac{1}{2 } } \nonumber\\[1ex ] & \qquad \times \left(1 + \frac{d}{d - 2 } \frac{p_a^2}{m_a^2c^2 } \psi_a^{-\frac{4}{d - 2 } } \right ) \bigg ) , \quad a=1,\,\ldots,\,\mathcal{n}.\end{aligned}\ ] ] this system of @xmath0 equations for @xmath0 unknowns provides in principle the values of the monopoles @xmath180 entering the expression of @xmath175 , eq . . the lapse function itself is then given by @xmath181 for @xmath182 and @xmath100 , we recover the result of ref . @xcite in the time - symmetric limit @xmath152 ( with @xmath183 ) . in order to solve eq . for the shift function @xmath184 , we take the first and second order space derivatives of both sides . we obtain : @xmath185 \label{eq : equation_divergence_ni_wilson } \frac{d-1}{d } \delta(\partial_i n^i ) & = - \partial_{ij } \left(\psi^{-\frac{2d}{d-2 } } n \pi^i_{\ , j } \right ) \ , . \end{aligned}\ ] ] the source term of the second equation is the sum of a rational function of @xmath186 , @xmath11 , namely @xmath187 , plus a distribution , @xmath188 . the poisson integral of the term with non - compact support is not explicitly known , but that of the distributional term is easy to compute . now , the new `` flesh '' term does not contribute at the newtonian level . it is therefore at least post - newtonian and will eventually play a role in the dynamics only at the 2pn order ; it will be consistently neglected in the present model . thus , the substitution will be supplemented by the replacement : @xmath189 = -\frac{8\pi g}{c^3 } \psi^{-\frac{2d}{d-2 } } n \sum_{a=1}^\mathcal{n } p_{ai } \delta_a \,.\end{gathered}\ ] ] taking eq . into account , eq . becomes @xmath190 the solution of can be obtained with the help of formula . it reads @xmath191 & \times \sum_{a=1}^\mathcal{n } \chi_a \psi_a^{-\frac{3d-2}{d-2 } } p_{ai } \partial_i \left ( \frac{1}{r_a^{d-2 } } \right ) \,.\end{aligned}\ ] ] the poisson equation after inserting the above expression for @xmath192 and performing the substitution takes the form @xmath193 & \quad \times \sum_{a=1}^\mathcal{n } \chi_a \psi_a^{-\frac{3d-2}{d-2 } } p_{aj}\partial_{ij}\left(\frac{1}{r_a^{d-2}}\right ) \nonumber\\[1ex ] & + \frac{16\pi g}{c^3 } \sum_{a=1}^\mathcal{n } \chi_a \psi_a^{-\frac{3d-2}{d-2 } } p_{ai } \delta_a \,.\end{aligned}\ ] ] by virtue of the formulas and , the solution of eq . is simply @xmath194 \times \bigg ( \frac{d-2}{(4-d)(d-1 ) } p_{aj } \partial_{ij } r_a^{4-d } - 4 \frac{p_{ai}}{r_a^{d-2 } } \bigg ) \,.\end{gathered}\ ] ] for a black - hole binary , the number of coupled equations composing the system that gives the effective masses reduces to two , and the sum @xmath195 contains only one term . it is now convenient to regard the two quantities @xmath196 as the unknowns : @xmath197 a = 1,2 , \quad b \neq a.\end{gathered}\ ] ] let us restrict ourselves to the space dimension @xmath198 . the constant @xmath199 reads then @xmath200 , and the system consists of two equations for the two unknowns @xmath201 : @xmath202 \quad b=1,2 , \quad a \neq b \ , .\end{gathered}\ ] ] it is clearly possible to decouple the two equations . in appendix [ app : polynomial ] we show that each @xmath201 is a root of a polynomial of order 200 . the equations of motion of a black - hole binary read [ eq : generic_equations_of_motion ] @xmath203 \dot{p}_{ai } & = -\frac{\partial h}{\partial x_{a}^i},\end{aligned}\ ] ] where the skeleton hamiltonian @xmath144 expressed in terms of the quantities @xmath196 equals @xmath204 using the theorem of implicit differentiation , we put eqs . in a form apparently more convenient for numerical integration . we start by differentiating eq . with respect to the positions . we find ( here @xmath205 and @xmath206 ) : @xmath207 \label{eq : psi_derivative_2_equation } \partial_{2i}\psi_b & \equiv \frac{\partial\psi_b}{\partial x_2^i } = \tilde{\eta}_{a2}^i - \zeta_a \partial_{2i } \psi_a,\end{aligned}\ ] ] where [ eq : tilde_variables ] @xmath208 & + \frac{g \partial_{1i } ( p_{aj } v_{aj}/r_{12})}{2 c^3 \psi_a^7 } , \\[1ex ] \label{eq : eta_a2 } \tilde{\eta}_{a2}^i \equiv & -\tilde{\eta}_{a1}^i , \\[1ex ] \label{eq : zeta_a } \zeta_a \equiv & -\frac{\partial \psi_b(x^i_c , p_{ci},\psi_c)}{\partial \psi_a } \nonumber\\[1ex ] = & \frac{g m_a}{2 c^2 r_{12 } \psi_a^2 } \frac{1 + 3 p_a^2/(m_a^2 c^2 \psi_a^4 ) } { [ 1 + p_a^2/(m_a^2 c^2 \psi_a^4)]^{\frac{1}{2 } } } \nonumber\\[1ex ] & + \frac{7 g \ , p_{aj } v_{aj}}{2 c^3 r_{12 } \psi_a^8 } \nonumber \\[1ex ] = & \frac{\chi_b - 1}{\chi_a}.\end{aligned}\ ] ] equality follows from the fact that the right - hand side of relation depends on @xmath209 and @xmath210 only through the black - hole separation @xmath211^\frac{1}{2}$ ] , and that we have @xmath212 . solving eqs . with respect to @xmath213 and eqs . with respect to @xmath214 , we find [ eq : psi_derivatives ] @xmath215 \partial_{2i}\psi_b & = -\partial_{1i}\psi_b \ , .\end{aligned}\ ] ] it remains to evaluate the momentum derivatives @xmath216 and @xmath217 . the system of equations for @xmath218 is given by @xmath219 where we have posed : @xmath220 & \quad + \frac{g\,\partial_{p_{1i}}(p_{aj}v_{aj})}{2 c^3 r_{12 } \psi_a^7}.\end{aligned}\ ] ] the coefficients @xmath221 are identical to those appearing in @xmath213 because they represent the result of the differentiation of the right - hand side of relation with respect to @xmath196 at @xmath12 and @xmath13 constant . the solution of eqs . with respect to @xmath218 reads @xmath222 from which we deduce @xmath223 by exchanging the particle labels . it is convenient to introduce the rescaled variables @xmath224 making use of eqs . , , , and , the equations of motion can be written in the form [ eq : equations_of_motion ] @xmath225 \dot{p}_{1i } & = -\frac{c^2}{1 - \zeta_1\zeta_2 } \big ( ( 1 - \zeta_2 ) \eta_1^i - ( 1 - \zeta_1 ) \eta_2^i \big ) \nonumber\\[1ex ] & \quad - \frac{2 c^4}{g } ( \psi_1 + \psi_2 - 2 ) n_{12}^i \nonumber \\[1ex ] & = \frac{c^2}{1 - \zeta_1 \zeta_2 } \big((1 - \zeta_1 ) \zeta_2 \eta_1^i - ( 1 - \zeta_2 ) \zeta_1 \eta_2^i \big ) \nonumber \\[1ex ] & \quad - c \left(\frac{1}{\psi_1 ^ 7 } + \frac{1}{\psi_2 ^ 7}\right ) \partial_{1i } ( p_{1j } v_{1j } ) \,,\end{aligned}\ ] ] where the last equality follows from relations and definitions . similar equations for @xmath226 and @xmath227 hold with the role of labels 1 and 2 exchanged . knowing the positions and momenta at a given time @xmath228 , we can solve the algebraic equations numerically , calculate @xmath229 , @xmath230 and @xmath221 by means of eqs . , , , and determine @xmath231 as well as @xmath232 from the evolution equations . as an example of application of our skeleton hamiltonian , we shall discuss in this section the parameters of the last stable ( circular ) orbit ( lso ) in the relative motion of two inspiraling compact objects . this class of systems is of primary importance , being one of the most promising sources of detectable gravitational waves for ground - interferometry experiments as ligo ( laser interferometer gravitational wave observatory ) , virgo , geo600 , and tama300 . because of gravitational - wave emission , the separation between the bodies decreases adiabatically in time while the frequency increases according to the relativistic version of the kepler law . the process ends as soon as the first instable orbit is reached before the coalescence phase . the point where this occurs changes according to the model ( see refs . @xcite and references therein for some comparisons ) . for binary black holes the skeleton hamiltonian reads @xmath233 , where @xmath234 and @xmath235 are solutions of the coupled system of equations . it is impossible to solve this system analytically , but it is not difficult to solve it perturbatively within the post - newtonian setting , treating @xmath236 as a small parameter . at the @xmath237pn approximation , we search the post - newtonian expansion of the coefficients @xmath125 in the form @xmath238 note that we need to carry the series up to the order @xmath239 in @xmath240 to get an @xmath237pn - accurate dynamics with respect to the exact skeleton evolution ( which first differs from the exact einstein s evolution at the 2pn order ) . we now substitute into eqs . and expand their right - hand sides up to the order @xmath239 . by identifying the coefficients of different powers @xmath241 , @xmath242 , we obtain a system of algebraic equations for @xmath243 . we display here the first three equations : @xmath244 \frac{\alpha_1^{(1)}}{m_1 } & = - \frac{(p_{1i}p_{1i})^2}{8m_1 ^ 4 } + \frac{g m_2}{2r_{12 } } \left ( - \frac{5 p_{1i } p_{1i}}{2 m_1 ^ 2 } + \frac{7 p_{1i } p_{2i}}{2 m_1 m_2 } \right . \\[2ex ] & \quad \left . + \frac{n_{12}^i n_{12}^j p_{1i } p_{2j}}{2 m_1 m_2 } - \frac{\alpha_2^{(0)}}{m_2 } \right ) + \frac{g^2 m_2 ^ 2}{4 r_{12}^2 } , \\[2ex ] \frac{\alpha_1^{(2)}}{m_1 } & = \frac{(p_{1i } p_{1i})^3}{16 m_1 ^ 6 } + \frac{g m_2}{2 r_{12 } } \left(\frac{9 ( p_{1i}p_{1i})^2}{8 m_1 ^ 4 } - \frac{5 p_{1i}p_{1i } \alpha_2^{(0)}}{2 m_1 ^ 2 m_2 } \right . \\[2ex ] & \quad \left . - \frac{\alpha_2^{(1)}}{m_2 } \right ) + \frac{g^2 m_2 ^ 2}{2 r_{12}^2 } \left ( \frac{15 p_{1i } p_{1i}}{4 m_1 ^ 2 } - \frac{49 p_{1i } p_{2i}}{4 m_1 m_2 } \right . \\[2ex ] & \quad \left . - \frac{7 n_{12}^i n_{12}^j p_{1i } p_{2j}}{4 m_1 m_2 } + \frac{\alpha_2^{(0)}}{m_2 } \right ) - \frac{g^3 m_2 ^ 3}{8 r_{12}^3 } \,;\end{aligned}\ ] ] similar equalities hold with labels 1 and 2 exchanged . it is straightforward to solve up this system to the required order . for convenience , we shall consider instead of @xmath144 the dimensionless hamiltonian ( after ruling out the constant rest - mass contribution ) @xmath245 @xmath246 denoting the reduced mass of the system . at the @xmath237pn level , we have @xmath247 we restrict our study to circular orbits in the center - of - mass reference frame . within the adm gauge , the frame shift is achieved by imposing the condition @xmath248 whereas circularity requires that @xmath249 the form of @xmath250 becomes particularly simple when use is made of the reduced relative position and momentum : @xmath251 when both relations and hold , the hamiltonian can be expressed with the help of @xmath252 and @xmath253 only . notice also that it depends on the masses @xmath147 and @xmath254 exclusively through the symmetric mass ratio @xmath255 . in the next step , we introduce the conserved ( reduced ) angular momentum @xmath256 along the cirular orbit of radius @xmath257 : @xmath258 after eliminating the linear momentum @xmath253 by means of eq . , the dimensionless energy @xmath259 of the binary system becomes a function of the variables @xmath257 and @xmath256 : @xmath260 the coordinate radius @xmath257 of the circular orbit and the angular momentum @xmath256 are related through one of the canonical equations of motion , namely @xmath261 a relation that may be regarded as expressing the quasi - stationarity of the configuration . remarkably , it can be shown in this case , i.e. , for circular motion , that the adm mass of the system equates the so - called komar mass @xmath262 , i.e. , the mass - like quantity proportional to the monopole part of @xmath263 at spatial infinity satisfying the relation @xmath264 @xcite . the proof of the equality @xmath265 is given in appendix [ app : komar ] for an arbitrary number of space dimensions . equation is solved perturbatively for @xmath266 . by plugging the result into eq . , we find the gauge invariant link between the energy @xmath267 and the angular momentum @xmath256 in the case of circular motion : @xmath268 = & -\frac{1}{2j^2 } - \frac{1}{8j^4 } ( 9+\nu ) \varepsilon \nonumber\\[1ex ] & - \frac{1}{16j^6 } \big(81 + 41\nu-5\nu^2\big ) \varepsilon^2 + { \cal o}(\varepsilon^3).\end{aligned}\ ] ] it is convenient to rewrite formula in terms of the dimensionless variable @xmath269 the parameter @xmath270 is now treated as a small quantity of the order @xmath240 [ cf . the definition establishes the link between @xmath270 and @xmath256 . we invert it order by order to obtain @xmath256 as a function of @xmath270 . inserting @xmath271 into eq . leads to a gauge invariant expression of the energy @xmath267 depending on the parameter @xmath270 . the calculation is performed at the 10pn approximation in order to come reasonably close to the exact result . we display here the explicit formula with the 3pn - accuracy . it reads @xmath272 the higher order post - newtonian coefficients @xmath273 ( for @xmath274 ) can be found in appendix [ app:10pn ] . we want to compare the result of eq . with the relation @xmath275 of the wilson - mathews model ( remember that in this model the post - newtonian expansion terminates at the 2pn order ) in eq . ( c4 ) of ref . @xcite and replacing there the angular momentum @xmath256 by @xmath276 according to eq . ( b1 ) of ref . @xcite . ] : @xmath277 the 2pn coefficient proportional to @xmath278 differs from that of the post - newtonian computations carried out in general relativity . indeed , the energy for circular orbits in einsteinian theory , up to the 3pn order , is @xcite @xmath279 where , according to ref . @xcite we set the parameter @xmath280 ( defined in paper @xcite ) equal to zero ( @xmath281 ) , getting the solution corresponding to the brill - lindquist initial data . by comparing eqs . and it is obvious that our energy @xmath267 starts to differ from the general - relativistic one , @xmath282 , at the 2pn order . cdddddddddd & & & & & & & & & & + @xmath283 & 0.5224 & 0.1077 & 0.0725 & 0.0624 & 0.0583 & 0.0563 & 0.0554 & 0.0548 & 0.0546 & 0.0544 + @xmath267 & -0.1622 & -0.0689 & -0.0566 & -0.0529 & -0.0514 & -0.0508 & -0.0505 & -0.0503 & -0.0503 & -0.0502 the position @xmath284 of the last stable orbit is determined from the requirement that @xmath285 . the corresponding energy @xmath286 [ computed by means of eq . ] is depicted in fig . [ fig : ex_plot ] for equal - mass binaries ( i.e. , for @xmath287 ) as a function of the dimensionless orbital frequency @xmath288 . numerical values are given in table [ tab : ex_table ] . the case of a test particle moving in a schwarzschild background is also plotted for comparison . similarly to what happens in einsteinian theory @xcite , the instabilities arise `` earlier '' ( i.e. , at lower frequency ) for higher post - newtonian orders in the skeleton model . the 2pn last stable orbit of our skeleton dynamics is quite far from that of general relativity in the sense that the absolute difference @xmath289 ( @xmath290 ) between the skeleton frequency @xmath291 ( the skeleton energy @xmath292 ) and the einsteinian one @xmath293 ( @xmath294 ) is not negligible with respect to the general - relativistic value : @xmath295% ( @xmath296% ) . the location of the wilson - mathews data suggests that modifications of the extrinsic curvature affect the parameters of the last stable orbit more than the choice of a conformally flat space metric @xcite . the 3pn skeleton values are even further away from the einsteinian 3pn ones , but they are closer to those of the effective one - body approach improved by pad approximants than the straight 3pn einsteinian values . the 10pn data are particularly interesting , considering that they may be regarded as good approximants to the full skeleton model . remarkably , their position in the @xmath297 plane relative to the 3pn data is nearly the same as that of the point referring to numerical simulations relative to the point referring to the 3pn data in general relativity . in fig . [ fig : ex_plot ] , the relative location of the last stable orbit indicated by the white circle obtained from numerical computations with respect to the 3pn result for corotating bodies ( black triangle ) is somehow plausible if we believe that it is essentially due to the wilson - mathews truncation . it is indeed comparable to the relative location of irrotational 2pn einsteinian and wilson - mathews data . we also want to point out that a realistic estimation of the last stable orbit does not necessarily follow from the standard 3pn calculation . it may well be close to the star - like point derived from the effective one - body model with pad approximant . there is currently no definite argument to decide what is the most likely statement . as a result , we regard the position of these points as defining the present uncertainty in the location of the last stable orbit derived analytically . p.j . thanks the theoretisch - physikalisches institut of the fsu jena for hospitality during the realization of the present work . is thankful to the spring-2003 numerical relativity group at caltech , particularly greg cook , lee lindblom , niall murchadha , and masaru shibata , for stimulating discussions . we also thank christian thierfelder for useful comments about the presentation of the evolution equations . this work is supported by the eu programme `` improving the human research potential and the socio - economic knowledge base '' ( research training network contract hprn - ct-2000 - 00137 ) , by the caltech visitors program , and by the deutsche forschungsgemeinschaft ( dfg ) through sfb / tr7 `` gravitationswellenastronomie '' . got an additional support from the polish kbn grant no . 5 p03b 034 20 . equation for @xmath298 , after multiplication by @xmath299 , can be written in the form @xmath300 , where @xmath301 and @xmath302 ^ 2 [ \psi_1 ^ 4 + p_1 ^ 2/(m_1 ^ 2 c^2)]$ ] are polynomials in @xmath303 with non - zero constant terms , of order 7 and 4 , respectively . a similar relation holds for @xmath304 . we now subtract @xmath305 from the left and right - hand side , and square the new identity . we obtain : @xmath306 = \left(\frac{g m_2}{2 r_{12 } c^2 } \right)^2 \psi_2 ^ 8 \left(\psi_2 ^ 4 + \frac{p_2 ^ 2}{m_2 ^ 2 c^2}\right ) \,.\end{gathered}\ ] ] next , we multiply the equality ( [ eq : equation_gamma1 ] ) by @xmath307 , make the substitution @xmath308 , and expand the terms @xmath309 , @xmath310 , in powers of @xmath311 with the help of the binomial formula . if @xmath237 is even , the contribution @xmath312 , with @xmath313 $ ] , is of polynomial kind , whereas that of @xmath314 has the form of a polynomial @xmath315 multiplied by the algebraic function @xmath316 ; @xmath317 and @xmath315 have degrees @xmath318= 7n$ ] and @xmath319 = 7n - 3 $ ] , respectively . the same decomposition holds for odd @xmath237 with different expressions for @xmath320 and @xmath321 . at last , we gather all polynomial quantities into the first member of the equality while the non polynomial contributions @xmath322 are recast to the second member . the final equation is @xmath323 \nonumber\\ & = - p_{1 \ , ( 4)}^{1/2 } \bigg[(\psi_1 - 1)^2 \sum_{k=0}^6 \genfrac{(}{)}{0pt}{}{14}{2k + 1 } p_{1\ , ( 7)}^{2k+1 } \psi_1^{4 ( 13 - 2k ) } p_{1\ , ( 4)}^{6-k } - \frac{g p_{2i } v_{2i}}{r_{12 } c^3 } ( \psi_1 - 1 ) \psi_1^{49 } \sum_{k=0}^3 \genfrac{(}{)}{0pt}{}{7}{2k } p_{1\ , ( 7)}^{2k } \psi_1^{4 ( 7 - 2k ) } p_{1\ , ( 4)}^{3-k } \nonumber\\ & \qquad \qquad - \left(\frac{g m_2 \psi_1 ^ 7}{2 r_{12 } c^2 } \right)^2 \bigg(\sum_{k=0}^5 \genfrac{(}{)}{0pt}{}{12}{2k+1 } p_{1 \ , ( 7)}^{2k+1 } \psi_1^{4 ( 11 - 2k ) } p_{1 \ , ( 4)}^{5-k } + \frac{p_2 ^ 2 \psi_1^{28}}{m_2 ^ 2 c^2 } \sum_{k=0}^3 \genfrac{(}{)}{0pt}{}{8}{2k+1 } p_{1\ , ( 7)}^{2k+1 } \psi_1^{4 ( 7 - 2k ) } p_{1\ , ( 4)}^{3-k } \bigg ) \bigg ] \,.\end{aligned}\ ] ] we see that the left - hand side of eq . is made of a sum of polynomials of degrees 100 , 99 , 98 , 98 , 98 , the first of them involving the only non - zero constant term . the right - hand side is the product of @xmath322 and polynomials of degrees 97 , 96 , 95 , 95 respectively that cancel when @xmath324 . as a consequence , @xmath303 is the root of a polynomial of degree 200 with a non - zero constant term . in this appendix , we restrict ourselves to a binary system . the black holes are thus labeled by @xmath325 . our starting point is the system of two sets of equations satisfied by the monopoles @xmath125 and @xmath180 entering the conformal factor @xmath49 and the auxiliary function @xmath171 : [ eq : alpha_beta_binary ] @xmath326 & \quad + m_a \psi_a^{-1 } \left(1+\frac{p_a^2}{m_a^2c^2}\psi_a^{-\frac{4}{d-2}}\right)^{\frac{1}{2 } } \ , , \\[2ex ] \label{eq : beta_a_binary } \beta_a & = \chi_a \bigg ( \frac{3d-2}{d-2 } \frac{p_{ai}v_{ai}}{c } \psi_a^{-\frac{4(d-1)}{d-2 } } \nonumber\\[2ex ] & \quad + m_a \psi_a^{-2 } \left(1+\frac{p_a^2}{m_a^2c^2}\psi_a^{-\frac{4}{d-2}}\right)^{-\frac{1}{2 } } \nonumber\\[2ex ] & \qquad \times \left(1 + \frac{d}{d-2 } \frac{p_a^2}{m_a^2c^2 } \psi_a^{-\frac{4}{d-2 } } \right ) \bigg ) \ , , \end{aligned}\ ] ] with ( here @xmath327 ) [ eq : psi_chi_binary ] @xmath328 \label{eq : chi_a_binary } \chi_a & = 1 - a_d \frac{\beta_b}{r_{12}^{d - 2 } } \,.\end{aligned}\ ] ] it is straightforward to show that for a binary system described by the functions @xmath49 and @xmath175 of the purely monopolar form given by eqs . and , the adm mass coincides with its komar mass if and only if the quantities @xmath125 and @xmath180 satisfy the simple relation : @xmath329 we shall assume that the relative orbits are circular . after introducing the reduced relative momentum @xmath253 and eliminating its square through the relation @xmath330 , all quantities become functions of @xmath331 and @xmath256 only . by virtue of eqs . , the effective mass @xmath125 can then be treated as a function of the _ three _ variables @xmath196 , @xmath331 , and @xmath256 : @xmath332 with the help of eqs . , it is not difficult to check that the following relations , crucial for our proof , are fulfilled by @xmath125 : [ eq : alpha_a_derivative ] @xmath333 \label{eq : alpha_a_derivative_r12 } \frac{\partial \alpha_a(\psi_a , r_{12 } , j)}{\partial r_{12 } } & = \frac{d - 2}{2r_{12 } } \bigg(\alpha_a - \beta_a \frac{\psi_a}{\chi_a } \bigg)\ , . \end{aligned}\ ] ] note that , writing down eqs . , we have already employed the relations . the adm hamiltonian , restricted to circular orbits , reads @xmath334 the circularity condition implies @xmath335 , which is equivalent to @xmath336 after substituting into we obtain the equality ( valid along circular orbits ) : @xmath337 according to eq . , it can be rewritten as @xmath338 on the other hand , the equations in the circular case take the form @xmath339 by differentiating both sides with respect to @xmath331 and resorting to relations , we obtain the system of two equations ( where @xmath340 ) @xmath341 which we solve with respect to the derivatives @xmath342 . after eliminating @xmath196 and @xmath343 by means of relations , the solution reduces to @xmath344 to complete the proof of eq . , it is enough to substitute eqs . into the left - hand side of equality and make use of eqs . . the coefficients @xmath273 ( for @xmath274 ) entering formula for energy @xmath267 along cirular orbits as a function of the parameter @xmath270 read @xmath345 e_6 & = & \frac{45927}{1024 } + \frac{662619}{1024 } \nu - \frac{233737}{128 } \nu^2 + \frac{1162513}{1024 } \nu^3 + \frac{56481}{1024 } \nu^4 - \frac{3777}{1024 } \nu^5 , \\[2ex ] e_7 & = & \frac{264627}{2048 } + \frac{16343855}{6144 } \nu - \frac{136513223}{18432 } \nu^2 + \frac{67182599}{27648 } \nu^3 + \frac{249565613}{31104 } \nu^4 - \frac{766814851}{497664 } \nu^5 - \frac{148888223 } { 13436928 } \nu^6 , \\[2ex ] e_8 & = & \frac{12196899}{32768 } + \frac{993480761}{98304 } \nu - \frac{403684801}{16384 } \nu^2 - \frac{19713422131}{884736 } \nu^3 + \frac{532659868169}{3981312 } \nu^4 - \frac{1096304004055}{11943936 } \nu^5 \nonumber\\[2ex ] & & + \frac{42034682027}{17915904 } \nu^6 + \frac{29197027769}{644972544 } \nu^7 , \\[2ex ] e_9 & = & \frac{70366725}{65536 } + \frac{2394150375}{65536 } \nu - \frac{13003419935}{196608 } \nu^2 - \frac{9560242585}{32768 } \nu^3 + \frac{248408863105}{196608 } \nu^4 - \frac{129128080475}{98304 } \nu^5 \nonumber\\[2ex ] & & + \frac{30481497595}{98304 } \nu^6 + \frac{189766295}{16384 } \nu^7 + \frac{1599565 } { 65536 } \nu^8 , \\[2ex ] e_{10 } & = & \frac{813439341}{262144 } + \frac{100457157571}{786432 } \nu - \frac{69694465361}{589824 } \nu^2 - \frac{44773413415657}{21233664 } \nu^3 + \frac{558727148465215}{63700992 } \nu^4 \nonumber\\[2ex ] & & - \frac{2097054471026245}{191102976 } \nu^5 + \frac{500159286898463}{143327232 } \nu^6 + \frac{858674754130903}{2579890176 } \nu^7 - \frac{233795515077109}{5159780352 } \nu^8 \nonumber\\[2ex ] & & - \frac{177721652120353}{417942208512 } \nu^9 , \\[2ex ] e_{11 } & = & \frac{4710988269}{524288 } + \frac{228788885795}{524288 } \nu + \frac{411952929451}{4718592 } \nu^2 - \frac{28395688669829}{2359296 } \nu^3 + \frac{3163918218071455}{63700992 } \nu^4 \nonumber\\[2ex ] & & - \frac{8137959523310875}{127401984}\nu^5 + \frac{6129421057080619}{1146617856 } \nu^6 + \frac{9328338011915891}{322486272 } \nu^7 - \frac{180325112745666065}{30958682112 } \nu^8 \nonumber\\[2ex ] & & - \frac{11728912802084585}{278628139008 } \nu^9 + \frac{1098645689138995}{2507653251072 } \nu^{10}.\end{aligned}\ ] ]

we study the combinatorics of tropical hyperplane arrangements , and their relationship to ( classical ) hyperplane face monoids . we show that the _ refinement _ operation on the faces of a tropical hyperplane arrangement , introduced by ardila and develin in their definition of a _ tropical oriented matroid _ , induces an action of the hyperplane face monoid of the classical braid arrangement on the arrangement , and hence on a number of interesting related structures . along the way , we introduce a new characterization of the _ types _ ( in the sense of develin and sturmfels ) of points with respect to a tropical hyperplane arrangement , in terms of _ partial bijections _ which attain permanents of submatrices of a matrix which naturally encodes the arrangement . marianne johnson and mark kambites school of mathematics , university of manchester , + manchester m13 9pl , england . [ multiblock footnote omitted ]

1604.03016

the set of faces of a ( central ) hyperplane arrangement , and more generally the set of covectors of an oriented matroid , provide an important source of examples in the class of finite monoids known as _ left regular bands_. such monoids have been the focus of much research interest , following the celebrated work of bidigare , hanlon and rockmore @xcite who showed that a number of well - known markov chains , including the tsetlin library and the riffle shuffle , are random walks on the faces of a hyperplane arrangement , and that the representation theory of the hyperplane face monoid could be used to analyze these markov chains . this observation created substantial interest in random walks on hyperplane face monoids and related semigroups @xcite and the theory of left regular band walks more generally @xcite , and in various aspects of the representation theory of hyperplane face monoids @xcite , general finite left regular bands @xcite and @xmath0-trivial monoids @xcite , as well as the study of several related ( quasi)varieties of semigroups and monoids @xcite . the representation theory of hyperplane face monoids is also closely connected to the solomon descent algebra ( see @xcite for further details ) and hyperplane face monoids themselves also arise in connection with combinatorial hopf algebras @xcite . the study of the tropical complex generated by the columns of a real @xmath1 matrix @xmath2 ( see section [ sect : tropfaces ] ) was initiated by develin and sturmfels @xcite . it has been observed @xcite that such a tropical complex can be viewed as the analogue of the abstract simplicial complex of faces resulting from a real hyperplane arrangement , by associating to each column of @xmath2 a min - plus linear form specifying a min - plus hyperplane ( this will be explained in detail in section 3 ) . in the case of real hyperplane arrangements , the corresponding face poset is a monoid with a natural geometrically ( or combinatorially ) defined product . there is no obvious analogous monoid structure on the face poset of a tropical hyperplane arrangement , but we shall show instead that the hyperplane face monoid @xmath3 of the classical _ braid arrangement _ ( see section [ sect : braid ] ) acts upon the set of faces determined by @xmath4 min - plus tropical hyperplanes in @xmath5 . the faces of the tropical complex generated by @xmath2 are labelled by _ types _ which are the tropical analogue of the _ sign sequences _ of a real hyperplane arrangement . we identify the collection of all types with the face poset @xmath6 of the tropical complex this is the analogue of the oriented matroid of covectors of a real hyperplane arrangement and hence provided the inspiration for the notion of a _ tropical oriented matroid _ introduced by ardila and develin @xcite . we shall show that their definition induced a natural right action of the monoid @xmath3 on the tropical face poset @xmath6 and there is a geometric interpretation of this action when viewing the elements of @xmath6 as faces . by viewing @xmath6 instead as a poset of boolean matrices ( the types ) , this action is just the restriction of the action of @xmath3 on the set of all @xmath7 boolean matrices described in section [ sect : action ] below . without the geometric intuition of the action of @xmath3 it is not so easy to see why each element of @xmath3 should act upon @xmath6 ; indeed , one needs to be able to determine whether a given @xmath7 boolean matrix is in fact a type labelling a face of @xmath6 . in this paper we give a purely combinatorial characterization of the types of @xmath6 , and use this to give a combinatorial argument showing that @xmath3 acts on the set of all types . to this end , we introduce another combinatorial object @xmath8 called the _ permanent structure _ of @xmath2 , and show that this provides an equivalent encoding of the combinatorial data of the tropical complex generated by @xmath2 ( corollary [ cor : sameinfo ] ) . this may be of independent interest . the paper is organised as follows . in section 2 we recall some facts about real hyperplane arrangements and the hyperplane face monoid @xmath3 , including its natural action on the power set of @xmath9 . in section 3 we recall the necessary background on tropical hyperplane arrangements . in section 4 we begin by making a distinction between boolean matrices whose corresponding order theoretic conditions can be _ satisfied _ by a point and those which can not . we show that a boolean matrix is _ satisfiable _ if and only if every partial bijection contained within it ( as a submatrix ) is satisfiable . in section 5 we show that the satisfiable partial bijections are precisely the max - plus permanent - attaining ones ( and hence a matrix is satisfiable if and only if every partial bijection contained within it attains the permanent ) . in section 6 we give a precise characterization of those matrices which correspond to types , in terms of permanent - attaining partial bijections . as an application , we obtain a combinatorial understanding of the action of @xmath3 on @xmath6 . we denote by @xmath10 $ ] the set @xmath9 . a key object of study will be the set of @xmath4-tuples of subsets of @xmath10 $ ] . it is convenient to identify each such tuple @xmath11 with the @xmath7 zero - one ( boolean ) matrix @xmath2 such that @xmath12 if and only if @xmath13 . we write @xmath14 for the set of all @xmath7 boolean matrices . we write @xmath15 and @xmath16 for the @xmath17th row and @xmath18th column respectively of a ( boolean or other ) matrix @xmath2 . for a boolean matrix @xmath2 we will also view each row @xmath15 ( respectively , column @xmath19 ) as the corresponding subset of @xmath20 $ ] ( respectively @xmath10 $ ] ) when convenient . notice that the _ dual _ @xmath21-tuple @xmath22 of subsets of @xmath20 $ ] , where @xmath23 if and only if @xmath24 corresponds to the transpose matrix , so with our identification , we have the natural notation @xmath25 . there is an obvious partial order on @xmath14 given by setting @xmath26 if and only if @xmath27 for all @xmath17 and @xmath18 . equivalently in terms of tuples , @xmath28 if and only if @xmath29 for each @xmath17 . by a _ partial bijection _ @xmath30 \dashrightarrow [ n]$ ] we mean a bijection @xmath31 between subsets @xmath32 $ ] and @xmath33 $ ] . we may identify such a partial bijection with the matrix @xmath34 defined by @xmath35 if and only if @xmath36 ; thus the partial bijections correspond to the elements of @xmath37 having at most one non - zero entry in each row and in each column . in this section we recall some necessary background on hyperplane face monoids ( introduced in section [ sect : faces ] ) . our main interest is in the hyperplane face monoid of the braid arrangement ( section [ sect : braid ] ) and a particular action of this monoid on the set of boolean matrices ( section [ sect : action ] ) . consider a finite collection @xmath38 of ( classical ) _ linear hyperplanes _ containing the origin in @xmath5 , each of which may be written in the form @xmath39 for some fixed choice of @xmath40 . the complement of a hyperplane arrangement in @xmath5 is a collection of open subsets of @xmath5 called _ chambers_. more generally , a _ face _ of the arrangement is defined to be a non - empty set which is an intersection of the form @xmath41 where @xmath42 and @xmath43 note that the sequence of signs @xmath44 therefore specifies on which ` side ' of each hyperplane @xmath45 the given face lies . for each @xmath46 and each hyperplane @xmath47 , we let @xmath48 denote the sign indicating the position of the point @xmath49 relative to the hyperplane @xmath47 , so that @xmath50 . it is then clear that the map @xmath51 defined by @xmath52 is constant on faces and that distinct faces have distinct images under @xmath53 . thus we may identify the set of all faces @xmath54 with the set of sign sequences @xmath55 . under this identification , the chambers are those elements of the image in which each component is non - zero , and the face @xmath56 corresponding to the common intersection of all hyperplanes in @xmath57 is identified with the all - zero sequence . it is well known ( see , for example , @xcite ) that the set of all faces @xmath54 of any finite linear hyperplane arrangement can be endowed with the structure of a monoid , whose identity element is @xmath56 , and whose product can be defined geometrically as follows . for faces @xmath58 and @xmath59 of a given hyperplane arrangement , choose points @xmath60 and @xmath61 . the product @xmath62 is the very first face one encounters after leaving point @xmath63 when walking in a straight line towards @xmath64 ( which could be the face @xmath58 itself ) ; this definition does not depend upon the particular points selected , and it can be shown that this product is associative . the resulting finite monoid is called the _ hyperplane face monoid _ of the arrangement and it is well - known and straightforward to check that each such monoid is a _ left regular band _ ( that is , @xmath65 and @xmath66 for all @xmath67 ) and that each chamber is a _ left zero _ of this monoid ( that is , for any chamber @xmath68 , we have @xmath69 for all @xmath70 ) . we refer the reader to @xcite and @xcite for general reference on properties of hyperplane face monoids . the set of signs @xmath71 can be considered as a left regular band consisting of an identity element @xmath72 and two left zeroes , @xmath73 and @xmath74 . it turns out that @xmath53 induces a monoid embedding of @xmath54 in the direct power @xmath75 ( see @xcite for example ) . in other words , upon identifying @xmath54 with the set of possible sign sequences @xmath55 , for each @xmath76 and all @xmath77 we have @xmath78 in fact , as observed in @xcite , the elements of @xmath79 satisfy the covector axioms of an _ oriented matroid _ ( we refer the reader to @xcite for general reference on oriented matroids and also ( * ? ? ? * section 2.3.1 ) for a comparison between the terminology of hyperplane face monoids and oriented matroids ) . in ( * definition 3.5 ) ardila and develin suggested a definition of _ tropical oriented matroid _ , based on their geometric understanding of both tropical hyperplane arrangements and the oriented matroids arising from ( classical ) hyperplane arrangements . we shall show that their definition induces an _ action _ of a certain classical hyperplane face monoid on the faces of a tropical hyperplane arrangement . in the case of coxeter arrangements , the geometric description of the product of faces @xmath62 described above can be viewed as the tits projection of @xmath59 on @xmath58 , denoted @xmath80 @xcite . one well - studied example is the linear hyperplane arrangement @xmath81 known as the _ braid arrangement _ @xmath82 , consisting of @xmath83 hyperplanes in @xmath5 intersecting in the one - dimensional real vector space @xmath84 . the corresponding hyperplane face monoid has received considerable attention in connection with shuffling schemes , random walks and the solomon descent algebra of the symmetric group @xcite . it is clear that the braid arrangement yields @xmath85 chambers of the form @xmath86 for @xmath87 in the symmetric group @xmath88 . moreover , since a face of this arrangement is defined to be a non - empty intersection of the form @xmath89 where @xmath90 , @xmath91 it is easy to see that the faces of the braid arrangement can be identified with ordered partitions of @xmath10 $ ] : for example , the face @xmath92 is identified with the ordered partition @xmath93 . let @xmath3 denote the set of all ordered set - partitions of @xmath10 $ ] . viewing faces @xmath58 and @xmath59 of the braid arrangement as ordered partitions @xmath94 and @xmath95 it is known ( see @xcite or @xcite for example ) that their product @xmath96 may be written concretely as @xmath97 where @xmath98 denotes the operation of deleting any empty sets . it is straightforward to check that the partition @xmath99 acts as an identity on both sides , and that the chambers @xmath100 of the braid arrangement correspond to ordered partitions @xmath101 , each of which is a left zero of the monoid . since @xmath3 is a monoid it acts ( by multiplication ) on both the left and the right of @xmath3 itself . it is clear that each chamber is a fixed point of the _ right action _ of faces , since each chamber is a left zero element of the monoid . we will show that the right action of the monoid @xmath3 on @xmath102 can be extended in a natural way to give a right action on a number of interesting combinatorial structures resulting from a tropical hyperplane arrangement in @xmath5 . we begin by showing that @xmath3 acts on the right of the power set of @xmath10 $ ] as follows : given any subset @xmath33 $ ] and any @xmath103 we define @xmath104 that is , @xmath105 is the right - most non - empty intersection of @xmath106 with a block of the partition @xmath58 . note that this operation is well - defined , since if @xmath107 , then @xmath106 must have non - empty intersection with _ some _ component of the ordered set partition @xmath58 of @xmath10 $ ] . [ prop_paction ] the above operation gives a right action of the face monoid @xmath3 on the power set of @xmath10 $ ] . moreover , @xmath108 for all @xmath33 $ ] and @xmath103 . recall that the identity of the monoid @xmath3 is the trivial partition @xmath109)$ ] ; it is immediate from the definition that @xmath110 ) = i \cap [ n ] = i$ ] . it remains to show that @xmath111 for any @xmath33 $ ] and any two faces @xmath112 . this is clear if @xmath113 , so suppose that @xmath106 is non - empty . by definition , @xmath114 where @xmath18 is the largest index of the product providing a non - zero intersection with @xmath106 . using the concrete interpretation of the product @xmath62 , we conclude that @xmath115 where @xmath17 is the maximal index of @xmath58 such that @xmath116 is non - empty for some @xmath117 , and @xmath118 is the maximal index of @xmath59 such that @xmath119 is non - empty . since @xmath59 is a partition , this is equivalent to saying @xmath120 where @xmath17 is the maximal index of @xmath58 such that @xmath121 is non - empty , and @xmath118 is the maximal index of @xmath59 such that @xmath119 is non - empty , which is precisely the definition of @xmath122 obtained by applying twice . that @xmath108 is immediate from the definition . the last part of proposition [ prop_paction ] should not be misinterpreted as saying that the action of @xmath3 is monotonically decreasing with respect to the containment order : if @xmath123 $ ] then we do _ not _ necessarily have @xmath124 . ( what is true is that either @xmath124 or @xmath125 . ) for any @xmath4 the action of @xmath3 on the powerset of @xmath10 $ ] obviously extends ( componentwise ) to an action on the set of @xmath4-tuples of subsets of @xmath10 $ ] and hence ( through the identification described in the introduction ) on the set @xmath14 of @xmath7 boolean matrices . it follows from proposition [ prop_paction ] that @xmath126 for all @xmath127 and all @xmath103 . in particular , the action of @xmath3 restricts to an action on any subset of @xmath14 which is downward - closed under the partial order . for @xmath128 write @xmath129 , @xmath130 and @xmath131 . the operations @xmath132 and @xmath133 give @xmath134 the structure of a _ semiring _ ( without zero element ) , called the _ max - plus semiring _ ( see @xcite for further reference ) . the min - plus semiring is defined similarly , and negation provides an isomorphism between these two structures , allowing one to easily interpret min - plus results in the max - plus setting and vice versa . the word _ tropical _ is used throughout the literature to describe a number of algebraic and geometric constructions involving either the max - plus or min - plus semiring . for example , it is clear that these operations extend naturally to give notions of addition and scalar multiplication of vectors , hence giving @xmath5 the structure of a ( semi)module over the max - plus or min - plus semiring . a subset of @xmath5 closed under max - plus ( or min - plus ) operations will be referred to as a max - plus ( respectively min - plus ) submodule of @xmath5 ; the finitely generated submodules have been termed max - plus ( min - plus ) `` tropical polytopes '' in the literature . the theory of tropical semimodules and convex sets has been developed by several authors , including cohen , gaubert and quadrat @xcite , and develin and sturmfels @xcite . by analogy with the case of real hyperplanes , each @xmath135 can be used to define a max - plus or min - plus hyperplane , by considering the linear forms @xmath136 in co - ordinate variables @xmath137 . the max - plus ( respectively , min - plus ) hyperplane is then defined to be the set of all points @xmath138 for which the maximum ( respectively minimum ) is _ attained twice _ in expression ( respectively expression ) above . in contrast to hyperplanes in usual euclidean space ( which split the containing space into two half - spaces ) , a tropical hyperplane divides the containing space @xmath5 into @xmath21 tropical `` half''-spaces or _ sectors _ , according to which co - ordinate is maximal ( respectively , minimal ) . it is clear that the point @xmath139 lies on the hyperplane defined by the linear form specified by @xmath140 , and that this point is adjacent to each of the sectors , in the sense that one can move to any given sector by making some small perturbation to the co - ordinates of @xmath140 . we therefore define a ( max - plus / min - plus ) tropical hyperplane with _ apex _ @xmath140 to be the ( max - plus / min - plus ) tropical hyperplane defined by the linear form specified by @xmath141 . note that this tropical hyperplane does not have a unique apex , however the apexes are precisely all tropical scalings @xmath142 . there is a natural partial order on @xmath5 given by @xmath143 if and only if @xmath144 for all @xmath17 . for @xmath145 we write @xmath146 this operation is a _ residuation _ operator ( see @xcite ) and plays an important role in tropical mathematics ( see for example @xcite ) . we say that @xmath49 _ dominates _ @xmath147 in position @xmath17 if @xmath148 . it is straight - forward to verify that domination is scale invariant , that is , if @xmath49 dominates @xmath147 in position @xmath17 , then @xmath149 dominates @xmath150 in position @xmath147 for all @xmath151 . the set @xmath152 consisting of all elements of @xmath5 dominated by @xmath49 in position @xmath17 can be seen to be a max - plus ( and min - plus ) submodule of @xmath5 as well as a closed convex set when considered as a subset of euclidean space @xmath5 ( see ( * ? ? ? * lemma 1.1 ) for example ) . if @xmath153 is the min - plus hyperplane with apex @xmath135 , then the set @xmath154 is the closed sector consisting of points @xmath49 such that @xmath155 for all @xmath118 . since domination is scale invariant , it follows that we may identify each hyperplane @xmath153 and the corresponding domination sets @xmath154 with their respective images in the ( classical ) quotient vector space @xmath156 , termed _ tropical projective space _ , as convenient . it will also be convenient ( for the purpose of drawing pictures ) to identify @xmath156 with @xmath157 , via the linear map @xmath158 . ] figure 1 illustrates a single min - plus tropical hyperplane , viewed in tropical projective space @xmath159 which is identified with @xmath160 ( consisting of three half rays corresponding to the three ways in which the minimum can be attained at least twice ) , together with a labelling of the sectors following the usual co - ordinate axis conventions . note that by setting @xmath161 and extending the half - rays to give full real hyperplanes , one obtains the braid arrangement @xmath162 . let @xmath2 be a real @xmath7 matrix . the columns of @xmath2 determine a min - plus _ tropical hyperplane arrangement _ @xmath163 in @xmath5 , where @xmath164 is the min - plus tropical hyperplane with apex @xmath16 , or in other words , the set of all @xmath165 such that the minimum over @xmath118 of @xmath166 is attained twice . the _ type _ of a point @xmath46 with respect to the hyperplane arrangement specified by @xmath2 is an @xmath7 boolean matrix ( or equivalently , an @xmath4-tuple of subsets of @xmath10 $ ] ) recording in which of the closed sectors relative to each @xmath164 the point @xmath49 lies . specifically , the type of @xmath49 with respect to @xmath2 is the boolean matrix @xmath167 where @xmath168 if and only @xmath169 . types may be viewed as the tropical analogue of the sign sequences associated with a real hyperplane arrangement ; the collection of all types ( whilst not a matroid in the usual sense ) is clearly the analogue of the oriented matroid of covectors of a real hyperplane arrangement , and motivated the notion of _ tropical oriented matroids _ studied in @xcite . develin and sturmfels ( * ? ? ? * theorem 15 ) have shown that the min - plus hyperplane arrangement given by ( the columns of ) @xmath2 induces a polyhedral cell complex structure on @xmath5 , called the _ tropical complex _ generated by @xmath2 , whose face poset @xmath6 is naturally labelled by the types . since domination is scale invariant we note that the type of @xmath149 is the same as the type of @xmath49 for all @xmath170 , @xmath46 . it follows that we may identify the hyperplanes @xmath164 and the faces of the resulting tropical complex with their respective images in tropical projective space @xmath156 , and hence by @xmath157 , when convenient . under this latter identification the faces of the max - plus tropical polytope are precisely the bounded faces in the tropical complex ( * ? ? ? * theorem 15 ) . by the _ dimension _ of a face we shall mean the dimension of the corresponding polyhedron in @xmath157 . consider the matrix @xmath171 the min - plus tropical hyperplane arrangement specified by the columns of @xmath2 is shown in figure 2 . as viewed in projective space , the resulting tropical complex has 12 cells of dimension 2 ( such as the bounded cell @xmath58 and the unbounded cell @xmath172 ) , 18 cells of dimension 1 ( such as the bounded cell @xmath59 , which is a face of both @xmath172 and @xmath58 ) and 7 cells of dimension 0 ( such as the point marked @xmath153 , which is a face of each of the other labelled cells ) . the bounded cells have been shaded , and these cells constitute the max - plus polytope generated by the columns of @xmath2 ( see ( * ? ? ? * theorem 15 ) ) . we identify the cells @xmath173 and @xmath153 with their corresponding types : @xmath174 . the bounded cells of the resulting tropical complex are shaded . the union of the bounded regions is the max - plus polytope generated by the points @xmath175 . ] let @xmath6 denote the set of all faces with respect to the hyperplane arrangement specified by @xmath2 . since the faces are in one - to - one correspondence with the types specified by this arrangement , we shall regard @xmath6 as a subset of @xmath37 . continuing with the notation of section [ sect : tropfaces ] , we fix a real @xmath7 matrix @xmath2 whose columns @xmath16 specify a tropical hyperplane arrangement in @xmath5 ; all types in this section are considered with respect to the columns of @xmath2 . more generally , for each @xmath176 we define @xmath177 and say that @xmath178 is _ satisfiable _ ( with respect to @xmath2 ) if @xmath179 . that is , @xmath178 is satisfiable if there exists @xmath46 such that @xmath16 dominates @xmath49 in position @xmath17 for all @xmath180 such that @xmath181 ; we say that such a point @xmath49 _ satisfies _ @xmath178 . ( for example , with respect to the hyperplane arrangement given in example 3.1 above , the ( column ) tuple @xmath182 is not satisfiable . the tuple @xmath183 is satisfiable , however it is not the type of any point . ) notice that if @xmath178 is satisfiable with respect to @xmath2 then the set @xmath184 , which by definition is a non - empty intersection of euclidean convex max - plus submodules of @xmath5 of the form , is itself a euclidean convex max - plus submodule of @xmath5 . it is clear from the definitions above that @xmath49 satisfies @xmath178 if and only if @xmath185 where @xmath167 is the type of @xmath49 , and that the cells of the tropical complex are the relative interiors of the sets @xmath186 indexed by types @xmath167 . it follows that the set @xmath187 of all satisfiable ( with respect to @xmath2 ) boolean matrices corresponding to the hyperplane arrangement specified by @xmath2 is a downward - closed subset of @xmath37 and hence ( by our observation at the end of section [ sect : action ] ) @xmath3 acts on the right of @xmath187 . the set @xmath6 of all types with respect to @xmath2 is clearly not downward - closed , but it turns out that the action of @xmath3 also restricts to an action on this set . in order to study this action , we begin with some foundational results concerning satisfiability which will enable us to provide a combinatorial characterization of the types . recall from the introduction that we identify each partial bijection @xmath188 \dashrightarrow [ n]$ ] with the boolean matrix @xmath34 defined by @xmath35 if and only if @xmath36 . we say that @xmath176 _ contains _ the partial bijection @xmath53 if @xmath189 . our main result in this section is the following `` local '' characterization of satisfiability . [ thm : bijectionsat ] let @xmath190 and @xmath176 . then @xmath178 is satisfiable with respect to @xmath2 if and only if all of the partial bijections contained in @xmath178 are satisfiable with respect to @xmath2 . to prove this result we require a number of technical lemmas . [ lem : sub ] let @xmath190 . an element @xmath191 is satisfiable with respect to @xmath2 if and only if @xmath192 is satisfiable with respect to @xmath2 for all @xmath193 with @xmath194 . as we have already observed , @xmath187 is a downward - closed subset of @xmath14 . it follows immediately that if @xmath178 is satisfiable , then so too is @xmath192 , where @xmath193 . suppose now that @xmath195 is satisfiable for each @xmath193 with @xmath194 . since domination is scale invariant we note that we may assume without loss of generality ( by rescaling if necessary ) that each column @xmath196 has 0 in the first position . for each @xmath197 choose @xmath198 satisfying @xmath199 . we shall show that @xmath200 satisfies @xmath178 . first note that we may choose each @xmath201 so that its first co - ordinate ( and hence the first co - ordinate of @xmath49 ) is @xmath72 . since @xmath201 satisfies @xmath199 , we have in particular that @xmath196 dominates @xmath201 in position 1 : @xmath202 from which it follows easily that @xmath203 . on the other hand , by definition of @xmath49 we see that @xmath204 and hence @xmath205 , and from the latter we can also deduce that @xmath196 dominates @xmath49 in position 1 for all @xmath206 : @xmath207 now let @xmath208 and consider the max - plus submodule @xmath209 defined via . since each @xmath201 satisfies @xmath210 , by definition it is contained in @xmath209 . thus the max - plus sum @xmath200 must be contained in @xmath209 too . now since @xmath196 dominates @xmath49 in position 1 for all @xmath206 and @xmath49 satisfies @xmath210 we conclude from that @xmath49 satisfies @xmath178 . it is clear that the analogues of lemma [ lem : sub ] concerning positions @xmath211 also hold . [ cor : satis ] let @xmath190 and @xmath212 . then @xmath213 if and only if @xmath214 for all @xmath215 with @xmath216 . since @xmath187 is downward - closed , the forward implication is immediate . on the other hand , if @xmath217 is satisfiable for all @xmath218 with @xmath216 , then by repeated application of lemma [ lem : sub ] ( and its analogues for positions @xmath211 ) we see that @xmath178 must be satisfiable too . the following lemma combined with corollary [ cor : satis ] will enable us to prove the desired result ( theorem [ thm : bijectionsat ] ) . [ prop : satis ] let @xmath190 and let @xmath219 be such that @xmath220 . then there exists @xmath221 such that : + ( i ) @xmath222 ; and + ( ii ) if @xmath223 , then @xmath224 . let @xmath225 be such that @xmath49 satisfies @xmath178 and @xmath147 satisfies @xmath167 . thus by assumption @xmath226 dominates @xmath49 in position @xmath18 and @xmath147 in position @xmath118 . we want to construct @xmath227 such that @xmath226 dominates @xmath228 in both positions @xmath18 and @xmath118 , and that in position @xmath229 , @xmath228 is dominated by any columns @xmath230 which dominate both @xmath49 and @xmath147 in that position . first of all we note that we may assume without loss of generality that @xmath231 , as translating the @xmath17th column of @xmath2 by the vector @xmath232 clearly does not alter the combinatorial structure of the resulting face poset . moreover , since domination is scale invariant , we may also assume without loss of generality that @xmath233 . recalling the definition of the residuation operator @xmath234 given in , we define @xmath235 and claim that @xmath228 satisfies the required conditions . notice that the @xmath17th term in the above definition of @xmath228 is @xmath236 , giving @xmath237 and hence @xmath238 for all @xmath239 $ ] . it is also easy to see that @xmath240 and @xmath241 , since for example by we have @xmath242 for all @xmath243 , and hence @xmath244 now @xmath245 from which we deduce that @xmath246 by an identical argument using @xmath147 instead of @xmath49 , we get @xmath247 . it now follows that @xmath248,\ ] ] or in other words , that @xmath228 is dominated by @xmath226 in both of these positions . now suppose that @xmath230 dominates both @xmath49 and @xmath147 in position @xmath229 , that is , @xmath249 and @xmath250 . but now looking at the @xmath243th term in the definition of @xmath228 gives @xmath251 since we know that @xmath252 and @xmath253 , giving @xmath254 , we must have @xmath255 . we are now in position to show that @xmath230 dominates @xmath228 in position @xmath229 . assume without loss of generality that @xmath256 ( the argument being so far symmetric in @xmath49 and @xmath147 ) . then because @xmath230 dominates both @xmath49 and @xmath147 in position @xmath229 we see that @xmath257 thus by the definition of @xmath228 we see that @xmath258 by this means that @xmath259 , giving @xmath260 or in other words , @xmath230 dominates @xmath228 in position @xmath229 . since @xmath187 is downward - closed , if @xmath178 is satisfiable , then so too is each @xmath261 . suppose then that all partial bijections contained in @xmath178 are satisfiable . we shall show that each @xmath218 with @xmath216 must be satisfiable ( by repeated application of lemma [ prop : satis ] ) . then by corollary [ cor : satis ] we can conclude that @xmath178 is satisfiable . given @xmath218 with @xmath216 we consider the collection @xmath262 of all partial bijections contained within @xmath263 . we apply lemma [ prop : satis ] to all pairs @xmath264 of _ maximal _ elements @xmath265 which differ in a single column , add the resulting satisfiable element @xmath266 to @xmath262 , and continue in this fashion . we note that at each step of this procedure @xmath262 contains only satisfiable elements that are less than or equal to @xmath263 in the order . since there are finitely many such elements , it is clear that this process must terminate . the resulting set @xmath262 then has unique maximum element @xmath263 . the _ max - plus tropical permanent _ of @xmath267 is the real number @xmath268 where @xmath269 denotes the symmetric group on @xmath270 $ ] . the matrix @xmath271 is said to be _ max - plus tropically singular _ if the expression for the permanent above is attained by at least two different permutations , and _ max - plus tropically non - singular _ otherwise . let @xmath30 \dashrightarrow [ n]$ ] be a partial bijection with domain of definition @xmath32 $ ] and image @xmath33 $ ] . we say that @xmath53 is _ permanent - attaining _ ( with respect to @xmath2 ) if @xmath53 attains the maximum in the calculation of the max - plus permanent for the corresponding @xmath272 submatrix of @xmath2 , that is , if for every bijection @xmath273 we have @xmath274 we define the _ permanent structure _ @xmath275 of @xmath2 to be the set of all permanent - attaining partial bijections . the permanent - structure @xmath8 is a downward - closed subset of @xmath14 . let @xmath276 and suppose @xmath277 . clearly @xmath87 is a partial bijection . suppose for a contradiction that it is not permanent - attaining . choose a permanent - attaining partial bijection @xmath278 with the same domain of definition and image as @xmath87 . since @xmath277 , @xmath53 maps the domain of @xmath278 onto the range of @xmath278 ; this means that we can extend @xmath278 to a partial bijection @xmath279 with the same domain and image as @xmath53 , by setting @xmath280 for all @xmath17 not in the domain of @xmath278 . but now it is easily seen that @xmath279 makes a greater contribution to the permanent calculation than @xmath53 , contradicting the assumption that @xmath53 is permanent - attaining . it follows from our observation in section [ sect : action ] that @xmath3 also acts on the right of @xmath8 . the following theorem shows that the permanent - attaining partial bijections are precisely the satisfiable partial bijections ( or in fact , the partial bijections which are satisfied by elements of the max - plus column space of @xmath2 ) . [ permanentsatis ] let @xmath281 and let @xmath188 \dashrightarrow [ n]$ ] be a partial bijection . then the following are equivalent : * @xmath53 is permanent - attaining with respect to @xmath2 ; * @xmath53 is satisfiable with respect to @xmath2 ; * there exists @xmath282 satisfying @xmath53 . let @xmath32 $ ] and @xmath33 $ ] be the domain and image respectively of @xmath53 . it is clear that ( iii ) implies ( ii ) . we shall show that ( ii ) implies ( i ) and ( i ) implies ( iii ) . let @xmath283 be the element defined by @xmath284 if and only if @xmath285 . suppose that ( ii ) holds . by definition there exists @xmath147 in @xmath286 thus for each @xmath287 we have @xmath288 or in other words @xmath289.\ ] ] suppose for contradiction that @xmath53 does not attain the permanent . then there is another bijection @xmath290 with @xmath291 such that @xmath292 giving @xmath293 but it follows from that the right hand side of this inequality is less than or equal to @xmath294 , giving a contradiction since the latter is 0 ( because @xmath53 and @xmath87 are partial bijections with the same domain and _ image _ ) . now suppose that ( i ) holds . suppose first that @xmath295 . for each @xmath296 let @xmath297 be the @xmath298 square matrix such that @xmath299 then the permanent of @xmath297 is attained uniquely by @xmath53 and hence @xmath297 tropically non - singular . it follows from ( * ? ? ? * theorem 4.2 ) and ( * ? ? ? * proposition 17 ) that there is a vector @xmath300 whose type _ with respect to the columns of @xmath297 _ is an @xmath21-tuple of singleton sets whose union is @xmath10 $ ] . moreover , it is not hard to see that @xmath301 - again , with respect to the columns of @xmath297 - must be @xmath302 , since ( having already shown that ( ii ) implies ( i ) ) we note that any other vector of singletons will yield another permanent - attaining permutation . for each @xmath303 we define @xmath304 thus @xmath305 is the set of all elements @xmath46 having @xmath306 . since @xmath302 is the type of a point in the max - plus column space of @xmath297 , we note that by ( * ? ? ? * theorem 15 ) @xmath305 corresponds to a closed and bounded subset of @xmath157 ( identified with tropical projective space @xmath307 as in ) . moreover , it is easy to see that as @xmath308 we obtain a decreasing sequence of nested closed and bounded subsets of @xmath157 . fixing @xmath309 , the set @xmath310 is therefore compact in projective space , with respect to the usual topology induced from @xmath157 . since each @xmath311 with @xmath312 is contained in @xmath313 , there is a convergent subsequence of @xmath311 s in projective space , the limit of which must satisfy the limiting set of inequalities . in other words , there exists @xmath165 whose type @xmath167 _ with respect to the columns of @xmath2 _ contains @xmath302 . by ( * corollary 12 ) we note that face corresponding to @xmath167 in the tropical complex generated by @xmath2 must be bounded ( each row of @xmath314 is non - zero ) and hence , by ( * ? ? ? * theorem 15 ) , this face is contained in the column space of @xmath2 . but now @xmath147 satisfies @xmath302 with respect to @xmath2 and @xmath282 . finally , suppose that @xmath315 . let @xmath271 denote the @xmath272 submatrix of @xmath2 whose rows are indexed by @xmath106 and whose columns are indexed by @xmath316 . by the argument above , there exists @xmath317 in the column space of @xmath271 satisfying @xmath53 , that is : @xmath318 since @xmath319 we can write @xmath320 for some @xmath321 , and one can deduce from definition that @xmath322 now set @xmath323 thus @xmath282 and @xmath324 for all @xmath325 . notice that for all @xmath287 @xmath326 thus for all @xmath239 $ ] and all @xmath287 we have @xmath327 giving @xmath328 for all @xmath287 and all @xmath239 $ ] . in other words , @xmath329 dominates @xmath147 in position @xmath330 for all @xmath287 , showing that @xmath147 satisfies @xmath302 as required . in light of the previous theorem we may rephrase theorem [ thm : bijectionsat ] as follows : [ cor : bijectionsat ] let @xmath190 and let @xmath176 . then @xmath178 is satisfiable with respect to @xmath2 if and only if every partial bijection contained in @xmath178 is permanent - attaining with respect to @xmath2 . it follows from theorem [ permanentsatis ] that one may deduce the permanent structure @xmath8 from the collection of all types @xmath6 . in the next section we shall prove the converse , hence showing that @xmath8 and @xmath6 encode exactly the same combinatorial data about the tropical complex . we conclude this section with the following crucial observation . [ ijtypes ] let @xmath190 and let @xmath314 be a type with respect to the columns of @xmath2 . if @xmath167 contains a partial bijection @xmath188 \dashrightarrow [ n]$ ] , then @xmath167 contains all of the permanent - attaining ( with respect to @xmath2 ) partial bijections that have the same domain of definition and image as @xmath53 . since @xmath167 is a type , there exists @xmath46 having type @xmath167 . by definition , @xmath49 satisfies @xmath331 only if @xmath332 , and so in particular @xmath49 satisfies @xmath53 . now if @xmath87 is any permanent - attaining partial bijection with the same domain of definition and image as @xmath53 it follows from theorem [ permanentsatis ] that @xmath87 is satisfiable . we show that @xmath49 also satisfies @xmath87 , from which we conclude that @xmath167 contains @xmath87 . since @xmath53 and @xmath87 are satisfiable it follows from theorem [ permanentsatis ] that these partial bijections attain the permanent @xmath333 in the submatrix of @xmath2 with rows indexed by @xmath106 and columns indexed by @xmath316 : @xmath334 suppose that @xmath49 satisfies @xmath53 . in other words , suppose that @xmath335.\ ] ] we note that since the image of @xmath87 is a subset of @xmath10 $ ] , we have in particular @xmath336 we claim that each of the last inequalities has to be an equality . indeed , suppose not . then summing them gives @xmath337 but since @xmath53 and @xmath87 are bijections with the same _ image _ @xmath106 , we see that this becomes @xmath338 we note that the first summations on either side are identical , whilst each of the remaining summations is equal to @xmath333 by assumption , giving a contradiction and proving the claim . thus we must have @xmath339 for all @xmath287 and all @xmath239 $ ] . or in other words , @xmath49 satisfies @xmath87 . given a matrix @xmath2 defining a tropical complex and an @xmath7 boolean matrix ( or equivalently , a @xmath4-tuple of subsets of @xmath10 $ ] ) @xmath178 , it is natural to ask whether @xmath178 is a type of the complex . our results so far suffice to establish some elementary properties which @xmath178 must possess if it is to be a type ; our aim in this section is to establish a collection of conditions which are necessary and sufficient for @xmath178 to be a type : * every column of @xmath178 is non - empty ; * every partial bijection contained in @xmath178 is permanent - attaining with respect to @xmath2 ; and * if @xmath178 contains a partial bijection @xmath53 , then @xmath178 contains all of the permanent - attaining ( with respect to @xmath2 ) partial bijections with the same domain of definition and image as @xmath53 . notice that the three conditions ( t1 ) , ( t2 ) and ( t3 ) depend only on the matrix @xmath178 and the permanent structure of @xmath2 ; there is no direct dependence on the matrix @xmath2 or its associated tropical complex . it follows that one can determine whether a given boolean matrix is a type by reference only to the permanent structure . conversely , by theorem [ permanentsatis ] a partial bijection is permanent - attaining if and only if it is contained in a type , so the permanent structure can be also be recovered from the types . this justifies the following claim which we made earlier : * since the point @xmath49 must lie in at least one of the sectors relative to the hyperplane specified by a column of @xmath2 , we see that each column of @xmath2 dominates @xmath49 in some position and thus each column of @xmath178 must be non - empty . * the point @xmath49 satisfies every @xmath167 such that @xmath340 , and hence in particular every partial bijection contained in @xmath178 . thus , by theorem [ permanentsatis ] , every such partial bijection must be permanent - attaining . * this is lemma [ ijtypes ] above . [ technical ] let @xmath341 satisfying conditions ( t1 ) and ( t3 ) above and choose a point @xmath49 satisfying @xmath178 . let @xmath342 be the type of @xmath49 and define a binary relation @xmath343 on @xmath10 $ ] by @xmath344 if @xmath345 . * if @xmath344 then @xmath346 for all @xmath347 . * if @xmath344 , @xmath147 satisfies @xmath178 and @xmath348 for some @xmath303 , then @xmath349 . * suppose that @xmath33 $ ] with the property that if @xmath350 and @xmath344 then @xmath351 . let @xmath296 and let @xmath311 be the element obtained from @xmath49 by adding @xmath352 to the entries indexed by @xmath106 . then there exists @xmath309 such that @xmath311 satisfies @xmath178 for all @xmath353 . \(i ) suppose that @xmath344 and let @xmath347 . since @xmath354 and @xmath49 satisfies @xmath178 , we have @xmath355 on the other hand , since @xmath356 and @xmath49 satisfies @xmath167 , we have @xmath357 it follows immediately from these two inequalities that @xmath358 \(ii ) suppose that @xmath344 and choose @xmath347 . by part ( i ) we know that @xmath346 . now , since @xmath147 satisfies @xmath178 we have @xmath359 rearranging this last inequality and using the fact that @xmath348 gives @xmath360 and hence @xmath361 as required . \(iii ) for any @xmath296 we have that @xmath362 with equality in positions not indexed by @xmath106 . it follows that @xmath311 must satisfy the conditions specified by the subsets @xmath363 with @xmath364 ; specifically , if @xmath365 for some @xmath364 , then @xmath366.\ ] ] moreover , @xmath365 for some @xmath367 , then @xmath368.\ ] ] in particular , this gives @xmath369 it remains to show that we can choose @xmath352 small enough so that @xmath311 satisfies the remaining inequalities of @xmath178 too . thus we need to be able to choose @xmath303 to simultaneously satisfy the inequalities @xmath370 or in other words , we want to show that there exists @xmath352 satisfying @xmath371 where the minimum takes place over all @xmath367 , @xmath365 and @xmath372 . suppose not . then , since there are only finitely many terms in the minimum , we must have @xmath373 rearranging this gives @xmath374 since @xmath365 we know that @xmath49 is dominated by @xmath16 in position @xmath118 , and so @xmath375 but then the last two inequalities combine to give @xmath376 showing that @xmath377 , and hence @xmath378 . since @xmath367 , by assumption we must have that @xmath351 too , contradicting @xmath379 . [ induct ] let @xmath213 satisfying conditions ( t1 ) and ( t3 ) above , and choose a point @xmath46 satisfying @xmath178 . let @xmath167 be the type of @xmath49 and suppose that @xmath380 . then there exists @xmath165 such that @xmath381 with @xmath382 . we shall construct @xmath147 by increasing @xmath383 by a small amount @xmath352 for all positions @xmath17 contained in some carefully chosen subset @xmath33 $ ] . it is straightforward to check that given any choice of @xmath106 we can choose @xmath352 small enough so that @xmath384 . suppose that @xmath106 has the property that if @xmath350 and @xmath344 then @xmath351 . then lemma [ technical ] ( iii ) tells us that we can choose @xmath352 so that @xmath178 is still satisfied . thus for any such a choice of @xmath106 we may choose @xmath352 small enough so that @xmath381 . it remains to show that some choice of @xmath106 gives @xmath385 . let @xmath386 and @xmath387 . since @xmath388 , we have @xmath389 , but @xmath390 for some @xmath391 $ ] and @xmath392 $ ] . since @xmath178 satisfies ( t1 ) , @xmath18 must occur somewhere in @xmath178 . suppose that @xmath393 . we first show that choosing @xmath106 so that @xmath394 and @xmath395 yields @xmath396 and hence @xmath385 . suppose for contradiction that @xmath397 . then @xmath16 dominates @xmath147 in position @xmath17 we have in particular @xmath398 and since @xmath394 and @xmath395 this gives @xmath399 but since @xmath393 and @xmath49 satisfies @xmath178 , we also know that @xmath16 dominates @xmath49 in position @xmath400 , giving in particular that @xmath401 it is easy to see that the last two inequalities combined contradict @xmath296 . suppose for contradiction that any subset @xmath106 satisfying condition ( a ) and containing @xmath17 must also contain @xmath400 . it follows that there must be a sequence @xmath402 by relabelling co - ordinates as necessary we may assume that @xmath403 , @xmath404 and @xmath405 , so that @xmath406 , but @xmath407 and @xmath408 thus for @xmath409 we may choose @xmath410 , giving : @xmath411 it is clear that any partial bijection contained in @xmath412 or @xmath413 will be satisfiable ( by theorem [ thm : bijectionsat ] it will be satisfied by @xmath49 ) , and hence must attain the permanent in the corresponding submatrix of @xmath2 . we claim that there are partial bijections @xmath414 and @xmath415 such that @xmath416 and @xmath417 have the same domain of definition and the same image , with @xmath418 and @xmath419 . we first note that this will give the desired contradiction . since @xmath178 satisfies condition ( t3 ) we know that the set of all partial bijections contained in @xmath178 contains either all or none of the permanent - attaining bijections on any given submatrix . thus it follows from the fact that @xmath416 is a partial bijection contained in @xmath178 that @xmath417 ( a partial bijection with the same domain and image ) must be contained in @xmath178 too . but then @xmath420 , giving a contradiction . the partial bijections @xmath416 and @xmath417 are constructed as follows . we know that @xmath421 , @xmath422 , @xmath423 and @xmath18 occurs in at least one row of @xmath412 . let @xmath424 be the first row of @xmath412 in which @xmath18 occurs and let @xmath425 denote the element contained in @xmath426 . since @xmath413 is just @xmath412 shifted one place to the right it is clear that @xmath427 . if @xmath428 , then we will have partial bijections of the form @xmath429 otherwise , @xmath430 . let @xmath431 be the first position of @xmath412 in which @xmath425 occurs . since @xmath413 shifts @xmath412 one place to the right it is clear that @xmath432 . let @xmath433 denote the element contained in @xmath434 . again , since @xmath413 shifts @xmath412 one place to the right it is clear that @xmath435 . if @xmath436 , then we will have partial bijections of the form @xmath437 otherwise @xmath438 . continuing in this way will yield a cycle of the required form . suppose then that @xmath178 satisfies the conditions ( t1)-(t3 ) . since @xmath178 satisfies ( t2 ) , it follows from corollary [ cor : bijectionsat ] that we may choose @xmath138 such that @xmath49 satisfies @xmath178 . choose @xmath49 so that the type of @xmath49 is minimal with respect to the partial order @xmath439 , and denote this type by @xmath167 . thus @xmath185 and if @xmath440 , then we are done . otherwise , we may apply lemma [ induct ] to find @xmath441 such that @xmath442 with @xmath385 , contradicting our choice of @xmath49 with @xmath167 minimal . ( t1 ) by definition we see that the @xmath18th column of @xmath446 is @xmath447 . since @xmath167 is a type , we know that @xmath448 $ ] and since @xmath333 is an ordered set partition of @xmath10 $ ] it is therefore clear that @xmath449 has non - empty intersection with at least one component of @xmath333 . thus by we conclude that @xmath447 is non - empty . ( t3 ) suppose that @xmath53 and @xmath87 are partial bijections contained in @xmath167 with the same domain @xmath32 $ ] and image @xmath33 $ ] . by our previous remarks , it suffices to show that if @xmath53 is contained in @xmath446 then @xmath87 is contained in @xmath451 too . assume by relabelling if necessary that @xmath452 $ ] , and that @xmath453 for all @xmath454 $ ] , with @xmath87 a non - identity permutation of @xmath270 $ ] . since @xmath53 ( the identity map on @xmath270 $ ] ) and @xmath87 are both contained in @xmath167 , by definition we know that : @xmath455 for each @xmath456 $ ] we define @xmath457 so that by we have @xmath458 . since @xmath53 ( the identity ) is contained in @xmath446 it follows that @xmath459.\ ] ] now fix @xmath18 and look at the orbit , @xmath460 say , of @xmath18 under @xmath87 . by applying and to each of the elements @xmath461 $ ] in turn , we obtain : @xmath462 note that if @xmath463 then by definition we see that @xmath464 . since @xmath465 is contained in this intersection , we must have @xmath466 . continuing in this way we note that @xmath467 from which we conclude that these indices all coincide and hence by combining the information above we see that the set @xmath468 contains the entire orbit of @xmath18 under @xmath87 . in particular , @xmath469 for all @xmath470 , hence showing that @xmath87 is contained in @xmath446 as required . since @xmath471 for all @xmath331 it follows that if a type @xmath167 is equal to @xmath472 for some other type @xmath473 , then @xmath473 must be a face of @xmath167 . moreover , it is easy to see that each maximal dimensional cell contained in @xmath6 is a fixed point of the right action of @xmath3 . thus ( just as in the case of usual hyperplane arrangements ) , this right action can not be used to generate a random walk on the maximal dimensional cells ( `` chambers '' ) of the tropical hyperplane arrangement . we also note that although @xmath3 acts upon @xmath6 , this action does not immediately restrict to give an action on the max - plus polytope generated by the columns of @xmath2 ( that is , the collection of bounded cells ) . for example , acting upon the face @xmath153 of the tropical hyperplane arrangement given in example 3.1 by the element @xmath474 gives the unbounded face @xmath172 , which is not contained in the max - plus column space of @xmath2 . 99 m. aguiar , s. mahajan , coxeter groups and hopf algebras , fields institute monographs , 23 . american mathematical society , providence , ri , 2006 . j. almeida , s. margolis , b. steinberg , m. volkov , _ representation theory of finite semigroups , semigroup radicals and formal language theory _ soc . , 361 , 14291461 , 2009 . f. ardila , m. develin , _ tropical hyperplane arrangements and oriented matroids _ z. , 262 , 795816 , 2009 . a. ayyer , a. schilling , b. steinberg , n. m. thierry , _ markov chains , r - trivial monoids and representation theory _ , j. algebra comput . , 25 , no . 1 - 2 , 169231 , 2015 . c. a. athanasiadis , p. diaconis , _ functions of random walks on hyperplane arrangements _ , adv . in appl . , 45 , no . 3 , 410437 , 2010 . c. berg , n. bergeron , s. bhargava , f. saliola , _ primitive orthogonal idempotents for r - trivial monoids _ , j. algebra , 348 , 446461 2011 . p. bidigare , p. hanlon , d. rockmore , _ a combinatorial description of the spectrum for the tsetlin library and its generalization to hyperplane arrangements _ , duke math . j. , 99 , no . 1 , 135174 , 1999 . a. bjrner , _ random walks , arrangements , cell complexes , greedoids , and self - organizing libraries _ , building bridges , 165203 , bolyai soc . math . stud . , 19 , springer , berlin , 2008 . a. bjrner , m. las vergnas , b. sturmfels , n. white , g. m. ziegler , oriented matroids , second edition , encyclopedia of mathematics and its applications , 46 . cambridge university press , cambridge , 1999 . t. s. blyth , m. f. janowitz . residuation theory . international series of monographs in pure and applied mathematics , vol . 102 , pergamon press , oxford , 1972 . k. s. brown , p. diaconis , _ random walks and hyperplane arrangements _ , ann . , 26 , no . 4 , 18131854 , 1998 . k. s. brown , _ semigroups , rings , and markov chains _ , j. theoret . probab . , 13 , no . 3 , 871938 , 2000 . k. s. brown , _ semigroup and ring theoretical methods in probability _ , in representations of finite dimensional algebras and related topics in lie theory and geometry , 326 , fields inst . , 40 , amer . soc . , providence , ri , 2004 . p. butkovic , max - linear systems : theory and algorithms . springer monographs in mathematics . springer - verlag london ltd . , london , 2010 . g. cohen , s. gaubert , j .- quadrat , _ duality and separation theorems in idempotent semimodules _ , linear algebra appl . , 379 , 395422 , 2004 . r. cuninghame - green , minimax algebra . lecture notes in economics and mathematical systems 166 . springer - verlag , berlin - new york , 1979 . m. develin , f. santos , b. sturmfels . on the rank of a tropical matrix . in _ combinatorial and computational geometry _ , volume 52 of _ math . _ , pages 213242 . cambridge univ . press , cambridge , 2005 . m. develin , b. sturmfels , _ tropical convexity _ , doc . , 9 , 127 ( electronic ) , 2004 . m. johnson , m. kambites , _ convexity of tropical polytopes _ , linear algebra appl . , 485 , 531544 , 2015 . s. margolis , f. saliola , b. steinberg , _ semigroups embeddable in hyperplane face monoids _ , semigroup forum , 89 , no . 1 , 236248 , 2014 . s. margolis , f. saliola , b. steinberg , _ cell complexes , poset topology and the representation theory of algebras arising in algebraic combinatorics and discrete geometry _ , preprint 2015 . v. reiner , f. saliola , v. welker , spectra of symmetrized shuffling operators , mem . soc . , 228 , no . 1072 , 2014 f. v. saliola , _ the quiver of the semigroup algebra of a left regular band _ , internat . j. algebra comput . , 17 , no . 8 , 15931610 , 2007 . f. saliola , _ the face semigroup algebra of a hyperplane arrangement _ , canad . j. math . , 61 , no . 4 , 904929 , 2009 . f. saliola , _ eigenvectors for a random walk on a left - regular band _ , adv . in appl . , 48 , no . 2 , 306311 , 2012 . m. schocker , _ the module structure of the solomon - tits algebra of the symmetric group _ , j. algebra , 301 , no . 2 , 554586 , 2006 . l. solomon , _ a mackey formula in the group ring of a coxeter group _ , j. algebra , 41 , no . 2 , 255264 , 1976 . j. tits , _ two properties of coxeter complexes _ , j. algebra , 41 , no . 2 , 265268 , 1976 ( appendix to l. solomon s article _ a mackey formula in the group ring of a coxeter group _ , j. algebra , 41 , no . 2 , 255264 , 1976 ) .

shell effects in nuclei about the stability line are investigated within the framework of the relativistic hartree - bogoliubov ( rhb ) theory with self - consistent finite - range pairing . using 2-neutron separation energies of ni and sn isotopes , the role of- and-meson couplings on the shell effects in nuclei is examined . it is observed that the existing successful nuclear forces ( lagrangian parameter sets ) based upon the nonlinear scalar coupling of-meson exhibit shell effects which are stronger than suggested by the experimental data . we have introduced nonlinear vector self - coupling of-meson in the rhb theory . it is shown that the inclusion of the vector self - coupling of-meson in addition to the nonlinear scalar coupling of-meson provides a good agreement with the experimental data on shell effects in nuclei about the stability line . a comparison of the shell effects in the rhb theory is made with the hartree - fock bogoliubov approach using the skyrme force skp . it is shown that the oft - discussed shell quenching with skp is not consistent with the available experimental data . = cmbx10 scaled2

nucl-th0002017

the relativistic mean - field ( rmf ) theory @xcite has proved to be successful in providing a framework for description of various facets of nuclear properties @xcite . in the rmf theory , the nuclear force is produced by a virtual exchange of various mesons . the nuclear saturation is achieved by a balance between an attractive @xmath0- and a repulsive @xmath1-field . the relativistic lorentz covariance of the theory allows an intrinsic spin - orbit interaction based upon exchange of @xmath0- and @xmath1-mesons . this has been shown to be advantageous for properties which depend upon spin - orbit potential @xcite . an immediate advantage of the proper spin - orbit potential has been the success @xcite of the rmf theory to be able to describe the anomalous kink in the isotope shifts of pb nuclei . the isotope shifts of pb have been measured with a high precision using atomic beam laser spectroscopy @xcite and are known to show a pronounced kink about the magic number n=126 . the non - relativistic approaches based upon the skyrme and gogny forces have been unable to reproduce this kink @xcite . it was shown @xcite that this difference in the predictions of the skyrme model and the rmf theory is due to the isospin dependence of the spin - orbit term . the skyrme model assumes an isospin independent two - body spin - orbit force . a strong isospin dependence is , however , introduced herein by the exchange term . on the other hand , the isospin dependence in the rmf theory is provided mainly by the coupling constant @xmath2 of the @xmath3-meson . however , as the strength of the spin - orbit term derives from a large sum of the absolute values of the scalar and vector fields , the @xmath3-field is rendered much weaker in comparison . consequently , the isospin dependence of the spin - orbit potential in the rmf theory is relatively weak . a weak isospin dependence of the spin - orbit potential in the rmf theory was also concluded from calculations employing asymmetric semi - infinite nuclear matter @xcite . the rmf theory has achieved significant success in respect of nuclei near the stability line as well as for nuclei far away from the stability line @xcite.the binding energies , charge radii , deformation properties and isotope shifts are some of the properties which are described successfully in the rmf theory . the nonlinear scalar self - coupling of @xmath0-meson has been the most successful model of the rmf theory used so far . with the need to describe highly exotic nuclei far away from the stability line , where the particles couple to the continuum , relativistic hartree - bogoliubov ( rhb ) theory with a self - consistent pairing has been developed @xcite . it has been shown @xcite that the rhb theory which uses the nonlinear scalar model provides a good description of nuclei away from the stability line . however , the level densities in the rmf or the rhb theory are perceived to be generally low , which lead to larger shell gaps . the large shell gaps imply that shell effects are stronger than in the experimental data . in the present work , we focus upon the shell effects and investigate the role of various meson couplings on the shell effects . the shell effects manifest strongly in nuclei with magic particle numbers . such an effect is well known to exist experimentally in many nuclei all over the periodic table , whereby a stronger binding is exhibited by these nuclei as compared to neighbouring nuclei . nuclei with a particle or two above a magic number show a significantly reduced binding for the extraneous particles above the shell closure . the origin of shell closure has long been understood @xcite due to spin - orbit coupling and to an ensuing splitting of levels . the spin - orbit potential and a bunching of levels create shell closures ( magic numbers ) which are predicted correctly by most of the models . in the non - relativistic density - dependent theory of the skyrme type @xcite , the spin - orbit interaction is added phenomenologically and its strength is adjusted to reproduce the spin - orbit splitting in @xmath4o . in the rmf theory , on the other hand , the spin - orbit interaction arises naturally due to exchange of @xmath0 and @xmath1 mesons by the nucleons . the strength of the interaction is determined by the spin - orbit splitting in @xmath4o and other nuclei , which is uniquely decided by the effective mass . this constrains an effective mass @xmath5 in the vicinity of @xmath6 in the rmf models @xcite . the nature of the shell effects about the stability line is well established through the experimental data available on the binding energies and on one- and two - particle separation energies over a large region of the periodic table . the experimental data encompass all the known magic numbers . using the experimental binding energies on 2-neutron and 2-proton separation energies have been plotted in ref . @xcite for nuclei all over the periodic table . an unambiguous kink appears in the separation energies across all the major magic numbers , demonstrating the existence of a major shell closure at these numbers . this also shows that the shell effects at most of the major magic numbers along the line of stability are strong . it is not clear how the magic numbers or the shell closures behave in going to extreme regions of the periodic table such as near the drip lines . experimental data about neutron drip - line in the region of light nuclei are gradually emerging @xcite . as the neutron drip - line for medium - heavy and heavy nuclei is predicted to lie very far away ( with abnormally large neutron excesses ) from the stability line , it is not expected that experimental data on such nuclei would become available in the near future . however , information on shell effects in such nuclei in the vicinity of drip lines are vital to understanding the r - process nucleosynthesis of heavy nuclei @xcite shell effects near the drip lines have been a matter of intense debate in the last few years . within the framework of the rmf theory , it was shown that shell effects about n=82 in the vicinity of the neutron drip - line remain strong @xcite . these conclusions were based upon the rmf theory with the non - linear scalar potential of @xmath0-meson . on the other hand , on the basis of the hartree - fock - bogoliubov approach with the skyrme interaction skp , it was contended that the shell effects at n=82 in the vicinity of the neutron drip - line show a strong quenching @xcite . however , in the absence of any data in this region in forseeable future , it is difficult to ascertain whether the shell effects near the drip - line remain strong or are quenched . therefore , this debate needs to looks into other observables or evidences which might support or refute the either contention . in this paper , we begin with our investigations of shell effects first along the stability line where experimental data is available . having founded the basis , it could then be extended to predict behaviour of the shell effects at the drip lines . with this in mind , we include the nonlinear vector self - coupling of @xmath1-meson in the lagrangian of the rmf theory . the non - linear vector self - coupling of @xmath1-meson was introduced by bodmer @xcite and properties of nuclear matter were discussed on adding a quartic term in @xmath1-meson potential . the vector self - coupling to the properties of finite nuclei was applied in ref . it was noticed @xcite that the inclusion of the vector self - coupling of @xmath1 meson in addition to the non - linear scalar self - coupling of @xmath0 meson has the effect of softening the high - density equation of state ( eos ) of the nuclear matter . the eos of nuclear matter has been investigated @xcite by taking expansion terms of higher orders in the @xmath0 and @xmath1 fields . recently , a comprehensive study of the vector self - coupling and its effect on properties of nuclei and nuclear matter has been undertaken @xcite . an emphasis has been given to various facets of finite nuclei . it is shown @xcite that an inclusion of the vector self - coupling of @xmath1-meson gives an improved description of nuclei as compared to the frequently used scalar self - coupling of @xmath0-meson only . two new forces nl - sv1 and nl - sv2 have been proposed @xcite with a view to improve the predictions of the ground - state properties of nuclei . we will employ these forces in the present work . details of this study will be provided elsewhere @xcite . in this paper , we discuss the details of the rhb theory preceded by a short review of the rmf approach . the calculational details are provided in the section thereafter . the main objective of this work is to investigate the shell effects in nuclei about the stability line using the vector self - coupling of @xmath1 meson and compare the results with those from the scalar self - coupling of @xmath0 meson . we have taken the isotopic chains of ni and sn nuclei for this purpose . experimental binding energies of these nuclei are available over a large range . this allows us to examine the shell effects in nuclei in the known region using the various models . the shell effects about the magic neutron numbers n=28 , 50 and 82 are explored . the rmf+bcs results are compared with those of the rhb theory . the results on binding energies and two - neutron separation energies are presented and discussed in section 5 . the single - particle levels with various approaches are presented for a few nuclei , with a view to reinforce the results on the shell effects . the last section summarizes the main conclusions . the rmf approach @xcite is based upon the lagrangian density which consists of fields due to the various mesons interacting with the nucleons . the mesons include the isoscalar scalar @xmath0-meson , the isovector vector @xmath1-meson and the isovector vector @xmath3-meson . the lagrangian density is given by : @xmath7 the bold - faced letters indicate the vector quantities . here m , m@xmath8 , m@xmath9 and m@xmath10 denote the nucleon- , the @xmath0- , the @xmath1- and the @xmath3-meson masses respectively , while g@xmath8 , g@xmath9 , g@xmath10 and e@xmath11/4@xmath12 = 1/137 are the corresponding coupling constants for the mesons and the photon , respectively . the @xmath0 meson is assumed to move in a scalar potential of the form : @xmath13 this was introduced by boguta and bodmer @xcite in order to make a substantial improvement in the surface properties of finite nuclei . this ansatz for the @xmath0-potential has since become a standard and necessary ingredient for description of the properties of finite nuclei . recently , several variations of the non - linear @xmath0 and @xmath1 fields have been proposed @xcite . the new proposals for the form of the lagrangian need to be investigated with a view to constrain the form in compliance with the properties of nuclear matter and finite nuclei . here we have incorporated the non - linear vector self - coupling of the @xmath1-meson , in addition to the non - linear scalar potential of eq . the coupling constant for the non - linear @xmath1-term is denoted by @xmath14 in the lagrangian ( 1 ) . the vector self - coupling of the @xmath1-meson was first proposed in ref . @xcite where properties of nuclear matter associated to this potential were also discussed . however , the coupling constant of the new ansatz can be constrained appropriately only within the framework of finite nuclei . one such attempt was made in ref . @xcite where parameter sets tm1 and tm2 were obtained . we have now performed a comprehensive investigation of the properties of nuclear matter and finite nuclei associated with the vector self - coupling @xcite . the field tensors of the vector mesons and of the electromagnetic field take the following form : @xmath15 the mean - field approximation constitutes the lowest order of the quantum field theory . herein , the nucleons are assumed to move independently in the meson fields . the latter are replaced by their classical expectation values . the ground - state of the nucleus is described by a slater determinant @xmath16 of single - particle spinors @xmath17 ( i = 1,2, .... a ) . the stationary state solutions @xmath17 are obtained from the coupled system of dirac and klein - gordon equations . the variational principle leads to the dirac equation : @xmath18 \ } ~\psi_{i } = ~\epsilon_{i } \psi_{i}\ ] ] where @xmath19 represents the @xmath20 potential : @xmath21 and @xmath22 is the @xmath23 potential @xmath24 which defines the effective mass as : @xmath25 the klein - gordon equations for the meson fields are time - independent inhomogeneous equations with the nucleon densities as sources . @xmath26 for the case of an even - even nucleus with time - reversal symmetry , the spatial components of the vector fields , , and vanish . for the mean - field , the nucleon spinors provide the corresponding source terms : @xmath27 where the sums are taken over the valence nucleons only . consequently , the ground - state of a nucleus is obtained by solving the coupled system of the dirac and klein - gordon equations self - consistently . the solution of the dirac equation is achieved by using the method of oscillator expansion @xcite . in the rmf approach , the pairing is included within the bcs scheme , where the pairing gaps are calculated from the experimental masses of neigbouring nuclei . nuclei which are known to show strong pairing correlations can be treated appropriately within the framework of the rhb approach . this is especially important when the pairing correlations in the middle of a shell become important . however , the pairing correlations for nuclei 2 neutrons less or more than a magic number can be expected to be similar in the rhb and the bcs approach . this is due to the reason that the pairing correlations in such nuclei are reduced to a minimal level . the rhb approach becomes more important while dealing with nuclei far away from the stability line and in particular for nuclei in the vicinity of the drip lines , whereby the fermi level is usually very near the continuum . the coupling between the bound states and the states in the continuum is taken into account in the rhb theory . thus , the rhb is a preferred scheme over the bcs in such cases . this seems to be necessary because we consider ni nuclei also about and above n=50 , which are already far away from the stability line . moreover , the sn isotopes being open - shell nuclei are known to exhibit strong pairing correlations . thus , we employ the rhb theory as being the most suitable one for the purpose . analogous to the hartree - fock - bogoliubov ( hfb ) theory @xcite , the rhb theory contains the self - consistent field @xmath28 which represents the particle - hole correlations , and a pairing field @xmath29 which includes all particle - particle correlations . for the self - consistent mean - field , the rhb equations are given by : @xmath30 where @xmath31 is the single - nucleon dirac hamiltonian and @xmath32 is the nucleon mass . @xmath33 and @xmath34 are the two dirac spinors and @xmath35 represents the pairing field . these equations are solved self - consistently , whereby the potentials are obtained in the mean - field approximation from the klein - gordon equations ( 8) . the corresponding source terms in the klein - gordon equations are sums of bilinear products of baryon amplitudes @xmath36 where the sums are taken over all positive energy states . for m degrees of freedom , one chooses the m positive eigenvalues @xmath37 for the solution that corresponds to a ground state of a nucleus with an even particle number . the negative energy solutions are ignored as it is forbidden to occupy the levels @xmath37 and @xmath38 simultaneously . the integral operator @xmath29 in eq . ( 10 ) acts on the wave function @xmath39 : @xmath40 for the pairing field , the kernel of the integral operator is given by @xmath41 where @xmath42 denote all quantum numbers , except the coordinate @xmath43 , that specify the single - nucleon states . here @xmath44 denotes matrix elements of a general two - body pairing interaction , and @xmath45 is the pairing tensor . the system of the dirac - hartree - bogoliubov equations ( 10 - 17 ) is solved self - consistently . the eigensolutions of the rhb equations ( 10 ) form a set of orthonormal single quasi - particle states and the corresponding eigenvalues are the single quasi - particle energies . the basis of the quasi - particles is then transformed into the canonical basis of the single - particle states . this basis determines the single - particle energies and occupations probabilities corresponding to the ground - state of a nucleus . it was realized earlier that using the standard parameter sets of the rmf theory for the pairing , one obtains highly unrealistic pairing correlations @xcite . due to this reason , a completely microscopic derivation of a pairing interaction from the first principles has not been feasible so far . instead , the finite - range pairing force of the gogny - type is used in the p - p channel as suggested in ref . the pairing interaction is approximated by a two - body finite range force of the gogny type , @xmath46 with parameters @xmath47 , @xmath48 , @xmath49 , @xmath50 and @xmath51 @xmath52 . we use the parameter set d1s @xcite for the pairing force . the gogny force is a sum of two gaussians with finite range . it has been shown @xcite that the gogny force is able to represent the pairing properties of a large number of finite nuclei including the sn and pb isotopes . we have performed the rmf+bcs and rhb calculations for ni and sn nuclei with a spherical symmetry . for both the rmf+bcs and rhb calculations , the wavefunctions are expanded into the oscillator basis . the number of shells taken into account is 20 for the fermionic as well as bosonic wavefunctions . in this study , we have used the forces nl - sh and nl3 for the lagrangian with the non - linear scalar self - coupling . in a large number of studies it has been shown that the force nl - sh @xcite is able to provide a very good description of the ground - state properties of nuclei all over the periodic table . it has been found to be especially useful for exotic nuclei near drip lines . recently , the force nl3 @xcite has been proposed with a view to modify the compression modulus of nuclear matter . studies have shown @xcite that the force nl3 describes the ground - state properties of nuclei as well as nl - sh and that nl3 results are found to be very similar to the nl - sh ones . we have also used the forces nl - sv1 and nl - sv2 @xcite for the lagrangian with the scalar and vector self - coupling . the most important point in the development of the forces nl - sv1 and nl - sv2 is to soften the equation of state ( eos ) for the nuclear matter as compared to that with the scalar self - coupling only and to improve the description of the ground - state properties of nuclei vis - a - vis the forces with the scalar self - coupling . we have also used the force tm1 @xcite for a comparative view of the results . we have chosen the chain of ni and sn isotopes for the present study . the ni isotopes are known to be spherical in the neighbourhood of the stability line and this trend continues well up to the neutron shell closure n=50 . the ni isotopes are favourable for the study of shell effects , as the shells in these nuclei are spread broadly . the large shells gaps are conducive to exploring the shell effects . the effects due to the large shell gaps appear more prominently in the experimental observables such as 2-neutron separation energies . this is exemplified by a large value of about 32 mev of the 2-neutron separation energy , s@xmath53 , for the nucleus @xmath54ni . with the addition of a pair of neutrons to form @xmath55ni , the s@xmath53 value shows a dramatic decrease to @xmath56 22 mev in the experimental value . this large difference in s@xmath53 values of the ni isotopes should serve as a good test - bench to examine the shell effects . for heavier mass nuclei such as sn isotopes , the shells become compressed and consequently the @xmath57 values show a lesser sensitivity to the shell effects . in the present case , we examine the shell effects in sn nuclei at n=82 . the experimental binding energies of of the neutron - rich nuclei @xmath58sn , @xmath59sn and @xmath60sn are available . this makes it possible to look into the shell effects at n=82 in sn nuclei away from the stability line . we show the binding energies of the ni isotopes with the forces in the non - linear scalar coupling of the @xmath0-meson . the binding energies ( @xmath61 ) of the ni isotopes as compared to the experimental values ( @xmath62 ) @xcite are shown in fig . 1 . the results of the rmf theory ( fig . 1a ) with the bcs pairing are compared to those with relativistic hartree - bogoliubov ( rhb ) approach using the self - consistent finite - range pairing ( fig . for the bcs calculations , the proton number z=28 is assumed to be a closed shell and hence the proton pairing gap is taken to be zero . the neutron pairing gap is calculated using the experimental binding energies of neighbouring nuclei . however , for the magic nuclei with n=28 and n=50 , the neutron pairing gap is taken to be zero . the results in fig . 1a show that the rmf+bcs approach reproduces the binding energies of the most of the ni isotopes well . only for a few nuclei about a=62 ( n=34 ) , the rmf theory underestimates the binding energy by about 3 - 4 mev ( @xmath63 ) . a comparison of the results of nl - sh and nl3 shows that the binding energies with nl3 are similar to those of nl - sh and that both the forces show a similar pattern of behaviour . the binding energies with nl - sh are slightly better than those with nl3 , especially in the region a=50 - 60 . the rhb results with the forces nl - sh and nl3 are shown in fig . it can be seen that the rhb approach with the self - consistent pairing produces results which are very similar to those with the rmf+bcs ones . the whole pattern of the bcs+rmf results is reproduced by the rhb , with both the forces nl - sh and nl3 . moreover , the rhb results show an improved agreement for nuclei in the region a=62 . a comparison of the results with nl - sh and nl3 shows that both the results are very similar to each other . for the region a=70 - 78 , both the forces overestimate the experimental binding energies slightly in the rhb approach . with the non - linear vector self - coupling of the @xmath1-meson , we have used the forces nl - sv1 and nl - sv2 @xcite . we have also included the force tm1 @xcite . the binding energy difference ( @xmath64 ) is shown for the rmf+bcs ( fig . 2a ) and the rhb approach ( fig . 2b ) for the three forces . for the bcs pairing we have used the experimental pairing gaps obtained from the masses of neighbouring nuclei as mentioned above . the results ( fig . 2a ) show that the behaviour of the binding energies with the forces with the scalar - vector self - coupling is similar to those with the scalar self - coupling only , except for a few minor differences . the region about a=60 shows some deviations from the experimental values . however , the degree of agreement with the experimental data varies with the forces . the force tm1 underestimates the binding energies by about 4 mev for several nuclei in the region a=54 - 62 . the force nl - sv1 shows an improvement in the predictions of energies of these nuclei over those of tm1 . in comparison , the force nl - sv2 provides the best description of the binding energies , where the deviations from the experimental energies are reduced considerably . for the heavier ni isotopes with a=68 - 78 , the results of nl - sv1 and nl - sv2 are very similar , whereas for tm1 deviations from the experimental energies start increasing at a=74 . we have incorporated the vector self - coupling of the @xmath1-meson in the rhb approach for the first time . the rhb results ( fig . 2b ) with the forces with the scalar - vector self - coupling show a high degree of similarity with the corresponding bcs results , as seen also for the non - linear scalar coupling . only a slight improvement is noticed in the rhb predictions . with the force nl - sv2 the deviations from the experimental binding energies are improved slightly in the rhb over the bcs for the region a=52 - 64 . for the heavier isotopes , rhb shows marginally higher deviations than in the bcs with nl - sv2 . the trend of the binding energies as a function of the mass number is smoothened in the rhb . the rmf+bcs and rhb results of figs . 1 and 2 demonstrate that the bcs results are able to mock the rhb results provided experimental pairing gaps are used . secondly , the scalar - vector self - coupling force nl - sv2 provides a description of the binding energies which is considerably improved over that with the non - linear scalar self - coupling only . the 2-neutron separation energies have been tabulated and plotted @xcite over the whole periodic table using the experimental masses of nuclei . the salient feature in the curves of @xmath57 values in ref . @xcite is a kink which can be observed conspicuously about all the major magic numbers . a kink implies that @xmath57 values for nuclei with 2 neutron numbers above a magic number exhibit a dramatic decrease as compared to a nucleus with the magic number or nuclei below the magic number . this sudden decrease in the @xmath57 values signifies a shell gap above the magic number and thus reflects the role played by the ensuing shell effects . the magnitude of a shell gap does also show a significant dependence on mass number of a nucleus . it is thus expected that the shell effects vary strongly depending upon the part of the periodic table one is considering . figure 3 shows the results obtained with the forces with the non - linear scalar self - coupling nl - sh ( fig . 3a ) and nl3 ( fig . the results for the rmf+bcs and the rhb are compared with the experimental @xmath57 values ( solid circles ) . a staggeringly high value of @xmath56 35.0 mev for the 2-neutron separation energy for the nucleus @xmath65ni can be seen . it decreases smoothly to 30.8 mev for the doubly magic nucleus @xmath54ni . however , on adding 2-neutrons to the magic core , the @xmath57 value shows a dramatic decrease to 22.5 mev for @xmath55ni . this difference in the @xmath57 values is a clear measure of the shell gap at the magic number . inevitably , the downward kink at n=28 provides a sensitive observable for an investigation of the shell effects in nuclei . therefore , we focus upon the 2-neutron separation energies to analyze the characteristics of the shell effects . a comparison of the bcs results with the rhb in fig . 3 shows that except for a few minor differences , the two approaches provide results which are very close to each other for most of the isotopic chain . this is also the case in the extremely neutron - rich region . essentially , these results mirror the accord in the binding energies in the two approaches as shown in fig . this implies that the experimental pairing gaps used in the bcs approach are adequately represented by the self - consistent gogny pairing in the rhb approach . for a comparative look at the various predictions , @xmath57 values for rmf+bcs are shown in fig . 4a and for rhb in fig . a striking similarity is observed in the results of fig . 4a and fig . it arises naturally due to the similarity in the predictions of the binding energies with rmf+bcs and rhb as already observed in fig . there are , however , two differences in the rhb results as compared to the bcs ones . firstly , the inclusion of the self - consistent pairing in rhb increases the kink at n=28 slightly . secondly , the rhb results show an improved agreement with the experimental data in the region a=64 - 70 as compared to the bcs results . the rhb results employ the self - consistent pairing and ought to be considered as more realistic than bcs . therefore , we focus our attention primarily upon the rhb results . as observed for the binding energies in fig . 1 , the @xmath57 values with rhb for nl - sh and nl3 are also close to each other . as far as the experimental data is concerned , the rhb is able to describe the @xmath57 data on the lighter ( a=52 - 56 ) ni isotopes very well . however , for nuclei just above the magic number n=28 , i.e. , for @xmath55ni and @xmath66ni , the rhb values are lower than the experimental values by about 2 - 2.5 mev . this implies that the shell gap at n=28 is higher in the theory than indicated by the experimental data . thus , the rmf theory with the non - linear scalar coupling shows shell effects which are stronger than suggested by the experimental data . it can be seen from fig . 4b that for nuclei heavier than @xmath67ni , the @xmath57 values for nl - sh are closer to the experimental data than the nl3 values . the @xmath57 values for nuclei with a=70 - 78 are reproduced successfully both by nl - sh and nl3 . it may be noted that the empirical data on @xmath57 exist only up to @xmath68ni . the mass of the nucleus @xmath69ni is not known experimentally . therefore , our knowledge about the shell effects far away from the stability line is severely constrained . in this context , the nucleus @xmath69ni constitutes one such nucleus which has a potential to expose shell effects in reaching out to the neutron drip - line . a measurement of the mass of the nucleus @xmath69ni would undoubtedly shed light on how the shell effects behave far away from the stability line . here , we provide our predictions only on this account . due to uncertainties of pairing and coupling of states to the continuum , the bcs results are not expected to be fully realistic in this region . the rhb approach is the most appropriate one for the highly exotic ni nuclei . the presence of a kink in the rhb results ( fig . 4b ) at n=50 signifies the existence of shell effects which are still strong far away from the stability line . comparatively , the shell effects with nl - sh are slightly stronger than with nl3 in the extreme region . we will also compare the shell effects in the rhb approach to those in the hartree - fock bogoliubov ( hfb ) approach with the skyrme force skp in the latter part of this paper . the results on 2-neutron separation energy for the model with the scalar - vector self - coupling are presented in fig . we show the results with the forces nl - sv2 and tm1 . the upper panel ( fig . 5a ) displays the @xmath57 values with the bcs pairing . a comparison of the bcs results with the rhb results ( fig . 5b ) in the region of experimentally known nuclei shows that the two approaches give results which are remarkably similar , provided the experimental pairing gaps are used in the bcs calculations . again , we focus upon the rhb results due to the self - consistency of the pairing . the results with nl - sv2 show a good agreement with the experimental data for most of the isotopic chain . for the region below n=28 , the @xmath57 values with nl - sv2 are closer to the experimental data than those with tm1 . the relative difference in the @xmath57 values of nuclei @xmath54ni and @xmath55ni with tm1 is less than the experimental data , whereas it is closer to the experimental difference for the force nl - sv2 . in the region a=62 - 68 , where deviations from the experimental data are most notable , the nl - sv2 results are closer to the experimental data than tm1 . this is due to a better degree of agreement with the experimental binding energies obtained by using the vector self - coupling force nl - sv2 as shown in fig . 2 . we compare in fig . 6 the @xmath57 values obtained with the scalar - self coupling ( nl - sh ) to those with the scalar - vector self - coupling ( nl - sv2 ) . in fig . 6a the @xmath57 values across the shell - closure n=28 are presented . a comparison with the experimental data about n=28 shows that the force nl - sh gives shell effects which are stronger than suggested by the empirical data . the relative difference between the @xmath57 values of @xmath54ni and @xmath55ni is 11.1 mev with nl - sh and 9.2 mev with nl - sv2 . the difference in the @xmath57 values of nuclei about a shell closure is an indicator of the shell gap at the magic number . the results show that the shell gap at n=28 is reduced considerably with the scalar - vector self - coupling ( nl - sv2 ) as compared to the scalar coupling . in comparison , the experimental difference amounts to about 8.4 mev . thus , the shell gap at n=28 with nl - sv2 is very close to the experimental one . shell effects with the hartree - fock bogoliubov approach using the skyrme force skp have been discussed often in the literature @xcite . in order to clarify the situation , we include the @xmath57 values @xcite obtained in the hfb approach with the force skp in fig . it is seen clearly that the skp values cross the shell - closure n=28 ( a=56 ) smoothly . this has an unambiguous implication that _ contrary _ to the experimental observation , the shell effects with skp are quenched strongly at n=28 and consequently the shell gap at n=28 is diminished considerably . this aspect of skp will become obvious in the following where we compare single - particle levels from various approaches . the 2-neutron separation energies across the n=50 shell closure are shown in fig . the experimental data on @xmath57 is available only up to a=78 ( n=50 ) as shown by the solid circles . how the shell effects behave about n=50 in the region very far away from the stability line is still an open question ? we will show elsewhere @xcite that the very few experimental data available for n=50 are consistent with strong shell effects . determination of the nature of shell effects for waiting - point nuclei in the vicinity of n=50 are crucial to understanding the r - process nucleosynthesis in the neighbourhood of a=80 . a future measurement of mass of the nucleus @xmath69ni would help resolve this issue . in the present work , we make our predictions on the basis of the extrapolation from the region of stability . a comparison of the @xmath57 values for @xmath68ni ( n=50 ) and @xmath69ni for the forces nl - sh and nl - sv2 show that the shell gap at n=50 is larger for nl - sh than for nl - sv2 . this situation is very similar to that we have seen in the known region of n=28 . thus , also at n=50 the shell effects with nl - sh are stronger than nl - sv2 . taking into account a good agreement of nl - sv2 results with the experimental data about n=28 , we predict that at n=50 the shell effects still remain strong . this is in accordance with a general perception that the magic number n=50 constitutes a strong shell closure . this feature is thus predicted to persist also far away from the stability line . these results are contrasted sharply ( fig . 6b ) with the results of the hfb with the skyrme force skp which predicts a smooth decrease in the @xmath57 values across the magic number n=50 . the strong quenching of the shell effects at n=50 with skp as seen here is consistent with the similar quenching shown by this force also at n=28 ( see fig . 6a ) . differences in the shell effects in various approaches are exhibited succinctly in the single - particle spectrum as shown in fig . the neutron single - particle spectra for the nucleus @xmath55ni are displayed for the forces nl - sh and nl - sv2 and a comparison is made with skp . the force nl - sv2 with the scalar - vector self - coupling shows a decrease in the shell gap at n=28 as compared to nl - sh indicating again that the shell effects with nl - sv2 are softer . this is is conformity with the corresponding @xmath57 values in fig . 6a . the single - particle spectrum with skp in fig . 7 reveals a striking difference with that of nl - sh and nl - sv2 . the levels with skp are compressed significantly as compared to nl - sh and nl - sv2 . the mean - field potential with skp appears to be much shallower than the corresponding one in the rmf theory . the most important difference about the shell effects is that with skp the shell gap at n=28 is reduced significantly . this is the reason that a strong quenching of the shell effects is exhibited by skp at n=28 . with a view to illustrate the shell effects at n=50 , the single - particle levels for the nucleus @xmath69ni are shown in fig . 8 for nl - sh , nl - sv2 and skp . the shell gap at n=50 with nl - sv2 is reduced as compared to that with nl - sh . this reduction in shell gap is similar to that also observed for n=28 with nl - sv2 in fig . thus , the behaviour of the single - particle levels for @xmath55ni and @xmath69ni and the reduced shell gaps at n=28 and n=50 serve to illustrate the point that the slightly - softer shell effects with nl - sv2 are consistent with the experimental data . the single - particle spectrum for @xmath69ni with skp in fig . 8 is strikingly different than that for nl - sh and nl - sv2 . the levels with skp are compressed significantly as compared to nl - sh and nl - sv2 . this feature is very similar to that portrayed in fig . in fact , such a feature seems to be a generic one for skp . the single - particle spectrum for skp shows that the shell gap at n=50 with skp is about a factor of 2 smaller than the corresponding one with nl - sv2 . therefore , with skp the shell closure n=50 is quenched strongly . this is consistent with the smooth decrease of @xmath57 values with skp across n=50 ( see fig . the strikingly different single - particle scheme with skp may be attributed to the effective mass @xmath70 assumed for this force . the effective mass has an important influence on the single - particle structure of nuclei . the quantities which lend their influence to a single - particle scheme are the effective mass and the compressibility of nuclear matter . the latter is intricately connected to the saturation density through the properties of finite nuclei . recently , it has been shown that in the density dependent skyrme approach , the shell effects show a significant dependence on the compressibility of nuclear matter @xcite . in fig . 9 we examine predictions on the shell effects from various mass models . the @xmath57 values are shown for the mass formulae the finite - range droplet model ( frdm ) @xcite and the extended thomas - fermi with strutinsky integral ( etf - si ) @xcite . the frdm data about n=28 ( fig . 9a ) are in close agreement with the experimental data , showing that the shell effects about n=28 are represented adequately in the frdm . on the other hand , in the etf - si which is based upon the skyrme ansatz , the shell effects are weakened significantly at n=28 . a comparison of the etf - si data with the skp predictions shows that in both the cases , the shell effects are quenched strongly . it may be recalled that in the etf - si approach , shell corrections to a smooth component of energy - functional are added using the strutinsky method . thus , the nature of the shell corrections as obtained in this approach is crucial to deciding the strength of the shell effects . at n=50 ( fig . 9b ) predictions of the etf - si for the shell effects are , however , in stark contrast to those of skp . whilst skp predicts a washing out of the shell effects at n=50 as discussed above , the etf - si , on the other hand , shows shell effects which are strong at n=50 . this difference between etf - si and skp at n=50 is not easy to understand notwithstanding the similar quenching at n=28 in both the approaches . the strength of the shell effects with skp derives from the high value of the effective mass . this tends to reduce shell gaps at most of the magic numbers . on the other hand , shell effects in etf - si stem from as to how the shell effects are evaluated in different regions . shell corrections in the etf - si based upon the experimental data in the region maintain a strong shell closure at n=50 . a point of comparison , however , may be made here that the magnitude of the shell gap as reflected by the @xmath57 values at n=50 in etf - si is similar to that projected by the force nl - sv2 . the frdm results , however , show shell effects which are stronger than those predicted by both etf - si and nl - sv2 . the sn isotopes provide a very long chain where the experimental masses are known over a large range of isospin . the experimental data is presently available above the neutron magic number n=82 up to the mass a=136 ( n=86 ) . this allows us to examine the shell effects about the neutron shell closure n=82 which in the present case lies reasonably away from the stability line . the nature of the shell effects about n=82 should reveal as to how the shell closure n=82 retains ( or looses ) it magic character in the region far away from the stability line . we have included sn isotopes with mass a=100 - 138 for the present study . it must be pointed out that the sn isotopes have long been known to possess strong pairing properties akin to superfluidity . thus , the role of the pairing is very important for the sn nuclei . as in case of the ni isotopes , we have performed rmf+bcs calculations using the pairing gap obtained from the experimental masses of neighbouring nuclei . we have also performed the rhb calculations using the self - consistent finite - range pairing . the binding energy differences @xmath71 obtained with the two approaches are shown in fig . a direct comparison is made between the results of the forces nl - sh and nl3 . the rmf+bcs results ( fig . 10a ) show strong deviations from the experimental binding energies in the vicinity of a=100 . for the other regions of the range , the nl - sh results are generally better than nl3 . however , in the vicinity of the magic number n=82 ( a=132 ) , nl3 predictions are slightly better than those of nl - sh . the bcs results show several undulations which arise from the inadequacy of the bcs pairing . a kink due to the similar reasons is seen at a=108 with both the forces . the rhb results with the self - consistent pairing ( fig . 10b ) are free from the fluctuations of fig . 10a . the binding energies show a smooth trend with mass number and the kink at a=108 has disappeared . the smoothness of the rhb results vis - a - vis the bcs ones is an indication that the self - consistent pairing is very important for nuclei like sn isotopes . the rhb results ( fig . 10b ) with nl - sh and nl3 follow a similar pattern in the binding energy . it is observed that for the light sn isotopes , nl - sh overestimates the binding energies by a few mev . this aspect is especially apparent at the doubly magic nucleus a=100 both for nl - sh and nl3 . in the region a=100 - 110 , the nl3 values show a better agreement with the experimental data than nl - sh . however , in the latter part of the chain where stable sn isotopes exist , the nl - sh and nl3 results are very similar , showing a good agreement with the experimental data within 1 - 2 mev . in the neighbourhood of a=132 disagreement of about 2 - 3 mev with the experimental data appear for both nl - sh and nl3 . for a direct comparison of the rmf+bcs with the rhb , we show in fig . 11 the results from the two approaches for the forces nl - sh ( a ) and nl3 ( b ) . the rhb results show a smooth behaviour with the mass number for both the forces . the change from bcs to rhb is moderate for nl - sh . however , for nl3 results the change from bcs values to the rhb values is much larger than nl - sh . thus , the self - consistent pairing brings about a significant change in the energies of nl3 especially in the light - mass sn nuclei . this shows that with the force nl3 , the experimental pairing gaps are not able to take into account the pairing correlations correctly for the sn nuclei in particular those with lower mass . on the whole , this situation of sn isotopes is unlike the results of ni isotopes ( see figs . 1 and 2 ) where the rhb results mirrored much the results from the bcs . in other words , within the bcs the experimental pairing gaps for the ni isotopes were adequate to represent the pairing correlations . in fig . 12 we show the results with the forces with the scalar - vector self - couplings , with both the rmf+bcs and the rhb theory . here the forces nl - sv1 , nl - sv2 and tm1 are included . the presence of a kink at a=108 in the bcs calculations indicates that in spite of taking the pairing gaps from the experimental data , there seems to be an inconsistency in representing the pairing correlations appropriately in the bcs formalism . the smoothness of the rhb results with the self - consistent pairing shows that this drawback is eliminated in the rhb approach . this is true for all the forces included here . examining the the bcs results only , it is seen that the binding energies with nl - sv1 and tm1 are closer to the experimental data than nl - sv2 for the light - mass sn nuclei . nl - sv2 results in this region overpredict the experimental data slightly . for the region a=110 - 130 the binding energies with all the forces show a good agreement with the data within about 1 mev . in general , the bcs results with nl - sv1 show an overall good agreement with the data , except about the doubly magic nucleus @xmath72sn . in comparison , the tm1 results show a slightly higher disagreement with the data . the divergence of tm1 with the data is , however , more pronounced in the extreme region about a=136 . we have compared the rhb results with the scalar - vector self - coupling forces in fig . 13 . it is seen that the results of nl - sv2 and tm1 are similar over most of the mass range except in the vicinity of the doubly magic nucleus @xmath72sn . both the forces overestimate the binding energies of sn nuclei by about 2 - 4 mev . this implies that the underlying pairing correlations with the given pairing force d1s are stronger for the forces nl - sv2 and tm1 . on the other hand , the force nl - sv1 shows the best agreement with most of the data . the nl - sv1 results are within 1 - 2 mev of the experimental energies except for the region about @xmath72sn . in the neighbourhood of the other doubly magic nucleus @xmath58sn , the nl - sv1 results are are similar to nl - sv2 and are in agreement with the data within about 1 - 2 mev . the 2-neutron separation energies for the sn isotopes with the rhb approach using the forces with the non - linear scalar self - coupling are shown in fig . 14 . for both the forces nl - sh and nl3 , there is a very good agreement of the results with the experimental data in the mass region a=110 - 132 . both the forces show a slight divergence from the experimental values as one moves below a=110 towards a=100 . the results in the vicinity of the doubly magic nucleus a=132 show that the shell gap at n=82 is larger than the experimental data suggests . consequently , the shell effects with nl - sh are stronger than the empirical data . this is similar to what was also observed for the case of ni isotopes at n=28 . in comparison , nl3 shows shell effects which are slightly less stronger than those of nl - sh . however , the experimental difference in @xmath57 values of @xmath58sn and @xmath59sn is still overestimated by nl3 . the 2-neutron separation energies obtained with the scalar - vector self - coupling are shown in fig . the results with all the forces are in very good agreement with the experimental data over the whole mass range except for nuclei in the vicinity of the doubly magic a=132 . both nl - sv1 ( a ) and tm1 ( c ) overestimate slightly the experimental @xmath57 values for nuclei lighter than @xmath58sn . in comparison , the nl - sv2 results show an impressive agreement with the data almost over the whole mass range . for nuclei heavier than @xmath58sn , however , all the forces show a small divergence with the data . considering the shell gap at n=82 , we observe that all the forces with the scalar - vector self - coupling in fig . 15 are in good agreement with the experimental gap . this demonstrates that these forces reproduce the experimental shell gaps about n=82 very well . further , the presence of a well developed kink in @xmath57 values at a=132 ( n=82 ) emphasizes that the shell effects in this region which is away from the stability line do remain strong . indeed , this leads to a sharp reduction in the @xmath57 value of the isotope @xmath59sn . for nuclei heavier than this , the 2-neutron separation energies show only a mild decrease with the mass number . this is supported by the only experimental data point available at a=136 . theoretically , this behaviour is expected to be continued for heavier sn isotopes as shown by another nucleus ( a=138 ) in fig . 15 . it may be mentioned that the neutron drip line in sn nuclei is surmised to occur at very large mass number about a=170 . it is expected that the 2-neutron separation energy would decrease very slowly to a vanishing value until the drip line is reached . figure 16 shows the neutron single - particle levels in the canonical basis for the nucleus @xmath73sn as obtained with the forces nl - sh and nl - sv2 . it can be seen that with nl - sv2 the shell gap at n=50 is reduced as compared to nl - sh . this reduction in the shell gap near the fermi surface with the scalar - vector self - couplings was also observed for the ni nuclei . a comparison of the single - particle levels obtained with the skyrme force skp shows that the level density with this force is generally higher than the rmf forces and that due to this reason the shell gaps are suppressed rather strongly . the shell gap at n=50 is about half the shell gap in the rmf theory . this is illustrative of the quenching with the force skp , which is much discussed in the literature @xcite . we show in fig . 17 the single - particle levels for the nucleus @xmath59sn . we have chosen this nucleus in order to examine the shell gaps about the magic number n=82 . the shell gap at n=82 with nl - sv2 is smaller than with nl - sh and is consistent with the reduction in shell gap as reflected by the @xmath57 values about n=82 in fig . 15 as compared to nl - sh . in comparison , with the force skp the shell gap at n=82 is reduced significantly . in order to review the situation on the shell effects at n=82 with various mass models , we present in fig . 18 the @xmath57 from frdm , the etf - si and the mass model based upon infinite nuclear matter ( inm ) @xcite . although the mass formulae describe the @xmath57 values of most of the sn isotopes correctly , there are differences in predictions about the shell closure . comparison of @xmath57 values in the vicinity of a=132 in fig . 18a shows that the gap in the @xmath57 values about n=82 is slightly smaller with frdm than the experimental one implying that in the frdm the n=82 shell closure is slightly weaker . however , the shell effects with frdm are still stronger as compared to the scenerio of the strong quenching with skp in fig . the etf - si , on the other hand , is able to come closer to the experimental data . thus , at n=82 the etf - si demonstrates the shell effects which are stronger than frdm . at n=50 the shell effects with etf - si were also found to be strong and consistent with the experimental data . it may be reminded that at n=28 the shell effects are strongly quenched in the etf - si as opposed to the experimental data . in fig . 18b we compare the predictions of the inm model @xcite and skp with the experimental data . the @xmath57 value from the inm model for @xmath59sn is much higher compared to the experimental one . the corresponding shell gap at n=82 is consequently too small , implying a quenching of the shell effects . for all other isotopes , the @xmath57 values from the inm agree well with the experimental data . it may be mentioned that in the inm mass formula , determination of the nuclear matter properties and properties of finite nuclei is achieved on the basis of a model independent analysis of the available experimental data . however , the shell effects in the model seem to be quenched as seen at n=82 . the skp results show agreement with the experimental data except for nuclei in the vicinity of the magic number . with skp the shell gap at n=82 is reduced as compared to the experimental data . the reduction in the shell gap at n=82 ( see fig . 17 ) leads to an @xmath57 value for @xmath59sn , which is higher than the experimental one . this is indicative of a reduced shell strength with skp at n=82 . in contrast to the situation in ni isotopes , it may , however , be stated that skp does show a semblance of some shell strength at n=82 . the quenching of the shell effects at n=82 with skp is not so strong as has been witnessed for the ni isotopes at n=28 . in comparison , a strong quenching was presented by skp also at n=50 . therefore , contrary to the available experimental data , the quenching of the shell effects seems to be a generic feature of the force skp . we have studied the shell effects in nuclei about the stability line in the rmf theory . the chains of ni and sn isotopes have been considered , where experimental data over a large range of isospin are available . first , employing the existing forces nl - sh and nl3 with the non - linear scalar self - coupling of @xmath0-meson , we have investigated the ground - state properties of the ni and sn isotopic chains . it is shown that in the rmf+bcs approach where the pairing gaps are taken from the experimental data , the experimental binding energies of ni isotopes are described well . calculations within the rhb approach with the self - consistent finite - range pairing show only a slight improvement over the rmf+bcs results in this isotopic chain . this shows that the rmf+bcs approach in conjunction with the pairing gaps obtained from the experimental data suffices to bring about a good description of the binding energies of the ni isotopes . however , the same can not be said for the sn isotopes . the pairing correlations for sn nuclei are not adequately described within the bcs scheme . it is shown that for these nuclei , the rhb approach with the self - consistent pairing is necessary in order to get a good agreement with the binding energies . in addition to the nonlinear scalar coupling of @xmath0 meson , we have introduced non - linear vector self - coupling of @xmath1 meson in the rhb approach . the self - consistent pairing is included by taking the finite - range gogny force in the p - p channel . we have employed the recently developed forces nl - sv1 and nl - sv2 and the force tm1 . it is shown that the rhb approach with the non - linear scalar - vector self - couplings provides a very good description of the binding energies of ni and sn isotopes . we have investigated the shell effects in nuclei about the stability line within the two model lagrangians , i.e. the rhb theory with the non - linear scalar self - coupling of @xmath0-meson only and the rhb theory with the non - linear self - couplings of both @xmath0 and @xmath1 mesons . the experimental data on 2-neutron separation energies of ni and sn isotopes have been used . the sharp decrease in the @xmath57 values about a magic number is taken as a measure of the shell gap . the experimental data on ni encompass the magic number n=28 , whilst the magic number n=82 is encompassed by the sn data . it is shown that the forces with the non - linear scalar self - coupling give shell gaps which are larger than suggested by the experimental data . this is found to be the case for n=28 in ni isotopes as well as for n=82 in sn isotopes . the results with the lagrangian with the scalar - vector self - couplings show that the force nl - sv2 describes the experimental data on the 2-neutron separation energies very well . the shell effects about the magic numbers in ni and sn isotopes with the scalar - vector self - couplings are shown to be consistent with the experimental data . the single - particle spectra for the ni and sn nuclei in the vicinity of magic numbers show that the shell gaps at n=28 , n=50 and n=82 are smaller in the model with the scalar - vector self - couplings as compared to the scalar self - coupling only . a comparison of the experimental data and the results of the scalar - vector self - coupling model with the hfb results using the skyrme force skp for the ni isotopes has shown that the shell effects with skp are quenched strongly at n=28 . this is contrary to the experimental data which show that the shell effects are substantially stronger . the weak or quenched shell effects are also shown in ni isotopes by skp at n=50 . the quenching of the shell effects with the force skp is demonstrated by the corresponding single - particle spectra for @xmath55ni , @xmath69ni and @xmath59sn , whereby the shell gaps at n=28 , n=50 and n=82 , respectively are considerably smaller than the rhb single - particle gaps . in summary , the quenching of the shell effects seems to be an artifact of the force skp and is not supported by the experimental data . the substantial reduction in the shell gaps with skp stems from the correspondingly high level density . this is due to a large effective mass being chosen for this force . it is expected that due to inherently different single - particle structure in each model the shell effects at the stability line would have varying consequences on the shell effects at the drip lines . it was shown earlier @xcite that with the non - linear scalar self - coupling in the rmf theory shell effects near the neutron drip line in the vicinity of n=82 were predicted to be strong . this conclusion was based upon rmf theory with the bcs pairing using the force nl - sh . it was also shown @xcite that the pairing itself does not create or destroy the shell effects . the conclusions of the strong shell effects near the neutron drip line were contested @xcite . it was suggested on the basis of hfb calculations with skp that the shell effects near the drip line quench strongly . in the light of the forgoing discussions , it is only expected that the shell gap at n=82 with the force skp would diminish . thus , the quenching of the shell effects near the drip line at n=82 with the force skp is in complete accord with the quenching shown by it near the stability line . this quenching is , however , not supported by the experimental data at the stability line as demonstrated in this work . as shown above , the force nl - sh with the scalar coupling gives shell effects which are slightly stronger than the experimental data . this fact notwithstanding , we have performed a study @xcite to investigate the role of the vector self - coupling on shell effects at the drip line . it is observed that with the scalar - vector self - couplings which describes the shell effects at the stability line well , shell effects near the neutron drip line are slightly less stronger than with the scalar self - coupling only . however , we do not find indications of any shell quenching . details of this work will be provided elsewhere @xcite . it may be noted that in the fits of the r - process abundances @xcite , hfb results based upon the force skp were used . it was shown @xcite that the hfb results give an improved fit to the global r - process abundances . since the force skp is known to quench shell effects as shown above , it was concluded on this basis that the shell effects at n=82 at the neutron drip line are quenched . however , as discussed in our work , this quenching with skp is not consistent with the experimental data . thus , an improved fit to r - process abundances using the skp results may be insufficient to assert the shell quenching at n=82 near the neutron drip - line , notwithstanding an overwhelming shell quenching with skp . this research was supported by the kuwait university research administration through the project no . we thank prof . j. dobaczewski for supplying the hfb results on ni isotopes . fig . 1 : : binding energy of ni isotopes with respect to the experimental values , in ( a ) rmf approach with bcs pairing and ( b ) rhb approach with self - consistent finite - range pairing , using the forces nl - sh and nl3 with the non - linear scalar self - coupling of @xmath0 meson . fig . 2 : : the same as in fig . 1 , with the forces nl - sv1 , nl - sv2 and tm1 with the non - linear scalar - vector self - coupling . fig . 3 : : the 2-neutron separation energy @xmath57 of ni isotopes using the forces with the non - linear scalar self - coupling ( a ) nl - sh and ( b ) nl3 . a comparison of the rmf+bcs and rhb values is made with the experimental data . fig . 4 : : the @xmath57 values for ni isotopes in ( a ) rmf+bcs and ( b ) rhb approach . within each approach , a comparison of the results of the forces nl - sh and nl3 is made with the experimental data . fig . 5 : : the @xmath57 values for ni isotopes in ( a ) rmf+bcs and ( b ) rhb approach with the forces with the non - linear scalar and vector self - coupling . a comparison of the results of the forces nl - sv2 and tm1 is made with the experimental data . : : a comparison of the 2-neutron separation energy @xmath57 of ni isotopes using the force nl - sh with the scalar self - coupling and using the force nl - sv2 with the scalar - vector self - coupling in rhb . the results are shown in the region of closed neutron shell ( a ) n=28 ( a=56 ) and ( b ) n=50 ( a=78 ) . a comparison with the experimental data is made . the results from the hfb approach with the skyrme force skp are also shown for comparison @xcite . : : the neutron single - particle spectrum for the nucleus @xmath55ni calculated in the canonical basis in rhb with the forces nl - sh and nl - sv2 . a reduction in the shell gap at n=28 with the force nl - sv2 can be seen , as compared to nl - sh . the corresponding spectrum for skp shows a drastic reduction in the n=28 shell gap . : : the same as fig . 7 , for the nucleus @xmath69ni . the shell gap at n=50 with the force skp is significantly reduced as compared to that with nl - sh and nl - sv2 . : : the @xmath57 values in the region ( a ) n=28 and ( b ) n=50 from the mass models frdm , etf - si and the hfb approach with skp . a comparison is made with the experimental data available . 10 : : the binding energies of sn isotopes in the ( a ) rmf+bcs and ( b ) rhb approach with the forces nl - sh and nl3 . 11 : : a comparison of the binding energies of sn nuclei in the rmf+bcs and rhb approach with the forces ( a ) nl - sh and ( b ) nl3 . a smoothening of the energies in the rhb is observed . 12 : : the binding energies of sn nuclei using the forces with the vector self - coupling of @xmath1 meson ( a ) nl - sv1 , ( b ) nl - sv2 and ( c ) tm1 , compared with the experimental values , in rmf+bcs and rhb approach . fig . 13 : : the binding energies of sn isotopes obtained with the rhb approach with the self - consistent pairing . a comparison is made between the results from the forces nl - sv1 , nl - sv2 and tm1 . 14 : : the @xmath57 values for sn isotopes obtained in the rhb approach with the forces ( a ) nl - sh and ( b ) nl3 . the experimental data about a=132 shows a strong kink indicating strong shell effects about n=82 . 15 : : the @xmath57 values for sn isotopes obtained in the rhb approach with the forces ( a ) nl - sv1 , ( b ) nl - sv2 and ( c ) tm1 . the experimenetal data are shown for the comparison . 16 : : the neutron single - particle spectrum of @xmath73sn obtained in the canonical basis in the rhb approach with the forces nl - sh and nl - sv2 . the levels with the force skp are also shown for comparison . fig . 17 : : the same as fig . 16 , for @xmath59sn . 18 : : a comparative view of the @xmath57 values for sn isotopes from the mass models frdm @xcite , etf - si @xcite , and inm model @xcite in the region about n=82 . l l l l l l l l l + & & & nl - sh & nl3 & & & nl - sv1 & nl - sv2 & tm1 + + & m & & 939.0 & 939.0 & & & 939.0 & 939.0 & 938.0 + & @xmath74 & & 526.05921 & 508.1941 & & & 510.03488 & 519.81202 & 511.198 + & @xmath75 & & 783.0 & 782.5010 & & & 783 . & 783.0 & 783.0 + & @xmath76 & & 763.0 & 763.0 & & & 763 . & 763.0 & 770.0 + & @xmath77 & & 10.44355 & 10.2169 & & & 10.12479 & 10.32001 & 10.0289 + & @xmath78 & & 12.9451 & 12.8675 & & & 12.72661 & 12.88233 & 12.6139 + & @xmath79 & & 4.38281 & 4.4744 & & & 4.49197 & 4.50144 & 4.6322 + & @xmath80 & & -6.90992 & -10.4307 & & & -9.24058 & -6.86061 & -7.2325 + & @xmath81 & & -15.83373 & -28.8851 & & & -15.388 & 0.36754 & 0.6183 + & @xmath82 & & 0.0 & 0.0 & & & 41.01023 & 72.38965 & 71.30750 +

the multivariate generalized pareto distribution arises as the limit of a suitably normalized vector conditioned upon at least one component of that vector being extreme . statistical modelling using multivariate generalized pareto distributions constitutes the multivariate analogue of peaks over thresholds modelling with the univariate generalized pareto distribution . we introduce a construction device which allows us to develop a variety of new and existing parametric tail dependence models . a censored likelihood procedure is proposed to make inference on these models , together with a threshold selection procedure and several goodness - of - fit diagnostics . we illustrate our methods on two data applications , one concerning the financial risk stemming from the stock prices of four large banks in the united kingdom , and one aiming at estimating the yearly probability of a rainfall which could cause a landslide in northern sweden .

1612.01773

peaks over threshold modelling of univariate time series has been common practice since the seminal paper of @xcite , who advocated the use of the asymptotically motivated generalized pareto ( gp ) distribution @xcite as a model for exceedances over high thresholds . the multivariate generalized pareto distribution was introduced in @xcite , ( * ? ? ? * chapter 8) , and @xcite , but still , statistical modelling using this approach has thus far received relatively little attention . partially this is because theoretically equivalent dependence modelling approaches , based on the so - called `` point process approach '' , have already been in existence for some time @xcite . nonetheless , the multivariate gp distribution has conceptual advantages over that of the point process representation , in so much as it represents a proper multivariate distribution on an `` l - shaped '' region , where at least one variable is extreme , @xmath0 ; see figure [ fig : supp ] . furthermore , the gp distribution permits modelling of data on this region without the need to perform any marginal transformation , which is common in other extremal dependence modelling approaches . the justification for the use of multivariate gp distributions as models for the tails of essentially arbitrary distributions is the limit property in . there is a growing body of probabilistic literature devoted to multivariate gp distributions ( e.g. @xcite , @xcite , @xcite , @xcite , @xcite ) . to our knowledge , however , there are only a few papers that exploit these as a statistical model @xcite . this paper forms a companion to @xcite , which collates new and existing probabilistic results on multivariate gp distributions . we briefly recall necessary elements from this paper , but our main focus is on statistical modelling and inference . the premise of statistical modelling using limiting distributions such as the gp distribution is that _ we assume the limit model holds sufficiently well in a sufficiently extreme region_. care is needed , however , to ensure that the postulated model is indeed appropriate . an important consideration when modelling multivariate extremes is the concept of asymptotic dependence . random variables @xmath1 and @xmath2 with distribution functions @xmath3 and @xmath4 , respectively , are said to be asymptotically dependent if @xmath5 = \frac{{\mathbb{p}}[f_1(y_1 ) > q , f_2(y_2 ) > q]}{1-q } \to \chi_{1:2 } > 0 , \qquad q\to1.\end{aligned}\ ] ] existence of the limit is an assumption , and the limit being positive characterizes asymptotic dependence , whereas @xmath6 defines asymptotic independence . an extension of @xmath7 to a general dimension @xmath8 is given by @xmath9}{1-q } \to \chi_{1:d}.\ ] ] when @xmath10 , there is a positive probability of the most extreme events occurring simultaneously in all @xmath11 variables . multivariate gp distributions are useful mostly when @xmath10 . when @xmath12 then the corresponding _ limiting _ gp distribution does not place any mass in @xmath11-dimensional space , with all the mass lying in some lower dimensional subspace instead . this situation is challenging to deal with , as in practice the data do belong to the full @xmath11-dimensional space . whilst modelling of such possibilities is not precluded in the censored likelihood framework that we adopt , we do not consider it further here . yet other subtleties arise , since even when @xmath10 , this does not rule out the possibility of a distribution placing mass on some lower - dimensional subspace . however , for the remainder of the paper , we consider the simplified situation where there is no mass on any lower - dimensional subspace . the contributions of this paper are the following . in section [ sec : background ] we present some of the key results and properties of multivariate gp distributions that are useful for statistical modelling . in section [ sec : models ] we introduce a construction device for gp - distributed random vectors and use it to develop a variety of new and existing models in section [ sec : examples ] . inference using a censored likelihood - based approach is detailed in section [ sec : inference ] , together with threshold selection and goodness - of - fit diagnostics . in section [ sec : applications ] we fit gp models to returns of four uk - based banks and to rainfall data in the context of landslide risk estimation , showing that multivariate gp modelling can be more useful for financial risk handling than one - dimensional methods , and that our models can respect physical constraints in a way which was not possible before . we conclude with a discussion in section [ sec : discussion ] . let @xmath13 be a random vector in @xmath14 with distribution function @xmath15 . a common and broadly - applicable assumption on @xmath13 is that it is in the so - called _ max - domain of attraction _ of a multivariate max - stable distribution , @xmath16 . this means that if @xmath17 are independent and identically distributed copies of @xmath18 , then one can find sequences @xmath19 and @xmath20 such that @xmath21 \to g(\bm{x } ) , \label{eq : maxstabconv}\end{aligned}\ ] ] with @xmath16 having non - degenerate margins . in and throughout , vectors are boldface , and operations involving vectors are to be interpreted componentwise , with shorter vectors being recycled if necessary . the resulting max - stable distribution @xmath16 has marginal location , scale and shape parameters denoted @xmath22 , @xmath23 and @xmath24 , respectively , and the lower endpoints of its support are determined by these parameters . if @xmath25 denotes this vector of lower endpoints , and @xmath26 , its components are @xmath27 if @xmath28 and @xmath29 otherwise . we assume @xmath30 , which is always possible through appropriate choice of @xmath31 and @xmath32 . max - stable distributions are not a primary concern of this paper , but the above mild assumption leads us to an analogous convergence theorem for multivariate threshold exceedances . specifically , if convergence holds , then @xmath33 where @xmath34 follows a multivariate gp distribution @xcite . we let @xmath35 denote the distribution function of @xmath36 , and @xmath37 its marginal distributions . typically the margins @xmath38 are not univariate gp , due to the difference between the conditioning events @xmath39 and @xmath40 in the one - dimensional and @xmath11-dimensional limits . however , the marginal distributions conditioned to be positive are gp distributions . that is , writing @xmath41 , we have @xmath42 = ( 1+\gamma_j x /\sigma_j)^{-1/\gamma_j}_+,\end{aligned}\ ] ] where @xmath43 and @xmath44 are as previously defined , giving the link between marginal parameterizations for the two convergences . the full link between @xmath35 and @xmath16 is @xmath45 , and we say that such a @xmath35 and @xmath16 are _ associated_. the support of the multivariate gp distribution @xmath35 is included in the set @xmath46 the dependence structure of @xmath35 does not have a finite - dimensional parameterization @xcite , but it does satisfy certain properties that can be used to construct flexible parametric models . in section [ sec : examples ] we give several examples , some of which are gp models that are associated to well - known max - stable models . following common practice in the statistical modelling of extremes , @xmath35 may be used as a model for data which arise as multivariate threshold excesses in the sense @xmath47 . in particular , if @xmath48 is a threshold that is `` sufficiently high '' in each margin , then from , @xmath49 can be approximated by a member of the class of multivariate gp distributions , with @xmath50 , @xmath24 , the marginal exceedance probabilities @xmath51 , and the dependence structure to be estimated . in practice the truncation by the unknown vector @xmath52 is only relevant when dealing with mass on lower - dimensional subspaces . following the discussion in section [ sec : extremaldependence ] , we suppose that @xmath53 is to be approximated by a gp distribution . a member of the class of gp distributions @xmath36 has a representation on @xmath54 as @xmath55 where @xmath56 is a `` standard form '' gp random vector , that is , a gp vector on a standardized scale @xmath57 and @xmath58 . its construction will be further discussed in section [ sec : models ] . for @xmath59 , the corresponding component of the right - hand side of equation is simply @xmath60 . the following are useful properties of the gp distributions ; for further details and proofs we refer to @xcite . [ [ sec : ts ] ] threshold stability . + + + + + + + + + + + + + + + + + + + + gp distributions are _ threshold stable _ , meaning that if @xmath61 follows a gp distribution with marginal parameters @xmath62 and @xmath63 then for @xmath64 such that @xmath65 and @xmath66 , @xmath67 this property states that if we increase , or at least do not decrease , the level of the threshold in each margin , then the distribution of conditional excesses is still gp , with a new set of scale parameters , but retaining the same vector of shape parameters . a special role is played by the levels @xmath68 : these have the stability property that for any set @xmath69 it holds that , for @xmath70 , @xmath71 = { \mathbb{p}}[{\bm{x}}\in a ] / t,\ ] ] where @xmath72 . this follows from equation along with the representation of @xmath73 to be given in equation . the @xmath74-th component of @xmath75 , @xmath76 , is the @xmath77 quantile of @xmath78 . equation provides one possible tool for checking if a multivariate gp is appropriate ; see section [ sec : diagnostics ] . [ [ sec : condmarg ] ] lower dimensional conditional margins . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + lower dimensional margins of gp distributions are typically not gp , as the conditioning event leading to the distribution involves all @xmath11 variables , and not only the variables in a lower dimensional margin . let @xmath79 , with @xmath80 , and similarly for other vectors . then @xmath81 does follow a gp distribution . combined with the threshold stability property above , we also have that if @xmath82 is such that @xmath83 and @xmath84 then @xmath85 follows a gp distribution . [ [ sec : chi ] ] constant conditional exceedances . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + [ [ sec : sum ] ] sum - stability under shape constraints + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + if @xmath34 follows a multivariate gp distribution , with scale parameter @xmath50 and shape parameter @xmath95 , then for weights @xmath96 such that @xmath97 with positive probability , we have @xmath98 that is , weighted sums of components of a multivariate gp distribution with equal shape parameters , conditioned to be positive , follow a univariate gp distribution with the same shape parameter and with the scale parameter equal to the weighted sum of the marginal scale parameters . the dependence structure of the gp distribution does not affect this result , but it does affect the probability of the conditioning event , i.e. , the probability that the sum of components is positive . further details can be found in @xcite . we focus on how to construct suitable densities for the random vector @xmath56 , which through equation , leads to densities for the multivariate gp distribution with marginal parameters @xmath62 and @xmath63 . let @xmath99 be a unit exponential random variable and let @xmath100 be a @xmath11-dimensional random vector , independent of @xmath99 . define @xmath101 . then the random vector @xmath102 has the required properties to be a gp vector with support included in the set @xmath103 and with @xmath104 and @xmath105 ( interpreted as the limit for @xmath106 for all @xmath74 ) . moreover , _ every _ such gp vector can be expressed in this way @xcite . the probability of the @xmath74-th component being positive is @xmath107 = { \mathbb{e } } [ e^{t_j - \max(\bm{t } ) } ] $ ] , which , in terms of the original data vector @xmath13 , corresponds to the marginal exceedance probability @xmath108 $ ] , i.e. , the probability that the @xmath74-th component exceeds its corresponding threshold given that one of the @xmath11 components does . suppose @xmath100 has a density @xmath109 on @xmath110 . by theorem 5.1 of @xcite , the density of @xmath56 is given by @xmath111 one way to construct models therefore is to assume different distributions for @xmath100 , which provide flexible forms for @xmath112 , and for which ideally the integral in can be evaluated analytically . one further construction of gp random vectors is given in @xcite . if @xmath113 is a @xmath11-dimensional random vector with density @xmath114 and such that @xmath115 < \infty$ ] for all @xmath116 , then the following function also defines the density of a gp distribution : @xmath117 } \int_{0}^\infty f_{\bm{u } } ( { \bm{x}}+ \log t ) { \ , \mathrm{d}}t . \end{aligned}\ ] ] the marginal exceedance probabilities are now @xmath107 = { \mathbb{e } } [ e^{u_j } ] / { \mathbb{e } } [ e^{\max(\bm{u } ) } ] $ ] for @xmath118 . formulas and can be obtained from one another via a change of measure linking @xmath109 and @xmath114 . where @xmath109 and @xmath114 take the same form , then the similarity in integrals between and means that if one can be evaluated , then typically so can the other ; several instances of this are given in the models presented in section [ sec : examples ] . what is sometimes more challenging is calculation of the normalization constant @xmath119 = \int_0^\infty \{1-f_{\bm{u}}(\log t \bm{1})\ } \,{\ , \mathrm{d}}t$ ] in . nonetheless , the model in has the particular advantage over that of that it behaves better across various dimensions : if the density of the gp vector @xmath36 is @xmath120 and if @xmath121 , then the density of the gp subvector @xmath122 is simply @xmath123 . this property is advantageous when moving to the spatial setting , since the model retains the same form when numbers of sites change , which is useful for spatial prediction . in section [ sec : standard ] we presented densities in the standardized form @xmath124 , @xmath125 . using transformation , we obtain a general form for the corresponding densities for threshold excesses , @xmath34 , which in terms of the original data , is an approximation for the conditional density of @xmath126 given that @xmath127 : @xmath128 in , @xmath129 may be either @xmath112 or @xmath120 . observe that it is always possible to transform this density to be suitable for large values of @xmath18 itself , i.e. , to a density for the conditional distribution of @xmath18 given that @xmath127 by replacing @xmath130 by @xmath131 in the expression above . the models detailed in section [ sec : standard ] are built on a standardized scale , and then transformed to the observed , or `` real '' scale . alternatively a model can be constructed directly on the real scale , which gives the possibility of respecting structures , say additive structures , in a way which is not possible with the other two models ; this approach will be used to model ordered data in section [ sec : rainfall ] . one way of presenting this is to define the random vector @xmath132 in terms of @xmath113 in through the componentwise transformation @xmath133 and develop suitable models for @xmath134 . this gives the gp density @xmath135 } \int_0^\infty t^{\sum_{j=1}^d \gamma_j } f_{\bm{r } } \left ( \bigl ( g(t;x_j,\sigma_j,\gamma_j ) \bigr)_{j=1}^d \right ) { \ , \mathrm{d}}t,\ ] ] where @xmath136 denotes the density of @xmath132 and where @xmath137 the @xmath11 components of @xmath113 are found by inverting equation . for @xmath58 and @xmath57 , the densities and are the same . here we provide details of certain probability distributions for @xmath100 , @xmath113 and @xmath132 that generate tractable multivariate gp distributions . below we refer to these probability distributions as `` generators '' . several articles have previously focused on the use of random vectors to generate dependence structures for extremes , for example @xcite and @xcite among others . the literature on max - stable modelling for spatial extremes also relies heavily on this device @xcite . in order to control bias when using a multivariate gp distribution , we often need to use censored likelihood ( see section [ sec : inference ] ) and thus not just to be able to calculate densities , but also integrals of those densities . these considerations guide our choice of models presented below . for each model we give the uncensored densities in the subsequent subsections , whilst their censored versions are given in appendix [ sec : censored ] . in sections [ sec : indep ] and [ sec : mvn ] we consider particular instances of densities @xmath109 and @xmath114 to evaluate the corresponding densities @xmath112 and @xmath120 in and . as noted in section [ sec : standard ] , even if @xmath138 , the gp densities @xmath112 and @xmath120 are still different in general . thus we will focus on the density of a random vector @xmath139 with density @xmath140 and create two gp models per @xmath140 by setting @xmath141 and then @xmath142 , in the latter case with the restriction @xmath115 < \infty$ ] . the support for each gp density given in sections [ sec : indep ] and [ sec : mvn ] is @xmath143 . in section [ sec : structured ] we exhibit a construction of @xmath144 in with support depending on @xmath24 and @xmath50 . in all models , identifiability issues occur if @xmath100 or @xmath113 have unconstrained location parameters @xmath145 , or if @xmath132 has unconstrained scale parameters @xmath146 . indeed , replacing @xmath145 or @xmath146 by @xmath147 or @xmath148 , respectively , with @xmath149 and @xmath150 , lead to the same gp distribution ( * ? ? ? * proposition 1 ) . a single constraint , such as fixing the first parameter in the vector , is sufficient to restore identifiability . let @xmath151 be a random vector with independent components and density @xmath140 , so that @xmath152 , where @xmath153 are unspecified densities of real - valued random variables . the dependence structure of the associated gp distributions is determined by the relative heaviness of the tails of the marginal distributions : roughly speaking , if one component has a high probability of being `` large '' compared to the others , then the dependence is weaker than if all components have a high probability of taking similar values . throughout , @xmath154 is such that @xmath155 . let @xmath156 , \qquad \alpha_j>0 , \ , \beta_j\in\mathbb{r}.\ ] ] [ [ case - f_bmt - f_bmv . ] ] case @xmath157 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + density is @xmath158 if @xmath159 then the integral can be explicitly evaluated : @xmath160 [ [ case - f_bmu - f_bmv . ] ] case @xmath161 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the marginal expectation of the exponentiated variable is @xmath162 = \begin{cases } e^{\beta_j } \gamma(1 - 1/\alpha_j ) , & \alpha_j>1 , \\ \infty , & \alpha_j \leq 1 . \end{cases}\ ] ] for @xmath163 , density is @xmath164 if @xmath159 then this simplifies to : @xmath165 observe that if in addition to @xmath159 , we also have @xmath166 , then this is the multivariate gp distribution associated to the well - known _ logistic _ max - stable distribution . let @xmath167 , \qquad \alpha_j>0,\,\beta_j\in\mathbb{r}.\ ] ] as the gumbel case leads to the multivariate gp distribution associated to the logistic max - stable distribution , the reverse gumbel leads to the multivariate gp distribution associated to the _ negative logistic _ max - stable distribution . calculations are very similar to the gumbel case , and hence omitted . let @xmath168 [ [ case - f_bmt - f_bmv.-1 ] ] case @xmath157 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + density is @xmath169 [ [ case - f_bmu - f_bmv.-1 ] ] case @xmath161 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the expectation of the exponentiated variable is @xmath115 = 1/ \left\{e^{\beta_j}(\alpha_j+1)\right\}$ ] , which is finite for all permitted parameter values . density is @xmath170}\int_0^{e^{-\max(\bm{x}+\bm{\beta } ) } } \prod_{j=1}^d \frac{1}{\alpha_j}(te^{x_j+\beta_j})^{1/\alpha_j } { \ , \mathrm{d}}t \notag\\ & = \frac{(e^{-\max(\bm{x}+\bm{\beta})})^{\sum_{j=1}^d 1/\alpha_j + 1}}{{\mathbb{e}}[e^{\max(\bm{u})}]}\frac{1}{1+\sum_{j=1}^d 1/\alpha_j } \prod_{j=1}^d \frac{1}{\alpha_j } ( e^{x_j+\beta_j})^{1/\alpha_j}. \label{eq : modrevexp}\end{aligned}\ ] ] the normalization constant may be evaluated as @xmath171 & = \int_0^\infty \left(1-{\textstyle\prod_{j=1}^d } \min(e^{\beta_j } t , 1)^{1/\alpha_j}\right ) { \ , \mathrm{d}}t \\ & = e^{-\beta_{(d)}}- \frac{\prod_{j=1}^d e^{\beta_j/\alpha_j}}{\sum_{j=1}^d 1/\alpha_j + 1 } e^{-\beta_{(1)}(\sum_{j=1}^d 1/\alpha_j+1 ) } \\ & \qquad + \sum_{i=1}^{d-1 } \frac{\prod_{j = i+1}^d e^{\beta_{(j)}/\alpha_{[j]}}}{\sum_{j = i+1}^d 1/\alpha_{[j ] } + 1 } \left(e^{-\beta_{(i+1)}(\sum_{j = i+1}^d 1/\alpha_{[j ] } + 1)}- e^{-\beta_{(i)}(\sum_{j = i+1}^d 1/\alpha_{[j ] } + 1)}\right),\end{aligned}\ ] ] where @xmath172 and where @xmath173}$ ] is the component of @xmath23 with the same index as @xmath174 ( thus the @xmath173}$]s are not in general ordered ) . as far as we are aware the associated max - stable model is not well known . let @xmath175 i.e. , @xmath176 gamma@xmath177 . [ [ case - f_bmt - f_bmv.-2 ] ] case @xmath157 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + density is @xmath178 [ [ case - f_bmu - f_bmv.-2 ] ] case @xmath161 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the marginal expectation of the exponentiated variable is @xmath115 = \alpha_j$ ] , hence finite for all permitted parameter values . density is @xmath179 } \prod_{j=1}^d \left ( \frac{e^{\alpha_j x_j}}{\gamma(\alpha_j ) } \right ) \int_0^\infty t^{\sum_{j=1}^d \alpha_j } e^{- t \sum_{j=1}^d e^{x_j } } { \ , \mathrm{d}}t \\ & = \frac{1}{{\mathbb{e}}[e^{\max(\bm{u } ) } ] } \frac{\gamma\left(\sum_{j=1}^d \alpha_j+1\right)}{\prod_{j=1}^d \gamma ( \alpha_j ) } \frac { e^{\sum_{j=1}^d \alpha_j x_j - \max(\bm{x})}}{(\sum_{j=1}^d e^{x_j})^{\sum_{j=1}^d \alpha_j+1}}. \ ] ] the normalization constant is @xmath180 & = \frac{\gamma \left ( \sum_{j=1}^d \alpha_j + 1\ \right)}{\prod_{j=1}^d \gamma ( \alpha_j ) } \int_{\delta_{d-1 } } \max ( u_1,\ldots , u_d ) \prod_{j=1}^d u_j^{\alpha_j - 1 } { \ , \mathrm{d}}u_1 \cdots { \ , \mathrm{d}}u_{d-1},\end{aligned}\ ] ] where @xmath181^d : u_1 + \cdots + u_d = 1\ } $ ] is the unit simplex , and the integral can be easily computed using the package . this gp distribution is associated to the so - called dirichlet max - stable distribution @xcite . let @xmath182 , where @xmath183 is the mean parameter , and @xmath184 is a positive - definite covariance matrix . as before , @xmath154 is such that @xmath155 . for these calculations , it is simplest to make the change of variables @xmath185 in equations and . [ [ case - f_bmt - f_bmv.-3 ] ] case @xmath157 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + density is @xmath186 with @xmath187 a @xmath188 matrix of rank @xmath189 . [ [ case - f_bmu - f_bmv.-3 ] ] case @xmath161 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as required , the expectation @xmath190 = e^{\beta_j + \sigma_{jj}/2}$ ] is finite for all permitted parameter values , where @xmath191 denotes the @xmath74th diagonal element of @xmath192 . density is @xmath193 } \int_{-\infty}^\infty \frac{(2\pi)^{-d/2}}{|\sigma|^{1/2}}\exp\left\{-\tfrac{1}{2}({\bm{x}}-\bm{\beta}-s \bm{1})^t\sigma^{-1}({\bm{x}}-\bm{\beta}-s \bm{1 } ) + s\right\ } { \ , \mathrm{d}}s\\ & = \frac{(2\pi)^{(1-d)/2}|\sigma|^{-1/2}}{{\mathbb{e}}[e^{\max(\bm{u})}](\bm{1}^t\sigma^{-1}\bm{1})^{1/2 } } \exp\left\{-\tfrac{1}{2}({\bm{x}}-\bm{\beta})^t a ( { \bm{x}}-\bm{\beta } ) + 2\frac{({\bm{x}}-\bm{\beta})^t\sigma^{-1}\bm{1}-1}{\bm{1}^t\sigma^{-1}\bm{1 } } \right\},\end{aligned}\ ] ] with @xmath88 as in . the distribution for this case is already known ; it is the gp distribution associated to the brown resnick or hsler reiss max - stable model @xcite . a variant of the density formula with @xmath115 = 1 $ ] ( equivalently @xmath194 ) was given in @xcite . although they focussed on poisson point process inference , one can use their expressions to derive the associated gp distribution in a straightforward manner . the normalization constant is @xmath195 { \ , \mathrm{d}}t$ ] , where @xmath196 is the zero - mean multivariate normal distribution function with covariance matrix @xmath192 . this normalization constant can also be expressed as a sum of multivariate normal distribution functions , see @xcite . observe that in high dimensions , the normalization constant for the density based on @xmath113 may be onerous to compute , whilst the density based on @xmath100 does not require this . in sections [ sec : indep ] and [ sec : mvn ] we considered several distributions for @xmath113 and @xmath100 . here we present a model for @xmath132 based on cumulative sums of exponential random variables . cumulative sums will lead to a vector @xmath132 whose components are ordered ; for the components of the corresponding gp vector to be ordered as well , we will assume that @xmath95 and @xmath197 . this model will be of interest in section [ sec : rainfall ] , where we will focus on modelling cumulative precipitation amounts which may trigger landslides . [ [ case - bmgamma - bm0 . ] ] case @xmath57 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + by construction , the densities @xmath198 and @xmath199 coincide since @xmath200 . let @xmath201 be the random vector whose components are defined by @xmath202 where @xmath203 are the rate parameters , i.e. , @xmath204 = \exp \left(-\lambda_j x_j \right)$ ] for @xmath205 . its density , @xmath136 , is given by @xmath206 where we set @xmath207 . in view of , @xmath208 ( or equivalently @xmath209 ) implies @xmath210 . the density of @xmath56 is given as follows : if @xmath211 , then @xmath212 } \left ( \prod_{j=1}^d \lambda_j e^{x_j } \right ) \int_0^\infty t^d \exp \left\ { - t \left ( \sum_{j=1}^d ( \lambda_j - \lambda_{j+1 } ) e^{x_j } \right ) \right\ } { \ , \mathrm{d}}t \notag \\ & = \frac{{\mathbbm{1}}(x_d > 0 ) \ , d ! \ , \prod_{j=1}^d \lambda_j e^{x_j } } { \left ( \sum_{j=1}^d \lambda_j^{-1 } \right ) \left ( \sum_{j=1}^d ( \lambda_j - \lambda_{j+1 } ) e^{x_j } \right)^{d+1}},\end{aligned}\ ] ] while @xmath213 is zero otherwise . the density @xmath214 is then directly obtained from . for @xmath215 , the dependence measure @xmath7 from takes the simple form @xmath216 . some properties of the structured components model can be inferred from this expression ; when @xmath217 , then @xmath218 regardless of the value of the parameter . if @xmath219 , then @xmath220 ; if @xmath221 , then @xmath222 . it is natural that this model can not approach asymptotic independence , since it is based on cumulative sums . [ [ case - bmgamma - bm0.-1 ] ] case @xmath223 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath224 be the random vector whose components are defined by @xmath225 where @xmath203 are the rate parameters . its density , @xmath136 , is given by @xmath226 where we set @xmath207 . then @xmath227 = { \mathbb{e}}\left [ \max_{1 \leq j \leq d } \left ( \frac{\gamma r_j}{\sigma } \right)^{1/\gamma } \right ] = \left ( \frac{\gamma}{\sigma } \right)^{1/\gamma } { \mathbb{e}}\left [ r_d^{1/\gamma } \right].\ ] ] the distribution of @xmath228 is called generalized erlang if @xmath229 for all @xmath230 @xcite , and , letting @xmath231 denote its density we get @xmath232 & = \left ( \frac{\gamma}{\sigma } \right)^{1/\gamma } \int_0^{\infty } r^{1/\gamma } f_{r_d } ( r ) { \ , \mathrm{d}}r \\ & = \left ( \frac{\gamma}{\sigma } \right)^{1/\gamma } \sum_{i=1}^d \lambda_i \left(\prod_{j=1 , j \neq i}^d \frac{\lambda_j}{\lambda_j-\lambda_i } \right ) \int_0^{\infty } r^{1/\gamma } e^{-r \lambda_i } { \ , \mathrm{d}}r \\ & = \left ( \frac{\gamma}{\sigma } \right)^{1/\gamma } \gamma \left(\frac{1}{\gamma } + 1 \right ) \sum_{i=1}^d \lambda_i^{-1/\gamma } \left(\prod_{j=1 , j \neq i}^d \frac{\lambda_j}{\lambda_j-\lambda_i } \right).\end{aligned}\ ] ] if @xmath233 , then @xmath228 follows an erlang distribution . by , the density of @xmath34 becomes , for @xmath234 , @xmath235 } \\ & = \frac{{\mathbbm{1}}(x_d > 0 ) \left ( \prod_{j=1}^d \lambda_j \right ) \left ( \frac{\gamma}{\sigma } \right)^{-1/\gamma } \gamma \left ( d + \frac{1}{\gamma } \right ) / \gamma \left(\frac{1}{\gamma } \right ) } { \left ( \sum_{j=1}^d ( \lambda_j - \lambda_{j+1 } ) x_j + ( \sigma/\gamma ) \lambda_1 \right)^{d + 1/\gamma}\sum_{i=1}^d \lambda_i^{-1/\gamma } \left(\prod_{j=1 , j \neq i}^d \frac{\lambda_j}{\lambda_j-\lambda_i } \right)}.\end{aligned}\ ] ] working within a likelihood - based framework for inference allows many benefits . firstly , comparison of nested models can be done using likelihood ratio tests . this is important as the number of parameters can quickly grow large if margins and dependence are fitted simultaneously , thus we can test for simplifications in a principled manner . secondly , incorporation of covariate effects is straightforward in principle . for univariate peaks over thresholds , such ideas were introduced by @xcite , but nonstationarity in dependence structure estimation has received comparatively little attention ; an exception is @xcite in the context of max - stable processes . thirdly , such likelihoods could also be exploited for a bayesian approach to inference if desired . the density is the basic ingredient in a likelihood , however , we will use as a contribution only when all components of the observed translated vector @xmath236 are `` large '' , in the sense of exceeding a threshold @xmath237 , with @xmath238 . the reasoning for this is twofold : 1 . for @xmath28 , the lower endpoint of the multivariate gp distribution is @xmath239 . we would like to avoid the situation where small values of a component can affect the fit of the upper tail of that component too strongly . 2 . without censoring , bias in the estimation of parameters controlling the dependence can be larger than for censored estimation , see @xcite . let @xmath240 and @xmath241 be the subset of indices denoting which components of @xmath236 fall below the corresponding component of @xmath237 , i.e. , @xmath242 for @xmath243 , and @xmath244 for @xmath245 , with at least one such @xmath246 . for each realization of @xmath13 , we use the likelihood contribution @xmath247 } h(\bm{y}-\bm{u};\bm{\gamma},\bm{\sigma } ) { \ , \mathrm{d}}\bm{y}_{c } , \qquad \bm{y}_{c}=(y_j)_{j \in c},\end{aligned}\ ] ] which is equal to with @xmath248 if @xmath249 is empty . appendix [ sec : censored ] contains forms of censored likelihood contributions for the models included in section [ sec : models ] . thus , for @xmath250 independent observations @xmath251 of @xmath252 , the form of the censored likelihood function to be optimized is @xmath253 where @xmath254 denotes the censoring subset for @xmath255 , which may be empty , and @xmath256 represents parameters related to the model that we assumed for the generator . when fitting multivariate gp distributions to data on the observed scale we have a large variety of potential models and parameterizations . for non - nested models , akaike s information criterion ( aic = @xmath257 log - likelihood + @xmath258 number of parameters ) can be used to select a model with a good balance between parsimony and goodness - of - fit . when looking at nested models , e.g. , to test for simplifications in parameterization , we can use likelihood ratio tests . because of the many possibilities for model fitting , to reduce the computational burden we propose the following model - fitting strategy , that we will employ in section [ sec : banks ] . a. standardize the data to common exponential margins , @xmath259 , using the rank transformation ( i.e. , the probability integral transform using the empirical distribution function ) ; b. select a multivariate threshold , denoted @xmath260 on the scale of the observations , and @xmath261 on the exponential scale , using the method of section [ sec : diagnostics ] ; c. fit the most complicated standard form model within each class ( i.e. , maximum number of possible parameters ) to the standardized data @xmath262 ; d. select as the standard form model class the one which produces the best fit to the standardized data , in the sense of smallest aic ; e. use likelihood ratio tests to test for simplification of models within the selected standard form class , and select a final standard form model ; [ test ] f. fit the gp margins simultaneously with this standard form model , to @xmath263 by maximizing ; g. test for simplifications in the marginal parameterization . although this strategy is not guaranteed to result in a final gp model that is globally optimal , in the sense of minimizing an information criterion such as aic , it should still result in a sensible model whilst avoiding enumeration and fitting of an unfeasibly large number of possibilities . the validity of the fit of the selected model can be assessed via diagnostics . an important issue that pervades extreme value statistics in all dimensions is the selection of a threshold above which the limit model provides an adequate approximation of the distribution of threshold exceedances . here this amounts to `` how can we select a vector @xmath260 such that @xmath264 is well - approximated by a gp distribution ? '' . there are two considerations to take into account : @xmath265 should be well - approximated by a univariate gp distribution , for @xmath266 , and the dependence structure of @xmath264 should be well - approximated by that of a multivariate gp distribution . marginal threshold selection has a large body of literature devoted to it ; see @xcite and @xcite for recent reviews . threshold selection for dependence models is a much less well studied problem , though recent literature that begins to address this topic includes @xcite who consider threshold selection via bayesian measures of surprise , and @xcite who examines how to make better use of so - called parameter stability plots , offering a method that can be employed on any parameter , pertaining to the margins or dependence structure . here we propose using the `` constant conditional exceedance '' property of multivariate gp distributions outlined in section [ sec : chi ] to guide threshold selection for the dependence structure . if @xmath267 and @xmath268 , then on the region @xmath269 , we have @xmath270 with @xmath271 . a consequence of this is that @xmath272 should be constant on the region @xmath273 , if @xmath260 represents a sufficiently high dependence threshold . hence if we identify @xmath274 , then @xmath275 should provide an adequate threshold for the dependence structure . once suitable thresholds have been identified for margins , @xmath276 , and dependence , @xmath277 , then a threshold vector which is suitable for the entire multivariate model is @xmath278 . to illustrate this idea , let @xmath279 denote the empirical version of @xmath272 , defined by @xmath280 where @xmath281 represent the empirical distribution functions . figure [ fig : datachi ] displays @xmath279 for data in the _ domain of attraction _ of a gp distribution . from the right - hand panel it appears that the estimate looks reasonably constant after @xmath282 . in addition to aiding threshold selection , such plots also aid determination of whether the multivariate gp distribution forms a suitable model for the data : if it appears that @xmath283 as @xmath284 then we should explore further whether it is reasonable to assume that the data are asymptotically dependent . having identified a multivariate gp model to fit , and a threshold above which to fit it , a key concern is to establish whether the goodness - of - fit is adequate . for the dependence structure , one diagnostic comes from comparing @xmath279 for @xmath285 to @xmath94 , which for models @xmath112 in has the form @xmath286,\ ] ] whilst for models @xmath120 in we get the form @xmath287.\ ] ] the formula of @xmath94 stemming from @xmath144 models then follows through equation . in some cases these expressions may be evaluated analytically , but they can always be evaluated with arbitrary accuracy by simulation ; see @xcite . a further diagnostic could exploit the fact that @xmath288 = { \mathbb{e}}[e^{t_j-\max(\bm{t } ) } ] = { \mathbb{e}}[e^{u_j}]/{\mathbb{e}}[e^{\max ( \bm{u } ) } ] $ ] ( see section [ sec : standard ] ) . thus one would compare @xmath289 $ ] with the relevant model - based probability . in the event that the @xmath290 are equal marginal quantiles , @xmath289 $ ] = @xmath291 / { \mathbb{p}}[\bm{y}\not\leq\bm{u}]$ ] is the same for each margin . equation suggests a model - free diagnostic of whether a multivariate gp model may be appropriate . to exploit this , one defines a set of interest @xmath88 , and compares the number of points of @xmath263 that lie in @xmath88 to @xmath292 times the number of points of @xmath293 lying in @xmath88 for various choices of @xmath294 . according to , the ratio of these numbers should be approximately equal to @xmath295 . we exploit this diagnostic in section [ sec : rainfall ] . note that setting @xmath296 is equivalent to computing @xmath297 with @xmath37 replaced by @xmath298 . finally , in the event that the margins can be modelled with identical shape parameters , one can test property by examining the adequacy of the implied univariate gp distribution from a multivariate fit . this will be used in section [ sec : banks ] below . we examine the weekly negative raw returns on the prices of the stocks from four large banks in the uk : hsbc ( h ) , lloyds ( l ) , rbs ( r ) and barclays ( b ) . data were downloaded from yahoo finance . specifically , letting @xmath299 , @xmath300 , denote the closing stock price ( adjusted for stock splits and dividends ) in week @xmath292 for bank @xmath74 , the data we examine are @xmath301 , so that large positive values of @xmath302 correspond to large relative losses for that index . the observation period runs from 29 october 2007 to 17 october 2016 , with a total of @xmath303 datapoints . figure [ fig : bankpw ] displays pairwise plots of the negative returns . there is evidence of strong extremal dependence from these plots , as the largest value of @xmath304 occurs simultaneously , with positive association amongst other large values . the largest value of @xmath305 occurs at a different time , but again there is positive association between other large values . figure [ fig : bankchi ] shows an estimate of the function @xmath306 for these data . as is common in practice the value of @xmath307 generally decreases as @xmath89 increases , but is plausibly stable and constant from slightly above @xmath282 . consequently , we proceed with fitting a gp distribution . ultimately , we wish to fit a parametric gp model to the raw threshold excesses @xmath308 . in view of the large variety of potential models and parameterizations , we use the model selection strategy detailed in section [ sec : modelchoice ] . throughout we use censored likelihood with @xmath309 . we begin by selecting a threshold and finding a standard form model for the marginally transformed data . on the basis of figure [ fig : bankchi ] , the 0.83 marginal quantile is used as the threshold in each margin ; there are a total of 149 observations with at least one exceedance of the corresponding marginal threshold . + with hsbc ( h ) , lloyds ( l ) , rbs ( r ) and barclays ( b ) . approximate @xmath310 pointwise confidence intervals are obtained by bootstrapping from @xmath311.,scaledwidth=30.0% ] we fit five dependence models from section [ sec : examples ] , specifically the models with densities , , , and , to the standardized data . the smallest aic is given by model , i.e. , where @xmath109 is the density of independent gumbel random variables . we therefore select this class and proceed with item of the procedure in section [ sec : modelchoice ] to test for simplifications within this class . the results are summarized in table [ tab : banktest ] , indicating that in fact a model with a single parameter is adequate , since both possible sequences of likelihood ratio tests between nested models lead to m4 when adopting a 5% significance level . this is a useful simplification , since with relatively few data points it places us in a better position for fitting a full gp distribution . .negative uk bank returns : parameterizations of model for the standardized data and their maximized log likelihoods . [ cols="<,<,<,<",options="header " , ] [ tab : fit ] for a visual test we consider qq - plots for each of the marginal conditional gp distributions @xmath312 , with @xmath313 as before . figure [ fig : qqplotexp ] shows that for the model with @xmath314 , the fit is less good for the first component due to the restriction @xmath197 . this restriction is used to ensure that the components are ordered . plots for the model with @xmath315 ( not shown ) are very similar . for the dependence structure , we start by looking at the goodness - of - fit diagnostic suggested by equation ( see also section [ sec : diagnostics ] ) . since we set @xmath314 , we look at the ratio @xmath316}{t \ , \mathbb{p } [ \bm{y } - \bm{u } - \bm{\sigma } \log t \in a \mid \bm{y } \nleq \bm{u}]},\ ] ] where @xmath50 denotes the vector of scale parameter estimates of the marginal gp models above @xmath313 . we consider the sets @xmath317 for @xmath318 . figure [ fig : gof ] shows that the result is satisfactory , i.e. , expression is near one , so that a mgp model seems appropriate for this dataset . the result for @xmath319 shows more variability than the other two since the number of exceedances in the first component is relatively low . next , we would like to compare the pairwise and trivariate @xmath320 ( formulas can be found in appendix [ sec : supporting ] ) to their empirical counterparts ; see section [ sec : diagnostics ] . figure [ fig : chiplots ] shows the results , where the horizontal lines represent the model - based @xmath320 s based on @xmath321 , @xmath322 and @xmath323 from the top row in table [ tab : fit ] . again , the fit is satisfactory . finally , we consider the goodness - of - fit diagnostic which consists of comparing the model - based probabilities @xmath324 = \frac{\mathbb{e}[e^{u_j}]}{\mathbb{e}[e^{\max(\bm{u } ) } ] } = \frac{\sum_{i=1}^j \lambda_i^{-1}}{\sum_{i=1}^d \lambda_i^{-1}},\ ] ] with the empirical probabilities @xmath325 $ ] for @xmath326 . we find @xmath327 = 0.34 $ ] @xmath328 and @xmath329 = 0.63 $ ] @xmath328 using the values from the top row in table [ tab : fit ] ; the standard errors are obtained using the delta method . for @xmath330 , the empirical probabilities are @xmath331 and @xmath332 respectively . plots of the empirical probabilities for a range of thresholds @xmath333 ( not shown ) confirm the chosen threshold value @xmath313 . a more formal way to assess the goodness - of - fit of the dependence structure of this model is by calculating the test statistic presented in ( * ? * corollary 2.5 ) . the test statistic proposed there is based on the difference between the value of @xmath334 and an empirical estimator thereof . it depends on a sequence @xmath335 where @xmath336 and @xmath337 as @xmath338 , which represents the threshold value used : a low value of @xmath339 corresponds to a high threshold . the test statistic converges to a chi - square distribution with @xmath340 degrees of freedom ; its @xmath310 quantile is equal to @xmath341 . computing the test statistic for @xmath342 , where we set again @xmath343 , we find the values @xmath344 , @xmath345 , @xmath346 , @xmath347 , and @xmath348 respectively , so that we can not reject the structured components model for any value of @xmath339 . we have outlined several new models for multivariate gp distributions , along with their uncensored and censored likelihoods . the models were applied to two types of data : stock price returns , where advantages in terms of consideration of a portfolio were demonstrated ; and rainfall data , where the context dictated that extremes of ordered cumulative data were the object of interest . methods to select a multivariate threshold and diagnostics for the fitted models have been considered and demonstrated through the applications . the threshold selection method suggested in section [ sec : diagnostics ] is relatively conservative in the sense that a multivariate gp model could hold with lower thresholds in some margins . better ways to select a multivariate threshold , ideally incorporating marginal and dependence considerations simultaneously , remains an interesting problem . a key issue that was highlighted in section [ sec : extremaldependence ] was the idea of asymptotic ( in)dependence , and how we can detect when multivariate gp distributions form an appropriate model . in particular , we have not dealt with the situation where the gp distribution places mass on a lower dimensional subspace , with the models outlined in section [ sec : examples ] placing no mass on lower - dimensional subspaces of the @xmath11-dimensional space @xmath349 . in principle , parameters assigning mass to such lower - dimensional spaces could be introduced and estimated through censored likelihood , although it seems likely that information on these parameters would be weak , and alternative ways of handling this situation are much needed . if certain subsets of variables display asymptotic dependence , then , depending on questions of interest , it may be worth considering these separately . however , if all pairwise @xmath350 , then no asymptotic dependence exists amongst any variables , and no multivariate gp model would be approporiate , since the limiting model places mass only on one - dimensional lines . in this context , methods from @xcite , @xcite or @xcite may prove useful . here we detail forms of censored likelihoods for the models detailed in section [ sec : models ] . for simplicity they are presented in standardized ( @xmath124 , @xmath125 ) form , i.e. , @xmath351 } h({\bm{x}};\bm{1},\bm{0 } ) { \ , \mathrm{d}}{\bm{x}}_c , \label{eq : gpcens}\end{aligned}\ ] ] for @xmath352 and @xmath129 corresponding to either @xmath112 or @xmath120 . the generalized form of a censored likelihood is easily obtained from as @xmath353 the support for each density is @xmath354 , and we let @xmath355 denote the cardinality of the set @xmath249 . [ [ case - f_bmt - f_bmv.-4 ] ] case @xmath157 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @xmath356 if all @xmath357 are equal to @xmath358 : @xmath359 [ [ case - f_bmu - f_bmv.-4 ] ] case @xmath161 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @xmath360 if all @xmath357 are equal to @xmath358 : @xmath361 [ [ case - f_bmt - f_bmv.-5 ] ] case @xmath157 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @xmath362 to evaluate this , consider two cases : ( i ) @xmath363 ; and ( ii ) let @xmath364 for @xmath365 and @xmath366 . in case ( i ) , we have @xmath367 since on the range @xmath368 the term @xmath369 in is equal to @xmath370 . in case ( ii ) this term will vary over that range , and one needs to split the integral as follows : @xmath371 an evaluation of each integral yields that @xmath372 is equal to @xmath373 \bigg\}\\ & \quad \mbox { } + \frac{\prod_{j\in c_{(k ) } } e^{(v_j+\beta_j)/\alpha_j } \prod_{j\in { d\setminus c } } ( 1/\alpha_j ) e^{(x_j+\beta_j)/\alpha_j } } { \sum_{j \in c_{(k ) } } 1/\alpha_j + \sum_{j \in { d\setminus c } } 1/\alpha_j } \\ & \qquad\qquad \mbox { } \times \left [ \left ( e^{\max_{j\in{d\setminus c}}(x_j+\beta_j ) } \right)^{-\sum_{j \in c_{(k ) } } 1/\alpha_j -\sum_{j \in { d\setminus c } } 1/\alpha_j}-\left(e^{v_{(k)}+\beta_{(k)}}\right)^{-\sum_{j \in c_{(k ) } } 1/\alpha_j -\sum_{j \in { d\setminus c } } 1/\alpha_j } \right]\end{aligned}\ ] ] with @xmath374 , i.e. , with the indices corresponding to the @xmath375 largest @xmath376 removed . [ [ case - f_bmu - f_bmv.-5 ] ] case @xmath161 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + is found similarly by noting the relation between these two approaches . [ [ case - f_bmt - f_bmv.-6 ] ] case @xmath157 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath377 denote the cumulative distribution function of a gamma@xmath177 random variable . then @xmath378 [ [ case - f_bmu - f_bmv.-6 ] ] case @xmath161 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + defining @xmath379 , we have @xmath380 [ [ case - f_bmt - f_bmv.-7 ] ] case @xmath157 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for the gaussian model , using abbreviated notation , the key observation is @xmath381 } h({\bm{x } } ) { \ , \mathrm{d}}{\bm{x}}_c = h_{{d\setminus c}}({\bm{x}}_{{d\setminus c } } ) \int_{\times_{j\in c}(-\infty , v_j ] } \frac{h({\bm{x}})}{h_{{d\setminus c}}({\bm{x}}_{{d\setminus c } } ) } { \ , \mathrm{d}}{\bm{x}}_c , \label{eq : loggaussint}\end{aligned}\ ] ] and the ratio in the second integral can be written as a proper gaussian density function ( with parameters that depend on @xmath382 ) . the integrand is @xmath383 \right\}\end{gathered}\ ] ] with @xmath384 firstly note that @xmath385 as the maximum will not occur among the censored components . by a completion of the square it can be shown that expression is in fact equal to @xmath386 with @xmath387 and @xmath388 where @xmath389 ( respectively @xmath390 ) is a @xmath391 [ respectively @xmath392 matrix of 0s with 1s in the @xmath393 position , for @xmath394 the @xmath339th index in @xmath249 and @xmath395 , @xmath396 ( similarly for @xmath390 ) . therefore equation resolves as @xmath397 with @xmath398 the cdf of a @xmath355-variate multivariate gaussian distribution with location vector @xmath22 and covariance matrix @xmath399 . [ [ case - f_bmu - f_bmv.-7 ] ] case @xmath161 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + again this can be found similarly to the above noting the relation between these two forms ; see also @xcite . recall that since this is a model on the random vector @xmath132 , we need to differentiate between @xmath57 and @xmath223 . we present the case @xmath400 only , since the case @xmath223 is very similar . moreover , we set @xmath401 as in section [ sec : rainfall ] . [ [ case - bmgamma - bm0.-2 ] ] case @xmath57 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the censored likelihood has an analytical expression but is tedious to write down . note that , since the density @xmath402 is non - zero only for @xmath211 , we censor in @xmath403 components if @xmath404 . if @xmath405 , then for @xmath406 and @xmath407 , @xmath408 where @xmath409 . expressions for @xmath410 follow naturally by repeated integration of the above result . pointwise confidence intervals are obtained by a transformation of the beta distributed order statistics of a uniform distribution.,scaledwidth=99.0% ] pointwise confidence intervals are obtained by a transformation of the beta distributed order statistics of a uniform distribution . , title="fig:",scaledwidth=35.0% ] pointwise confidence intervals are obtained by a transformation of the beta distributed order statistics of a uniform distribution . , title="fig:",scaledwidth=35.0% ] + for the three - dimensional structured components model fitted in section 6.2 , the dependence measures @xmath411 , @xmath412 and @xmath413 are @xmath414 the authors gratefully acknowledge the financial support from the following agencies and projects : the knut and alice wallenberg foundation ( a. kiriliouk , h. rootzn , and j. wadsworth ) , the `` fonds de la recherche scientifique - fnrs '' ( a. kiriliouk ) the contract `` projet dactions de recherche concertes '' no . 12/17 - 045 of the `` communaut franaise de belgique '' ( a. kiriliouk and j. segers ) , iap research network grant p7/06 of the belgian government ( j. segers ) , and epsrc fellowship grant ep / p002838/1 ( j. wadsworth ) . we thank the abisko scientific research station for providing access to their rainfall data , which we used in section [ sec : rainfall ] . these data may be obtained by following the instructions at http://polar.se/en/abisko-naturvetenskapliga-station/vaderdat . data for section [ sec : banks ] can be downloaded from http://finance.yahoo.com . programs for the analyses in sections [ sec : banks ] and [ sec : rainfall ] are available from the authors on request .

we use the kpno international spectroscopic survey ( kiss ) for emission - line galaxies to identify and describe a sample of local analogues to the luminous compact blue galaxies ( lcbgs ) that are observed to be abundant at intermediate and high redshift . the sample is selected using criteria believed effective at isolating true examples of lcbgs :(b - band ) mag arcsec , ( for km s mpc ) , and . additionally , all lcbg candidates presented are selected to have star - formation as their dominant form of activity . we examine the properties of our lcbgs and compare them to those of other kiss star - forming galaxies of the same absolute magnitude range . we find that the kiss lcbgs lie on the extreme end of a fairly continuous distribution of `` normal '' star - forming galaxies in the plane of surface brightness versus color . this result differs from the results of previous studies that show lcbgs at higher - z to be more separate from the `` normal '' ( usually non - active ) galaxies they are compared against . on average , lcbgs have a higher tendency to emit detectable flux in the radio continuum , have higher h luminosities by a factor of 1.6 , indicating strong star - formation activity , and have slightly lower than expected metal abundances based on the luminosity - metallicity relation for kiss galaxies . we calculate the volume density of our low- ( ) sample to be , approximately 4 times lower than the volume density of the lcbgs at 0.4 z 0.7 and times lower than the volume density of the population at 0.7 1.0 .

astro-ph0410127

common at intermediate and high redshifts , yet rare in the local universe , luminous compact blue galaxies ( lcbgs ) typically have luminosities equal to or greater than that of the milky way but are considerably smaller in size @xcite . as evidenced by such properties , they are among the most extreme galaxies known in the universe . although these galaxies have been the subject of numerous comprehensive studies in the past decade , their evolution and nature , to a large extent , remains in dispute @xcite . at the heart of the issue lies the difficulty of identifying nearby analogues that would allow a more complete exploration of the mechanisms responsible for their extreme properties at high redshifts , as well as their evolutionary paths . characterized by high surface brightnesses , small half - light radii , and vigorous star - formation , lcbgs were initially identified as stellar objects from 4-m plate surveys for qso candidates @xcite . the formal definition of lcbgs is developed by @xcite from examining regions of six - dimensional parameter space ( color , luminosity , asymmetry , concentration , size , and surface brightness ) occupied by three classes of compact galaxies chosen to represent the population of luminous , blue , compact galaxies as a whole . the three classes are : ( 1 ) compact narrow emission - line galaxies ( cnelgs , koo _ etal . _ 1994 , 1995 ; guzmn 1996 , 1998 ) , ( 2 ) a subset of galaxies very similar to cnelgs studied by @xcite and @xcite called faint blue galaxies , and ( 3 ) higher redshift blue nucleated galaxies ( bngs , schade _ etal . _ 1995 , 1996 ) . while the classification of an lcbg relies on the galaxy s falling within three of five defined regions in the aforementioned parameter space , recently groups have adopted three cut - offs that roughly correspond to the criteria laid out by @xcite : @xmath14 , @xmath15 mag arcsec@xmath2 , and @xmath16 , with @xmath17 km s@xmath5 mpc@xmath5 @xcite . these selection criteria will be further examined in subsequent sections . with a volume density that drops off substantially from @xmath18 to the present ( previously estimated to be a change of roughly a factor of ten ) , lcbgs appear to be the most strongly evolving galaxy population @xcite . @xcite estimate that they contribute nearly @xmath19 of the fractional increase in the sfr density between @xmath20 and @xmath18 . furthermore , luminous , compact star - forming galaxies appear to dominate high - redshift galaxy samples . @xcite have suggested that the lyman break galaxies at @xmath21 , due to their very compact cores and high surface brightnesses , may be the high - redshift counterparts of the intermediate - z lcbgs . @xcite , in a sub - millimeter galaxy survey , found a @xmath22 galaxy population that may contribute a substantial fraction of the sfr density in the early universe , nearly half of which are very luminous and compact . evidently , lcbgs play a prominent , if not paramount , role in the formation and evolution of galaxies in the young universe . a drove of unanswered questions hovers around the subject of lcbgs , mainly owing to the lack of a representative local sample . one of these questions involves the mechanism(s ) responsible for the observed intense starbursts in lcbgs . in theory , these triggering mechanisms could be determined by examining the h@xmath7 velocity fields and morphologies of local lcbg examples @xcite . this sort of study relies on the assumption that one can select a sample of local lcbgs to have the same properties as the more distant lcbgs . @xcite use a sample of six galaxies with @xmath23 between -17.5 and -20 selected from local blue compact galaxies ( bcgs ) that are bright in h@xmath7 to infer certain kinematic properties of the higher-@xmath8 lcbgs . they conclude that , in most cases , mergers or strong interactions trigger the intense starbursts of lcbgs . however , one should use caution in interpreting these results given the size and properties of their sample . indeed , a statistically complete sample of low - z lcbgs will provide more clues as to the the origin of their activity . another crucial topic in the study of lcbgs has been the question of what sort(s ) of galaxy they become as they evolve . the determination of reliable mass estimates for the intermediate-@xmath8 lcbgs is key to establishing these evolutionary connections to present - day galaxies . however , accurate stellar mass estimates are difficult to determine and measurements of emission - line widths used to indicate the virial mass may underrepresent the gravitational potential for the intermediate - redshift lcbgs @xcite . as a consequence of these difficulties , there is not good agreement for the mass estimates of lcbgs and there is a corresponding uncertainty as to their proposed evolutionary path(s ) . @xcite , @xcite , and @xcite argue for an evolutionary scenario in which lcbgs represent the formation of the bulges of objects that will eventually evolve into massive spiral disks of today . etal . _ ( 1995 ) , and guzmn _ etal . _ ( 1996 , 1997 , 1998 ) maintain that while lcbgs represent a heterogeneous class of galaxies , the majority will evolve into local low - mass dwarf elliptical galaxies . a possible solution to the mass debate may lie in the hi line widths of local lcbgs . these line widths can be used to infer the total dynamical mass . accordingly , @xcite set out to measure the hi line widths of a nearby sample of galaxies they believe are representative of the more distant lcbgs . they infer the masses of the intermediate-@xmath8 galaxies from these local examples , and find that lcbgs have a variety of dynamical masses , and thus a variety of evolutionary paths . yet , they recognize that their local sample may not have been truly representative of the the high-@xmath8 lcbg population . of their 11 local galaxies , only three meet the criteria proposed by @xcite . in a study of 20 local lcbg candidates selected to more nearly meet the the criteria of @xcite , @xcite found results similar to those of @xcite . specifically , they show that while most lcbgs are nearly ten times less massive than local galaxies of the same luminosity , some are just as massive . the implication is that lcbgs are a heterogeneous class of galaxy . clearly , defining reliable local samples of lcbgs is central to the process of understanding their nature and evolutionary paths . the @xcite and @xcite samples are recent attempts to construct samples of nearby lcbgs . the latter study is based on a sample of lcbgs discovered by @xcite who analyzed data from the sloan digital sky survey . another recent study by @xcite utilized the complete spectroscopic study of a 12 deg@xmath24 area centered on the fornax cluster to identify a sample of 13 lcbgs with redshifts less than 0.21 . in the current study , we take a slightly different approach . using the kpno international spectroscopic survey ( kiss , salzer _ etal . _ 2000 ) , we aim to identify and describe a statistically complete local sample of star - forming galaxies that best resemble those lcbgs observed in the more distant universe . our low - z sample will provide a basis for future work aimed at resolving the debates over the evolutionary histories and kinematics of lcbgs . due to the limited resolution of the survey images , the classification scheme developed by @xcite , that uses morphological parameters , can not be employed . instead , we apply selection criteria consistent with the parameters of @xcite as a way to isolate the lcbgs from other kiss star - forming galaxies . while sharp boundaries in parameter space do not always effectively isolate one type of galaxy from another , lcbgs in the color - surface brightness plane occupy a well - defined region of parameter - space @xcite . therefore , we employ criteria in _ _ b__@xmath25_v _ color , in surface - brightness , and in luminosity to select a sample of local lcbg candidates . still , none of the selection criteria we use can guarantee that the kiss lcbg candidates , like other putative local samples , will contain _ bona fide _ lcbgs ; they only insure a similarity . regardless , we believe that the sample we define will provide an adequate starting point for future investigation . in @xmath26 2 , we describe kiss , our measurements , and selection criteria . an analysis of the properties of our local lcbgs along with the volume density calculation appears in @xmath26 3 . section 4 contains a discussion of the implications our sample has for high-@xmath8 studies and potential future work on these galaxies beyond kiss . in @xmath26 5 , we summarize our results . throughout this paper we adopt @xmath4 km s@xmath5 mpc@xmath5 . the kpno international spectroscopic survey ( kiss ) aims to find a quantifiably complete , well - defined sample of extragalactic emission - line sources @xcite . although this survey is not the first to look for galaxies that display unusual activity , it reaches substantially deeper than other wide - field objective - prism surveys , mainly because it @xmath27 the first survey of this type to employ a large - format ccd as its detector . the survey data , composed of both objective - prism and direct images , were taken with the 0.61 m burrell schmidt telescope located on kitt peak . the objective - prism images cover a spectral range of either 4800 - 5500 or 6400 - 7200 and the direct images are observed through standard _ b _ and _ v _ filters . our local sample of lcbgs is derived from the first three kiss survey strips : kiss red list 1 ( kr1 ) , described in @xcite ; kiss blue list 1 ( kb1 ) , described in @xcite ; and kiss red list 2 ( kr2 ) , described in @xcite . kr1 and kr2 use the h@xmath7 line for selection whereas kb1 detects objects by their [ oiii ] emission line . the kr1 portion of kiss finds 1128 elg candidates in a survey area of 62.2 deg@xmath24 ( or 18.1 kiss elgs per square degree ) ; the kb1 portion of kiss finds 223 elg candidates in 116.6 deg@xmath24 ( or 1.91 kiss elgs per square degree ) ; and the kr2 portion of kiss finds 1029 elg candidates in 65.8 deg@xmath24 ( or 15.6 kiss elgs per square degree ) . kr1 and kb1 overlap with part of the century survey ( geller _ etal . _ 1997 ; wegner _ etal . _ 2001 ) while kr2 runs through the center of the botes void . the kiss data are run through several steps of processing using a series of image reduction and analysis facility(iraf ) scripts : object detection and inventory , photometry and object classification , astrometry , spectral image coordinate mapping and background subtraction , spectral extraction and overlap correction , emission line detection , and spectral parameter measurement . measurements of the extracted objective - prism spectra yield estimates of the redshifts , line fluxes , and equivalent widths of the elgs . for a complete description of the data processing , see @xcite . it should be noted that many objects are unresolved in the direct images ( which have a resolution of @xmath28 ) ; it is not necessary for an object to be classified morphologically as a galaxy for inclusion in the elg list . the only criterion for inclusion is the presence of an emission line in the objective - prism spectrum . in other words , there is no bias against compact , stellar appearing objects in kiss . although the survey data themselves provide considerable information about the sources , obtaining follow - up spectra is an important part of the kiss project . these spectra provide higher quality redshifts , confirm elg candidates status as actual elgs , and provide the means by which to identify the type of activity powering the elg . in most cases , emission lines such as h@xmath7 , h@xmath29 , [ o iii ] @xmath304959,5007 , and [ n ii ] @xmath306548,6583 are present in the follow - up spectra . with the higher - accuracy line strengths of these spectra , one can distinguish between star - forming galaxies , seyfert 1 galaxies , seyfert 2 galaxies , liners , and qsos . furthermore , determination of the metallicity @xcite and the star - formation rate is possible with accurate line - strength measurements . follow - up spectra have been obtained for 83% of the kr1 elgs , 100% of the kb1 elgs , and 30% of the newer kr2 elgs at various telescopes including the hobby - eberly telescope @xcite , the michigan - dartmouth - mit 2.4 m telescope @xcite , the wisconsin - indiana - yale - noao 3.5 m , the 2.1 m on kitt peak , the apache point observatory 3.5 m , and the shane 3.0 m telescope at lick observatory @xcite . lcbgs typically are identified by their small half - light radii ( @xmath31 ) and/or their high surface brightnesses ( @xmath0 ) . therefore , measuring these quantities in the kiss data was an essential first step in our effort to isolate a sample of nearby lcbgs . we wrote an iraf procedure called kcompact to find @xmath31 for every object in the kiss survey images . the procedure performs a curve of growth analysis using circular apertures to determine @xmath31 . the formal uncertainty in @xmath31 is estimated by comparing the separate measurements of elgs present in two survey fields ( the overlap in adjacent kiss fields is approximately 3 - 5 arcminutes ) . based on 14 objects , we determine the uncertainty in @xmath31 to be 0.21 arcseconds . combining our estimates of @xmath31 with the total apparent magnitude measured during the kiss data processing ( see @xmath26 [ kiss_sum ] ) allows us to compute @xmath0 using the following formula : @xmath32 as the formal errors in the kiss apparent magnitudes are always below 0.1 magnitude , we can find the upper limit on the uncertainty of @xmath0 due to the errors in @xmath33 and @xmath31 . we find @xmath34@xmath35 @xmath11 0.14 mag arcsec@xmath2 . the reliability of @xmath31 and @xmath0 is inextricably linked to the resolution of the image on which they are measured . both the image scale and the effective seeing significantly affect the accuracy of these quantities . resolution limits restrict the precision with which these quantities can be determined in the more distant portion of kiss . since _ all _ objects in each survey image are analyzed by kcompact , we use the locus of the @xmath31 values for objects previously identified as stars to define a rough resolution limit for each image . marked interactively , the resolution limit represents not the mean fwhm of the stellar distribution but rather is selected to encompass the entire locus of stellar objects . as we will discuss in the following paragraph , variable image quality across our survey images causes this limit to underestimate the true image quality for a modest fraction of our elg candidates . for kb1 and kr1 , the mean resolution limit ( i.e. , the lower limit on @xmath31 ) of the images is 2.50 compared to an image scale of 2.03/pixel ; for kr2 , this mean resolution limit of the images is 2.13 compared to an image scale of 1.43/pixel . therefore , objects with @xmath31 under @xmath36 are almost always unresolved ( or , at best , marginally resolved ) in the kiss survey direct images . moreover , as the distance to an object increases , so does the physical diameter that corresponds to this resolution limit . that is to say , at greater distances , galaxies with progressively larger diameters dominate any resolved sample . figure [ fig : distsel ] shows this effect , plotting diameter in kpc vs. distance in mpc , where the diagonal lines represent the nearly constant resolution limits ( converted to kiloparsecs ) below which an object is unresolved . everything but galaxies with large diameters will be unresolved at the redshift limit of the survey . ultimately , this bias leads us to adopt a distance - limited sample of kiss lcbgs ( see section 2.3 ) . = 6.0 in -0.2 in the rough resolution limit discussed in the previous paragraph suffices for a determination of an appropriate distance - cut ( see below ) , but fails to account for the rather extreme image quality variability of the kiss direct images . the delivered image quality of the burrell - schmidt 0.61-m at the time kiss was carried out could vary between @xmath36 and @xmath37 over a single survey field . hence , rather than use a single value to describe the image quality on a given survey image , it is necessary to estimate the local value for the image quality in the vicinity of each kiss elg . an iraf script was written for this purpose . by locating the ten closest stars to each elg and measuring the fwhm of their profiles , this script accounts for the variable image quality of the kiss direct images . the local fwhm ( average seeing ) for each kiss elg is the mean of the fwhm from the ten stellar profiles . as our study relies upon accurate measurements of @xmath31 and @xmath0 , we were forced to exclude those objects for which the local image quality was severely degraded . a ratio between the mean local fwhm and the half - light diameter acted as our main diagnostic : if the ratio were above 1 , we could not trust the measurement of @xmath31 , and therefore we exclude the galaxy . in addition , all objects with the local stellar fwhm above 5 and a diagnostic ratio above 0.7 were also excluded . in total 96 galaxies were cut out of the survey area , decreasing the area over which we could search for lcbgs by only @xmath38 . with these exclusions , we feel confident in the completeness of our sample . that is , in the remaining portions of kiss we should not miss any apparently small , potential lcbg candidate simply because it is under - resolved . of course , the galaxies remaining in our sample of kiss elgs are not all well - resolved , but as we will describe , we can account for the inaccuracies in their measurements . = 5.0 in -0.2 in in essence , our measurements of marginally resolved galaxies act as upper limits on @xmath31 and lower limits on @xmath0 . we explore the nature of our unresolved and marginally resolved lcbg candidates using higher - resolution images from the wisconsin indiana yale noao ( wiyn ) @xmath39 m telescope on kitt peak . in may 2003 and march 2004 we observed 11 kiss elgs from kr1 and 12 kiss elgs from kr2 under non - photometric conditions with the s2 kb ccd , for which the image scale is 0.6 arcsec / pixel . the pre - existing kiss direct images for the galaxies , that were taken under good conditions , provided the photometric calibration for our 0.9-m images . despite the different image scales of kr1 and kr2 , we treat them the same in the following analysis due to the variable image quality described above . for the kiss compact , star - forming elgs in each survey strip , we re - measured the half - light radii on the 0.9-m images using the same curve - of - growth routine described above , and compared them with the values of @xmath31 obtained in the same way for the kiss images . the kiss @xmath31 is plotted against the ratio of the 0.9-m @xmath31 to the kiss @xmath31 in figure [ fig : hlradcorr]a to show the results of these re - measurements . the triangles represent the elgs from kr2 , while the circles represent the elgs from kr1 . the relationship is equivalent for both strips : those objects with larger @xmath31 have a ratio close to unity , while those objects with @xmath31 near our resolution limits have ratios that deviate more considerably from unity . we sought to quantify the correlation seen in figure [ fig : hlradcorr]a in order to develop a correction of @xmath31 for all unresolved and marginally resolved elgs in the kiss database . analyzing our sets of measurements for the 0.9-m images and the kiss images , we can predict the way in that the difference between a galaxy s 0.9-m @xmath31 and its kiss @xmath31 ( @xmath40@xmath31 ) increases as its @xmath31 decreases below the resolution limits of kiss . instead of re - observing all unresolved and marginally resolved kiss elgs to achieve more accurate measurements of their @xmath31 , we can infer the values we would obtain . two of the eleven kr1 and two of the twelve kr2 elgs observed with the wiyn telescope had @xmath31 greater than 6.0 arcseconds , and were thus very well - resolved in the kiss direct images . figure [ fig : hlradcorr]a reassures us that our measurements of @xmath31 on the 0.9-m images are very close to our measurements on the kiss images for these four well - resolved galaxies . these four galaxies were excluded from the determination of our correction . upon closer examination of the 0.9-m images , one galaxy in kr1 was observed to have a merging partner , making suspect any measurement of its @xmath31 . an additional two galaxies from each survey strip were excluded as well because they suffered from noticeably poorer image quality in the kiss direct images , skewing the relationships shown in the following plots . figure [ fig : hlradcorr]b shows @xmath40@xmath31 versus the kiss @xmath31 for the 14 remaining galaxies . in this figure , we notice the same effect as in figure [ fig : hlradcorr]a : apparently smaller galaxies have greater @xmath40@xmath31 . we based the final correction for @xmath31 upon the linear - least squares fit of the 14 data points plotted in figure [ fig : hlradcorr]b . the fit is shown by the dashed line . the equation for this line is : @xmath40@xmath31 @xmath41 0.22(kiss @xmath31 ) @xmath25 1.19 , where @xmath40@xmath31 is in arcseconds , and the scatter about the fit is 0.11 arcseconds . = 5.0 in -0.2 in we also compared the apparent b - band magnitudes ( @xmath33 ) measured on the 0.9-m and kiss images . since our photometric calibration of the 0.9-m images made use of the original kiss calibration , our 0.9-m values for @xmath33 agree well with the kiss values for @xmath33 . this re - measurement of @xmath33 combined with the correction for @xmath31 , described above , is relevant for probing the effects of low - resolution upon the measurement of @xmath0 ( see equation 1 ) . figure [ fig : sbcorr ] plots the observed relation between @xmath40@xmath0 ( 0.9-m @xmath0 @xmath25 kiss @xmath0 ) and @xmath40@xmath31 . since the kiss and 0.9-m apparent magnitudes show good agreement , the difference in the measured values of @xmath0 must be due to solely the change in @xmath31 between the kiss and 0.9-m images . therefore we can compute the correction for the effective surface brightness , @xmath40@xmath0 , such that it depends on only @xmath40@xmath31 . @xmath42 figure [ fig : sbcorr ] plots the observed values of @xmath40@xmath0 and @xmath40@xmath31 for the 14 kiss elgs along with a curve illustrating the analytical correction given by equation 2 . the observed values of @xmath40r@xmath43 and @xmath40@xmath0 are shown by the filled triangles ( kr2 ) and the filled circles ( kr1 ) . in the equation , @xmath40@xmath31 corresponds to the linear - least - squares fit shown in figure [ fig : hlradcorr]b for the @xmath31 correction . figure [ fig : sbcorr ] shows that the actual measurements of @xmath40r@xmath43 and @xmath40@xmath0 follow the correction defined by equation 2 very well . we apply these corrections for @xmath31 and @xmath0 to all non - excluded ( see above ) kiss star - forming galaxies with an @xmath31 less than 5 arcseconds . among all kr1 star - forming elgs with @xmath31 under 5 arcseconds ( i.e. , those affected by the correction ) , the mean change in @xmath31 is -0.50 arcseconds , the mean change in diameter is -4.91 kpc , and the mean change in @xmath0 is -0.44 mag arcsec@xmath2 . among all kr2 star - forming elgs with @xmath31 under 5 arcseconds , the mean change in @xmath31 is -0.56 arcseconds , the mean change in diameter is -5.31 kpc , and the mean change in @xmath0 is -0.57 mag arcsec@xmath2 . the ability of our local lcbg candidates to provide meaningful mass and sfr comparisons with the higher - z lcbgs requires that these galaxies be genuine lcbgs . typical methods of identifying nearby lcbg analogues usually employ cut - offs in surface brightness , luminosity , and color , and can not guarantee that the galaxies are true examples of lcbgs . indeed , we only can pick out those galaxies that are _ likely _ to be lcbgs . considering the nature of kiss , we believe that we are identifying those elgs that possess the same properties as the lcbgs at intermediate and high redshifts , and thus , those galaxies that are most likely to be true lcbgs . hence , we call our galaxies lcbg candidates . in an attempt to match our kiss lcbg sample with the higher - redshift samples , we endeavor to employ the same cut - offs as those used in earlier work . the sample is based on criteria proposed by @xcite , who uses cut - offs in surface brightness , rest - frame color , and absolute magnitude for the selection of lcbgs . all criteria are limited by the angular resolution of the kiss survey direct images . certainly , lcbgs are _ not _ agn , and are most often characterized by vigorous star - formation . therefore the sample was selected from a subsample of kiss galaxies known from follow - up spectra to have star - formation as their primary form of activity . all values given for @xmath44 color and all magnitudes used in the kiss sample selection are corrected for galactic absorption using the values of @xcite . additionally , to select this sample , we used the corrected values of @xmath0 determined from the analysis outlined in the previous section . a total of 17 kiss star - forming galaxies meet the selection criteria proposed by r. guzmn : @xmath0 ( _ b_-band ) @xmath1 mag arcsec@xmath2 , @xmath3 , @xmath6 with @xmath17 km s@xmath5 mpc@xmath5 . we use this absolute magnitude limit for the selection of kiss lcbgs despite our use of @xmath4 km s@xmath5 mpc@xmath5 by kiss . since this absolute magnitude cut corresponds to @xmath45 with @xmath4 km s@xmath5 mpc@xmath5 , we apply a slightly more stringent cut for absolute magnitude by 0.15 magnitudes when we use @xmath3 as one of our selection criteria . this sample of lcbgs is partially determined by the resolution selection effects described in section [ kcomp ] and shown in figure [ fig : distsel ] . these effects dictate that the majority of identified lcbg candidates will be located at distances far closer than the kiss redshift limit of @xmath46 . considering that the median half - light diameter of the 17 lcbgs is 4.21 kpc , figure [ fig : distsel ] indicates that likelihood of finding an lcbg beyond @xmath47 mpc is low . a further implication of this effect is that any comparison of the kiss lcbg sample with other kiss star - forming galaxies will be skewed . kiss star - forming galaxies detected at greater distances will be , on average , considerably larger in diameter and more luminous . to account for these effects , and to improve the completeness of our lcbg sample , we impose on it a distance limit of 185 mpc ( @xmath48 ) . 16 of the 17 kiss lcbgs lie within this boundary , and according to the resolution thresholds marked on figure [ fig : distsel ] , any galaxy within 185 mpc of the milky way with a diameter equal to or greater than @xmath49 kpc will be resolved in the kiss images . figure [ fig : properties ] shows kiss elgs that lie within 185 mpc of the milky way and have @xmath50 . the 16 kiss elgs that remain in the sample after the application of the distance cut - off are marked as filled circles in the plot , and the dashed lines indicate two of the selection criteria . the stars in the plot represent a subset of `` normal '' galaxies from the century survey ( cs ) carried out by @xcite that lie within the distance limit and have @xmath51 . the century survey is a traditional magnitude - limited survey of `` normal '' field galaxies that are used here to represent the general galaxian population . the open circles represent a comparison sample of kiss star - forming elgs that have @xmath3 and lie within 185 mpc of the milky way . henceforth , when we refer to our samples of lcbg candidates and comparison samples , we refer to the _ distance - limited _ samples . it should be noted that all kiss elgs selected as lcbg candidates have been identified as having star - formation as their primary form of activity from follow - up spectra . = 5.0 in -0.2 in = 5.0 in -0.2 in on the plot of @xmath0 versus color in figure [ fig : properties ] , two features stand out . while the kiss lcbg candidates were selected to have surface brightnesses higher than the typical star - forming galaxy ( sfg ) , even the highest surface brightness lcbg candidates do not appear to be clearly separated from the distribution of kiss sfgs . instead , the star - forming galaxies form a more - nearly continuous distribution in color and @xmath0 , as one might expect . at intermediate redshifts , @xcite find that in the parameter space of @xmath0 and b@xmath25v color , lcbgs separate themselves more cleanly than in any other parameter space from ordinary irregular , elliptical , and spiral galaxies . a plot from @xcite is shown in figure [ fig : anna ] , illustrating this separation . when we compare lcbgs with the other kiss star - forming galaxies , we do not see this effect . in the @xmath9 local universe , the bluest , highest surface brightness star - forming elgs do not appear to be separate from a distribution of otherwise regular star - forming galaxies , but rather represent the most extreme galaxies in this class of star - forming elgs . however , the kiss lcbg candidates are more clearly separate from the cs galaxies . interesting to note is that while only three of the cs galaxies meet the selection criteria for an lcbg , 29 cs galaxies redder than the color cut meet both the surface brightness and absolute magnitude cut . we believe these 29 galaxies to be local , normal elliptical galaxies . only 6 kiss elgs fall in this region of the diagram ( i.e. , compact but red ) . the properties of the 16 elgs selected as local lcbg candidates are listed in table 1 . columns 1 and 2 give the survey identification numbers for each galaxy ( kissr and kissb numbers ) . the apparent b - band magnitude is listed in column 3 , the _ b - v _ color in column 4 , the half - light diameter in kiloparsecs in column 5 , the b - band effective surface brightness ( @xmath0 ) in column 6 , the distance to the galaxy ( in mpc ) in column 7 , the absolute b - band magnitude in column 8 , the logarithm of the h@xmath7 luminosity in column 9 , and the logarithm of the 1.4 ghz radio power from @xcite in column 10 . those galaxies without a listed radio power were not detected as radio emitters in the van duyne _ etal . _ study . images of the 16 lcbg candidates made from the kiss direct images appear in figure [ fig : finders ] . although the resolution of kiss is too low for measurements of parameters such as asymmetry and concentration , we can see some of the morphological characteristics of our lcbg candidates on these images . at least three of the lcbg candidates , kissr 147 , 1274 , and 1870 , appear to have disturbed morphologies , evidence for recent interactions and/or mergers . however , the majority of the 16 lcbg candidates appear to be small , compact , and quite symmetric , suggesting that interactions are not the main driver of the observed star - formation activity in most cases . = 7.0 in -1.5 in the kiss project obtains photometrically calibrated b and v direct images along with the objective - prism spectra , that provide accurate measurements of each galaxy s h@xmath7 flux . in addition , all of our lcbg candidates have been observed as part of our follow - up spectroscopy program ( e.g. , wegner _ etal . _ 2003 ; gronwall _ etal . _ hence , there is a large amount of information readily available for all kiss elgs that allows us to examine the properties of the kiss lcbg candidates in some detail . in this subsection , we present the properties of the lcbg candidates and a comparison sample of kiss star - forming galaxies ( sfgs ) with [email protected] . when the necessary data are available , we also present the cs comparison sample of `` normal '' galaxies that have [email protected] . all three of these samples are taken from the same volume of space ( same area of the sky , and d @xmath11 185 mpc ) . one implication of this distance limit is that the galaxies in our kiss comparison sample have lower luminosities and bluer colors than the overall sample of star - forming kiss galaxies . in general , the most distant galaxies in any sample tend to be the most luminous , and galaxies that are more luminous tend to be redder . the b - band absolute magnitudes of the kiss lcbgs , kiss star - forming galaxies , and cs magnitude - limited sample are plotted against sb@xmath53 in figure [ fig : sbmag ] . together with figure [ fig : properties ] , these two plots exhibit the selection criteria of the lcbg candidates , and also illustrate nicely the properties of the three galaxy samples . as pointed out in the previous section , the kiss sfg comparison sample exhibits a fairly continuous distribution in all three parameters sb@xmath53 , m@xmath54 , and b@xmath25v color with the lcbg candidates representing the extreme cases in terms of color and surface brightness ( figure [ fig : properties ] ) . the cs and kiss sfg samples overlap substantially in both figures [ fig : properties ] and [ fig : sbmag ] , although the latter are on average much bluer . the lcbg candidates are seen to be distributed fairly evenly in absolute magnitude between @xmath2518.6 to @xmath2520.6 . they define the upper envelope of the kiss elgs in figure [ fig : sbmag ] . = 5.0 in -0.2 in figure [ fig : histdiam ] shows histograms of the half - light diameters of the lcbg candidates , the kiss comparison sample sfgs , and the cs comparison sample `` normal '' galaxies . the median diameters for the three samples , 4.11 kpc , 6.63 kpc , and 6.43 kpc , respectively , are indicated in the figure . as expected , the kiss lcbgs are , on average , 60% smaller than the cs galaxies and sfgs . the largest lcbgs have sizes comparable to the medians for the other two distributions . since the samples have comparable luminosities , the difference in diameter accounts for most of the difference in sb@xmath53 . = 5.0 in -0.2 in figure [ fig : hamag ] shows the h@xmath7 luminosity plotted against m@xmath54 for the two kiss samples . the sample of 16 kiss lcbg candidates exhibits a higher median h@xmath7 luminosity by a factor of 1.6 than the comparison sample sfgs . because we consider only actively star - forming galaxies in both samples , we can draw an important conclusion from this result . in general , the h@xmath7 luminosity indicates the activity level of a galaxy , telling us how many ionizing stars a galaxy contains . the kiss lcbg candidates have , on average , 60% higher current star - formation rates than the less compact sfgs . there is a clear tendency for the lcbg candidates to be located near the top of the diagram , again illustrating that they are above average relative to the comparison sample . note that not all of the lcbg candidates are in the high star - formation rate category . one of the kiss elgs with the _ lowest _ h@xmath7 luminosity , kissr 147 , happens to be included among the lcbgs . we note in passing that the measurement of the h@xmath7 flux in this object may well be underestimated , since it is so extended on the sky and since the star - formation appears to be spread over a large fraction of the galaxy . the kiss objective - prism flux measurement is made over an 8 arcsec wide region centered on the galaxy . hence , our current estimate for the h@xmath7 luminosity of kissr 147 may be significantly lower than the actual value . we do not believe that this problem occurs for any of the other kiss lcbgs . = 5.0 in -0.2 in = 5.0 in -0.2 in it is also relevant to consider the spectral properties of the lcbg candidates . do their spectra reveal any systematic differences relative to the less compact sfgs of the kiss sample ? we try to explore this question in two ways . first , in figure [ fig : diagplot ] , we plot an emission - line ratio diagnostic diagram showing the locations of all of the kiss sfgs ( regardless of distance ) . we plot the logarithms of the [ ] /h@xmath29 ratio vs. the [ ] /h@xmath7 ratio . the smaller open circles are the regular " star - forming galaxies , while the larger filled symbols indicate the location of the 15 lcbg candidates with measured line ratios . the diagram reveals the characteristic sequence , where the sfgs fall along an arc in the diagram stretching from the upper left ( typically lower luminosity , lower metallicity galaxies ) to the lower right ( typically more luminous , higher metallicity objects ) . the lcbgs are distributed fairly uniformly along the lower 2/3rds of the arc . our selection against lower luminosity galaxies prevents them from falling in the upper left portion of the sequence . we detect no obvious trends amongst the lcbgs that would lead us to believe that their spectral properties are systematically different from the rest of the kiss sfgs . we also consider the metal abundances ( log(o / h ) + 12 ) of these 15 lcbg candidates compared to the sample of all kiss sfgs within 185 mpc of the milky way . using the linear luminosity - metallicity ( l - z ) relation of @xcite , we plot metal abundance versus absolute magnitude for the kiss sfgs . figure [ fig : abun ] shows a weak tendency for the lcbgs , shown as filled circles on the plot , to lie at or below the mean l - z relation . the mean abundance of the 15 lcbgs is 8.65 , compared to the mean abundance of the comparison sample ( kiss sfgs with d @xmath11 185 mpc and [email protected] ) of 8.75 . the statistical significance of this 0.1 dex difference is minimal , since the formal error in the difference of the two means is 0.08 . still , this small difference may suggest that either the lcbgs are under - producing heavy elements compared to the comparison sfgs , are systematically losing a larger fraction of their metals via outflows , or the enhanced star - formation of the lcbgs is producing a modest luminosity enhancement relative to the rest of the sfgs . while it is not possible to definitively select between these options , the high current star - formation rates exhibited by the lcbgs would appear to be inconsistent with the first of these choices , while the higher central mass densities implied by the high sb@xmath55 of the lcbgs is inconsistent with the second ( see discussion in @xcite ) . hence , we conclude that the lcbg candidates appear to be offset slightly to the left of the kiss sfgs , indicating that their luminosities may not be representative of their masses . while this latter point is likely true in general , for all star - forming galaxies ( e.g. , lee _ etal . _ 2004 ) , it appears to be even more the case for the lcbg sample . = 5.0 in -0.2 in to some extent , all galaxies exhibit radio emission , but it is those that are the most radio - luminous that are known historically as `` radio - galaxies . '' because star - forming galaxies are often radio sources , we investigate the radio properties of the lcbg candidates as compared to a sample of all kiss sfgs , expecting to detect radio emission for some subset of our samples . all radio fluxes cited in this work come from @xcite , who used the first @xcite and nvss @xcite 20-cm radio continuum surveys to study the radio emission associated with the 2158 kiss elgs from kr1 and kr2 . because that study did not include the kb1 portion of kiss , we exclude the kb1 elgs from this part of our analysis . 13.7% of all kiss elgs known to be star - forming galaxies in the kr1 and kr2 survey strips are detected by first and nvss in the radio . this is an upper limit to the fraction of radio - detections , however , since only 57.6% of the kr1 and kr2 galaxies currently have follow - up spectra ( note that _ all _ of the radio - detected kiss galaxies possess follow - up spectra ) . if we assume that @xmath1290% of the galaxies lacking follow - up spectra are also sfgs , then the fraction of sfgs with radio continuum detections drops to only 7.4% . this latter value is likely to be closer to the actual percentage of radio - detected sfgs . we compare this number to the fraction of radio galaxies within our lcbg candidate sample ( from only kr1 and kr2 ) and to the fraction of radio galaxies within our distance - limited comparison sample of kr1 and kr2 sfgs with [email protected] . fully 53% of the 15 lcbg candidates have detectable radio continuum emission , while 28% of the comparison sample galaxies have detected radio fluxes . the lcbgs clearly have a stronger propensity to possess detectable radio emission . we hypothesize that the intense star - forming activity indicated by the high @xmath0 of the lcbg candidates is the primary contributor to this effect . in star - forming galaxies , the energy given to the relativistic electrons which cause synchrotron radiation comes from the supernova explosions occurring over the past @xmath123@xmath5710@xmath58 years @xcite . presumably , the kiss lcbgs produce more supernova in this time period then regular star - forming galaxies due to their above - average star formation rates ( previous section ) . additionally , due to their compact nature , lcbgs may have stronger magnetic fields than regular star - forming galaxies . these properties , in concert , enhance the detectability of synchrotron radiation in lcbgs relative to other star - forming galaxies . despite the high proportion of radio detections among the lcbgs , their radio powers do not stand out . for the eight detected lcbgs , the mean radio power is 10@xmath59 w hz@xmath5 . this is statistically indistinguishable from the mean of the distance - limited comparison sample ( 10@xmath60 w hz@xmath5 ) , and significantly less that the average of all kiss star - forming galaxies detected by first and nvss ( 10@xmath61 w hz@xmath5 ) . the majority of the overall radio - detected sfgs lie at distances beyond 185 mpc , so it is no surprise that their radio luminosities are higher . however , it is perhaps a bit surprising that the lcbgs do not have higher radio powers than the comparison sfgs . of course , the number of lcbgs in this analysis is small , so it may be premature to draw strong inferences regarding the radio powers at this time . what is clear , however , is that the lcbgs are more commonly detected as radio emitters than are the overall population of star - forming galaxies in the same volume . the space density of the intermediate and high redshift lcbgs compared with that of low redshift lcbgs acts as an essential constraint on their density evolution . nearly all previous studies of lcbgs stress the apparent extreme density evolution of lcbgs . for our sample of lcbg candidates with z @xmath11 0.045 we compute the volume density using the 1/@xmath62 method . a complete description of the method along with the specific completeness parameters used in the calculations can be found in @xcite . in brief , we first determine whether the detectable volume of an individual galaxy is limited by the effective volume set by the kiss redshift limit or by its own emission - line flux . then , we sum the inverse of the maximum volume within which each galaxy is detectable for all galaxies in this sample . accordingly , we determine the space density of local lcbgs to be @xmath10 . in order to arrive at an estimate of the uncertainty in the volume density , we will assume that the main source of error is the poissonian uncertainty in the number of lcbgs ( i.e. , @xmath34 = @xmath63 ) . this is reasonable , since the uncertainties in the line flux completeness limits and the relevant volumes are fairly small . hence , since there are 16 lcbgs in the distance - limited sample , the computed volume density carries with it a 25% uncertainty . a handful of previous studies have derived estimates for the space densities of lcbgs at intermediate and high redshifts . for example , @xcite find the volume density of lcbgs at redshifts between 0.4 and 0.7 with m@xmath64 -18.5 to be @xmath65 , and that of lcbgs at reshifts between 0.7 and 1.0 with m@xmath64 -18.5 to be @xmath66 . @xcite estimate a total space density of the high redshift ( @xmath67 ) population of lcbgs to be @xmath68 . this latter density estimate must be taken as a lower limit , since the lowenthal _ etal . _ sample is at such high redshifts that only the most luminous sources are detectable . a correction to account for the incompleteness at the lower luminosities would likely raise the density estimate substantially . a large ( and probably rather uncertain ) incompleteness correction to account for missing lower luminosity lcbgs was applied by phillips _ _ to their higher redshift subsample . because of the large uncertainties associated with the lcbg density estimates in these higher redshift samples , one should use caution in interpreting the numbers . we estimate that the uncertainties in these densities , including any errors associated with the incompleteness corrections , must be at least 33% for the lower redshift value of phillips et al . , and closer to 50% for the two higher redshift estimates . the uncertainty that arises from different absolute magnitude cut - offs at different redshifts can be alleviated somewhat by applying a * brighter * cut - off to our sample . by making our local sample s luminosity limit consistent with the absolute magnitude limit of the higher - z samples , rather than extrapolating the high - z absolute magnitudes to lower luminosities ( as was done by @xcite as described above ) , one avoids introducing some large and very uncertain corrections . for example , @xcite had an absolute magnitude limit of -20 ( using h@xmath69 km s@xmath5 mpc@xmath5 ) for their sample at @xmath70 , which corresponds to an absolute magnitude limit of @xmath71 for h@xmath72 km s@xmath5 mpc@xmath5 . we derive a volume density of @xmath73 for our sample of local lcbgs with when we adopt [email protected] as our luminosity limit . for comparison , @xcite find the volume density of lcbgs at reshifts between 0.7 and 1.0 with m@xmath64 -19.12 to be @xmath75 . we note that this method is impractical to use for the lowenthal _ etal . _ sample , and unnecessary when comparing to the lower - redshift phillips _ etal . _ sample . with the above caveats in mind , we compare the volume densities of lcbgs from our local ( z = 0 ) sample to those of the higher redshift studies . the value @xcite obtain for their lower redshift sample implies that the volume density of lcbgs has declined by a factor of @xmath124 between z @xmath12 0.5 and the present , while their higher redshift value suggests that the density of lcbgs has dropped by a factor of @xmath1216 between z @xmath12 0.85 and today . this latter factor becomes @xmath1210 for the comparison using the volume density of the kiss lcbgs with m@xmath64 -19.12 . the density drop implied by @xcite for the high reshift lcbgs ( @xmath67 ) is at least a factor of @xmath127 , and likely to be much higher . the work of @xcite and @xcite all suggest that the volume density of lcbgs drops off by a factor of @xmath1210 from @xmath18 to the present . this prediction is in general agreement with the results presented here , given the large errors associated with the density estimates . that is , the space density derived for our sample of lcbgs compared with that of the higher redshift samples supports the claim that these galaxies are indeed a strongly evolving population . however , the uncertainties associated with the volume density of lcbgs at higher redshifts precludes an accurate assessment of exactly how extreme their evolution is . hopefully , future studies of high redshift lcbg populations will constrain their volume densities more definitively , and lead to a better understanding of their evolution with look - back time . this paper deals with the problem of lcbgs by approaching only one of its facets . namely , we have sought to identify local examples of lcbgs in an effort to constrain the properties of those lcbgs present in such profusion at intermediate- and high-@xmath8 . however , we must be cautious in accepting that our sample is composed of genuine lcbgs @xmath25 indeed , we call the galaxies in our sample lcbg candidates . the classification scheme for lcbgs presented by jangren _ defines a six - dimensional parameter space in which lcbgs are effectively isolated from other types of galaxies . however , we could not apply this method for our lcbg selection , as kiss direct images did not have the resolution required to perform the prescribed tests . and , while jangren _ etal . _ conclude that the parameter space of b@xmath25v color vs. @xmath0 is the best at separating the lcbgs from other types of galaxies , they acknowledge that there is still no single cut that will eliminate all non - lcbgs nor include all _ bona fide _ lcbgs . specifically , employing a color limit to select local lcbgs based on the corrected rest - frame colors of higher - redshift lcbgs is a bit problematic . the many corrections that must be applied to color , ( i.e. , k - corrections , galactic reddening , internal extinction , emission - line contamination ) make it a difficult parameter to use for defining a sample of galaxies . additionally , the rest - frame colors of star - forming galaxies at high redshift will be inherently biased to be bluer than those of lower redshift star - forming galaxies . this is because host galaxies at higher redshift have not had as much time to assemble as those in the nearby universe . hence , they will be intrinsically less luminous , and a star - formation event at high-@xmath8 will tend to dominate the galaxian color more than at lower redshifts . accounting for this effect , however , would be a difficult task , mainly because of the uncertainty in our understanding of galaxy evolution . the necessity of a color parameter , however , is manifest in figure [ fig : properties ] . luminous elliptical galaxies from the century survey meet both the luminosity and surface brightness criteria for lcbgs . however , lcbgs are not similar to local elliptical galaxies . a color cut acts as the means for excluding early - type galaxies in traditional , magnitude - limited surveys . by contrast , in emission - line galaxy surveys such as kiss that contain few , if any , elliptical galaxies , the color criterion becomes less pertinent . limiting the samples of lcbgs to actively star - forming galaxies should ensure their relative blueness . while not extremely blue compared to other star - forming galaxies , the kiss lcbg candidates are considerably bluer than `` normal '' galaxies . the century survey `` normal '' field galaxies within the distance limit of 185 mpc have a median b@xmath25v color of 0.77 , considerably higher than the median color of the lcbg sample , 0.44 . were we to have selected a sample of lcbgs using only the luminosity and surface - brightness criteria , and limiting our sample to only actively star - forming galaxies , our sample would have contained 22 kiss lcbg candidates with a median b@xmath25v color of 0.53 , still considerably bluer than the cs galaxies . we select our local samples of lcbgs from only actively star - forming galaxies not only as a way to ensure their relative blueness , but also to further constrain their properties to match those of the higher - redshift samples . indeed , all known lcbgs at high-@xmath8 are vigorously forming stars , as evidenced by their strong emission lines . this property gives rise to important questions regarding the nature of lcbgs . is active star - formation always present in galaxies meeting the lcbg selection criteria ? can an lcbg be an lcbg without its starburst ? what do lcbgs look like when and if the star - formation fades ? do the lcbg criteria allow for non - active galaxies ? in an attempt to address these questions , we looked at the galaxies located in the kiss database that are @xmath76 elgs , but rather galaxies detected by the century survey that have measured redshifts , colors , and absolute magnitudes . since the first two survey lists of kiss ( kr1 and kb1 ) overlap parts of the cs , we possess direct images taken through b and v filters of essentially all of the cs galaxies . by combining our imaging photometry with the cs redshifts @xcite , we are able to carry out the same analysis on the cs galaxies as we previously did for the kiss elgs . applying the same cut - offs in @xmath0 , @xmath23 , and b@xmath25v color to these galaxies yields a sample of seven cs lcbg candidates . of these , only three were not also detected by kiss . all three lie outside the area of sky covered by kr1 . one of these galaxies has been identified using the nasa extragalactic database as a nearby seyfert 2 with very weak emission lines . kiss could not have detected this galaxy due to its strong continuum and the weakness of its h@xmath7 line . the others were initially detected by the case low - dispersion northern sky survey @xcite , with published follow - up spectroscopy in @xcite . one galaxy , cg 60 , is a star - forming galaxy with strong h@xmath7 and weak . the other , cg 90 , appears to be a weak - lined star - forming galaxy , possibly a post - starburst system . this latter galaxy appears to be the closest thing to a `` normal '' ( i.e. , not actively star - forming ) galaxy in the @xmath12100 square degrees of the century survey that meets the lcbg selection criteria . however , even cg 90 has _ some _ current star - formation . the study by @xcite which used the fornax spectroscopic survey to find a sample of bright , compact galaxies that are unresolved on photographic plates obtains a similar result . of their 13 local ( @xmath77 ) galaxies that resemble the higher-@xmath8 lcbgs , they find only four that do not have strong emission lines . one of the galaxies has a post - starburst spectrum and another has a spectrum indicating an old population . this paucity of non - active lcbgs suggests that they are , in general , extremely rare , and that perhaps the star - formation that seems always to be present in lcbgs is a key factor to getting labeled as an lcbg in the first place . clearly the star - formation activity helps to make the lcbg . the starburst will both increase the central luminosity of the galaxy , and lower the half - light radius r@xmath43 . these two effects combine to increase sb@xmath55 and lower the color . is a central starburst _ required _ for a galaxy to become an lcbg ? perhaps not , but it certainly appears to be strongly favored . while it is possibly premature to attempt to resolve the questions regarding the evolution of lcbgs discussed in 1 , we can at least use the properties of the local candidates to speculate a bit . the distribution of the kiss elgs in figure [ fig : properties ] suggests that they exhibit a continuum of physical characteristics in the surface brightness , color , and luminosity parameter space . galaxies classified as lcbgs using the jangren _ _ selection criteria represent those galaxies that currently lie at the extreme end of that continuum . clearly , however , there are plenty of kiss elgs that are located just below the surface brightness threshold that distinguishes our lcbg candidates from the remainder of the star - forming galaxy population . since the kiss lcbgs do , on average , exhibit higher h@xmath7 luminosities , it is likely that at least some of them will evolve out of the lcbg region in figure [ fig : properties ] as their starbursts fade . will they fade substantially , as proposed by @xcite and @xcite , dropping in luminosity by 4 magnitudes or more , becoming substantially redder and more gas poor and eventually resembling dwarf elliptical galaxies ? we feel that it is unlikely that any of our lcbg candidates will evolve into early - type galaxies . the evolutionary scenario proposed by @xcite and @xcite requires that the gas present in the lcbg be removed , presumably by strong supernovae - driven outflows . hydrodynamic models ( e.g. , mac low & ferrara 1999 ) suggest that this complete blow - away of a galaxy s ism does not occur for any but the least massive systems . therefore , we do not expect any of our current sample of lcbgs to become gas - poor and fade into red ellipticals . of course , it remains possible that the higher redshift lcbgs represent a more heterogeneous sample of objects @xcite . if some of the high-@xmath8 lcbgs did lose their gas and evolve into dwarf elliptical galaxies by the present time , our selection method would not be sensitive to them . however , based on the models of mac low & ferrara , we would expect that the majority of lcgbs should retain their gas to the current epoch . this appears to be the case with our sample . the continuum of physical characteristics evident in figure [ fig : properties ] suggests that luminous galaxies possess a range in their central surface brightnesses / central mass densities . this has previously been observed in dwarf galaxies ( e.g. , papaderos _ etal . _ 1996 , salzer & norton 1999 ) , where the most actively star - forming dwarf galaxies ( a.k.a . blue compact dwarfs , or bcds ) are seen to lie at the extreme end of a continuum that stretches to very low central surface brightnesses at the other extreme . furthermore , the gas distributions in the bcds are seen to be much more centrally concentrated than those in more quiescent dwarf irregulars ( e.g. , van zee , skillman & salzer 1998 ) . the combination of high central mass densities and high central gas densities make bcds particularly efficient at forming stars . we assume that this is no accident : bcds are by their nature well equipped to be making stars . by analogy , one might argue that lcbgs are predisposed to make stars efficiently because they too represent the extreme high surface brightness , high central mass density end of a continuous distribution , albeit for more luminous galaxies . in this picture , lcbgs are preferentially in a star - forming phase for a large fraction of the time . when not undergoing a strong starburst , they will fade somewhat ( mostly in sb@xmath55 , but also slightly in color and luminosity ) and join the population of galaxies lying below the lcbg selection region in figure [ fig : properties ] . this would help to explain the lack of _ any _ non - star - forming lcbgs present among the cs galaxies . even when not strongly bursting , post lcbgs may exhibit a modest level of star formation , such as that seen in cg 90 . the question of where the large population of lcbgs seen at @xmath78 0.5 to 1.0 has gone is then resolved , at least in part , by the population of less actively star - forming galaxies seen in figure [ fig : properties ] . the large number of kiss elgs located below the lcbg selection region in this plot appear to be able to account for a large fraction of the missing " lcbgs . the drop in density of lcbgs may be directly linked to the fall in the star - formation rate density between @xmath18 to the present ( e.g. , madau _ etal . _ 1996 , 1998 ) . the galaxies still exist , they are simply in a somewhat less active state . in this scenario , lcbgs are not pathological outliers , but rather the most compact galaxies at the extreme end of a continuous distribution of gas - rich star - forming galaxies . it should be relatively simple to test the above hypothesis . a combination of high spatial resolution surface photometry in the optical and vla hi mapping in the radio could be used to evaluate the stellar- and gas - mass distributions of the local lcbg candidates . a direct comparison of these properties between the lcbgs and a sample of normal " galaxies would reveal whether our lcbgs are structurally different from the non - lcbgs . these same observations would be extremely helpful in establishing their dynamical masses as well . this latter parameter is essential for showing clearer linkage between the local samples of lcbgs with the intermediate- and high - redshift analogues . efforts are already underway in this regard . as a first step in determining gas and dynamical masses , we have obtained single - dish hi observations with the arecibo 305-m radio telescope for the sample of lcbgs accessible from arecibo , along with a control sample of normal " star - forming galaxies . with these data , we will derive hi masses and provide at least a rough estimate for the dynamical masses ( werk _ etal . _ 2005 ) for our lcbgs . we will also be able to ascertain which of our galaxies would be suitable for follow - up vla mapping . in addition , we have begun acquiring higher resolution optical images of our lcbgs , with the goal of being able to perform detailed surface photometry . in addition , these more detailed images will allow us to carry out a morphological analysis of our sample of local lcbgs , that will provide clues as to whether their intense starbursts are caused by mergers and/or interactions , or are simply the result of their intrinsically compact nature . finally emission line - widths measured from high - resolution spectra combined with near - infrared images and our hi mass estimates could settle the debate about the masses of lcbgs . in this paper , we present a sample of local analogues to the intermediate- and high-@xmath8 lcbgs . we selected our sample from only those kiss galaxies that are actively forming stars . due to resolution effects , we limit our sample to only those galaxies within 185 mpc of the milky way . the sample of lcbgs is defined by the following criteria : a surface brightness limit of @xmath79 mag arcsec@xmath2 , an absolute magnitude limit of @xmath80 , and a color cut - off of @xmath6 with @xmath4 km s@xmath5 mpc@xmath5 . a total of 16 kiss sfgs meet these criteria . we have compared some of the properties of the kiss lcbg candidates to those of a comparison sample of kiss sfgs , and a comparison sample of century survey ( cs ) galaxies that are supposed to represent `` normal galaxies . '' on average , the kiss lcbg candidates have a higher tendency to emit detectable radio continuum flux , have higher h@xmath7 luminosities indicating strong star - formation properties , have slightly lower than expected metal abundances for kiss galaxies of their luminosity , and considerably smaller size than those galaxies in both kiss and cs comparison samples . we have calculated the volume density for the sample of local lcbg candidates to be @xmath10 . this value indicates a substantial density evolution of lcbgs , which is in general agreement with the predictions made by @xcite and @xcite . however , precise estimates of the density evolution of the lcbgs are difficult , due to the large uncertainties in the values of the volume density for higher - redshift populations . we have utilized the cs database in an attempt to select a sample of local non - active lcbgs in order to constrain the density of non - star - forming lcbgs . we found no additional non - star - forming lcbgs , although one weakly star - forming ( possibly post - starburst ) galaxy was recognized . we postulate that lcbgs are structurally predisposed for star - formation , and that their compact nature may be indicative of a high central gas concentration that drives ongoing star - formation . the need for a local sample of lcbgs is well documented . much of the published work on these galaxies speaks to the need for a representative local comparison sample that will help settle the debates over their size and mass , morphology , and low- redshift counterparts . we conclude that our sample of local lcbg candidates contains galaxies that are likely to be local versions of the more distant lcbgs . this local sample provides the starting point in our ongoing study of the properties and evolution of lcbgs . funding for this work was provided by an nsf presidential faculty award to jjs ( nsf - ast-9553020 ) . additional support for our ongoing follow - up spectroscopy campaign came from continued funding from the nsf ( nsf - ast-0071114 ) and wesleyan university . we are grateful to the anonymous referee for several helpful suggestions that improved the presentation . we thank the numerous kiss team members who have participated in spectroscopic follow - up observations during the past several years , particularly caryl gronwall , drew phillips , gary wegner , jason melbourne , laura chomiuk , kerrie mckinstry , robin ciardullo , and vicki sarajedini . special thanks to rafael guzmn for providing us with information about selection criteria for lcbgs , and several useful discussions . finally , we wish to thank the support staffs of kitt peak national observatory , lick observatory , the hobby - eberly telescope , mdm observatory , and apache point observatory for their excellent assistance in obtaining both the survey data as well as the spectroscopic observations . 113 & & 15.36 & 0.40 & 1.53 & 18.52 & 93 & -19.48 & 40.46 & + 147 & & 12.44 & 0.55 & 6.53 & 20.68 & 33 & -20.18 & 39.59 & 22.19 + 218 & & 15.34 & 0.49 & 4.21 & 20.93 & 84 & -19.29 & 41.20 & 21.60 + 242 & 128 & 16.66 & 0.35 & 4.02 & 20.84 & 153 & -19.26 & 41.72 & 21.62 + & 169 & 16.56 & 0.51 & 3.39 & 20.86 & 122 & -18.87 & 41.14 & + 707 & 189 & 14.99 & 0.40 & 5.42 & 20.85 & 94 & -19.87 & 41.64 & 21.41 + 1094 & 218 & 15.24 & 0.41 & 6.08 & 20.60 & 134 & -20.40 & 41.91 & 21.86 + 1271 & & 17.33 & 0.50 & 2.82 & 20.74 & 153 & -18.60 & 40.40 & + 1274 & & 14.37 & 0.57 & 6.91 & 20.58 & 99 & -20.61 & 41.65 & 21.89 + 1400 & & 16.45 & 0.17 & 4.49 & 20.57 & 176 & -19.78 & 41.24 & + 1438 & & 16.52 & 0.33 & 3.01 & 20.81 & 109 & -18.67 & 40.66 & + 1578 & & 15.85 & 0.40 & 3.11 & 20.12 & 114 & -19.43 & 41.64 & 21.71 + 1637 & & 16.30 & 0.33 & 4.46 & 20.89 & 141 & -19.44 & 41.14 & + 1664 & & 15.33 & 0.60 & 5.96 & 20.62 & 136 & -20.34 & 41.66 & 22.22 + 1870 & & 15.66 & 0.59 & 2.99 & 20.71 & 76 & -18.75 & 41.09 & + 2079 & & 16.80 & 0.52 & 3.17 & 20.84 & 129 & -18.75 & 41.20 & +

in theories in which different regions of the universe can have different values of the the physical parameters , we would naturally find ourselves in a region which has parameters favorable for life . we explore the range of anthropically allowed values of the mass parameter in the higgs potential , . for , the requirement that complex elements be formed suggests that the higgs vacuum expectation value must have a magnitude less than 5 times its observed value , for , baryon stability requires that , the planck mass . smaller values of may or may not be allowed depending on issues of element synthesis and stellar evolution . we conclude that the observed value of is reasonably typical of the anthropically allowed range , and that anthropic arguments provide a plausible explanation for the closeness of the qcd scale and the weak scale . # 1#1 + 0.3 in * v. agrawal , s.m . barr , john f. donoghue and d. seckel * + a)_bartol research institute + university of delaware , newark , de 19716 + _ b)_department of physics and astronomy , + university of massachusetts , amherst , ma 01003 + _ 0.4 in

hep-ph9707380

some of the major puzzles of particle physics and cosmology concern parameters which are much smaller than their expected natural " size [ 1 ] . for example , the yukawa coupling constant of the electron , @xmath8 , is about @xmath9 . the qcd vacuum angle " , @xmath10 , is experimentally bounded to be less than about @xmath11 . perhaps the mother of all such small - number puzzles is the cosmological constant problem [ 2 ] . the cosmological constant , @xmath12 , in natural gravitational units ( _ i.e. _ in units of @xmath13 , where @xmath14 gev is the planck mass " ) is known to be less than about @xmath15 . all of these dimensionless ratios would naturally be expected to be of order unity in the absence of some dynamical mechanism or symmetry principle that determined them to be small . for some of these small numbers fairly simple conventional particle physics explanations are available . in particular , viable models exist for explaining the smallness of @xmath10 and @xmath8 . other small numbers are harder to explain conventionally . for example , although one can find ways ( involving supersymmetry ) to explain why the cosmological constant is as small as ( 1 tev)@xmath16 ( or @xmath17 in natural units ) , that is still @xmath18 orders of magnitude larger than the observational bound . in recent work , weinberg has addressed the question of whether the anthropic principle can explain the smallness of the cosmological constant [ 3 ] . roughly stated , the anthropic principle [ 4 ] says that the parameters of the universe that we observe are governed by the requirement that they must be able to support intelligent life , as otherwise we would not exist to observe our universe . if there is only one single universe " , with the same laws and parameters everywhere , this is far from a satisfactory physical explanation . however , it has been realized that some physical theories can support the existence of separated domains in the universe in which different parameters and even different gauge groups are applicable . for example , in chaotic inflation [ 5 ] different domains have different higgs vacuum expectation values , selecting different effective particle physics theories . such domains could be regarded as , effectively , different universes . this idea that multiple universes " can exist takes the copernican revolution to the ultimate limit even our universe may not be unique . in a multiple universe theory , the anthropic requirement that we live in a universe with viable parameters is as natural as is the good fortune that we happen to live on a planet that has a temperature ideal for life . discussions of the anthropic principle often end up dangerously close to being non - scientific . however , weinberg shows how calculations can be done based on mild forms of the the anthropic principle or perhaps more accurately the multiple universe hypothesis which allow one to assess whether they are able to explain the smallness of a parameter such as the cosmological constant [ 3 ] weinberg examines the requirement that the evolution of the universe be such that matter clumps into galaxies , and shows that this only occurs for a range of values of the cosmological constant . he then calculates the mean value of viable @xmath12 s . if the actual value of @xmath12 in our universe turns out to be very much smaller than this mean value , one would conclude that this form of the anthropic principle does _ not _ provide sufficient explanation for the magnitude of @xmath12 . unfortunately , the results are not yet conclusive , as neither the mean viable value nor the experimental value of @xmath12 is yet well determined . the greater part of this paper will be devoted to applying anthropic arguments to a single parameter of the standard model of particle physics , namely @xmath0 , the mass parameter of the higgs potential . like the cosmological constant , this parameter is many orders of magnitude smaller than its natural scale " and has so far no completely satisfactory conventional particle physics explanation . its smallness ( about @xmath19 in planck units ) is considered one of the major puzzles in particle physics , and is often called the `` fine tuning problem '' or ( in the context of grand unified theories ) the gauge hierarchy problem " [ 6 ] . we will assume that the standard model is the correct theory of particle interactions in the limit that effects from the planck scale or the unification scale are neglected . in the next several sections of the paper we will see what the consequences are of varying a single parameter , @xmath0 , of that model [ 7 ] . in particular we will ask whether for different ranges of @xmath0 complex elements , which are presumably required for the emergence of life , can ( 1 ) exist , and ( 2 ) be actually formed in significant quantities in the evolution of the universe . it is important that we do not attempt to modify the standard model by adding new physics at low energies . first , to do so would make the space of possibilities too large to make meaningful anthropic arguments . and , second , the need for new low - energy physics would be unclear if the magnitude of @xmath0 could be explained without it . the reasons for focusing particularly on @xmath0 will be explained more fully below , but it is also interesting to look at certain other parameters of the standard model from an anthropic perspective , and this we do in section 6 , where we examine the ratios of the quark masses . of all the parameters of the standard model , @xmath0 stands out in a number of ways . first , it is the only one which is dimensionful , and because of that it sets the scale for the masses of all the known elementary particles [ 8 ] . all the elementary particles of the standard model which have mass quarks , leptons , @xmath20 , and @xmath21 derive these masses from coupling to the expectation value of the higgs . this expectation value , which is denoted @xmath22 , is determined by the minimization of the effective potential of the higgs field , @xmath23 . since @xmath24 one has that @xmath25 . the observed value of @xmath2 is @xmath26 gev . we will call this value @xmath27 , and henceforth the subscript ` 0 ' will denote the value a parameter takes in our universe . the masses of the weak - interaction gauge bosons , @xmath20 and @xmath21 are given by @xmath2 times the gauge coupling constants , which are of order @xmath28 . the masses of the quarks and leptons are given by @xmath2 times yukawa couplings which range from about @xmath29 for the electron to about @xmath30 for the top quark . the mass of the higgs particle is @xmath31 , where @xmath32 is as yet unknown , but is roughly of order unity [ 9 ] . the second way in which @xmath0 stands out among the parameters of the standard model is its extreme smallness . as noted above , it is of order @xmath33 , or about @xmath19 in natural planckian units . this is to be compared to the next smallest parameters , the electron yukawa coupling , @xmath34 , and the qcd vacuum angle , @xmath35 . in the simplest grand unified models @xmath0 receives contributions of order @xmath36 gev@xmath37 from each of several terms , which must therefore cancel to fantastic accuracy . in such fine - tuned " models , a very small change in the other parameters would disturb this cancellation and cause @xmath0 to vary over an enormous range . ( this may be another justification for our approach here of only considering variations of @xmath0 and keeping the other parameters essentially fixed . ) finally , @xmath0 stands in contrast to the other parameters of the standard model in that fairly plausible explanations in terms of symmetry principles or other conventional particle physics considerations are available to account for their magnitudes . for example , it is generally regarded as likely that the relative values of the gauge couplings are the result of unification of the gauge groups at or below the planck scale [ 10 ] . likewise , there are many ideas for explaining the ratios of quark and lepton masses in terms of grand unification , family symmetries , horizontal interactions , radiative hierarchies , or a combination of these [ 11 ] . the smallness of the qcd vacuum angle can be explained by the axion mechanism [ 12 ] or by approximate cp invariance [ 13 ] . by comparison , the smallness of @xmath0 is very hard to account for in conventional ways . there have been two main approaches to doing this , technicolor " [ 14 ] and supersymmetry [ 15 ] . the technicolor approach is fraught with difficulties and has fallen into disfavor . the supersymmetry approach is more promising , but a completely satisfactory and simple explanation of the smallness of @xmath0 does not yet exist . it will be assumed , then , that @xmath0 can be of either sign and can vary between @xmath38 and @xmath39 . to understand the behavior of @xmath2 as @xmath0 is varied over this range , one must look at the potential for @xmath40 including the effect of its coupling to the quark - antiquark condensates ( which can ordinarily be neglected ) . @xmath41 the @xmath42 condensates for light quarks have a value of order @xmath43 , where @xmath44 is the strong - interaction chiral - symmetry - breaking scale [ 8 ] . ( @xmath45 mev . ) for @xmath0 negative and much larger in absolute value than @xmath46 , as in our universe , one can ignore the last term in eq.(1 ) and obtain @xmath47 . as @xmath0 becomes smaller in absolute value than @xmath46 , and of either sign , one can neglect the @xmath0 term and obtain @xmath48 . finally , when @xmath0 is positive and larger than @xmath46 one can neglect the quartic term and obtain @xmath49 . note that in the @xmath50 world , the longitudinal components of the weak interaction gauge bosons come from the pions " not from the higgs field , and @xmath51 . in universes with @xmath24 and greater in absolute value than in our universe , one has @xmath52 . one is dealing , then , with large @xmath2 " universes . we will examine the possibilities for life in this case , our basic assumption being that for life to exist complex chemistry " must be possible . in our universe , protons and neutrons combine to form a variety of nuclei . these are dressed with electrons to form atoms which may be bound into a variety of simple and complex molecules . it seems plausible , therefore , that in order for life to develop it is necessary that a variety of nuclei be ( a ) stable , and ( b ) formed in either primordial or stellar nucleosynthesis . one of the things that can go wrong when @xmath2 becomes larger than @xmath27 is that nuclei , or even the protons and neutrons , can become unstable . for example , in our universe neutrons can be stable within nuclei ; but , for sufficiently large @xmath2 , as we shall see , the reaction energy for neutron decay , @xmath53 , becomes larger than the binding energy per nucleon in nuclei of about @xmath54 mev . at that point neutrons even in nuclei will decay and the only stable nuclei will be protons . we argue that such a proton universe " would be sterile . to analyze the stability of nucleons and nuclei in large @xmath2 universes , we must understand how masses and binding energies depend on @xmath2 . we turn first to this question . the quark and lepton masses simply scale with @xmath2 ( ignoring the relatively small effect of the logarithmic running of the yukawa couplings ) . thus , we take @xmath55 mev , @xmath56 mev , and @xmath57 mev . we model baryon masses by @xmath58 , where @xmath59 is the sum of the quark masses in the baryon , @xmath60 is the color energy , and @xmath61 is the electromagnetic energy . since for the neutron and proton the color energy is the same , the neutron - proton mass splitting is given by @xmath62 . as long as the quark masses are small compared to the qcd scale ( which will be true for @xmath63 less than a few hundred ) the size of nucleons , and therefore the electromagnetic energy , will be relatively insensitive to @xmath2 . [ the size of a nucleon will scale as @xmath64 . for the dependence of this on @xmath2 see below . ] thus we can take @xmath65 to have the same value which it has in our universe , namely about @xmath66 mev . thus we have that @xmath67 mev , and the @xmath68 value for neutron beta decay , @xmath69 is @xmath70 mev . for @xmath71 most of the mass of the @xmath72 , @xmath73 , and @xmath74 baryons is due to color energy . the splitting between the @xmath75 baryons ( @xmath73 and @xmath72 ) and the @xmath76 baryons ( @xmath74 ) is in our universe about @xmath77 mev . ( since the lightest baryons will be made purely of @xmath78 and @xmath79 quarks , we need be concerned only with isospin and not with flavor @xmath80 . ) we will assume that both @xmath81 and @xmath82 are proportional to the qcd scale , @xmath83 . @xmath83 depends only indirectly and fairly weakly upon @xmath2 . ( this dependence arises because the renormalization group running of the strong coupling constant " , @xmath84 , depends on quark thresholds , which in turn depend on the quark masses . ) we find that @xmath85 , where @xmath86 for @xmath87 . thus we take @xmath88 mev . of course , for very large @xmath2 ( larger than a few hundred @xmath27 ) the quark masses will become larger than the color energy , and the proton , neutron , and @xmath74 will become non - relativistic bound states in which the color energy will go as @xmath89 and thus be proportional to @xmath2 . the dependence of the nuclear force on @xmath2 is a much more complicated matter . the long - range part of the nucleon - nucleon potential is due primarily to one - pion exchange , and therefore has a range of @xmath90 . as long as the @xmath78 and @xmath79 masses are small compared to the qcd scale , the mass of the pion is well approximated by @xmath91 . assuming that @xmath92 , one has that @xmath93 . we will take @xmath94 to go as @xmath95 , which is adequate for our discussions . the shorter - range part of the nucleon - nucleon potential comes from multi - pion exchange , the exchange of heavier states , and more complicated effects . it is difficult to estimate how these will depend on @xmath2 . however , our qualitative conclusions do not depend upon this issue . with these general considerations we map out the nature of baryons and nuclei in universes where @xmath52 . * @xmath96 . in our world , the splitting between isospin multiplets is large , @xmath97 mev @xmath98 . the lightest baryons are thus the proton and neutron . of these the proton is lighter because the quark mass splitting ( @xmath99 mev ) wins out in competition with the electromagnetic energy splitting ( @xmath100 mev ) . * @xmath101 . as @xmath2 increases the neutron becomes more unstable because @xmath102 increases , and the nuclear potential between nucleons gets weaker ( since @xmath94 is getting larger ) . the combined effect is to render nuclei less stable . we estimate that for @xmath103 there will be no stable nuclei , as the mass excess of the neutron is greater than the nuclear binding energy . for @xmath104 a variety of nuclei will continue to exist , with fewer and fewer stable isotopes surviving as @xmath63 increases . + even if nuclei are stable there is the question of whether or not they may form through nucleosynthesis . relevant to this question is the fact that one of the first nuclei to become unstable as @xmath63 increases above 1 will be the deuteron , which even in our universe is very weakly bound . this is a particularly important case as all primordial and stellar nucleosynthesis ultimately begins with deuterium . + the critical reaction for decay of the deuteron is @xmath105 which occurs whenever @xmath106 { \rm mev}$ ] . for the binding energy of the deuteron as a function of @xmath2 we consider two models , both based on the knowledge that the deuteron is a weakly bound and rather extended nuclear state , sensitive to the long - range pion - exchange component of the nuclear potential . + in the simpler model we treat the nuclear potential as a square well with a hard core . the hard core mocks up the short range repulsion , whereas the square well of depth 35 mev and width 2 fm represents the one pion exchange potential . to model the effects of changing @xmath2 we decrease the width as @xmath107 to account for the increase in @xmath108 . the deuteron binding energy can be solved for analytically ( see appendix ) , yielding an approximate relation @xmath109 { \rm mev},\ ] ] where @xmath110 , for small @xmath111 . + the shortcoming of this model is that all the deuteron binding is attributed to one - pion - exchange , which probably overstates its importance . we therefore tried a more sophisticated approximation [ 16 ] , using a one - boson - exchange - potential ( obep ) based on deuteron binding and scattering phase shifts . this model includes 6 bosons , and also includes @xmath112 and @xmath79 wave mixing for the deuteron . we varied @xmath108 proportionally to @xmath95 ( we neglect the scaling of @xmath83 ) , but it is not clear how the other meson parameters should vary . the problem is not well defined , as several of the `` mesons '' do not correspond to physical particles but are mockups of the short range exchange of two or more pions in channels with the same quantum numbers . the masses for these mesons reflect the momentum distribution of the multiple pions as much as the mass of the pion itself . further , the mass and coupling parameters are arrived at only after fitting , thus to change their relative values in an ad - hoc way can destroy some sensitive cancellation , and is of questionable value . faced with this problem we chose to vary the pion mass , but kept all other parameters of the obep unchanged . solving for the deuteron bound state we find a nearly linear relation between @xmath113 and @xmath2 of the same form as eq . 2 , with @xmath114 mev [ 16 ] . + for both our models , the deuteron binding energy is explicitly a function of @xmath108 and implicitly a function of @xmath63 . the deuteron becomes unstable to weak decay for @xmath115 with @xmath116 in mev . for the square - well model and the obep model , the deuteron is unstable at @xmath117 and 1.4 respectively . + if the deuteron lifetime against weak decay is long enough , then a chain of nuclear reactions involving intermediate unstable deuterons may be possible . in this case the critical reaction is the strong decay @xmath118 , which becomes possible if @xmath119 . the corresponding values of @xmath63 are 1.4 and 2.7 for the square - well and obep models , respectively . + the anthropic argument in the case where deuterons are unstable is not airtight there exists the possibility that nuclei may form in neutron - rich regions following stellar collapse . such a scenario would require significant rates for three - body processes , or a long - lived deuteron as may exist for the regime of @xmath120 for the obep model . * @xmath121 . for this range of @xmath63 nuclei are unstable to decay of constituent neutrons , @xmath122 . hypothetically , there is the possibility of stable proton - rich ( @xmath123 ) nuclei , but this seems unlikely . in our world the depth of the nuclear potential is of order 50 mev , but the binding energy per nucleon in nuclei is only about 10 mev . the difference is primarily due to kinetic energy in the form of nucleon degeneracy energy and coulomb energy due to the protons . in a nucleus with @xmath123 , with the same value of @xmath124 and the same nuclear density , the fermi energy will be greater by a factor of about @xmath125 , and the coulomb energy will be greater by roughly a factor of 4 . given , also , an expected decrease in nuclear binding due to an increase in @xmath108 and stable proton - rich nuclei seem unlikely even in the absence of inverse @xmath126 decay . + even if stable proton - rich nuclei do exist , it will be difficult to form them . the most stable nuclei will occur for intermediate @xmath124 , but there will be significant gaps in the sequence of stable nuclei necessary for nucleosynthesis . for example , either from direct experiment or by comparison to other unstable isotopes we may conclude that he@xmath37 , li@xmath127 , and be@xmath16 , are all strongly unstable in our universe , and therefore also in universes where @xmath128 . it is unclear , then , how compound nuclei would form if @xmath103 . + thus , the only stable nucleus will be the proton , and the only element hydrogen . we expect such proton universes " to be sterile . it is interesting that the existence of neutrons close enough in mass to the proton to be stable in nuclei plays an important role in making life in our universe possible . * @xmath129 . for large enough @xmath63 , the mass difference between @xmath78 and @xmath79 quarks is greater than the penalty in color energy , @xmath130 , that must be paid to have three identical quarks in a baryon . exactly where this occurs is not too important , but at some point @xmath131 . comparing @xmath132 mev to @xmath133 mev suggests that equality takes place at around @xmath134 . + on either side of the critical value there is a range of order 20% in @xmath63 where both @xmath72 and @xmath135 are stablized by the electron mass . this leads to what we shall call proton plus @xmath135 universes " . + as @xmath2 is increased above the narrow range of values where @xmath72 and @xmath135 can coexist , the proton becomes unstable to the decay @xmath136 , and at this point the only stable baryon is the @xmath135 . we refer to such a universe as a @xmath135 universe " . + it is fairly clear that the @xmath135 universes are sterile , as in them it seems quite unlikely that two @xmath135 s would bind . since @xmath63 is large enough that @xmath137 , the pion is not a goldstone - boson - like particle , and the nuclear force has only a short range part . here , in addition , there is a substantial coulomb repulsion between the @xmath135 s . + if the @xmath135 s can not fuse to form heavier nuclei , then there is only a single kind of element , which will have a single @xmath135 as its nucleus . but this element is chemically equivalent to helium , and is therefore chemically inert as well . therefore , in the @xmath135 universes one expects neither nuclear nor chemical reactions to occur . it is hard to conceive , then , what kind of reactions could form the basis of life . the proton plus @xmath135 universes " are only marginally more interesting . from what we know of their molecular states , it seems plausible that hydrogen and helium alone could not form the basis of biochemistry . we conclude , therefore , that the whole range from @xmath129 to @xmath138 can be excluded anthropically . * @xmath139 finally , although it is not strictly an issue for the hierarchy problem , we examine the nuclear consequences of @xmath140 in universes where @xmath1 . as @xmath2 decreases from @xmath27 , the neutron becomes stable , then @xmath141 , followed eventually by a region where the proton is unstable to decay @xmath142 . the stability criteria are determined using a @xmath2 dependent electron mass @xmath143 . + one result of increasing neutron stability is to increase the primordial nucleosynthesis yield of he@xmath16 . once the neutron is lighter than the proton we can no longer reliably estimate the results of nucleosynthesis . with no coulomb barrier to suppress reactions , we anticipate that all single nucleons are bound into compound nuclei , but we can not calculate the distribution of heavy elements . the value of @xmath144 never gets larger than 1.7 mev , so it is never large enough to destablize nuclei in a manner similar to neutron decay in the @xmath145 regime . these compound nuclei are therefore stable and we see no reason why values of @xmath146 could not support life . for @xmath24 the anthropic argument based on viable chemistry worked very effectively . the situation is not as simple for @xmath50 . in this case , @xmath147 , and therefore for @xmath148 one has @xmath149 , and all the quark and lepton masses are extremely small . for example , @xmath150 . also , symmetry breaking of @xmath151 is driven by quark condensates , so that @xmath152 . these facts have dramatic consequences for the chemical energy scale of life , the structure of elements and stellar evolution ; all of which play a role in the genesis of life . our critical assumption is that life requires the chemistry of complex molecules . a typical biochemical energy is @xmath153 , where @xmath154 is a numerical factor , which in our universe is of order @xmath155 . ( this gives , not coincidentally , @xmath156k , the average surface temperature of the earth . ) what @xmath154 would be in the bizarre small-@xmath2 universes we shall consider is hard to say . we shall keep it as a parameter in our formulae . we think it unlikely that it is large compared to one . the crucial point is that @xmath157 is proportional to @xmath158 , which in @xmath50 universes is tiny compared to its value in our universe . it is clear that chemical life can not emerge until at least the time , which we will call @xmath159 , when the temperature of the cosmic background radiation cools below @xmath157 . to put it picturesquely , there is the problem of life being fried by the cosmic background radiation . when this radiation has temperature @xmath160 , the matter density will be @xmath161 , where @xmath162 is the baryon to photon ratio . we assume @xmath162 to be set by physics that is insensitive to the value of @xmath0 , such as the interactions of a grand unified theory , and therefore to have a value similar to that in our universe , namely @xmath163 . using @xmath164 , @xmath165 , and @xmath166 , one derives @xmath167 \epsilon^{-\frac{3}{2 } } \left ( \frac{\mu^2}{\mu_0 ^ 2 } \right)^{\frac{3}{2}},\ ] ] or @xmath168 by this time one of several disasters may occur which prevent the emergence of life . we will discuss two of these : the burning out of stars , and the decaying away of all baryons . before we do so , however , we must reexamine the assumptions that went into eq . first , we note that this is a lower limit for @xmath159 as other contributions to the radiation density ( e.g. baryon decay , see below ) may be greater than the primordial cosmic background radiation . second , we have assumed that chemistry is based on electrons , an assumption which we now examine . it is not hard to show that if we replace @xmath158 by either @xmath169 or @xmath170 in the formula for @xmath159 the resulting time is still short compared to the lifetimes of the @xmath171 or @xmath172 . this is because the @xmath126-decay lifetime of a lepton goes as one over the fifth power of its mass , and therefore as @xmath173 . ( recall that the fermi constant , @xmath174 in this limit rather than as @xmath175 . ) since @xmath171 s and @xmath172 s are stable on the relevant time scales , it is possible that the valence leptons of biochemistry would be some lepton other than ( or in addition to ) electrons . this can not be ruled out , and in fact whether it is so depends on the details of baryo / leptogenesis . whatever the mechanism of baryo / leptogenesis , sphaleron processes will certainly have the effect of reshuffling baryon and lepton numbers in such a way that there will be non - zero values of all three lepton numbers . further , there is no reason for all three lepton numbers to be negative , even though sphaleron processes will try to minimize @xmath176 . let us consider two cases to see some of what is possible . ( 1 ) if primordially only @xmath172-number was produced , and it was negative , then sphaleron - processes , weak scattering processes , and beta decay would lead to a distribution with a net positive charge in @xmath172s , @xmath177 . additionally , baryon s ( see below ) tend to be positively charged , so @xmath178 . these charges are balanced by @xmath179 and @xmath180 . this means that baryons and @xmath181 s would become clothed with @xmath182 s and @xmath183 s . since the @xmath183 s would be much more tightly bound , chemistry in this case would be electron chemistry . ( 2 ) if , on the other hand , primordially only electron number were produced , and it were negative , then the baryons would end up being clothed exclusively with @xmath184 s as long as @xmath185 . in any baryonic atom , a @xmath172 could replace a @xmath171 , with a gain in binding energy . the remaining @xmath172s and @xmath171s would bind , albeit more weakly , with electrons . if the chemistry relevant for life is dominated by lepton @xmath186 , then one simply multiplies the formula for @xmath159 by @xmath187 . this result is to be compared with the baryon lifetime . if @xmath188 , we then consider the structure of nuclei , and stellar evolution . * ( i ) the decay of baryons . * the unification of gauge couplings suggests grand unification at a scale of order @xmath189 gev , and therefore the existence of gauge bosons of that mass whose exchange leads to violation of baryon number . but even if there is no unification at that scale , it is plausible to suppose that at least at the planck scale there will be states whose exchange violates baryon number . thus we will parametrize the baryon decay rate as @xmath190 , where @xmath191 is assumed to lie between @xmath189gev and @xmath192gev . as mentioned above , if baryons are not stable the radiation density may be dominated by the energy released in their decay . we equate the age of the universe , @xmath193 , to the baryon lifetime . at this time the radiation density is given by the relation @xmath194 , where @xmath195 is a numerical factor that we take to be about @xmath196 [ 17 ] . this gives @xmath197 . requiring that this be less than @xmath198 , and using the parametrization of @xmath199 , yields the constraint @xmath200 thus if @xmath201gev , the constraint that baryons still exist when the universe is cooled down to @xmath157 tells us that @xmath0 must be at least @xmath202 orders of magnitude smaller than the natural scale of @xmath203 . the value in our universe is about @xmath204 orders of magnitude smaller than the natural scale . if biochemistry is controlled by @xmath171 s or @xmath172 s , then the above constraint on @xmath0 would be weakened by a factor of @xmath205 , or @xmath206 . there are a number of considerations that may actually strengthen the bound on @xmath0 . in the first place , we have neglected the possibility that baryons might decay predominantly into channels that are blocked in our universe by the heaviness of the higher generation quarks and leptons . for example , in some supersymmetric theories the exchange of superheavy colored higgsinos would produce very fast proton decay ( in our universe ) if it were not for the fact that the final states include flavors heavier than the proton . even more interesting is the possibility that baryons could decay _ via _ sphalerons [ 18 ] . since in @xmath3 universes the weak interactions are broken not at @xmath27 but at @xmath44 , non - perturbative weak baryon decay would not be as strongly suppressed as in our universe . finally , there is the question of baryogenesis . if it occurs through standard drift and decay " mechanisms [ 19 ] involving grand unified interactions , then one would expect the asymmetry to be essentially independent of @xmath0 . but if baryogenesis takes place at the weak scale in our universe [ 20 ] , it is probable that in the @xmath207 universes the asymmetry would be very different . * ( ii ) the structure of the elements * with the higgs system not generating a vacuum expectation value , the breaking of the standard model occurs through the goldstone bosons of the strong interactions . the @xmath208 mass is of order 50 mev . however , the quark masses are very small since they are sensitive only to the higgs vacuum expectation value . with six essentially massless quarks , the lightest baryon multiplet will be a 70-plet , with 27 members which are neutral , 27 which have charge + 1 and eight each with charge -1 and charge + 2 . the lightest meson multiplet starts out with 35 members , but three are eaten " to become the longitudinal gauge bosons , leaving 32 pseudoscalar mesons , which are the goldstone bosons of the dynamically broken chiral su(6 ) symmetry . electroweak interactions will split the multiplets and our estimates for these effects are based on the understanding of electromagnetic mass splittings of the observed hadrons . for the goldstone bosons the mass splittings can be understood using effective lagrangians with chiral symmetry , which in the world under consideration would involve a @xmath209 chiral invariance . the purely lefthanded interactions of the charged @xmath208 s will not produce masses for the goldstone bosons , since such interactions lead to an effective chiral lagrangian which involves derivatives [ 21 ] , and hence vanishes at @xmath210 . however vectorial interactions do generate masses , in analogy to the electromagnetic mass shifts of pions and kaons . the electromagnetic and @xmath21 interactions will have such vectorial effects , and display an @xmath211 invariance for separate rotations of up - type and down - type quarks . by a generalization of dashen s theorem [ 22 ] , these interactions will leave the 16 neutral goldstone bosons massless , while giving a common mass to all 16 charged mesons . using the observed pion mass splitting we can estimate that the charged mesons will have a mass of about @xmath212 mev . the neutral mesons will develop very small masses due to chiral symmetry breaking , @xmath213 . even for @xmath171 as small as 100 gev , some meson masses will be less than a kev . the tiny mass of the neutral mesons implies that nucleons will have a long - ranged neutral mesonic cloud , and that the nuclear force will have a correspondingly long - range potential . the baryon mass splittings are less amenable to an analysis based only on symmetry , and we must include quark model ideas about the effects of electroweak interactions . the short - distance electroweak effects preserve the quark chirality , which implies that they will not generate quark mass shifts . the longest range effects of the vectorial interactions will be proportional to the square of the baryon s electric or weak charges . describing these by the electric charge q and the strong isospin t , and ignoring the masses of the @xmath214 , we would expect that the lightest baryons should be the 19 @xmath215 states , a group which contains the neutron . with splittings of order an mev , the 19 @xmath216 charged baryons ( containing the proton and related states ) could likely be the next grouping . hyperfine interactions and quark masses could further split the states within these groups . all of the baryon and meson states which are shifted up in mass are unstable and can decay down to the ground states via weak or electromagnetic interactions , although perhaps with very long lifetimes . given these building blocks we outline the nature of nuclei . in our world , consisting of just protons and neutrons , any substantial nucleus has comparable numbers of neutrons and protons , as dictated by a desire to minimize the fermi energy . there is a tendency to have fewer protons so as to simultaneously lower the coulomb energy . in the @xmath3 worlds , there are many species of neutral and charged nucleons , so the effects of degeneracy energy will only come into play for larger nuclei . at the same time , for modest size nuclei ( @xmath217 of a few fermi ) the weak bosons are effectively long range , @xmath218 , so the one must take into account weak - interaction energy as well as electromagnetic coulomb energy . a nucleus consisting of @xmath219 baryons of type @xmath220 and @xmath221 of type @xmath222 will have strong isospin @xmath223 and weak hypercharge @xmath224 . ( since right - handed quarks have vanishing weak isospin , the effective weak isospin of a nucleon is half of its strong isospin , and @xmath225 . ) the coulomb energy will have contributions proportional to @xmath226 . for large @xmath227 we can approximate @xmath228 by @xmath229 . clearly , this is minimized by making @xmath227 as small as possible , namely @xmath230 . this allows us to express the coulomb energy in terms of @xmath219 and @xmath221 , and minimizing with respect to @xmath221 subject to the constraint that @xmath231 , one finds easily that @xmath232 this relation holds for intermediate @xmath124 ; for large @xmath124 the range of the weak gauge bosons will saturate and eventually we must include degeneracy effects ( filling fermi levels ) as well as other nucleon states , while for small @xmath124 we must include the nucleon mass differences . coulomb energy may cause nuclei larger than some critical size to be unstable to spontaneous fission , as in our universe . however , above the value of @xmath124 for which the weak force saturates ( which we estimate to be on the order of a few hundred ) the ratio @xmath233 is determined predominantly by minimizing electric coulomb energy rather than weak - interaction coulomb energy . this would lead ( in these @xmath50 universes ) to large nuclei having @xmath234 . the smallness of @xmath233 may well mean that there is no maximum size set to nuclei by spontaneous fission . the situation is complicated by the longer range of the strong nuclear force and the effects of degeneracy energy with many degrees of freedom . therefore , we can not say with certainty whether or not there is a maximum nuclear size in the @xmath50 worlds . similarly , the spectrum of nuclei that result from primordial nucleosynthesis is not certain . early stages of nucleosynthesis will occur with neutral baryons combining to form light nuclei , but as nuclei grow in size and charge , it is not clear to us whether or not coulomb barriers will result in a termination of nucleosynthesis at small or modest nuclei , or whether nuclear burning will run away " to give only very massive nuclei . the difficulty lies in estimating how screening and thermal contributions to meson masses will affect the long - range nuclear potential in its competition with the coulomb barrier between light nuclei . with the uncertainty in these issues , it seems possible that light nuclei with small charge may exist and provide a basis for chemical life . it is also possible that nucleosythesis will result in a small number of low - charge , superheavy nuclei , which does not seem conducive to the development of chemical life . * ( iii ) stars burning out . * even if an interesting mix of elements develops during nucleosynthesis , the time @xmath159 is so large that it is natural to wonder if there would be any stars left by the time biochemistry could take place in the small-@xmath2 universe . on the other hand , with extremely small lepton masses , stellar cooling may be so obstructed that stars never contract to the nuclear burning phase at all . we give an overview of these issues . stars are supported by either gas pressure , radiation pressure , or degeneracy pressure . if the support is either gas , radiation , or degeneracy of a relativistic species then as the star cools ( loses energy ) it contracts and heats up ( @xmath235 increases ) . if supported by degeneracy of non - relativistic fermions , then it cools but does not contract . a system of the first kind contracts until it becomes hot enough for nuclear burning to proceed against the coulomb barriers of the nuclei . once the fuel is gone contraction continues until either another fuel burning stage is reached , the object is supported by non - relativistic degeneracy pressure ( white dwarf or neutron star ) , or the object contracts within its schwarzschild radius and disappears as a black hole . systems supported by degeneracy pressure before nuclear burning cool into planets . whether a cloud of gas turns into a star or a planet depends on its initial mass and ( to a lesser extent ) composition . initially the cloud is non - degenerate and is supported by gas pressure . as it contracts degeneracy pressure increases as @xmath236 while gas or radiation pressure increase as @xmath237 . when the star reaches a size @xmath238 degeneracy pressure will halt further contraction . from dimensional analysis , @xmath239 where @xmath240 , @xmath219 is the number of baryons in the star , and @xmath241 is the mass of the degenerate fermion [ 23 ] . the temperature at this point is @xmath242 and the fermi momentum of the degenerate fermions is @xmath243 if @xmath244 is greater than the temperature necessary for nuclear burning , @xmath245 1 kev [ 24 ] , then a star is born before degeneracy occurs . in our world , @xmath246 and , after including numerical factors , @xmath247 for @xmath248 . low mass stars develop , burn , and then turn into white dwarfs . in the @xmath3 world , the lepton may be @xmath249 , @xmath171 , or @xmath172 , but in any case @xmath250 ev . @xmath244 will therefore be too cool to support nuclear burning . protostars with @xmath251 turn into planets . if @xmath252 the leptons become relativistic before they become degenerate , and the collapse can not be halted until nuclear fusion takes place . after a burning phase contraction continues until either another nuclear burning phase occurs or the core contracts inside its schwarzschild radius and a black hole results . if @xmath253 is not too much greater than 1 , a neutron star may form . since @xmath252 for nuclear fusion , stellar burning will take place in a plasma of relativistic leptons and anti - leptons . this plasma is not degenerate : the degeneracy parameter is @xmath254 . as the leptons are relativistic opacities for photons will be large , dominated by the photon - lepton cross - section @xmath255 . on the other hand neutrino interactions are also much larger , so neutrinos dominate stellar cooling in the nuclear burning phase . at temperatures less than @xmath256 cross - sections for neutrino pair production , scattering from leptons or baryons , absorption , etc , will be of order @xmath257 . for stars of @xmath258 the density of baryons and thermal leptons will be comparable , but for larger stars thermal particles will dominate , so we may estimate mean free paths and emissivities from thermal pair processes . mean free paths for weak interactions are @xmath259 cm ; ie , solar type stars are likely to cool by volume emission or have a very deep neutrino sphere . the energy loss rate per lepton from pair annihilation to neutrinos is @xmath260 gev s@xmath261 ; ie , solar mass stars cool on time scales of roughly a year , and larger stars in much less time . this is very much less than @xmath159 . the next issue is star formation . of particular concern are the cooling rates during collapse , when temperatures are too low for efficient neutrino emmission , and cooling is regulated by the photon opacity either in the interior or at the surface . opacities are determined by the chemistry of the lightest charged lepton . for us this is the electron mass , although in some scenarios the active species will in fact be positrons bound in either @xmath171 or @xmath172-onium . there are two stages in the collapse of a cloud to the point of nuclear ignition . at first , temperature gradients are not large and convection provides the energy transport . the photosphere is essentially held at a fixed temperature @xmath262 at which material is no longer ionized and the opacity drops . the luminosity is @xmath263 . cooling is initially very fast but slows as the star shrinks . eventually , temperature gradients increase to where radiative transport is effective and convection is cut off . at that point the cooling time scale is the kelvin - helmholtz time for photon diffusion to the surface . @xmath264 where @xmath32 is the mean free path and @xmath265 is the cross - section for photon - lepton scattering . to derive the last relation : note that if @xmath266 then @xmath267 but for higher temperatures @xmath268 , and that @xmath269 is roughly constant during collapse . the cooling time is dominated by the epoch when @xmath270 , or @xmath271 yr . this is less than @xmath159 ( see eq . ( 5 ) ) , but not by so much that the energy from stellar burning may not be available for life to form . thus , within this crude treatment of stellar evolution , stars are expected to form slowly , but burn nuclear fuel very quickly . the actual stage of nuclear burning seems too short to benefit life , but there are other possible energy sources than surface heating of planets by stars . for example , planets may be volume heated by radioactivity , residual gravitational energy , baryon decay , or even absorption of the background of neutrinos produced by stellar burning . the energy flow to the surface would in principle be usable for the evolution of life . we have been discussing theories in which the higgs mass parameter @xmath0 can vary in different regions of the universe , under the assumption that the yukawa couplings do not change from the values that they have in our portion of the universe . it is possible that the underlying theory also allows the values of the yukawa couplings to vary in different regions of the universe . in this section , we discuss some of the possible implications of this situation . however , we stress that without knowing the details of the fundamental theory we do not know whether these masses are in fact subject to variation or whether they occur in fixed ratios due to some other mechanism . for example , we can see no anthropic argument which would force neutrinos to be as light as they seem to be ( if indeed they have a mass ) . however , a see - saw " mechanism [ 9 ] would make them very light automatically , and we would not need consider anthropic reasoning . the masses of the quarks and leptons of the second and third generations have very little impact on the particles and reactions which occur naturally in our universe . therefore anthropic arguments would not place constraints on their masses . in an ensemble of anthropically allowed universes , these masses could be randomly distributed . in practice , the observed masses appear to be distributed without any obvious pattern , and randomness appears as good an explanation " as anything else . the masses of the light quarks are constrained by the physics which has been discussed above . in particular , the requirement that the deuteron exist yields an upper bound on the sum of the up and down quark masses , @xmath272 if the @xmath79 quark mass was lighter , or the both the masses were small , the proton would be heavier than the neutron , and could decay into it . hydrogen would then not exist as a stable atom ( at least for small values of the electron mass ) . however , the complex elements would still exist , and there would seem to be sufficient building blocks for some form of life . thus , we will use the above inequality as the sole anthropic constraint on the light quark masses . the anthropic constraints on the electron mass come from other sources . an electron mass larger that the binding energy of a nucleon , around 10 mev , would lead to the decay of atoms , through the process @xmath273 , with the final neutron ejected from the nucleus . however , stronger constraints can be obtained in nucleosynthesis . if the electron mass is higher than the temperature at the time of nucleosynthesis , the electrons will have all disappeared , converting via @xmath273 leaving only neutrons behind . likewise , the reactions which would burn neutrons , such as @xmath274 which would be the neutron equivalent of the start of the pp cycle , use weak interaction transitions and would be shut off if the electron mass is large . the precise constraint depends on the neutron and proton masses , but is generically in the range of a few mev . in a universe with domains which can have different values of some of the underlying parameters , life may only be able to develop in some of those domains . if this is the case , we would expect that the parameters of our domain should be typical of anthropically allowed range . if the anthropic principle accommodates such a large range that our values of the parameters are unnaturally small within this range , then the anthropic principle fails to help us understand the sizes of these parameters . however , we have found that within the overall structure of the standard model there is a relatively small acceptable range for the higgs parameter @xmath0 and the light quark masses . the physical values of these parameters are quite typical of this range , raising the possibility that the anthropic principle could be an explanation " of these magnitudes . the arguments behind this conclusion are summarized in fig . 1 . for @xmath0 negative , as in our universe , it seems that the whole range of values for the vacuum expectation value from @xmath275 down to about @xmath276 ( or perhaps even down to @xmath277 ) times the value in our universe can be excluded . for most of that range ( down to about @xmath278 ) the universe would consist of sterile , helium - like atoms whose nuclei were @xmath135 . there would be essentially no reactions either chemical or nuclear . for the lower part of the excluded range , there would be virtually no nuclei other than protons , and the @xmath279 and @xmath280 processes that are needed for nucleosynthesis would be endothermic as the deuteron would not be stable . for positive values of @xmath0 the condition that baryons still exist at the time biochemistry becomes possible forces @xmath0 to be many orders of magnitude smaller than the natural " planck scale . our arguments for smaller values of @xmath0 are less certain . it may be that long range nuclear forces cause all baryons to clump into superheavy nuclei with small charge , which does nt appear to be promising for life . if these forces are screened in a mesonic plasma , then light nuclei will continue to exist , and can burn in stars , and stars may ignite at sufficiently late times to fuel life . individual stars , however , will be extremely short lived compared to the cosmological time scales . if life is to develop in such a universe , the energy source is not likely to be the photoluminescence of an individual star . one can thus plausibly argue that for life to exist the @xmath0 parameter has to be negative and has to be close to the value it has in our universe . one of the interesting features of this argument is that it explains as no other approaches do at present the curious fact that the weak scale is near to the qcd scale . in order for protons and neutrons ( rather than @xmath135 ) to be the lightest baryons , @xmath281 has to be less than the chromodynamic energy which splits the baryon decuplet from the octet . for the deuteron to have large enough binding energy to save neutrons from virtual extinction in the early universe and also to allow the @xmath279 reactions to be exothermic , the pion has to have a long compton wavelength compared to the nucleon , and this in turn means that the @xmath78 and @xmath79 masses have to be not only less than but small compared to the qcd energy scale by about the amount seen in nature . this provides a possible resolution to the `` fine - tuning problem '' in an ensemble of different domains of the universe , the higgs mass parameter will occasionally fall into the anthropically allowed region without having to be fine - tuned in general . if the cosmological constant is confirmed to have a non - zero value close to present estimates , and no new physics is found in the tev energy region , we may be faced with de - facto evidence for the presence of fine - tuning . in such a situation , the anthropic or multiple - universe considerations become highly attractive . finally , let us comment on the ability of these ideas to be tested . negatively , we can say that if the weak scale is what it is for anthropic reasons , there would be no need to invoke supersymmetry or technicolor or other structure at the weak scale to make the fine - tuning natural " [ 1 ] . if no such structure is found , then , it would be a point in favor of anthropic explanations ; indeed , in that case there would be few if any alternatives to an anthropic explanation . positive evidence is harder to come by . of course , we are not able to explore other domains in the universe . however , theories which generate multiple domains may be testable by other , more conventional means . for example , the community is hoping to be able to test the details of inflationary theories through cosmological measurements . likewise , direct physical experimentation has the potential to eventually sort out the correct underlying theory . through standard means we may be able to learn if the fundamental theory in fact produces multiple domains , in which case anthropic considerations automatically become relevant . until the time that this happens , our conclusion must be modest : the observed values of the mass parameters are reasonably typical of the anthropically allowed ranges . let us make a simple qualitative estimate of the deuteron binding energy . a well - known pedagogical model of the deuteron involves a square well of depth @xmath282 mev and range @xmath283 fm , with a hard core of radius @xmath284 . let us neglect the @xmath285-wave component , so that the @xmath286-wave solutions are @xmath287 and @xmath288 with @xmath289 , and @xmath290 . the boundary condition is therefore @xmath291 there will be a range , @xmath292 , at which the binding energy goes to zero , _ i.e. _ @xmath293 if we let @xmath294 , we can solve for the present value of @xmath217 _ via _ a taylor series . to first order in @xmath295 we have @xmath296 now , the outer range of the potential is determined by the pion mass , @xmath297 . if we search for the value of the pion mass at which the deuteron becomes unbound , we equate @xmath298 thus , according to this calculation , only a 20% increase in the pion mass would cause an unbound deuteron . and since the pion mass and the quark masses very nearly obey the relation @xmath299 a 20% increase in the pion mass corresponds to a 40% increase in @xmath2 , or a factor of @xmath300 increase in @xmath5 . we shall parametrize the vacuum expectation value as @xmath301 { \rm mev},\ ] ] for small @xmath111 , where @xmath116 is some positive constant . the square - well calculation gives @xmath302 . g. t hooft , in _ recent developments in gauge theories _ , proceedings of the cargse summer institute , cargse , france , 1979 , edited by g. t hooft _ et al . _ , nato asi series b : physics vol . 59 ( plenum , new york , 1980 ) ; h. georgi and s.l . glashow , _ phys . rev . _ * d6 * , 2977 ( 1972 ) . 2 . for a comprehensive review see s. weinberg , _ rev . phys . _ * 61 * , 1 ( 1989 ) . s. weinberg , _ phys . lett . _ * 59 * , 2607 ( 1987 ) ; s. weinberg , astro - ph/9610044 , to be published in the proceedings of the conference _ critical dialogues in cosmology _ at princeton university , june 1996 ; h. martel , p. shapiro , and s. weinberg , astro - ph/9701099 . b. carter , in i.a.u . symposium , vol 63 , ed by m. longair ( reidel , dordrecht , 1974 ) ; j. barrow and f. tipler , _ the anthropic cosmological principle _ ( clarendon press , oxford , 1986 ) ; b. j. carr and m.j . rees , nature * 278 * , 605 ( 1979 ) ; a. vilenkin , phys . lett . * 74 * , 846 ( 1995 ) . a. linde , _ phys . lett . _ * b129 * , 177 ( 1983 ) ; _ ibid . _ * b175 * , 395 ( 1986 ) , _ ibid . _ * b202 * , 194 ( 1988 ) , phys . scri.,*t15 * , 169 ( 1987 ) . 6 . e. gildener and s. weinberg , _ phys . * d13 * , 3333 ( 1976 ) ; e. gildener , _ phys . rev . _ * d14 * , 1667 ( 1976 ) . 7 . changes in @xmath0 can indirectly make small changes in the dimensionless parameters of the standard model through the renormalization group . these are generally negligible . the cases where they are not will be indicated explicitly . 8 . even though @xmath0 is the only dimensionful parameter that appears explicitly in the lagrangian of the theory , a scale , @xmath83 , is dynamically produced by the strong interactions through dimensional transmutation " . this sets the scale for chiral - symmetry breaking and for the baryon masses . however , the masses of the elementary particles ( quarks , leptons , higgs particle , and weak gauge bosons ) are set by the higgs potential the light neutrinos are generally believed to get a see - saw " mass of order @xmath303 , where @xmath304 is the scale of lepton - number violation . but these are negligible for our purposes . h. georgi , h. quinn , and s. weinberg , _ phys . lett . _ * 33 * , 451 ( 1974 ) . 11 . see , for example , k.s . babu and s.m . barr , _ phys . lett . _ * 75 * , 2088 ( 1995 ) ; g. anderson , s. raby , s. dimopoulos , and l.j . hall , _ phys . _ * d49 * , 3660 ( 1994 ) ; z. berezhiani , hep - ph/9602325 ; and references therein . 12 . r. peccei and h. quinn , _ phys . lett . _ * 38 * , 1440 ( 1977 ) ; s. weinberg , _ phys . _ * 40 * , 223 ( 1978 ) ; f. wilczek , _ phys _ * 40 * , 279 ( 1978 ) . 13 . the strong cp problem : solutions without axions " , s.m . barr , in _ cp violation in particle physics and astrophysics _ j. tran thanh van ( editions frontires , 1989 ) , and references therein . s. weinberg , _ phys . rev . _ * d13 * , 974 ( 1976 ) ; s. dimopoulos and l. susskind , _ nucl . phys . _ * b155 * , 237 ( 1979 ) . 15 . s. dimopoulos and h. georgi , _ nucl . _ * b193 * , 150 ( 1981 ) ; _ phys . lett . _ * 117b * , 287 ( 1982 ) . 16 . v. agrawal , ph.d . thesis , univ . of delaware , 1996 . scherrer and m.s . turner , _ phys . * d31 * , 681 ( 1985 ) . 18 . v. kuzmin , v. rubakov , and m. shaposhnikov , _ phys . lett . _ * b155 * , 36 ( 1985 ) ; * b191 * , 171 ( 1987 ) ; p. arnold and l. mclerran , _ phys . rev . _ * d36 * , 581 ( 1987 ) ; * d37 * , 1020 ( 1988 ) . 19 . s. dimopoulos and l. susskind , _ phys . lett . _ * 81b * , 416 ( 1979 ) ; d. toussaint , s. treiman , f. wilczek , and a. zee , _ phys . rev . _ * d19 * , 1036 ( 1979 ) ; s. weinberg , _ phys . _ * 42 * , 850 ( 1979 ) ; 20 . for a recent review see a.g . cohen , d.b . kaplan , and a.e . nelson , _ ann . rev . nucl . part . sci . _ * 43 * , 27 ( 1993 ) . j. f. donoghue , e. golowich , and b. r. holstein , _ dynamics of the standard model _ , ( cambridge university press , 1992 ) r. dashen , _ phys . * 183 * , 1245 ( 1969 ) ; j. gasser and h. leutwyler , _ phys . _ * 87*,77 ( 1982 ) ; j. f. donoghue , in * building blocks of creation - tasi 1993 * , ed . by s. raby , ( world scientific,1995 ) ( hep - ph/9403263 ) we have explicitly ignored questions of composition and taken the number of baryons to be equal to the number of fermions . we have also ignored numerical factors that arise from geometry ( e.g. @xmath305 ) or from more realistic solutions to hydrostatic equilibrium which would result in central concentration of the material . 24 . with light mesons , and long range nuclear reactions , burning may take place at slightly cooler temperatures . this is the same issue faced in primordial nucleosynthesis . if screening of the nuclear force is effective , then light nuclei exist and stars burn with @xmath306kev . if screening is not efficient then the whole discussion of stars must be greatly modified . * figure 1 : * the figure shows a summary of arguments that @xmath307 is necessary for life to develop . for @xmath1 large values of @xmath308 imply large values of @xmath2 , and hence larger masses for leptons , quarks , and baryons . the increasing difference between the light quark masses , @xmath309 , implies universes with but a single species of stable nuclei , which we argue would not allow for chemistry rich enough to support life . there is a narrow band where both @xmath72 and @xmath135 are stable , but the chemical equivalent of a mix of hydrogen and helium is plausibly also sterile . for @xmath3 , quark chiral condensates lead to @xmath310 and quark and lepton masses become very small . light lepton masses imply that biochemical processes can not occur until cosmologically late times , when baryons may have already decayed . we show a constraint for a baryon lifetime estimated from exchange of intermediate gut scale ( @xmath311 gev ) particles . even if baryons are stable , formation of a biologically acceptable mix of elements or the nature of stellar evolution may make development of life improbable . what is left is a rather narrow range of @xmath312 which includes the physical values in our universe .

a novel type of solitary wave is predicted to form in spin torque oscillators when the free layer has a sufficiently large perpendicular anisotropy . in this structure , which is a dissipative version of the conservative droplet soliton originally studied in 1977 by ivanov and kosevich , spin torque counteracts the damping that would otherwise destroy the mode . asymptotic methods are used to derive conditions on perpendicular anisotropy strength and applied current under which a dissipative droplet can be nucleated and sustained . numerical methods are used to confirm the stability of the droplet against various perturbations that are likely in experiments , including tilting of the applied field , non - zero spin torque asymmetry , and non - trivial oersted fields . under certain conditions , the droplet experiences a drift instability in which it propagates away from the nanocontact and is then destroyed by damping .

1008.1898

the concept of a soliton as a localized particle - like excitation that preserves its shape can be extended to systems that are far from thermodynamic equilibrium through the concept of a `` dissipative soliton '' @xcite . this allows us to analyze a broad range of physical , chemical , and biological nonlinear systems in which localized excitations are observed . driven magnetic systems , especially those of technological interest , exhibit strongly nonlinear dynamics and are an ideal experimental domain for exploring the dissipative soliton model . in this paper , we provide an analytical theory for a novel , localized oscillation mode in a spin torque oscillator with a free layer having perpendicular magnetic anisotropy . the salient features of this mode include a frequency well below that of uniform ferromagnetic resonance , a weak dependence of frequency on bias current , and a precession angle at the maximal value of @xmath0 . combining numerical micromagnetic simulations with an asymptotic analysis of the equations of motion , we identify this mode as a dynamic , dissipative magnetic soliton that is closely related to the `` magnon droplet '' predicted by ivanov and kosevich in 1977 @xcite . the mode central region exhibits magnetization pointing nearly opposite to its equilibrium direction with a perimeter manifesting @xmath0 precession . from our asymptotic analysis , we derive conditions on perpendicular anisotropy and bias current for the nucleation and existence of the dissipative droplet . using our numerical simulations , we analyze the stability of the dissipative droplet soliton as a function of applied magnetic field , bias current , and spin torque asymmetry . solitons in conservative systems occur when nonlinear terms in the equation of motion balance the effects of dispersion @xcite . a classic example is a light pulse moving in a lossless optical fiber : the change of refractive index with frequency ( dispersion ) tends to make the pulse spread out , but for a certain pulse shape , the change of refractive index with light intensity due to the optical kerr effect ( nonlinearity ) exactly balances the dispersion . pulses having this shape can propagate without spreading and are called solitary waves or solitons . the balance between nonlinearity and dispersion typically allows for the existence of a continuous family of solitons that can be excited in the system , rather than a single solution . in the optical fiber example , the family can be parametrized , for example , by pulse amplitude , and there is a continuous range of amplitudes that satisfy the soliton balancing condition . dissipative solitons @xcite are characterized by an additional balancing condition between gain and loss that typically allows only a single solution for a given set of external parameters . although conservative soliton models can explain weakly nonlinear behavior seen in magnetic systems of exceptionally low damping @xcite , damping is not a small effect for many magnetic systems of both fundamental and technological interest . by combining classical soliton theory with bifurcation theory of nonlinear dynamics and concepts of self - organization @xcite , the dissipative soliton concept provides a framework for describing a broad range of soliton - like behaviors . here , we apply this concept to a nanoscale ferromagnetic system in which both damping and a driving force ( spin torque ) are important . spin torque @xcite occurs when a current is driven through a structure with alternating magnetic and nonmagnetic layers in which spin - dependent conductance at the interfaces results in a spin - polarized electron flow . when the polarized electrons enter a ferromagnetic layer whose magnetization @xmath1 is not collinear with the electron spins , the transmitted spins are rotated toward @xmath1 and the angular momentum absorbed by the ferromagnetic layer is known as the spin torque . typical devices have two ferromagnetic layers through which current is driven : a thick `` fixed '' layer that determines the direction of electron polarization , and a thin `` free '' layer whose orientation can be readily changed by spin torque . for current of the appropriate polarity , spin torque opposes the intrinsic damping torque in the system , and currents above a threshold produce dynamic states in which @xmath1 of the free layer can be manipulated without applying a magnetic field . this effect has been used to control switching of nanoscale magnetic elements @xcite , with potential applications in computer memory and data storage . the effect has also been used to produce coherent , frequency - tunable microwave oscillations @xcite in a nanoscale device known as a spin torque oscillator ( sto ) , with potential applications in integrated microwave circuits for mixing and active phase control . recent reviews cover the physics of spin torque @xcite and its possible applications @xcite . the equations of motion for @xmath1 in the presence of spin torque ( presented below ) are inherently nonlinear , and their full solution for a general case is often studied by use of numerical methods . analytical methods can sometimes be applied by invoking restrictions such as high symmetry , spatially uniform @xmath1 ( the `` macrospin '' model ) , and small precession amplitude ( small angle between @xmath1 and its equilibrium direction ) . these restricted cases have been used to explain experimental results with mixed success . the local nature of spin torque allows it to drive large amplitude excitations in which @xmath1 varies on the scale of the magnetic exchange length ( typically a few nanometers ) , something that applied magnetic fields can not do . as we show here , this regime of strongly nonlinear , strongly nonuniform , sustained magnetodynamics is amenable to theoretical examination using numerical and analytical approaches . this regime is also experimentally accessible in stos . we note that a different type of magnetic soliton generated by spin torque , called a spin wave `` bullet '' , was predicted by slavin and tiberkevich in 2005 to occur in the point - contact geometry with magnetic films exhibiting in - plane oriented anisotropy and in - plane applied magnetic field @xcite . for this case , the precession frequency decreases with increasing current , which can result in localization if the frequency falls below the bottom of the spin wave band at the ferromagnetic resonance ( fmr ) frequency . the weakly nonlinear bullet soliton is a solution to a nonlinear schrdinger type equation with third - order nonlinearity in the excitation amplitude . as such , its predicted experimental signature consists of subtle shifts in microwave output frequency and threshold current relative to that expected for a non - localized mode . in contrast , the droplet soliton studied here exhibits dramatic differences in behavior from that of a non - localized mode . this is due to the fact that it is a strongly nonlinear solution of the full equations of motion , rather than simply a third order expansion . domain walls @xcite , magnetic bubbles @xcite , and vortices @xcite are examples of well studied , strongly nonlinear , localized structures that occur in magnetic materials . the droplet differs from these static structures in that it is inherently dynamic ; the frequency of spin precession within the droplet is always greater than zero . in this work , we focus on the two - dimensional ( 2d ) , non - topological droplet , but we note that droplets in two and three - dimensions come in topological flavors as well @xcite . we begin in the next section by presenting an asymptotic analysis of the model equations for the dissipative droplet in a high - symmetry case . we will also derive the droplet s frequency _ vs. _ current relation in this section . section [ sec : dropl - phys - pert ] is devoted to the study of droplets in physically realistic situations incorporating the current - induced oersted field as well as canting of the applied field and fixed layer . section [ sec : excitation - droplet ] details experimentally accessible nucleation conditions for a droplet that take advantage of a small amplitude instability . in sec . [ sec : discussion ] , we discuss possible extensions of the theory and we relate the droplet to other excitations in thin magnetic films . we conclude in sec . [ sec : conclusion ] with a summary of the droplet s unique properties . appendices [ sec : modul - inst ] and [ sec : numerical - method ] provide details of our stability calculation and numerical method , respectively . we consider the landau - lifshitz - slonczewski equation in non - dimensional form@xcite describing the free layer magnetization in polar coordinates ( @xmath2 ; bold symbols represent vectors in @xmath3 or @xmath4 , contextually dependent ; e.g. , @xmath5 $ ] ) @xmath6 figure [ fig : coordinates ] is a schematic of the nanocontact geometry and coordinate systems considered in this work . the magnetization @xmath7 and fields ( e.g. , @xmath8 ) are normalized by the saturation magnetization @xmath9 , time is normalized by @xmath10 ( @xmath11 is the gyromagnetic ratio and @xmath12 is the permeability of free space ) , and space is normalized by the exchange length @xmath13 ( @xmath14 is the exchange parameter and @xmath15 is the reduced planck s constant ) . we consider a free layer whose thickness @xmath16 is much smaller than the magnetic excitation wavelengths , so that a 2d model with local dipolar fields is justified @xcite . the precessional term is driven by the effective field @xmath17 incorporating exchange @xmath18 , a uniform applied field @xmath19 $ ] ( @xmath20 ) , the current - induced oersted field @xmath21 ( the definition of @xmath22 is given in ref . and summarized in sec . [ sec : oersted - field ] ) , the perpendicular field @xmath23 due to crystal anisotropy ( @xmath24 ) , and the demagnetizing field for a thin film @xmath25 . the landau - lifshitz damping coefficient is @xmath26 . the spin torque term involves @xmath27 , where @xmath28 is the current , @xmath29 ( @xmath30 and @xmath31 are the polarization and spin torque asymmetry parameters , respectively @xcite , @xmath32 is the electron charge , @xmath33 is the nanocontact radius ) , @xmath34 ( @xmath35 ) , @xmath36 , where @xmath37 is the heaviside step function , @xmath38 is the nondimensional contact radius , and @xmath39 $ ] is the uniform fixed layer magnetization . ) , the domain of @xmath40 ( @xmath4 in polar coordinates with radius @xmath41 , azimuthal angle @xmath42 ) , the range of @xmath40 ( unit sphere @xmath43 with polar angle @xmath44 , azimuthal angle @xmath45 ) , orientations of the applied field @xmath46 and fixed layer @xmath47 . ] numerical computations of eq . will be presented in sec . [ sec : dropl - phys - pert ] by means of a method discussed in appendix [ sec : numerical - method ] . for the rest of this section , we will focus on a high - symmetry case where @xmath48 for the analysis of this case , it is helpful to consider eq . in spherical coordinates @xmath49 $ ] ( see fig . [ fig : coordinates ] ) @xmath50 - \alpha g[\theta,\phi ] + \sigma v(\rho ) p_\theta[\theta,\phi ] , \\ \label{eq:3 } \sin \theta { \frac{\partial \phi}{\partial t } } & = g[\theta , \phi ] + \alpha f[\theta , \phi ] + \sigma v(\rho ) p_\phi[\theta,\phi ] , \end{aligned}\ ] ] where @xmath51 & = \sin \theta \nabla^2 \phi + 2 \cos \theta \nabla \phi \cdot \nabla \theta , \\ g[\theta,\phi ] & = -\nabla^2 \theta + \frac{1}{2 } \sin 2\theta ( |\nabla \phi |^2 + { h_\textrm{k}}- 1 ) \\ \nonumber & + h_0 \sin \theta , \\ p_\theta[\theta,\phi ] & = \frac{-\cos \theta \cos \phi \sin { \theta_{\textrm{f}}}+ \sin \theta \cos { \theta_{\textrm{f}}}}{1 + \nu ( \cos \phi \sin \theta \sin { \theta_{\textrm{f}}}+ \cos \theta \cos { \theta_{\textrm{f } } } ) } , \\ p_\phi[\theta,\phi ] & = \frac{\sin \phi \sin { \theta_{\textrm{f}}}}{1 + \nu ( \cos \phi \sin \theta \sin { \theta_{\textrm{f}}}+ \cos \theta \cos { \theta_{\textrm{f}}})}.\end{aligned}\ ] ] the polar angle @xmath44 satisfies @xmath52 , while the azimuthal angle @xmath45 is interpreted modulo @xmath53 . in the symmetric case of eq . , we can remove the explicit dependence on @xmath54 from eqs . and by introducing the following rescaling : @xmath55 recall that we are assuming @xmath24 . then , with the scalings in eq . and dividing eqs . and by @xmath56 , we can , without loss of generality , take @xmath57 . for the rest of this section [ sec : diss - dropl - solit ] , we will use the scalings in eq . so that @xmath58 . in this section , we consider localized magnetization configurations that satisfy @xmath59 with sufficiently rapid decay . as such , we define the magnetic energy in terms of exchange and anisotropy energy via @xmath60 = \frac{1}{2 } \int_{{\mathbb{r}}^2 } \big [ & \underbrace{|\nabla \theta|^2 + \sin^2 \theta |\nabla \phi |^2}_{\textrm{exchange } } \underbrace{+ \sin^2 \theta}_{\textrm{anisotropy } } \big ] d { \mathbf{r } } ' . \end{split}\ ] ] note that the damping and spin torque terms break energy conservation @xmath61 = \\ \int_{{\mathbb{r}}^2 } \bigg \ { & \sigma ' \big [ ( g[\theta,\phi ] - h_0 ' \sin \theta ) p_\theta[\theta,\phi ] - f[\theta,\phi ] p_\phi[\theta,\phi ] \big ] \\ & - \alpha \big [ ( g[\theta,\phi ] - h_0 ' \sin \theta)^2 + f[\theta,\phi]^2 \big ] \bigg \ } d{\mathbf{r } } ' . \end{split}\ ] ] in the absence of damping and spin torque ( @xmath62 , @xmath63 ) , eqs . and admit a continuous family of exponentially localized , non - topological solutions known as magnon droplet solitons @xcite . these solutions can be parametrized by the frequency @xmath64 and written as @xmath65 t . \end{split}\ ] ] they satisfy a balance between exchange ( dispersion ) and anisotropy ( nonlinearity ) through the nonlinear eigenvalue problem @xmath66 = 0 $ ] , @xmath67 = ( \omega_0 + h_0 ' ) \sin \theta_0 $ ] or @xmath68 with the boundary conditions @xmath69 . ] the conservative droplet solutions in eqs . ( [ eq:55 ] ) , , and have the following properties . the polar angle @xmath70 varies with radial distance and is independent of time ; thus the spatial distribution of @xmath71 is static and azimuthally symmetric . the azimuthal angle @xmath45 is independent of position and varies linearly in time ; thus all points precess at a common frequency and in phase . it has been shown that @xcite @xmath64 satisfies @xmath72 therefore , the total precessional frequency @xmath73 varies between the zeeman frequency @xmath74 and the frequency of spatially uniform precession about @xmath75 , @xmath76 ( the fmr frequency ) . while the conservative droplet does not have a closed - form analytical expression , we calculate it by numerically solving eq . ( [ eq:6 ] ) subject to the boundary conditions ( [ eq:7 ] ) with the function ` bvp4c ` in matlab@xmath77 . a plot of several conservative droplet profiles is shown in fig . [ fig : droplet_profiles ] . we see that the amplitude at the origin @xmath78 decreases as the frequency is increased . while it may appear in fig . [ fig : droplet_profiles ] that @xmath79 , in fact all conservative droplets with @xmath80 satisfy @xmath81 owing to their non - topological structure . the energy for the droplet @xmath82 $ ] satisfies @xcite @xmath83 the fact that the energy is a decreasing function of frequency has been used to argue that the conservative droplet is stable in a 2d material @xcite . the 2d conservative droplet embedded in an infinite , 3d magnet is known to be unstable . however , preliminary studies suggest that , for sufficiently thin films , the 2d conservative droplet is stabilized . this work is beyond the scope of this paper and will appear elsewhere . as we now show , the conservative droplet soliton can be generalized to damped / driven systems such as a nanocontact . whenever we refer to `` droplet '' , we mean the dissipative droplet studied in the future sections . we will always use `` conservative droplet '' to describe the solution of eq . ( [ eq:6 ] ) that is monotonically decaying to zero as @xmath84 . we now extend the analysis of kosevich , ivanov , and kovalev @xcite to the case of the dissipative droplet solution , where damping is no longer assumed to be negligible , and spin torque is included in the analysis . in addition to the balance between exchange and anisotropy that was required in the conservative droplet case , a further balance between uniform damping ( loss ) and spatially localized spin torque ( gain ) will be derived for the droplet to be sustained . we will assume that the spin torque and damping , while small , are not zero and are of the same magnitude so that @xmath85 we look for an asymptotic solution in the following form @xmath86 we have set @xmath87 to distinguish the frequency of the droplet @xmath88 from that of the conservative droplet @xmath64 . we will conclude that there is no frequency shift due to damping and spin torque so that @xmath89 here . however , other perturbations beyond those considered here could lead to a frequency shift . inserting the expansions into eqs . [ eq:2 ] , [ eq:3 ] , and equating like orders in @xmath26 gives the following equations for the perturbations @xmath90 and @xmath91 : @xmath92 , \\ \label{eq:13 } { \frac{\partial \phi_1}{\partial t ' } } + l_\theta \theta_1 & = - \frac{\sigma'}{\alpha } v(\rho ' ) p_\phi[\theta_0,(\omega + h_0')t'],\end{aligned}\ ] ] where the self - adjoint , schrdinger operators @xmath93 and @xmath94 are @xmath95 \frac{1}{\sin \theta_0 } \\ \nonumber & = -\nabla'^2 - \frac{d \theta_0}{d \rho'}^2 + \cos^2 \theta_0 - \omega \cos \theta_0 , \\ l_\theta & \equiv \frac{\delta g}{\delta \theta}[\theta_0,(\omega + h_0 ' ) t ' ] - ( \omega + h_0 ' ) \sin \theta_0 \\ \nonumber & = - \nabla'^2 + \cos 2 \theta_0 - \omega \cos \theta_0 .\end{aligned}\ ] ] note the following important property : @xmath96 the rest of this section is concerned with the solution of the perturbative equations ( [ eq:12 ] ) and ( [ eq:13 ] ) in two separate cases . we use standard solvability arguments for forced , linear differential equations to remove secular terms ( see e.g. ref . ) in order to derive an expression for the current @xmath97 at which the balancing condition for the droplet is achieved . we first consider the case where the fixed layer is oriented perfectly normal to the film plane so that @xmath98 . in this regime , we can study the effect of variable spin torque asymmetry @xmath99 . in addition to the scalings in eq . ( [ eq:52 ] ) , we can also scale out the dependence on the applied field @xmath74 when @xmath98 with the substitution @xmath100 then , @xmath44 and @xmath101 satisfy eqs . ( [ eq:2 ] ) and ( [ eq:3 ] ) , as before , but with @xmath102 . this transformation shows that the vertical applied field simply shifts the precessional frequency by @xmath74 . we seek a solution to eqs . ( [ eq:12 ] ) and ( [ eq:13 ] ) in the form @xmath98 , @xmath102 , @xmath103 , and @xmath104 ( i.e. , linear phase evolution with time ) which results in the following non - homogeneous equation for @xmath105 : @xmath106 this equation is solvable if and only if the nonhomogeneous terms are orthogonal to the kernel of @xmath93 . by use of eq ( [ eq:8 ] ) , multiplying the right - hand side of eq . ( [ eq:17 ] ) by @xmath107 and integrating from @xmath108 to infinity we obtain the existence condition for a dissipative droplet , @xmath109 , where @xmath110 the choice @xmath109 singles out a specific value for the current as a function of the droplet frequency @xmath111 . this value of the current provides the necessary balance between damping and spin torque , in addition to the balance between exchange and anisotropy , in order to sustain the droplet . therefore , we call @xmath112 the _ sustaining current_. we can also understand the choice for the sustaining current ( [ eq:18 ] ) by computing the rate of change in the energy for the droplet from eq . ( [ eq:22 ] ) : @xmath113 = \\ \omega \int_{{\mathbb{r}}^2 } \big [ & { \sigma_\textrm{sus}}(\omega ) v(\rho ' ) \frac{\sin^2 \theta_0}{1 + \nu \cos \theta_0 } \\ & - \alpha \omega \sin^2 \theta_0 \big ] d{\mathbf{r } } ' + \mathcal{o}(\alpha^2 ) = \mathcal{o}(\alpha^2 ) . \end{split}\ ] ] therefore , the energy of the droplet is approximately conserved for the choice @xmath114 . figure [ fig : droplet_symmetric_contour ] shows a snapshot in time of a dissipative droplet . . the color scale corresponds to @xmath115 , while the vector field corresponds to the in - plane components @xmath116 . the circle here and in future plots represents the boundary of the nanocontact . parameters are @xmath117 , @xmath118 , @xmath119 , and @xmath120 . ] figure [ fig : sustaining_current ] represents the numerical evaluation of eq . ( [ eq:18 ] ) and shows the dependence of @xmath121 on @xmath88 . the droplet frequency has two branches as the sustaining current is varied , but only one branch is stable . for a given droplet frequency @xmath88 , consider increasing the current slightly above the sustaining value , @xmath122 . from eq . , the droplet energy will increase , and from eq . , this increase in energy corresponds to a _ decrease _ in droplet frequency for a stable droplet @xcite . thus , the upper branch in fig . [ fig : sustaining_current ] , for which increasing current causes an _ increase _ in frequency , is unstable . ) . parameters are @xmath119 and @xmath120 . the dash - dotted line is the fmr frequency . ] figure [ fig : sustaining_current ] shows that there is a minimum sustaining current @xmath123 , maximum frequency @xmath124 , and minimum @xmath115 in the center of the droplet @xmath125 where @xmath126 while the specific choice of spin torque cutoff function @xmath127 ( e.g. , we could have used a smooth cutoff , as opposed to a sharp , heaviside cutoff ) may be important in numerical applications @xcite , it has only a slight effect on the droplet sustaining current . for the heaviside cutoff considered here , eq . ( [ eq:18 ] ) simplifies to @xmath128 equation ( [ eq:56 ] ) reveals the explicit dependence of the sustaining current on two key spin torque parameters : the contact radius @xmath129 and the spin torque asymmetry @xmath35 . we now investigate properties of the dissipative droplet as a function of these two parameters . figures [ fig : sus_min](a - d ) show the dependence of @xmath123 , @xmath125 , @xmath124 , and the droplet radius @xmath130 on the contact radius for various choices of the spin torque asymmetry . the droplet radius is defined to be the radius at half maximum : @xmath131 figure [ fig : sus_min](a ) shows that droplets excited by smaller contacts require larger sustaining currents . the dependence of @xmath132 on @xmath133 and @xmath99 for small contact radii can be made explicit by an asymptotic evaluation of the denominator in eq . ( [ eq:56 ] ) giving @xmath134 \int_0^\infty \sin^2 \theta_0(\rho;\omega ) \rho d \rho}{\rho_*'^2 \sin^2 \theta_0(0;\omega ) } + \mathcal{o}(1 ) , \quad 0 < \rho _ * ' \ll 1 .\ ] ] therefore , @xmath135 grows like @xmath136 for small contact radii in agreement with fig . [ fig : sus_min](a ) . interestingly , we find that the droplet is constrained to have a frequency @xmath137 ( fig . [ fig : sus_min](b ) ) , significantly below the fmr frequency of 1 . also , since @xmath138 ( fig . [ fig : sus_min](c ) ) , the droplet always has some region near its center that is partially inverted with respect to the surrounding magnetization . as shown in fig . [ fig : sus_min](d ) , the droplet is well localized within the nanocontact ; e.g. @xmath139 , when @xmath140 . there is a minimum droplet radius of about @xmath141 in dimensional units . finally , spin torque asymmetry has only a small , perturbative effect and does not substantially alter the droplet solution . and spin torque asymmetry @xmath99 : ( a ) minimum sustaining current , ( b ) maximum frequency , ( c ) maximum @xmath115 at origin , and ( d ) droplet radius ( dotted line is @xmath142 , plotted for comparison ) . ] we consider now the case without spin torque asymmetry ( i.e. @xmath143 ) in which case we can study the effect of varying the angle of the fixed layer @xmath144 on the droplet dynamics . in this section , we use the rescalings in eq . ( [ eq:52 ] ) but do not use the transformation ( [ eq:54 ] ) . we solve eqs . ( [ eq:12 ] ) and ( [ eq:13 ] ) with @xmath120 , @xmath145 , and @xmath146 , leading to the following system of non - homogeneous equations : @xmath147 note that , in contrast to the case @xmath98 , the overall phase @xmath45 no longer evolves linearly in time . applying @xmath94 to eq . ( [ eq:27 ] ) and @xmath93 to eq . ( [ eq:28 ] ) gives the decoupled system : @xmath148 , \\ ( l_\phi l_\theta + \omega^2 ) { \theta_\textrm{s}}= & ~\frac{\sigma'}{\alpha } \sin { \theta_{\textrm{f } } } [ l_\phi \ { v(\cdot ) \}(\rho ' ) \\ \nonumber & - \omega v(\rho ' ) \cos \theta_0(\rho ' ) ] .\end{aligned}\ ] ] these equations are always solvable if @xmath149 and @xmath150 are strictly positive operators . one can show that @xmath151 , so that @xmath152 . one can also show that @xmath153 , where @xmath154 is the smallest eigenvalue of @xmath94 and is strictly negative . then , @xmath155 . we have verified by numerical computation of @xmath156 that @xmath157 so that @xmath158 as required , and eqs . ( [ eq:27 ] ) and ( [ eq:28 ] ) are solvable . we are interested in the solvability condition for @xmath159 in eq . ( [ eq:29 ] ) , which is @xmath160 note that the applied field appears only as a shift in the droplet frequency , as in the case @xmath98 studied in the previous section . the expression ( [ eq:30 ] ) for the sustaining current agrees with the previously derived sustaining current in eq . ( [ eq:18 ] ) when @xmath120 and @xmath98 , as required . thus , the dominant effect of rotating the fixed layer is to increase the sustaining current in proportion to @xmath161 . so far , we have considered the dissipative droplet solution only for a simplified geometry where asymptotic methods can be applied . in these cases , the external field is both uniform and oriented perfectly perpendicular to the film plane . by making this geometrical simplification , we were able to factor out the contribution of the external field from the droplet solution . however , in a real point - contact system , the external field is neither uniform nor perfectly perpendicular . in particular , the current flowing through the contact is an additional source of spatially inhomogeneous magnetic field , the oersted field , and the applied uniform magnetic field in actual experiments is usually tilted away from the perpendicular axis . we employ micromagnetic simulations to investigate how these physically important perturbations to the external field alter the ideal droplet solution . the numerical details used for our simulations are presented in appendix [ sec : numerical - method ] . we find that the combination of external field tilt and the oersted field breaks the symmetry of the solution such that the droplet is no longer centered in the middle of the contact . as a result , the solution takes on a nontrivial inhomogeneous phase structure where the phase of the spin precession closer to the center of the contact precedes the phase further from the center of the contact . in addition , the spatial structure of the droplet is no longer perfectly circular . for some particular combinations of simulation parameters , the droplet breaks away from the contact altogether and dissipates , a behavior we call a drift instability . when this occurs , the droplet may maintain its form for many precession cycles , but it eventually decays , since outside the contact there is no spin torque excitation to balance damping . first , we consider the effect of the current induced oersted field while keeping the applied field and fixed layer oriented almost normal to the film plane ( canted by @xmath162 and @xmath163 , resp . ) . the reason for this slightly asymmetric configuration is to test whether high - symmetry solutions are structurally stable to small changes in the system parameters . such a configuration is experimentally possible , in principle . our model for the oersted field was presented in ref . and takes the form @xmath164 where @xmath165 the function @xmath166 given in ref . involves integrals of bessel functions and depends on the geometry of the current - density distribution . the parameters defining @xmath167 in ref . are @xmath168 , @xmath169 , and @xmath170 . the other , more dominant , term in eq . ( [ eq:58 ] ) is the field generated by an infinitely long conducting wire . as an example , for the simulation in fig . [ fig : droplet_oe_perturb ] , we have @xmath171 , while @xmath172 , an order of magnitude difference . figure [ fig : droplet_oe_perturb ] illustrates how the oersted field changes the structure of the droplet . in contrast with the symmetric case in fig . [ fig : droplet_symmetric_contour ] , the azimuthal angle @xmath45 shows significant spatial variations . the droplet is also slightly shifted off - center . the strong phase variations are indicative of a tendency for the droplet to propagate @xcite . in some cases , although not for the simulation in fig . [ fig : droplet_oe_perturb ] , the droplet breaks free from the nanocontact . this drift instability is discussed further in sec . [ sec : canted - applied - field ] . in fig . [ fig : droplet_f_i_oe ] we compare the numerically computed perturbed droplet frequency as a function of current with the sustaining current from eq . ( [ eq:56 ] ) . we find that the droplet frequency is approximately shifted down by an overall amount of @xmath173 , but otherwise follows the same trend as the symmetric result . this behavior demonstrates that the analysis of the previous section , despite the necessary high symmetry restrictions , yields relevant , qualitative information about localized structures excited in a physically realizable nanocontact . as shown in sec . [ sec : excitation - droplet ] , when the current exceeds a threshold value ; e.g. , the slonczewski critical current@xcite , a droplet can nucleate for sufficiently large anisotropy . the vertical line in fig . [ fig : droplet_f_i_oe ] is the numerically computed threshold current . it differs from slonczewski s result because @xmath174 , @xmath175 , and @xmath176 . this threshold for droplet nucleation suggests a hysteretic effect that will be discussed in sec . [ sec : hysteresis ] . our micromagnetic simulations also show that the perturbed droplet , for certain parameter choices ( e.g. , the crosses in fig . [ fig : droplet_f_i_oe ] ) , undergoes a drift instability . this behavior will now be investigated further . , @xmath177 , @xmath178 , @xmath179 , @xmath180 , @xmath181 , @xmath182 , @xmath183 . dash - dotted : the zeeman ( @xmath184 ) and fmr ( @xmath185 ) frequencies . triangle : the slonczewski critical current and onset frequency for high symmetry @xcite ( see appendix [ sec : modul - inst ] ) . solid vertical line : numerically computed threshold current in the presence of the oersted field and the canted fixed layer . parameter values are the same as in fig . [ fig : droplet_oe_perturb ] . ] in this section , we investigate the combined effects of the oersted field as well as canting of the applied field and fixed layer . figure [ fig : droplet_strong_perturb ] is a time sequence showing the evolution of a strongly perturbed droplet over one precessional period . in contrast to the nearly symmetric configuration of fig . [ fig : droplet_oe_perturb ] , where the droplet was slightly shifted to the left , the droplet is slightly shifted down , toward the region of lower in - plane field . as the system parameters are changed , the shifting of the droplet center can be large enough to actually dislodge the droplet from the nanocontact . an example of this drift instability is shown in the panels of fig . [ fig : droplet_drift ] . the current was taken to be less than the current for fig . [ fig : droplet_strong_perturb ] , which did not experience a drift instability . a number of precessional periods pass before the droplet breaks free . once free , it propagates , but because it no longer satisfies the required balancing condition between damping and spin torque , it loses amplitude and decays . once the droplet has drifted outside of the nanocontact , a new one is formed if the nucleation conditions are satisfied ( see sec . [ sec : excitation - droplet ] ) . we have also observed droplets that drift several nanocontact diameters before decaying , i.e. the central magnetization lifts up so that @xmath186 . the manifold in parameter space in which the droplet manifests a drift instability appears to be complicated . nevertheless , we readily find parameter regimes where the droplet apparently does not experience the drift instability , as in figs . [ fig : droplet_oe_perturb ] , [ fig : droplet_f_i_oe ] , and [ fig : droplet_strong_perturb ] . notice that the droplet propagates down , in the @xmath187 direction . recall that the canting direction of the applied field is along @xmath188 . due to the symmetry of the oersted field , the direction of propagation of the drifting droplet appears to track the azimuthal angle of the applied field minus @xmath0 , if @xmath189 is sufficiently large . for example , if the applied field is canted along @xmath190 , then the droplet will drift along @xmath188 if unstable . this can be understood as a consequence of magnetostatic interactions between the effective dipole moment of the droplet and the field gradient associated with the oersted field . given the canting of the applied field , the effective dipole moment of the droplet acquires an in - plane component that is drawn to the edge of the contact where the oersted field is also in the @xmath191 direction such that the oersted field gradient acts to trap the droplet . in the case of fig . [ fig : droplet_oe_perturb ] , where the applied field is canted only @xmath162 , the droplet is observed to drift to the left rather than down . therefore , the strength of the oersted field and the in - plane component of the applied field have a strong effect on the existence and dynamics of a drift instability . figure [ fig : droplet_birth ] shows the birth of the droplet pictured in fig . [ fig : droplet_strong_perturb ] starting from a state pointing uniformly in the @xmath192 direction . for sufficiently large perpendicular anisotropy @xmath54 and current , the magnetization inside the nanocontact reverses and nucleates a droplet . in this section we show that the reversal mechanism is caused by an instability of small amplitude waves . we will study eq . ( [ eq:1 ] ) in the weakly nonlinear regime and find that the small amplitude slonczewski mode that exists near the threshold for the onset of dynamics is stable / unstable depending on whether @xmath54 is less than / greater than a critical value @xmath193 . we find that @xmath194 due to exchange effects and converges to 1 as the contact size is increased . the value of @xmath193 is important for the possible experimental observation of a dissipative droplet . before studying the pde ( [ eq:1 ] ) , we consider the macrospin model where spatial variation is neglected . the stability analysis for this model is suggestive and mathematically simpler . however , we find that the critical anisotropy field in the macrospin case @xmath195 satisfies @xmath196 and depends strongly on the applied field , which differs from the result obtained by analyzing the full pde model , where we find @xmath194 with weak applied field dependence . we consider eqs . ( [ eq:2 ] ) and ( [ eq:3 ] ) neglecting all spatial variation and inter - layer dipole coupling in the symmetric regime @xmath197 and @xmath98 ; i.e. , we are assuming that the initial condition for the system is in the parallel state . in this case , the equation for @xmath44 is decoupled from @xmath45 , so we can just study the scalar , first - order ode @xmath198 + \frac{\sigma \sin \theta}{1 + \nu \cos \theta } .\end{aligned}\ ] ] by linearizing eq . ( [ eq:46 ] ) about the equilibrium @xmath199 , we find that it becomes unstable when @xmath200 in agreement with previous numerical and mathematical analyses of this system contained in ref . . we seek a periodic equilibrium solution @xmath201 just above threshold by taking @xmath202 then , we have @xmath203 } \right ] ^{1/2 } .\ ] ] this solution exists ( is real valued ) as long as @xmath204 furthermore , one can show that this equilibrium is stable . therefore , when @xmath205 , the equilibrium @xmath206 undergoes a supercritical hopf bifurcation as the current @xmath207 is increased beyond @xmath208 . when @xmath209 , there is no periodic solution , and the system switches from @xmath206 to the fully reversed state @xmath210 when @xmath207 exceeds @xmath208 . we consider eq . ( [ eq:1 ] ) in the symmetric regime with @xmath211 , @xmath212 , @xmath98 , and the substitutions @xmath213 where @xmath214 . then @xmath215 approximately satisfies a complex nonlinear schrdinger type equation @xmath216 |u|^2 u . \end{split}\ ] ] similar to the macrospin case , when @xmath207 is increased past a threshold value , the zero solution becomes unstable . the threshold , critical current and onset frequency were found by slonczewski as a solution to a linear eigenvalue problem @xcite ( see eq . ( [ eq:34 ] ) ) . incorporating weak nonlinear effects , ref . showed that a small amplitude periodic solution exists as a modulation of the slonczewski mode for @xmath217 and @xmath120 . this time - periodic , weakly nonlinear mode was found to be numerically stable . in this section and appendix [ sec : modul - inst ] , we extend these results to @xmath218 and @xmath35 , where @xmath193 is defined through @xmath219 and @xmath220 is given by eq . . this behavior is analogous to the supercritical hopf bifurcation for the macrospin , studied in the previous section . in this section , we show that weakly nonlinear modulations of the slonczewski mode are no longer stable when @xmath221 . this behavior is analogous to the switching exhibited by the macrospin for @xmath209 and @xmath207 above threshold . while we are unable to analytically follow the dynamics of the instability , numerical simulations such as the droplet birth sequence shown in fig . [ fig : droplet_birth ] demonstrate that the magnetization reverses inside the nanocontact and develops into a dissipative droplet . for the stability analysis , we seek a multiple scales solution representing a modulation of the slonczewski mode in the form @xmath222 , \\ \sigma = { \sigma_\textrm{s}}+ { \varepsilon}^2 \sigma_1 \qquad \quad \ , & \end{split}\ ] ] where @xmath223 is the amplitude of the mode at the origin , which is modulated by @xmath224 with @xmath225 the `` slow '' time , @xmath226 is the frequency of the slonczewski mode @xmath227 with threshold current @xmath228 , and @xmath229 represents a deviation from @xmath228 . the explicit form for @xmath230 and the implicit equations for @xmath226 and @xmath228 are given in appendix [ sec : modul - inst ] . invoking a solvability condition at @xmath231 , eq . gives the nonlinear amplitude equation @xmath232 with complex linear and nonlinear coefficients @xmath233 and @xmath220 . there is a time - periodic solution of eq . ( [ eq:40 ] ) for a specific choice of @xmath234 @xmath235 this solution represents the nonlinear frequency shift @xmath236 to the slonczewski frequency @xmath226 . the stability analysis in appendix [ sec : modul - inst ] shows that the solution ( [ eq:43 ] ) is unstable when @xmath237 . an explicit formula for @xmath220 is given in eq . ( [ eq:42 ] ) . we evaluate the integrals numerically and plot @xmath238 as a function of @xmath54 for specific parameter values in fig . [ fig : xi_hk ] . as a function of @xmath54 . when @xmath237 , the weakly nonlinear slonczewski mode is modulationally unstable . parameter values are @xmath177 , @xmath178 , @xmath179 , @xmath239 . ] there is a critical value of the anisotropy field @xmath193 satisfying @xmath240 = 0,\ ] ] above which the weakly nonlinear slonczewski mode is modulationally unstable . in other words , weak modulations of the slonczewski mode will grow exponentially in time when @xmath241 . figure [ fig : hkc ] shows the dependence of @xmath242 , computed numerically by solving eq . with eq . , as the contact radius and spin torque asymmetry are varied and @xmath179 . we see that @xmath193 is strictly greater than one and that larger perpendicular anisotropy is required to enable the nucleation of a droplet in smaller nanocontact devices . for a given contact size , larger spin torque asymmetry permits nucleation of a droplet at smaller values of @xmath54 . as a function of contact radius @xmath243 for various spin torque asymmetries @xmath99 . other parameters are @xmath177 and @xmath179 . ] we have also investigated the dependence of @xmath193 on the applied field magnitude @xmath184 . since @xmath184 appears in @xmath220 only multiplied by @xmath26 , we have @xmath244 , which is small . this is confirmed by numerical calculations of the @xmath193 dependence on @xmath245 for the values of @xmath243 and @xmath99 plotted in fig . [ fig : hkc ] . we find that @xmath193 varies by at most 3 % for @xmath246 $ ] with @xmath247 . importantly , the lower bound for @xmath193 is preserved : @xmath248 when @xmath20 . this behavior stands in stark contrast to the macrospin result for @xmath249 in eq . that strongly depends on the applied field . from eq . ( [ eq:43 ] ) , we see that @xmath229 is negative when @xmath250 . numerically , we find that @xmath251 , so that @xmath252 when @xmath241 , which corresponds precisely to the instability criterion . furthermore , when @xmath253 , any finite - amplitude excitation will tend to grow , thus the time - periodic solution is unstable . this behavior is similar to the macrospin model discussed in the previous section where , for @xmath241 , a small amplitude periodic solution did not exist . numerical simulations of eq . ( [ eq:1 ] ) confirm the foregoing analysis , even in the non - symmetric cases with @xmath175 , @xmath254 , and @xmath174 . there is a critical value of the current above which large amplitude dynamics ensue . when @xmath54 is above @xmath193 , we find that the magnetization inside the nanocontact reverses to form a localized , coherently precessing , fully nonlinear magnetic solitary wave . we identify this solitary wave with the dissipative droplet soliton found in the asymptotic analysis of sec . [ sec : diss - dropl - solit ] . due to the numerically robust formation of the droplet for a variety of initial data and across a large parameter regime , we view it as a global attractor . as long as the current is above threshold , a droplet is observed to nucleate . we have shown in this section that a dissipative droplet will form when the current exceeds the threshold for instability of small amplitude waves inside the contact . in the high symmetry case of eq . ( [ eq:53 ] ) and @xmath98 , the threshold is the slonczewski critical current @xmath228 , which is plotted in fig . [ fig : droplet_f_i_oe ] ( triangle ) along with the numerically calculated threshold current from micromagnetic simulations that incorporate the oersted field and a canted fixed layer ( @xmath255 , vertical line ) . we observe that the minimum droplet sustaining current is below the threshold for droplet nucleation . this suggests a hysteretic effect , whereby a droplet can be nucleated at a current above threshold and remains stable when followed by an adiabatic decrease of the current below threshold . we have performed this experiment numerically for the perturbed droplets of fig . [ fig : droplet_f_i_oe ] . we use an already nucleated droplet ( at @xmath256 ) as the initial condition for a new simulation with @xmath257 , below threshold . we find that sufficiently close to , but below , threshold , the droplet s frequency slightly increases but remains stable . for further decrease of the current to @xmath258 , however , the droplet undergoes a drift instability similar to behavior shown in fig . [ fig : droplet_drift ] . since the current is below threshold a new droplet does not form . the slonczewski form for the spin torque term we are considering here , eq . ( [ eq:1 ] ) , was derived strictly for the case of spatially uniform magnetization distributions @xcite . nonlocal generalizations of the spin torque term to nonuniform magnetization distributions have been derived in the small amplitude regime @xcite and applied to single layer nanocontact simulations @xcite . because the spin torque and damping terms are treated as perturbations giving rise to a dissipative droplet solution , we expect that the specific form for the spin torque term , and the damping term for that matter , will not yield qualitative changes in the structure of the droplet . the asymptotic analysis presented here is applicable to an arbitrary spin torque term as long as its net effect is to oppose the inherent damping in the system . due to the symmetries in the problem when @xmath259 , one might expect the oersted field to prefer the excitation of a localized vortex structure ; e.g. , a topological soliton @xcite . we observe no such excitations in our simulations . two - dimensional , conservative topological solitons are stable but they have higher energy than the conservative non - topological soliton @xcite , providing one possible explanation for this behavior . another possible explanation lies in the form of the spin torque term . in contrast to the _ local _ form for the spin torque considered here , single - layer nanocontact simulations that incorporate a _ nonlocal _ spin torque have demonstrated the formation and numerical stability of a precessional vortex in high - symmetry configurations without perpendicular anisotropy @xcite . a more realistic model of the spin torque may favor a topological soliton in certain cases . unfortunately , we as yet have no detailed physical understanding of either the dissipative droplet asymmetry or the drift instability in the presence of the combined effects of an oersted field and a tilted applied field . in part , our lack of understanding stems from the limited theory for the spatial propagation of the droplet @xcite . extension of the theory presented here to cases where the droplet is accelerated by field gradients and other such forces may provide more physical insight to aid in understanding why the droplet is subjected to displacing forces . the physical appearance of the droplet is reminiscent of the magnetic bubble that was once the subject of intense investigation as a possible alternative to ferrite - core computer memory @xcite . indeed , we can identify the zero - frequency droplet as a topologically trivial magnetic bubble @xcite with a winding number of zero , though in the theory presented here , the droplet frequency approaches zero only in the limit of infinitely large diameter . however , we expect that the droplet will converge to the bubble structure at a finite diameter , contingent on the inclusion of non - local magnetostatic energy in the calculation ; even when considering free layer films as thin as 3 nm , such a term will eventually overcome the positive contribution of exchange energy to the oscillator frequency , stabilizing the droplet as a static structure . in this sense , then , we can think of the conservative droplet soliton , a delicate balance between exchange and perpendicular anisotropy , as a dynamically collapsing bubble , and the dissipative droplet soliton as an imminently collapsing bubble that is critically stabilized by the localized injection of spin torque . we have derived equations for a dissipative droplet soliton through an asymptotic analysis of the landau - lifshitz - slonczewski equation for a point - contact spin torque oscillator with perpendicular anisotropy in the free layer . the droplet soliton is a localized , dynamic , solitary wave solution consisting of partially reversed magnetization directly under the contact and a zone of large amplitude precession in a region bounding the reversed magnetization . the diameter of the precessing boundary is approximately the contact diameter . the droplet frequency is always strictly less than the ferromagnetic resonance frequency for the film , and is also a monotonically decreasing function of droplet diameter . the balance between spin torque and damping required to sustain the droplet determines the relation between bias current and frequency . the instability of small amplitude solutions that leads to formation of the reversed domain in the droplet requires a minimum perpendicular anisotropy that is a function of the contact radius and spin torque asymmetry . some of the unique , identifying properties of the dissipative droplet that could be observed experimentally include * the droplet s frequency is well below the ferromagnetic resonance frequency . * sufficiently far from the minimum sustaining current , the droplet s frequency has a weak dependence on current . * the droplet may manifest a drift instability , which would reveal itself as a transitory cessation in ac oscillations until the droplet again forms . * hysteresis in current is expected , unless a drift instability results in a finite droplet lifetime . we have investigated the nucleation and stability of the dissipative droplet soliton through numerical simulations . we find that droplet formation begins once the current in the point - contact is sufficient to instigate the small amplitude slonczewski mode , characterized by spin waves that radiate away from the point - contact . for sufficiently strong perpendicular anisotropy , this mode is subject to a modulational instability and rapidly evolves into the reversed magnetization profile of the droplet soliton . we find that the droplet is stable in certain parameter regimes with regard to the inhomogeneous oersted field and to variations in spin torque asymmetry and applied field angle . finally , the droplet is subject to a drift instability that is a complicated function of the parameters employed in this theory . here we provide the details of our stability analysis for small amplitude , modulated waves excited in a nanocontact . inserting the ansatz ( [ eq:33 ] ) into eq . ( [ eq:32 ] ) and considering the leading order behavior in @xmath260 gives the linear slonczewski eigenmode @xmath227 satisfying @xcite @xmath261 slonczewski considered both @xmath226 and @xmath228 as eigenvalues for this equation and found the @xmath262 solution @xmath263 where @xmath264 is a bessel function and @xmath265 is a hankel function . the inner and outer wavenumbers @xmath266 , @xmath267 are @xmath268}{1 + i \alpha } } , \\ { k_\textrm{o}}&= -\sqrt{\frac{{\omega_\textrm{s}}- i\alpha(h_0 + { h_\textrm{k}}- 1)}{1 + i \alpha } } . \end{split}\ ] ] since @xmath269 for @xmath270 , the sign of @xmath267 has been chosen so that @xmath271 , and @xmath230 experiences exponential decay due to material damping @xmath272 . the decay length is weak compared to the contact radius . the two real eigenvalues @xmath226 and @xmath228 are determined by solving the complex valued transcendental equation @xmath273 which results from continuity of the first derivative of @xmath230 . continuing the asymptotic analysis to the next order gives the following nonhomogeneous equation @xmath274|f |^2 f \\ & + f \big [ \frac{d^2}{d\rho^2 } + \frac{1}{\rho } \frac{d}{d\rho } \big ] |f|^2 \\ & - |f|^2 \big [ f '' + \frac{1}{\rho } f ' \big ] + 2 i \alpha | f ' |^2 f \bigg \ } . \end{split}\ ] ] since the kernel of the adjoint of @xmath275 is spanned by @xmath276 ( @xmath277 denotes complex conjugation ) , we invoke the solvability condition of eq . ( [ eq:38 ] ) @xmath278 to determine the dynamical equation for @xmath224 in eq . ( [ eq:40 ] ) . the complex valued linear and nonlinear coefficients @xmath233 and @xmath220 are @xmath279 ^ 2}{1 - [ { k_\textrm{i}}/{k_\textrm{o}}]^2},\end{aligned}\ ] ] and @xmath280 \int_0^\infty |f|^2 f^2 \rho d \rho - i \frac{{\sigma_\textrm{s}}}{(1 + \nu)^2 } \int_0^{\rho _ * } |f|^2 f^2 \rho d\rho - ( 1 - 2i \alpha ) \int_0^\infty |f'|^2 f^2 \rho d \rho}}{\rho_*^2 j_1({k_\textrm{i}}\rho_*)^2[1 - ( { k_\textrm{i}}/{k_\textrm{o}})^2 ] } .\ ] ] we now perform a linear stability analysis of the time - periodic solution @xmath281 in eq . ( [ eq:43 ] ) by inserting @xmath282 into eq . ( [ eq:40 ] ) . keeping only the terms linear in @xmath283 and @xmath284 gives the simple , decoupled dynamical system @xmath285 which experiences exponential growth when @xmath237 . the numerical method we have used to solve eq . ( [ eq:1 ] ) is similar to the method tersely presented in ref . but adapted to a trilayer nanocontact . in this appendix , we describe our method in detail . in brief , we use a polar coordinate grid and cylindrical magnetization basis . angular derivatives are computed by use of a pseudospectral , fourier method while radial derivatives on a nonuniform grid are computed by use of finite differences . for time - stepping , we use an explicit , hybrid @xmath286-order runge - kutta time stepper with error control , ` ode23 ` from matlab@xmath77 . the polar coordinate system is a particularly efficient and accurate choice for nanocontact simulations . the discretization we use is non - uniform in radius ( `` inner '' and `` outer '' grids ) @xmath287 \\ & + { d\rho_\textrm{in}}\big \ } \,d\xi , \end{split}\ ] ] where @xmath288 , and @xmath289 , @xmath284 are parameters determining the location and width of the smooth change from the fine inner grid spacing @xmath290 to the coarser outer grid spacing @xmath291 . we typically have @xmath292 . for the angular , pseudospectral discretization , we expand the cylindrical magnetization basis in a truncated fourier series with @xmath293 fourier modes @xmath294 . \end{split}\ ] ] the transformation from an equispaced angular grid @xmath295 with magnetization coefficients @xmath296 evaluated at the discrete grid @xmath297 , to the fourier representation in eq . ( [ eq:60 ] ) is achieved by use of the fast fourier transform . the advantage of this discretization is that we can solve on a uniform computational grid while the physical grid is clustered in and around the nanocontact , where the majority of the fine scale dynamics occur . the outer grid supports the propagation of spin waves of the appropriate wavelength away from the nanocontact . in order to simulate an infinite domain , we choose a finite domain large enough , @xmath298 , so that spurious waves are naturally damped to a negligible amplitude . then their reflection off the boundary does not affect the strongly localized dynamics near the nanocontact . computing the vector laplacian @xmath18 is the computationally intensive portion of the algorithm . angular derivatives are approximated by multiplication in fourier space @xmath299 . the approximation of radial derivatives using finite differences requires some care , especially near the origin . we discuss the details now . due to the non - uniform radial grid ( [ eq:64 ] ) , radial derivatives in computational space require appropriate factors of @xmath300 and its derivatives . for example @xmath301 the radial derivatives in computational space are approximated using @xmath302 order finite differences . the boundary condition at the artificial boundary @xmath303 for eq . is the neumann condition @xmath304 , which is implemented by use of a standard ghost point method and one - sided differences . for radial derivatives near the origin at @xmath305 , we define symmetry conditions to obtain @xmath306 by taking @xmath307 : @xmath308 where we have used @xmath309 and @xmath310 when @xmath311 . therefore , @xmath312 are even / odd functions of @xmath313 as @xmath314 is odd / even while @xmath315 is even / odd as @xmath314 is even / odd . at @xmath316 , we take ( see ref . ) @xmath317 the derivative conditions in eq . ( [ eq:63 ] ) are approximated by use of one - sided differences to obtain an estimate of @xmath318 and @xmath319 . the symmetries in eq . ( [ eq:61 ] ) and the conditions in eq . ( [ eq:63 ] ) enable the use of centered finite differences , even near the origin . an explicit , runge - kutta @xmath286-order time - stepping method is used to advance the discretized version of equation forward in time while renormalizing the magnitude of @xmath40 after every time - step to preserve the constraint @xmath320 . to avoid severe time - step restrictions due to the small grid spacing near the origin ( @xmath321 ) , we apply a smooth , radial grid dependent angular mask that effectively reduces the number of angular modes at @xmath322 from @xmath293 to @xmath323 : @xmath324 the mask s parameters are the radial grid dependent wavenumber cutoffs @xmath325 , @xmath326 and the cutoff width @xmath327 . application of the mask at every time step filters out numerically induced small wavelengths near the origin @xcite . the cutoffs @xmath325 are chosen so that the approximate grid spacing @xmath328 , hence the grid near the origin has an effective spacing of @xmath290 . the mask applied to @xmath315 takes the form @xmath329 care must be taken when applying the mask to the in - plane fourier coefficients @xmath312 because @xmath330 and @xmath331 depend on the grid location . we use @xmath332 \widehat{m}^{(\rho)}_{n , k}(t ) \\ & + i [ -g_{n}(k-1 ) + g_{n}(k+1 ) ] \widehat{m}^{(\varphi)}_{n , k}(t ) \big \ } , \\ \mathcal{g}^{(\varphi ) } \left \ { \widehat{m}^{(\varphi)}(t ) \right \}_{n , k } & = \frac{1}{2 } \big \ { i [ g_{n}(k-1 ) - g_{n}(k+1 ) ] \widehat{m}^{(\rho)}_{n , k}(t ) \\ & + [ g_{n}(k-1 ) + g_{n}(k+1 ) ] \widehat{m}^{(\varphi)}_{n , k}(t ) \big \ } . \end{split}\ ] ] numerical parameters we use are : @xmath333 , @xmath334 , @xmath335 , @xmath336 , @xmath337 , @xmath338 , @xmath339 , @xmath340 , and @xmath341 . we find no significant change in the presented results for more accurate grids and filtering parameters . in order to nucleate a droplet , we use initial conditions that are saturated in the @xmath342 direction with a current that is above threshold . the small amplitude instability investigated in sec . [ sec : excitation - droplet ] leads to the formation of a droplet . for the computation of the frequencies of the droplet in fig . [ fig : droplet_f_i_oe ] found from micromagnetics , we take the fourier transform of the spatially averaged magnetization time series from the simulation and extract the frequency with the largest power . the frequency resolution is finer than the size of the dots in fig . [ fig : droplet_f_i_oe ] .

nv centers in diamond are generally recognized as highly promising as indefinitely stable highly efficient single - photon sources . we report an experimental quantification of the brightness , radiative decay rate , nonradiative decay rate and quantum efficiency of single nv centers in diamond nanocrystals . our experiments show that the commonly observed large spread in fluorescence decay rates of nv centers in nanodiamond is inconsistent with the common explanation of large nanophotonic mode - density variations in the ultra - small high - index crystals at near - unity quantum efficiency . we report that nv centers in 25 nm nanocrystals are essentially insensitive to local density of states ( ldos ) variations that we induce at a dielectric interface by using liquids to vary the refractive index , and propose that quantum efficiencies in such nanocrystals are widely distributed between 0% and 20% . for single nv centers in larger 100 nm nanocrystals , we show that decay rate changes can be reversibly induced by nanomechanically approaching a mirror to change the ldos . using this scanning mirror method , for the first time we report calibrated quantum efficiencies of nv centers , and show that different but nominally identical nanocrystals have widely distributed quantum efficiencies between 10% and 90% . our measurements imply that nanocrystals that are to be assembled into hybrid photonic structures for cavity qed should first be individually screened to assess fluorescence properties in detail .

1212.5172

the large promise of quantum - optical technologies to enable secure communication and novel computation architectures sets stringent targets for the quality of photon sources and detectors . in order to meet these demands many efforts are currently devoted to realizing bright sources of single photons @xcite . these developments at the same time require novel nanophotonic engineering designs around emitters to enhance light - matter interaction strength as well as indefinitely stable two - level systems that neither bleach , blink , nor spectrally jump . on the nanophotonic engineering side , many different systems have been proposed to control whereto , how fast and with which polarization an emitter emits provided that one manages to locate it exactly in the right location . these systems include whispering - gallery - mode cavities @xcite , micropillars and cylindrical wires @xcite , photonic crystal microcavities @xcite , photonic crystal waveguides near cut - off @xcite , anderson localizing systems @xcite , as well as ultrabroadband plasmonic waveguides and antennas @xcite . essentially , all these techniques modify the photonic environment of an emitter via the local density of optical states ( ldos ) @xcite . the ldos quantifies the light - matter interaction strength that appears in fermi s golden rule for spontaneous emission . placing the emitter at a position where the ldos is enhanced , first implies a much higher fluorescence decay rate . second , if the ldos enhancement is due to a select set of designed modes as in a cavity or waveguide , the enhanced rate is accompanied by extraction of photons preferentially via these enhanced modes . thus , photonic engineering promises optimum brightness , dynamics , and collection efficiency . as regards the choice of emitter , a wide variety of systems have been used . unfortunately , all choices appear to carry large disadvantages when going beyond pilot studies : most dye molecules photobleach @xcite , ii - vi quantum dots @xcite blink as well as bleach , and many systems , such as iii - v emitters , only have desirable properties when cooled to cryogenic temperatures @xcite . a promising candidate to provide an indefinitely stable source @xcite that furthermore allows room - temperature spin - control @xcite is the nitrogen - vacancy ( nv ) color center in diamond . when one chooses diamond nv centers as emitters for quantum optics in nanophotonic devices , one can either aim to manipulate emission by fabricating photonic structures directly in diamond @xcite , or one can assemble diamond nanocrystals with photonic structures of a different material @xcite recently , pick - and - place strategies @xcite were reported in which a single nanoparticle from a diluted powder of diamond nanocrystals dispersed on a substrate is selected and pushed to a desired location by a scanning probe , such as an atomic - force - microscopy tip , or a manipulator in a scanning electron microscope . reports of this technique span from the coupling of nanodiamonds to photonic - crystal cavities @xcite to coupling to plasmonic antennas and nanowires @xcite . as a variation of the pick - and - place strategy , a few groups recently developed so called ` scanning - emitter ' near - field microscopes in which nanosources are not deposited irreversibly inside a nanostructure , but actually remain attached to a scanning- probe @xcite . the advantage of such a scanning probe approach is that one can first construct a full map of the ldos using near - field lifetime imaging to determine where one should ultimately place the nanosource @xcite . microscopy with a light - source as a near - field tip is interesting as a microscopy technique , but only viable if the nanosource is indefinitely stable and does not blink , for which nanodiamonds appear the sole candidate @xcite . moreover nanodiamond scanning probes offer the possibility to create nanometer sized local magnetometry probes read out optically by nv spectroscopy @xcite . in view of the scanning probe microscopy and assembly efforts that seek to combine nanodiamonds and photonic structures , one question stands out as of key importance : given an ensemble of individual nanodiamonds , how does one recognize the ideal nanocrystal ? this question is especially relevant given that both fabrication of the photonic structure _ and _ the intended scanning - probe procedure is highly laborious @xcite . naively , one might think that all nanocrystals of subwavelength size that contain a single nv center will be equally suited , since an nv center is a defect of uniquely defined composition and geometry in the diamond crystalline matrix . however , many workers on diamond nanocrystals have established that nv centers in diamond nanocrystals actually show a distribution of photophysical properties , such as brightness , stability , and decay rates @xcite . this distribution is usually ascribed to the fact that even though all nv centers are expected to have unit quantum efficiency and the same oscillator strength , these identical unit - quantum - efficiency sources are each differently placed inside their nanoscopic high - index diamond grains @xcite . this is anticipated to cause different decay rates , due to the fact that the ldos even in an isolated nanoscopic object varies as a function of position and dipole orientation @xcite . in this paper we address the question how to recognize the ideal nanocrystal from an ensemble of nanocrystals on basis of the requirement that the ideal nanocrystal must contain a single nv@xmath0 center that fluorescences with a high quantum yield , so that it can be useful as a reporting probe of ldos in nanophotonic systems . we report that the commonly made assumption that nv centers are unit - efficiency dipoles randomly distributed in high index nanoscopic objects is not reconcilable with measured brightness and decay rate histograms data for nanocrystals in the frequently used size ranges around 2050 nm and 50150 nm . on basis of the wide distribution of brightness and rate that can not be explained by ldos variations between nanoparticles alone we conclude that both the radiative and nonradiative rates are broadly distributed . to quantify this distribution , we have performed experiments in which we measure emission rate changes of individual nanodiamonds as we controllably vary the ldos of their environment . for small nanocrystals we find a decay rate distribution indicative of low - quantum - efficiency emitters . furthermore , consistent with this low quantum efficiency , we find no evidence that small single nv centers are responsive ldos probes when applying moderate ldos variations . for the larger nanocrystals ( 100 nm diameter ) , we for the first time manage to induce reversible changes of up to 25% in the total decay rate of single nv centers using a calibrated ldos change induced by a nanomechanically moved mirror . on basis of these measurements we argue that the apparent quantum efficiency of nanodiamonds in the size range 50150 nm ranges widely from about 10% to 90% . this wide range of quantum efficiencies implies that prior to constructing a photonic structure , it is necessary to screen the nanocrystals using a calibrated quantum- efficiency measurement on the single nv center level , as shown in this paper . this paper is organized in the following manner . first we report on our experimental methods in section [ sec : setup ] . next , we discuss the distributions of brightnesses and rates for large nanocrystals in section [ subsec : distriblarge ] and for small nanocrystals in section [ subsec : distribsmall ] . finally we report on efforts to tune rates of nv centers in small diamonds by controlled ldos changes . the two methods used are liquid tuning of ldos in section [ sec : liquidldos ] , and nanomechanically approaching a mirror as reported in section [ sec : mirrorldos ] . since single nv centers in nanodiamonds are comparatively dim emitters that require high excitation powers , it is essential to avoid background fluorescence when performing fluorescence microscopy . therefore , we use intrinsically low - fluorescent quartz coverslips ( esco products ) as sample substrates that were cleaned by @xmath1 sonication in water followed by a @xmath1 bath in base piranha ( nh@xmath2(aq.,@xmath3):h@xmath4o@xmath4(aq.,@xmath3):h@xmath4o mixed at ratio 1:1:5 , at @xmath5c ) . in our work it is essential that we can unambiguously pinpoint the position of nanodiamonds containing nv centers in the course of the measurement , so that we can revisit the same color center after , for instance , material - deposition steps . to this end we define an array of markers on top of the coverslips using electron - beam lithography . to remove any organic residues after the lift - off process , we treated the coverslips with a mild @xmath6-plasma descum ( oxford instruments plasmalab 80 + , using a 5mtorr , 25sccm @xmath6 plasma ) . as sources for nv centers , we use solutions of monocrystalline synthetic nanodiamonds ( microdiamant msy ) with a narrow size tolerance obtained from microdiamant ag , lengwil , switzerland . we note that these nanoparticles have been used recently by a large number of groups in experiments that rely on fluorescence @xcite . we use these nanodiamonds exactly in the manner as described by schietinger , i.e. without further washing or purification steps @xcite . we prepared samples from diamond slurries with two different size distributions by spin - coating aqueous solutions of nanodiamonds that are diluted to a concentration of 1% of the as - received stock solutions on the cleaned and patterned coverslips . the first type of sample , from here on referred to as ` 25 nm diamond sample ' was made from microdiamant msy 00.05 , which nominally has sizes from 0 to 50 nm . these diamonds have a median diameter of 26 nm , with fewer than 1% of the diamonds above 50 nm in size , according to particle sizing performed on this batch of diamonds by the manufacturer . the second type of sample ( ` 100 nm diamond sample ' ) was made from microdiamant msy 00.2 , which nominally has crystal sizes ranging from 0 to 200 nm . this ensemble has crystals with median diameter 108 nm , and fewer than 1% of particles above 175 nm in size , again according to the size - distribution histogram supplied by the manufacturer . scanning electron microscope ( sem ) inspection indicated a size distribution of nanodiamonds consistent with these specifications . the average density of spin - coated nanodiamonds on the coverslips as checked through the sem images was about 1 to 8@xmath7 depending on the sample , assuring that only a few nanodiamonds are illuminated at the diffraction - limited focus of the objective . only a small fraction of the nanodiamonds are fluorescent , and an even smaller fraction fluoresces due to an nv center . we identify a diamond as containing at least one nv center , if we can clearly identify the characteristic zero phonon line ( zpl ) @xcite in its emission spectrum according to the criterion we specify in section [ single_nv_charac ] . with this criterion , for 100 nm nanodiamonds at an average density of 1@xmath7 , we found on average one nv center in an area equal to @xmath8@xmath9 , while for 25 nm nanodiamonds at an average density of 5@xmath7 we found on average 2 nv centers in a @xmath10@xmath11 area . these numbers translate to identifying fewer than 0.05% of the 100 nm nanodiamonds , and fewer than 0.001% of the 25 nm diamonds as containing a ( single ) nv center beyond doubt . schietinger @xcite have reported a higher density of nv centers of about 1% for msy 00.05 nanodiamonds . a difference in reported nv center densities could be either due to a different degree of strictness in labeling a fluorescent emitter as an nv center beyond doubt ( see section [ single_nv_charac ] for our criteria ) , or alternatively to batch - to - batch variations in nanodiamond slurries . dry objective . the emitted fluorescence can be guided to spectrometer or to apds to measure the lifetime and second order correlation function @xmath12 of the emitted photons . [ fig : setup ] , width=302 ] the optical setup , sketched in figure [ fig : setup ] , consists of an inverted confocal fluorescence microscope equipped with a sample scanning piezo stage . nv centers are optically excited at 532 nm using either a frequency - doubled nd : yag pulsed laser ( time - bandwidth , 10mhz repetition rate ) or a continuous - wave ( cw ) diode laser ( cni ) . the laser beam is focused through the coverslip to a diffraction - limited spot on top of the sample , using a @xmath13 dry objective with a numerical aperture of 0.9 ( nikon cfi plan fluor ) . the same objective collects the luminescence and guides it through a long - pass filter ( 580 nm cut - off ) before it is imaged on a ccd camera ( nikon ds - qi1mc ) or confocally detected on one or two avalanche photodiodes ( apd ) with single - photon sensitivity ( both apds : i d quantique id100 - 20uln ) . the apds in combination with a sub - nanosecond - resolution 16 channel correlator ( becker & hickl , dpc-230 ) allow for single - photon counting . the correlator can perform time - correlated single - photon - counting ( tcspc ) lifetime measurements by correlation of detection events with laser pulse arrival times , or photon - photon correlations using multiple apds . we use a second apd in a hanbury - brown and twiss ( hbt ) configuration to measure photon correlation statistics ( antibunching ) at cw excitation . single - nanocrystal spectra are collected using an imaging spectrometer ( spectrapro 2300i ) equipped with a thermoelectrically cooled back - illuminated si ccd array detector ( princeton instruments pixis:100b ) . of an nv center obtained by cw excitation . the zero dip in @xmath12 curve confirms the single nature of the center . ( d ) background - corrected decay traces of photons emitted from an nv center ( red circles , total acquisition time 20 s ) with a single - exponential fit ( black curve ) to extract the lifetime . [ fig : fig_1 ] ] in order to find nanodiamonds containing nv centers , the sample is first pumped with wide - field laser illumination and the fluorescence is imaged on the ccd camera . under these conditions , any potential nv center will appear as a diffraction - limited fluorescent spot . subsequently , we switch to diffraction - limited illumination and position the potential candidate at the laser focus using a piezostage ( figure [ fig : fig_1](a ) ) . the fluorescence intensity of the emitter is quantified by integration over a fixed ccd area which is indicated by the green line in figure [ fig : fig_1](a ) . at this stage , we spectrally resolve the fluorescence from the isolated emitter with the spectrometer to specifically find the characteristic zero phonon line of nv centers at a wavelength of about 638 nm ( figure [ fig : fig_1](b ) ) @xcite . on basis of the zpl , we either discard the emitter ( absence of zpl ) , or identify the nanodiamond as having one or more nv centers ( zpl well above the detector and photon - counting noise ) . for each identified nv center , we record a second - order correlation function @xmath14 to verify that the emission originates from a single center . to this end , we pump the center with the cw laser at a power of about 1mw to achieve count rates on the order of about @xmath15 counts per second on each apd and integrate for about 1000s to obtain reasonable correlation statistics of around 30 coincidences per 658ps bin . figure [ fig : fig_1](c ) shows a typical background - subtracted @xmath14 curve , with a clearly resolved minimum at zero delay @xmath16 that is well below 0.5 , indicating that emission is from a single nv center . here , the background is subtracted via the procedure reported in @xcite , based on a measurement directly next to the nv center where we expect identical fluorescence background from the substrate . the background measured only from the substrate shows a poissonian emission statistics . the bunching in the @xmath14 in figure [ fig : fig_1](c ) at longer delay times ( visible at @xmath17 50ns ) is well known to be due to the presence of a shelving state @xcite . once we have identified a nanodiamond as containing a single nv center based on its spectrum and antibunching signature , we measure its fluorescence lifetime by switching to pulsed laser excitation . the lifetime of each center is extracted by fitting a single - exponential decay incorporating a small constant background to the decay histogram of emitted photons as shown in figure [ fig : fig_1](d ) . we note that the substrate itself has a weak background fluorescence that shows a time dependence . we correct for this artifact by collecting decay traces from the sample pumped directly next to the nv center . we parameterize the background by a tri - exponential fit ( with a dominant sub - nanosecond lifetime ) which we subtract from the nv center decay histogram prior to fitting . the residual background is around 10 counts per bin for a total acquisition time of about 20s . in this section we report on statistical distributions of the various fluorescence characteristics , i.e. brightness , @xmath12 and lifetime , that we collected on both nanodiamond size ranges . we first discuss the ` 100 nm diamond ' sample , containing 0200 nm size nanodiamonds with 108 nm median diameter , and subsequently we discuss the ` 25 nm diamond ' sample with 050 nm size nanodiamonds and 26 nm median diameter . for each sample , our statistics is based on identifying 30 to 40 nv centers as described in section [ single_nv_charac ] . as a first step in the characterization process , we have measured the brightness of nanodiamonds that show a clear zpl line . figure [ fig : fig_2](a ) shows a histogram of measured brightnesses as quantified by intensity on the ccd at fixed illumination intensity ( solid bars ) . the intensity on the ccd is obtained by summing all pixels within the diffraction - limited image of each diamond ( green line in figure [ fig : fig_1](a ) ) . we observe a wide distribution of intensities , spanning from @xmath19 to @xmath20 adu s ( analog digital units ) on the ccd per 300ms exposure time . at an estimated photon - to - adu conversion factor of 6 for our ccd camera , these brightnesses correspond to @xmath21 to @xmath22 collected photons per second , at a pump power around @xmath23 supplied by the cw laser . first , these numbers show , at least assuming that nv centers are reasonably efficient emitters , the absorption cross sections of nanodiamond nv centers at @xmath24 nm pump wavelength are an order of magnitude below those of dyes and ii - vi quantum dot nanocrystals . second , the wide distribution of brightnesses of over a factor 3 to 4 from nominally identical emitters is highly surprising . as described in section [ single_nv_charac ] , some nanodiamonds may have multiple nv centers , especially in the case of big nanoparticles . to exclude that brightness variations are due to multiple nv centers , we screen the nanodiamonds on basis of @xmath12 minimum values . figure [ fig : fig_2](b ) shows a correlation plot , plotting the fluorescence intensities of all identified nv centers on 100 nm nanodiamond sample versus the recorded value of @xmath12 at zero delay ( @xmath25 ) . we find that not just the brightness , but also the minimum in @xmath12 is distributed with minima in @xmath12 ranging from 0 to 0.75 . the minimum in @xmath18 is only weakly correlated with collected intensity , especially via an apparent stepwise increase in the fluorescence intensity as @xmath18 reaches above @xmath26 . such a stepwise increase would be expected since the minimum in @xmath12 scales with the number of emitters @xmath27 as @xmath28 . for @xmath29 more than a single fluorescent center is involved in the emission , resulting in higher fluorescence intensities on average . in the remainder we concentrate our analyses on single nv centers , i.e. those nanocrystals for which @xmath30 . in figure [ fig : fig_2](a ) , we overplot a histogram of fluorescence intensities of the subset of centers with @xmath30 as patterned bars . the histogram shows a wide fluorescence distribution , with a factor of three difference between brightest and dimmest nv centers , and with a relative distribution of about 30% around the most frequent value . figure [ fig : fig_2](b ) shows that for the single nv centers the wide distribution in fluorescence intensities notably does not show any correlation to @xmath18 values . previously , beveratos @xcite attributed the distribution of @xmath18 values to inseparable background emission from the nanodiamond crystal host . the lack of correlation between brightness and @xmath18 suggests that this assertion is not valid , i.e. that differences in the nv emission itself causes the brightness distribution . it should be noted that brightness differences between nv centers could occur due to the fact that the linearly polarized pump field projects differently on the nv center absorption dipole moment . according to ref . @xcite , an nv center has two dipole moments of equal size in the plane perpendicular to the nv axis . for a random distribution of nv orientations @xmath31 , the dipole projection @xmath32 on a linearly polarized pump field @xmath33 would lead to a brightness distribution that is strongly asymmetric , skewed strongly to high brightness . in contradistinction , the brightness histrogram we find for nv centers is not skewed to high brightness . to conclude , the single nv centers display a variation in brightness over a factor 3 , which can not be attributed to either background emission of the crystal host or variations in absorption cross - section through dipole orientation . in ( b ) . ( b ) fluorescence intensity as a function of @xmath34 . ( c ) bars : histogram of decay rates of emitters with @xmath35 . solid line : calculated distribution of @xmath36 according to nanodiamond size distribution , normalized to the rate of nv centers in bulk diamond . dotted line : free scaling of the calculated ldos distribution ( solid line ) to match the experimental histogram ( bars ) . ( d ) fluorescence intensity as a function of decay rate for emitters with @xmath35 ( decay rate error bars are within symbol size ) . error bars are comparable to plot symbol size.[fig : fig_2],width=566 ] on basis of the large brightness distribution that can neither be explained by dipole orientation nor by background fluorescence , we hypothesize that there is a substantial variation in quantum efficiency ( qe ) of the nv centers . to examine whether quantum efficiency effects may be at play , we have measured fluorescence decay rates for each identified single nv center . we plot a histogram of the measured decay rates in figure [ fig : fig_2](c ) . we observe a very wide distribution , with the slowest emitters decaying almost 4 times more slowly than the fastest ones . the most frequently occurring decay rate is around @xmath37ns@xmath38 ( corresponding to about 33ns ) . the time constant of 33ns is appreciably slower than the fluorescence lifetime of nv centers in bulk diamond , for which the accepted literature value is 11.6 ns @xcite . this much slower decay as well as the occurrence of a distribution of rates is in agreement with previously reported values @xcite . if the hypothesis that quantum efficiency variations are responsible for the large variability in brightness in figure [ fig : fig_2](a ) is valid , one might expect a correlation between the brightness and decay rate of the emitters . we plot the measured brightness as a function of decay rate for each nv center in figure [ fig : fig_2](d ) which , however , displays no clear correlation . importantly , we note that the measured quantity here is the total decay rate @xmath39 which reflects both variations of radiative decay rate @xmath40 and nonradiative decay rate @xmath41 , whereas the quantum efficiency is given by @xmath42 . based on the collection efficiency of our microscope objective ( @xmath43 % ) , the count rates in figure [ fig : fig_2 ] correspond to an intermediate ( most data ) to strong ( data point at and above 150 kadu/300 ms))excitation regime , where the photon emission rate is approximately 0.2 to 0.9 times the total decay rate @xmath44 . both if one assumes to be in saturation , and below saturation , one expects the fastest decay to imply highest brightness if one assumes @xmath41 to be approximately constant but @xmath40 to be distributed . this conclusion is not strongly supported by the data . if conversely we assume @xmath40 to be approximately constant across nv centers while @xmath41 is distributed , one would expect highest brightness to correlate with the slowest decay rate ( lowest @xmath41 ) , a hypothesis also not strongly supported by the data . specifically , we find that the subset of crystals around the most frequent brightness show the full spread of decay constants , while conversely also the subset of crystals that have decay constant around the most frequent rate , contain the full range of brightnesses . we conclude that the common assumption that nv centers in such large nanocrystals as we study are near - unity quantum efficiency sources is not supported by the data . instead _ both _ the radiative and the nonradiative decay rates could be widely distributed . many workers had already noticed that nanodiamond decay rates are widely distributed @xcite . the variation is commonly attributed to variations solely in the radiative rate @xmath40 due to a local - density - of - photonic - states effect , assuming zero nonradiative decay ( i.e. unit quantum efficiency ) . the hypothesis , explained in detail by inam @xcite , is that variations are in large part due to the fact that the ldos experienced by an nv center is influenced by the nanoscale geometry of its environment , i.e. the fact that the source is situated inside , and close to the surface of , a very high index nano - object that is embedded in a low index environment . a distribution in ldos can arise from the size distribution of nanodiamonds and from the fact that different nv centers have different dipole - moment orientation and positions within the crystals @xcite . here we assess if this ldos hypothesis is quantitatively reasonable . considering that the radiative decay rate @xmath45 is proportional to ldos , we evaluate the @xmath45 distribution by calculating the ldos distribution assuming spherical nanoparticles , for which the ldos is analytically known @xcite . we modeled nanodiamond nv centers as point dipoles with randomly oriented dipole moments homogeneously distributed in position in a set of dielectric spheres with dielectric constant equal to the bulk diamond value ( @xmath46 ) . we furthermore take into account the distribution of particle size as specified according to the nanodiamond size - distribution histogram provided by the manufacturer . we histogram the occurrence of ldos values to find its probability distribution over all sphere sizes , dipole positions and dipole configurations . the resulting ldos distribution can be converted into a distribution directly comparable to experimentally measured decay rates by scaling the ldos to the previously reported decay rate of ( 11.6 ns)@xmath47ns@xmath38 of nv centers in bulk diamond @xcite . we note that this entire procedure involves no adjustable parameter or any fit to data . the final result is plotted in figure [ fig : fig_2](c ) as the blue solid line . the calculated histogram correctly predicts that emission is significantly decelerated compared to decay in bulk diamond , consistent with the fact that decay in small dielectric spheres is decelerated both compared to vacuum , and bulk dielectric . however , we observe that the calculated histogram peak falls at much lower decay rate than the experimentally measured histogram peak , with a discrepancy amounting to a factor of 3 . a similar discrepancy between the calculated and experimentally measured decay rates of nanodiamond nv centers was recently reported for samples with a much wider size distribution , centered at much larger median size , measured by inam @xcite . we note that this discrepancy can not be attributed to the fact that we have taken particle shape to be simply spherical , and that we have neglected the presence of a substrate . these effects cause only small changes in the expected decay rate histogram , as verified in fdtd simulations by inam @xcite . one might further argue that the occurrence of a degenerate in - plane dipole moment @xcite could skew the histogram of expected decay rates towards higher values , if one assumes the dipole moment is free to diffuse prior to de - excitation . however , if we just histogram the fastest rate instead of the average rate at each possible nv center position , the resulting histogram also does not lead to a consistent explanation ( not shown ) . we conclude that ldos variations in nanodiamond under the hypothesis of unit quantum efficiency and a bulk decay rate of ( 11.6 ns)@xmath38 do not explain the variation in measured decay rates . inam proposed that although calculated and experimental absolute decay values are inconsistent on basis of ldos theory and the bulk rate in diamond , the calculated and experimental results can be scaled onto each other . indeed , if we scale the reference rate that sets the rate axis for the calculated histogram peak not by 11.6ns ( rate constant in bulk diamond @xcite ) , but by a factor of 3 shorter , calculated and measured histogram coincide reasonably ( figure [ fig : fig_2](c ) black dotted line ) . such a scaling would imply as hypothesis a unit quantum efficiency , together with an as yet hidden explanation that introduces a multiplicative correction factor in the calculated rate distribution . a second hypothesis could be that no adjustment should be made of the bulk rate that enters the comparison , but that the quantum efficiency of nv centers , while unity in bulk , is not unity in nanocrystals . indeed , an additive offset to the calculated histogram is introduced by nonradiative decay channels that do not occur for bulk diamond , but could occur for nanocrystals due to defects and the presence of large surface area that could contain quenching sites . so far , significant nonradiative decay was evidenced only for very small nanocrystals ( 5 nm @xcite ) . a distribution of nonradiative rates around @xmath48ns@xmath38 would shift the calculated histogram to the measured rates . the magnitude of the required @xmath41 implies that quantum efficiencies should be around 30% to 50% for the slowest nanocrystals in the measured ensemble ( assuming unit efficiency in bulk ) . a third hypothesis could be that the assumption that the bulk rate is entirely radiative to start with , is incorrect , i.e. that the quantum efficiency of nv centers in bulk is significantly below unity contrary to common assumption . assuming a bulk quantum efficiency of 70% instead of 100% would overlap the calculated histogram peak with the measured most frequent rate . however , this explanation would severely underestimate the width of the decay rate distribution unless a distribution of @xmath41 is at play . we conclude that ldos variations in nanodiamond alone do not explain the variation in measured decay rates , and that a distribution of radiative and nonradiative decay constants must be at play for nv centers in nanodiamond . furthermore we conclude that an actual experimental calibration of quantum efficiency of individual nv centers is highly desired , which we will return to in sections [ sec : liquidldos ] and [ sec : mirrorldos ] . ( shown in the inset ) according to nanodiamonds size distribution , normalized to the rate of nv centers in bulk diamond . solid black line : free scaling of the calculated ldos distribution ( blue line ) so to match the experimental histogram ( bars ) . [ fig : fig_3],width=566 ] for experiments in which nanodiamonds are intended as probes of , or sources to be embedded in , nanophotonic environments , a size smaller than that of the 100 nm nanodiamonds would be advantageous , as smaller size implies higher spatial resolution and a lower perturbative effect on the modes of the nanophotonic system @xcite . therefore , we repeated the brightness and decay rate statistical measurements for the 25 nm sample , i.e. the batch of crystals with median diameter 26 nm . for 25 nm nanodiamonds , background - subtracted @xmath12 measurements show a zero dip for all identified nv centers ( for a typical 25 nm nanodiamond @xmath12 see e.g. figure [ fig : fig_1](c ) ) . the fact that we found no nanodiamonds with multiple nv centers for this sample is commensurate with the smaller average crystal size . for these single nv centers , figure [ fig : fig_3](a ) shows a histogram of the fluorescence intensities measured on the ccd using cw illumination at 1 mw/@xmath49m@xmath50 . as in the case of the 100 nm sample , the histogram exhibits a wide fluorescence intensity distribution , in this case spanning a factor of about two . the most frequent brightness is approximately a factor two below that of the 100 nm sample . the fact that the histogram is skewed towards lower intensities suggests that the intensity distribution is not due to the projection of the dipole moment orientation on the pump - field polarization , as this would cause a skewing in the opposite direction . as in the case of the 100 nm nanocrystals , we have also measured the decay rate for all 25 nm nanocrystals that we identified . the decay rate distribution , plotted as a histogram in figure [ fig : fig_3](c ) , again shows a very wide distribution , with a factor of 4 contrast in decay rate . we find higher decay rates on average compared to 100 nm nanodiamonds , with a most frequent decay rate around @xmath51ns@xmath38 ( corresponding to about 28ns ) . the most frequent decay rate is approximately 25% faster than for the 100 nm nanodiamonds . exactly as in the case of 100 nm nanodiamonds , the wide distribution of fluorescence brightness and decay rates does not imply a correlation between the two . as confirmation , in figure [ fig : fig_3](b ) we plot the fluorescence intensities and decay rates of 25 nm diamond nv centers , where we find no clear correlation . we hence conclude that as for the larger diamonds , a distribution in absorption cross - sections , radiative and nonradiative rates is likely at play . to assess whether the common hypothesis that the distribution in rate is due to a distribution in ldos is valid for small nanodiamonds , we calculated the ldos distribution ( plotted in the inset of figure [ fig : fig_3](c ) ) also for the 25 nm sample taking into account the size distribution histogram measured by the nanodiamond supplier . the rate distribution expected from the calculated ldos scaled with the bulk rate ( blue solid line in figure [ fig : fig_3](c ) ) is first , considerably narrower than the experimental decay rate distribution , and second , at considerably reduced rate compared to the measured decay rates . the magnitude of the discrepancy in decay rate is approximately a factor 6 , i.e. twice larger than for the 100 nm nanodiamonds . a scaling of the calculated distribution by a multiplicative factor ( here a factor of about 6 ) as proposed by inam @xcite does not lead to a good correspondence as in the 100 nm case , as the relative width of the measured histogram far exceeds that of the scaled calculation ( black line in figure [ fig : fig_3](c ) ) . taking our data on both 100 nm and 25 nm nanodiamonds together , we hence do not find support for the hypothesis by inam that a hitherto hidden effect multiplies the radiative rate of nv centers in nanocrystals compared to bulk diamond . a more likely explanation that does not involve a scaling of @xmath40 due to an unknown origin is that nv centers are subject to a distribution of nonradiative rates on top of the ldos - induced radiative rate distribution . the measured distribution of 25 nm nanodiamond decay rates points at a wide distribution of nonradiative rates . according to this hypothesis , those crystals with decay around the most - frequent decay rate of @xmath52ns@xmath38 must have a quantum efficiency below 15 - 20% , twice lower than for the 100 nm nanocrystals . the overall lower quantum efficiency estimate is commensurate with the reduction in average brightness , and is also consistent with the fact that a larger sensitivity to nonradiative decay channels is potentially associated with the increased nanocrystal surface - to - volume ratio . our data indicate that nv centers even in nanocrystals as large as 100 nm across do not have unit quantum efficiency , and that there is a distribution of nonradiative and radiative rates . therefore , it is desirable to measure quantum efficiency , @xmath40 and @xmath41 independently on individual nv centers in order to decide if /which nv centers in nanodiamonds are suitable for use as ldos probes for integration in nanophotonic devices . in literature , several reports have appeared that evidence lifetime changes of nv emission that are induced by placing nanocrystals in photonic environments of varying index and topology @xcite . to our knowledge all these measurements of lifetime changes for nanodiamonds were performed by comparing the _ mean rate _ from an ensemble of single - center measurements in one system , to measurement on a different ensemble of single - centers in a second system @xcite . ruijgrok @xcite in particular report changes for a system in which the induced ldos change is exactly known , and report changes in the _ mean _ rate consistent with those expected for unit - quantum - efficiency emitters . however , we note that in these ensemble measurements , the change of the _ mean _ decay rate was far smaller than the width of the rate distribution . therefore , those measurements do not allow to ascertain whether for _ any given _ single nv center the rate actually varies with varying ldos in a manner that is consistent with the expectations for efficient emitters . we perform measurements in which we change the ldos of single nv centers in two ways . first , we have immobilized the nanocrystals by evaporating a thin layer of sio@xmath4 , and then introduced liquids with different refractive indices to systematically modify the ldos around the nv centers ( section [ sec : liquidldos ] ) . second , we have used nanomechanical tuning of ldos using a piezo - controlled mirror to effectuate a drexhage experiment @xcite on single nanodiamonds ( section [ sec : mirrorldos ] ) . the nanomechanical method is preferable since it rapidly allows continuous variation of ldos . due to their smaller apparent brightness , however , we were unable to apply the technique also to 25 nm diamond nanocrystals as the much higher pump energy required to reach appreciable count rates caused thermal breakdown of the micro - mirror . therefore we applied the liquid tuning method ( section [ sec : liquidldos ] ) to the small nanodiamonds , and report on the nanomechanical method for the large nanodiamonds ( section [ sec : mirrorldos ] ) . for the small nanodiamonds only the liquid tuning method could be applied due to experimental constraints . the idea and first implementation of this method was pioneered by snoeks @xcite , and later adopted by e.g. @xcite . sources are placed in close proximity to a planar interface between a dielectric and a half - space that can be filled with liquids of different refractive index . the advantage of using liquid tuning of refractive index for sources near an interface is that the ldos changes near an interface are nearly independent of emission frequency , and are known to be excellently described by the theory explained in full by urbach and rikken @xcite . in view of the broad emission spectrum of nv centers in diamond , it is important to apply a broadband ldos change when seeking to measure lifetime changes . a narrowband ldos variation as obtained with a high - q microcavity that would for instance be tuned to the zero phonon line , would not necessarily affect the rate , but rather only the branching ratio between the zero phonon line and the rest of the spectrum . we prepared the sample by evaporating a 60 nm thick layer of sio@xmath53 on one of the 25 nm nanodiamond samples on which we had identified single nv centers . this step immobilizes the nanodiamonds , and ensures that liquid application to tune ldos does not add new chemically induced nonradiative decay channels . next , we defined a 3 mm deep liquid reservoir on top of the sample using a ring - shaped enclosure cut from polydimethylsiloxane ( pdms ) bonded to the substrate . figure [ fig : fig_4](a ) depicts an schematic of the sample . in order to investigate the decay dynamics of the nv centers in response to the ldos variations , we randomly selected six nv centers from the ensemble of figure [ fig : fig_3 ] and measured their lifetimes in three conditions : before adding any liquids where the top half space of nanodiamonds consists of air ( @xmath54 ) , after adding water with refractive index of @xmath55 , and after adding isopropyl alcohol ( ipa ) with refractive index of @xmath56 to the liquid bath on top of the nanodiamonds . in the lower panel of figure [ fig : fig_4](b ) we plot the decay rate of each nv center as a black bar measured when the sample is dry . the color bars in the upper panel of figure [ fig : fig_4](b ) show the difference between decay rates of each nv center after and before introducing a liquid . we observe that for most centers , the decay rate either barely varies , or varies non - monotonically with applied index , although generally , one expects to observe an increase in the decay rate of a dipole emitter in a homogenous medium by increasing the refractive index of the environment due to an increase of the ldos @xcite . for comparison to the data we calculate the normalized decay rate of a dipole positioned in the middle of a 60 nm thick sio@xmath4 slab ( @xmath57 ) which is sandwiched between a semi - infinite slab of quartz with @xmath58 at the bottom and a semi - infinite slab with varying refractive index at the top . figure [ fig : fig_4](c ) shows the calculated decay rate of the dipole as a function of the refractive index of the top layer for different dipole - moment orientations . the calculation shows that as the refractive index increases the decay rate change is expected to be monotonic and increasing and is the largest for a dipole oriented perpendicular to the interface while the change is slight for a dipole - moment parallel to the interface . comparing the calculated rate changes to the measured decay rates ( figure [ fig : fig_4](b ) ) , we find no systematic rate variation for the nv centers within the error bars of our experimental data . the error bar for measurements on individual nv centers is composed of two contributions . first , the error bar contains the uncertainty in the fit to the fluorescence decay trace . second , the experiment is hampered by the fact that the decay rates of some of the single nanodiamonds was observed to jump slightly between observations , even if no change in environment was applied . these jumps amount to approximately 5% , and occur on time scales of hours to days . to exclude this effect , we have measured the rate for each liquid and a reference in air on the same day for each nv center to generate the plot of rate changes in figure [ fig : fig_4](b ) . , width=566 ] we find that there is no evidence that the rate of _ individual _ nv centers in 25 nm diamond nanocrystals actually varies in accordance with the known applied ldos changes . this observation appears to be at variance with earlier work that evidenced a small change in the mean decay rate of an ensemble of nv centers in a liquid - tuning experiment @xcite . as two caveats , we note that first , in that previous work it was not verified if any such change occurred for each source individually , and second , that a shortcoming of our experiment for quantititave interpretation is that for dipoles located close to an interface ( as is the case here ) , the decay rate enhancement significantly depends on the unknown dipole - moment orientation of the emitter with respect to the interface . while the fact that we do not know the dipole orientations for each nv center makes it impossible to conclude with certainty that an ldos effect is absent , two hypotheses for absence of an ldos effect could be proposed . first , that the quantum efficiency of 25 nm diamond nanocrystals is low , or second , that the very high index diamond shell around the nv centers intrinsically prevents the nv center from responding to ( moderate ) changes in the ldos of the environment it is supposed to probe . we exclude the latter explanation on basis of a set of finite element ( comsol ) simulations , in which we calculate the decay rate of a point dipole randomly located in a diamond nanosphere , which in turn is placed at varying distances from a planar interface with materials of various dielectric constants . we find that in all cases , the decay rate of the source simply follows the theory of urbach and rikken @xcite for a source without dielectric shell , multiplied by a pre - factor that is essentially a quasistatic local field correction factor due to the diamond shell and is independent of the varying ldos . to conclude , our observation that no nv center responds to ldos changes together with our earlier correlation plots of brightness versus decay rate ( figures [ fig : fig_2 ] and [ fig : fig_3 ] ) means that the quantum efficiencies of 25 nm diamond nanocrystals are low . a second important conclusion is that single nv centers of 25 nm size appear unsuitable to measure ldos changes due to the slight jumps in lifetime , unless the entire measurement scheme ( including reference measurements ) takes less than a few hours . in this section we report on the results of a second method to assess if single nv centers are suited for ldos measurements , and to calibrate emission rates . this second method has the advantage that we can apply calibrated ldos changes rapidly so that jumps in the intrinsic rate constants can be avoided that hamper experiments where macroscopic sample changes are required to modify ldos , as in a liquid immersion experiment . the scheme is similar to a measurement procedure reported by buchler @xcite , based on a drexhage experiment @xcite . in a drexhage experiment , a mirror is used to impose a large ldos change , and rates are measured as a function of the emitter - mirror separation . while this method is usually implemented by creating a set of macroscopic samples where mirrors are coated with spacers of calibrated thicknesses @xcite or by creating a single sloping wedge between a mirror and an ensemble of emitters @xcite , buchler realized a nanomechanical version that can be applied to a single emitter . buchler used a silver - coated curved fiber end as a mirror attached to a piezo - stage to precisely tune the mirror distance to an underlying single emitter on the sample @xcite . here , we use a similar method in a shear - force - feedback near - field microscope . however , instead of a vertical mirror displacement , we use a lateral displacement of the curved mirror while staying in shear - force feedback to keep the mirror and sample substrate in near - contact . as the contact point moves sideways , the spherical mirror shape ensures that the mirror - to - emitter distance varies ( figure [ fig : fig_5](a ) ) . to fabricate the curved mirror , we glued 45@xmath49 m polystyrene beads ( polysciences , inc . ) to the end of cleaved optical fibers and coated the beads with about 200 nm of silver . ) as a function of mirror lateral displacement ( c ) corresponding decay rate map of the nv center as a function of mirror lateral displacement . decay rates are extracted from a single - exponential fit to the time - trace data for each pixel . ( d ) dots : decay rate as a function of the emitter distance to mirror , extracted from the decay rate map shown in ( c ) . solid line : fitted ldos on the measured rate oscillations with fit parameters @xmath40 , @xmath41 and a set of dipole moment orientations . the shown qe is the most - likely value . ( e ) and ( f ) decay rate as a function of emitter - mirror distance for two other nv centers ( number 4 and 5 in table [ table : table_1 ] , respectively ) and the fitted rate from ldos calculation . [ fig : fig_5],width=566 ] we have applied the scanning - mirror ldos changing technique to 100 nm nanodiamonds containing single nv centers , prepared as described in section [ subsec : smaple_prep ] , and subsequently embedded in a 200 nm thick layer of planarizing spin - on glass ( fox-14 , dow corning ) . the spin - on glass immobilizes the nanocrystals , so that they are not moved during shear - force scanning . in order to vary the distance between the mirror and the emitter , we scan the mirror bead laterally on top of an identified nv center . for each position of the mirror bead , we collect the fluorescence emission of the nv center , positioned at the focus of the pump laser , through the confocal microscope setup described in section [ subsec : exp_setup ] . by scanning the mirror bead , we obtain a confocal fluorescence intensity map as shown in figure [ fig : fig_5](b ) . here , each pixel represents the relative position of the mirror bead with respect to the nv center with the false color representing the collected fluorescence intensity . we clearly observe interference rings in the fluorescence intensity map . these rings mainly stem from the fact that the mirror imposes a standing - wave pattern on the 532 nm pump field , which subsequently results in a modulation of the fluorescence intensity . for each pixel in the fluorescence intensity map , we stored absolute photon arrival times , as well as laser pulse arrival times , allowing us to extract the decay dynamics . we use a single - exponential fit to obtain the corresponding decay rate of the emitter for each mirror position , as plotted in figure [ fig : fig_5](c ) . here , the color scale represents the fitted decay rate corresponding to a defined mirror position . interestingly , we observe a radial modulation of the decay rates varying between about 0.02ns@xmath38 to 0.04ns@xmath38 , which we attribute to the varying ldos in front of the mirror . to quantify the variation as a function of the distance from emitter to mirror , we extracted the decay rates as a function of the lateral distance to the mirror central position . to this end , we bin pixels in concentric rings of equal lateral distance to the mirror center , and concatenate the photon - correlation time traces of all pixels in each bin to obtain a single fluorescence decay trace per radial distance . using the spherical form factor of the bead , we convert lateral position of the mirror to normal distance of the mirror to emitter . the decay rate fitted to the fluorescence decay for each mirror - emitter separation is depicted in figure [ fig : fig_5](d ) ( black dots ) . we observe a distinct oscillation of the rate around 0.028ns@xmath38 , with a 15% amplitude . to our knowledge , this is the first report of a reversible change in the decay rate of a single nv center in a calibrated ldos experiment . due to the large radius of the mirror compared to its distance to emitters , it is reasonable to consider the experimental configuration as a planar glass - air - mirror system in which we vary the air thickness . for this system , the ldos is exactly known @xcite for any dipole orientation and position . we fit the experimental decay rate data using the theoretical ldos @xmath59 according to @xmath60 , where the fit parameters are the nonradiative decay rate @xmath61 , the radiative decay rate @xmath62 , and the dipole orientation @xmath63 relative to the normal to the plane . finally a small offset @xmath64 appears that is due to the unknown distance - of - closest approach in shear - force microscopy , of the order of 15 nm . care must be taken that a correlation exists between the dipole orientation @xmath63 , and the apparent quantum efficiency @xmath65 returned by the fit routine , due to the fact that the ldos for different dipole - moment orientations is similar in qualitative @xmath66-dependence , but different in oscillation contrast . to overcome this dependency , we examine fits of the rate oscillations for a set of dipole orientations and evaluate the goodness of fits individually . the goodness of fit is established by measuring in how far the residuals of the fit ( point - by - point deviation between data and fit function ) are within the data point error bar . this gives a range of quantum efficiency values for which the fit is consistent with the data . in this paper we report the range of quantum efficiency values consistent with the data given that the dipole orientation should be treated as an unknown parameter . in case the dipole moment orientation of the nv center is known , which could in principle be realized using fourier microscopy @xcite , one can extract a more precise value for the quantum efficiency . for the particular data set shown in figure [ fig : fig_5](d ) , we find a most likely quantum efficiency of 26% and a dipole moment orientation within @xmath67 along the sample plane . the range of quantum efficiencies consistent with the data for this nv center is bounded from below by 26% , and from above by 50% . to our knowledge this is the first experimental calibration of the quantum efficiency of a single nv center in a diamond nanocrystal . in order to investigate the typical quantum efficiencies of nv centers , we selected five random nv centers and examined them with the moving mirror experiment as explained above . each nanodiamond was selected to be a single nv center containing nanocrystal according to the criteria we outlined in section [ single_nv_charac ] , without further post - selection for inclusion in the drexhage experiment . table [ table : table_1 ] summarizes the confidence intervals for the fitted values of the quantum efficiencies , given that we do not know the dipole orientation a priori . the total decay rate @xmath44 , i.e. the sum of @xmath40 and @xmath41 varies almost over a factor 2 , and can be fitted accurately . in table [ table : table_1 ] we also report radiative and nonradiative decay rates . values are reported as fitted to the data while fixing the dipole orientation at its most likely value , in which case the data can be fit with small error bars on @xmath40 and @xmath41 . the wide range of quantum efficiencies consistent with the data of course imply that in a completely free fit @xmath40 and @xmath41 both have a large error bar , though their sum @xmath44 does not . we find that the different nv centers we probed , while having reasonably comparable brightness , in fact have widely different quantum efficiencies . for instance , nv center 4 ( figure [ fig : fig_5](e ) ) has an efficiency certainly below 14% , while nv center 5 ( figure [ fig : fig_5](f ) ) has a quantum efficiency certainly above 58% , in the range 58% to 90% . furthermore , we note that the fits neither result in the conclusion of a fixed nonradiative rate at varying radiative rate , nor conversely in the conclusion that the radiative rate is a constant while the nonradiative rate varies . instead , both the radiative and the nonradiative decay rates are distributed . even with the wide error bars on quantum efficiencies , we establish that both distributions span at least around a factor two in range . our measurements hence show that the common tacit assumption of near - unit quantum efficiency and nearly identical emission characteristics barring those due to depolarization factors in the nanodiamonds should be discarded . .extracted quantum efficiencies , rates and brightnesses for five randomly chosen nv centers in 100 nm nanodiamonds . quantum efficiencies are quoted as a range of values , taking into account that the dipole orientation is a free parameter . in this fit procedure , the proper fit parameters are the quantum efficiency and the total decay rate , which has small error bar . for completeness we also show the two derived quantities @xmath40 and @xmath41 . these rates and their error bars are given as fitted when fixing the dipole orientation to its most likely value . fitted brightnesses were converted to ccd response units to be comparable with fig . 3[table : table_1 ] [ cols="<,^,^,^,^,^",options="header " , ] in order to assess the suitability of nanodiamonds as photonic ldos probes and as building blocks for hybrid photonic systems , we have investigated the brightness , decay rate , and quantum efficiency of single nv centers in nanodiamonds of 25 nm , and of 100 nm median size . for both size ranges a wide distribution in brightnesses and rates is found , consistent with reports by earlier workers . we conclude that the wide distribution of rates is due to a distribution in radiative rates , nonradiative rates and quantum efficiencies . this conclusion contradicts earlier work , that interpreted the wide distribution of decay rates as mainly due to a photonic effect that causes a distribution in radiative decay rates via the ldos . instead , our measurements show that even nv centers in large nanocrystals show a wide range of quantum efficiencies when probed in a controlled ldos experiment . for nv centers in 100 nm nanocrystals we find quantum efficiencies distributed between 10% and 90% , while for the smaller nv centers the quantum efficiency for fluorescence is likely a factor of two smaller on average . for applications in nanophotonic experiments using single nv centers , a highly problematic property is that quantum efficiency does not correlate with brightness , due to the fact that both the radiative and nonradiative rate are distributed . in other words , screening nanodiamonds to pick the best one to probe ldos or for incorporation in a photonic device can not rely on a simple metric such as spectrum or brightness . instead , we argue that future work in the hybrid assembly of nanodiamonds in plasmonics and photonic crystals to realize accelerated spontaneous emission decay always requires an experimental protocol in which nanodiamonds are first individually calibrated in terms of quantum efficiency . while we have shown a method for such calibration , this is a highly tedious procedure that is not easily implementable and requires dedicated near - field manipulation equipment . in this work , we have not speculated on the origin of the apparent low quantum efficiency of single nv centers in nanodiamonds . we close with two remarks on the origin of the low apparent quantum efficiency . firstly , we have treated the nv center as a quasi two - level system . our experiments thus address the question what the apparent quantum efficiency is when attempting to use an nv center as a two - level ldos probe . from the nv center spectrum it is clear that the spectrum is very wide with a large vibrational broadening spectrum . moreover it is well known that the nv@xmath68 center is not a two level system . instead , the nv@xmath68 has different spin sublevels , and may experience spin - flip intersystem crossing between allowed spin transition manifolds . the rates for these transitions were recently characterized in detail for nv centers in bulk diamond in ref . a further complication is that the nv@xmath68 defect may infrequently transition to an uncharged nv center that also luminesces , yet at different rate and efficiency . the many rates involved in these transitions can further vary between nv centers due to , e.g. , variations in crystal strain . a full treatment of the response of nv centers to ldos changes would hence have to treat the full rate equations in which radiative transition rates are , and intersystem crossing rates are not , affected by ldos . an important step beyond our work will be to perform the measurements we have described here but employing the spin selective techniques reported by robledo @xcite to establish what the nonradiative and radiative rates are for each transition separately , instead of lumping rates into effective two - level parameters . while the ldos changes we have applied using a mirror are broadband ldos changes that modify radiative transitions roughly equally across the emission spectrum , ldos changes that have strong spectral features could be used advantageously to enhance or suppress the importance of intersystem crossing . having established that further work is required to separate the quantum efficiencies reported here into parameters per transition in a more complete level scheme , we turn to possible reasons for the below - unity values of quantum efficiency that we find . the lower brightness of the smaller nv centers , as well as their lack of response to ldos changes suggest that the surface , i.e. , surface contamination with carbon , or surface defects may play a role in providing quenching sites . indeed , it has already been suggested for very small ( 5 nm ) nanodiamonds that nv centers may suffer quenching due to graphite shells on the diamonds . while the nanodiamonds we used have been employed by several groups in spontaneous emission control experiments in untreated form , exactly as in our experiments , additional surface treatments such as by prolonger immersion in boiling sulphuric acid , or cleaning in kningswasser have been proposed by several workers . whether or not such treatments actually affect quantum efficiency is as yet unclear , as is whether quenching can be completely suppressed . we propose that the quantum efficiency calibration method that we demonstrated will be an indispensable tool to evaluate such cleaning methods , as well as to screen other color centers in diamond for advantageous fluorescence properties . we thank s. schietinger for communications regarding sample preparation methods , m. frimmer for experimental help and discussions , and c. osorio for suggestions to improve the manuscript . this work is part of the research program of the `` foundation for fundamental research on matter ( fom ) '' , which is financially supported by the `` the netherlands organization for scientific research ( nwo ) '' . afk gratefully acknowledges a nwo - vidi grant for financial support . 99 santori c , fattal d and yamamoto y 2010 _ single - photon devices and applications _ ( john wiley & sons ) sprik r , van tiggelen b a and lagendijk a 1996 _ europhys . lett . _ * 35 * 265270 brouri r , beveratos a , poizat j p , and grangier p 2000 _ opt . lett . _ * 25 * 1294 kurtsiefer c , mayer s , zarda p , and weinfurter h 2000 _ phys . rev . lett . _ * 85 * 290 beveratos a , brouri r , gacoin t , poizat j p , and grangier p 2001 _ phys . rev . a _ * 64 * 061802 tisler j , balasubramanian g , naydenov b , kolesov r , grotz b , reuter r , boudou j p , curmi p a , sennour m , thorel a , borsch m , aulenbacher k , erdmann r , hemmer p r , jelezko f , and wrachtrup j 2009 _ acs nano _ * 3 * 1959 babinec t m , hausmann j m , khan m , zhang y , maze y r , hemmer p r , and lonar 2010 _ nature nanotechn . _ * 5 * 195 michaelis j , hettich c , mlynek j and sandoghdar v 2000 _ nature _ * 405 * 325328 cuch a , drezet a , sonnefraud y , faklaris o , treussart f , roch j f and huant s 2009 _ opt . express _ * 17 * 1996919980 cuch a , mollet o , drezet a and huant s 2010 _ nano lett . _ * 10 * 45664570 degen c. l. 2008 _ appl . lett . _ * 92 * , 243111 taylor j m , cappellaro p , childress l , jiang l , budker d , hemmer p r , yacoby a , walsworth r and lukin m d 2008 _ nature phys _ * 4 * 810816 balasubramanian g , chan i y , kolesov r , al - hmoud m , shin c , kim c , wojcik a , hemmer p r , krueger a , hanke t , leitenstorfer a , bratschitsch r , jelezko f and wrachtrup j 2008 _ nature _ * 455 * , 648651 maletinsky p , hong s , grinolds m s , hausmann b , lukin m d , walsworth r l , loncar m , and yacoba a 2012 , _ nature nanotechnology _ * 7 * , 320 - 324 inam f a , gaebel t , bradac c , stewart l , withford m j , dawes j m , rabeau j r , and steel m j 2011 _ new j. phys . _ * 13 * 073012 ruijgrok p v , wuest r , rebane a a , renn a , and sandoghdar v 2010 _ opt . exp . _ * 18 * 6360 van der sar t , hagemeier j , pfaff w , heeres e , thon s , kim h , petroff p , tjerk o , bouwmeester d , and hanson r 2012 _ j. opt . soc . b _ * 29*(4 ) 698703 drexhage k h 1970 _ j. lumin . _ * 1 - 2 * 693 buchler b c , kalkbrenner t , hettich c , and sandoghdar v 2005 _ phys lett . _ * 95 * 063003

we study the hard- spectral properties of ten highly luminous radio - quiet ( rq ) active galactic nuclei ( agns ) at , including new observations of four of these sources . we find a significant correlation between the normalized accretion rate ( ) and the hard- photon index ( ) for 35 moderate high luminosity rq agns including our ten highly luminous sources . within the limits of our sample , we show that a measurement of and can provide an estimate of and black - hole ( bh ) mass ( ) with a mean uncertainty of a factor of 3 on the predicted values of these properties . this may provide a useful probe for tracing the history of bh growth in the universe , utilizing samples of -selected agns for which and have not yet been determined systematically . it may prove to be a useful way to probe bh growth in distant compton - thin type 2 agns . we also find that the optical spectral slope ( ) depends primarily on optical uv luminosity rather than on in a sample of rq agns spanning five orders of magnitude in luminosity and over two orders of magnitude in . we detect a significant compton - reflection continuum in two of our highly luminous sources , and in the stacked spectrum of seven other sources with similar luminosities , we obtain a mean relative compton reflection of and an upper limit on the rest - frame equivalent width of a neutral line of 105ev . we do not detect a significant steepening of the power - law spectrum below rest - frame 2kev in any of our highly luminous sources , suggesting that a soft - excess feature , commonly observed in local agns , either does not depend strongly on , or is not accessible at high redshifts using current detectors . none of our highly luminous sources displays flux variations on timescales ofhr , supporting the idea that the timescale of variability depends inversely on and does not depend on .

0804.0803

the spectrum of an unobscured , radio - quiet ( rq ) agn in the energy band is best characterized by a single power - law continuum of the form @xmath8 , where @xmath2 , hereafter the photon index in the energy band , typically lies in the range @xmath71.52.5 . a corona of hot electrons is assumed to produce the hard- emission via compton upscattering of uv soft- photons from the accretion disk , and @xmath2 is predicted to be only weakly sensitive to large changes in the electron temperature and the optical depth in the corona ( e.g. , haardt & maraschi 1991 ; zdziarski et al . 2000 ; kawaguchi et al . 2001 ) . the relatively narrow range of @xmath2 values in rq agns has been reported in numerous studies ( e.g. , nandra & pounds 1994 , reeves & turner 2000 ; page et al . 2005 ; shemmer et al . 2005 ; vignali et al . 2005 ; just et al . 2007 ) , and typically no strong dependence of @xmath2 on redshift or luminosity has been detected across the widest possible ranges of these parameters . on the other hand , a strong anticorrelation between @xmath2 and the full width at half - maximum intensity ( fwhm ) of the broad emission - line region ( belr ) component of h@xmath9 has been found , first by brandt et al . ( 1997 ) . the remarkable dependence between and optical spectroscopic properties has been suggested to arise from a more fundamental correlation between @xmath2 and the accretion rate ( e.g. , brandt & boller 1998 ; laor 2000 ) . a high accretion rate is expected to soften ( steepen ) the hard- spectrum by increasing the level of disk emission , resulting in the production of softer photons , which increase the compton cooling of the corona . using recent scaling relations for the belr size , luminosity , and the width of the broad h@xmath9 emission line from reverberation - mapping studies , it is clear that the normalized accretion rate ( i.e. , @xmath10/@xmath11 , hereafter @xmath1 , where @xmath10 is the bolometric luminosity ) is proportional to fwhm(h@xmath9)@xmath12 , at least for low moderate luminosity agns in the local universe ( e.g. , kaspi et al . 2000 ) . subsequent studies of nearby ( @xmath130.5 ) unobscured rq agns have confirmed the brandt et al . ( 1997 ) anticorrelation ( e.g. , leighly 1999 ; reeves & turner 2000 ; porquet et al . 2004 ; piconcelli et al . 2005 ; brocksopp et al . 2006 ) , and others have found significant correlations between @xmath2 and @xmath1 ( e.g. , lu & yu 1999 ; porquet et al . 2004 ; wang et al . 2004 ; bian 2005 ) . however , all these studies were not able to disentangle the strong dependence . recently , shemmer et al . ( 2006 ; hereafter s06 ) have suggested that this degeneracy can be removed if highly luminous sources are included in the analysis . this can be achieved by obtaining high - quality near - ir spectroscopy of the h@xmath9spectral region , to obtain @xmath1 ( e.g. , shemmer et al . 2004 ) , as well as accurate measurements of @xmath2 using and _ chandra _ for highly luminous agns found at 1@xmath133 . within the limits of their sample of 30 sources , spanning three orders of magnitude in luminosity , s06 have shown that @xmath2 does not depend on luminosity or black - hole ( bh ) mass ( @xmath4 ) . they have also shown that the @xmath2 values of the five highly luminous sources in their sample , while consistent with the values expected from their normalized accretion rates ( @xmath1 ) , are significantly higher than expected from the widths of their broad h@xmath9 emission lines . this has enabled , for the first time , breaking of the fwhm(h@xmath9)-@xmath1degeneracy and has provided evidence that @xmath2 depends primarily on the accretion rate . however , the number of highly luminous sources was small , and as explained below , some uncertainties remained . in this work , we double the number of highly luminous sources at high redshift and reinforce the s06 results . we show that @xmath2 can be considered a reliable accretion - rate indicator for moderate high luminosity rq agns , and that the combination of @xmath2 and luminosity may provide a useful probe for tracing the history of bh growth in the universe . we also discuss the spectral and temporal properties of luminous , high - accretion rate agns at high redshift as well as the dependence of the optical spectral energy distribution ( sed ) on luminosity and @xmath1 . in [ sec_observations ] we describe our sample selection , and present the new observations and their analysis . our results are presented and discussed in [ sec_results ] , where we focus on the correlation between the hard- photon index and the normalized accretion rate in rq agns and its implications for probing bh growth in the universe . a summary of our main findings is given in [ sec_conclusions ] . throughout this work we consider only rq agns to avoid any contribution from jet - related emission to the spectra . luminosity distances are computed using the standard cosmological model with @xmath14 , @xmath15 , and @xmath16 ^ -1 km s@xmath17mpc@xmath17 . our sample is composed of the 30 sources studied in s06 as well as five highly luminous sources ( @xmath18 { \raisebox{-.5ex}{$\;\stackrel{<}{\sim}\;$}}48 $ ] ) whose hard- spectra are analyzed in this work for the first time . motivated by the s06 hypothesis that only highly luminous agns may break the degeneracy , we have primarily searched for highly luminous agns that have high - quality h@xmath9 and hard- spectroscopy . our search of the literature and archive yielded ten such sources , five which were studied in s06 and five that are analyzed below . our new core sample of 35 sources includes unabsorbed , type 1 rq agns with @xmath19 { \raisebox{-.5ex}{$\;\stackrel{<}{\sim}\;$}}48 $ ] ( i.e. , a moderate high luminosity range ; see fig . [ fig_hist ] ) . rest - frame optical data for this sample was obtained from neugebauer et al . ( 1987 ) , boroson & green ( 1992 ) , nishihara et al . ( 1997 ) , mcintosh et al . ( 1999 ) , shemmer et al . ( 2004 ) , and sulentic et al . hard- data were obtained from page et al . ( 2004b ) , piconcelli et al . ( 2005 ) , s06 , and from this work ( see below ) . all of our sources were selected to have high - quality hard- spectra ( except for he 0926@xmath200201 that was serendipitously observed with _ chandra _ ) . we have intentionally limited our sample to the moderate high luminosity range in order to minimize potential effects of optical and variability ( since the optical and data are not contemporaneous ) as well as potential spectral complexities due to compton reflection , as discussed further below ( see also s06 for more details ) . in light of our selection criteria , we caution that our sample is neither complete nor fully representative of the agn population as a whole ; our main results may therefore be subject to selection biases over the wide range of agn properties , which were not fully explored in this work . lcccccccc lbqs0109@xmath210213 & 01 12 16.91 & @xmath2102 29 47.6 & 2.349 & 3.25 & 2007 jan 8 & 33.7 / 211 & 33.8 / 218 & 26.3 / 554 + 2qzj023805.8@xmath20274337 & 02 38 05.80 & @xmath2027 43 37.0 & 2.471 & 1.73 & 2006 dec 22 & 34.6 / 206 & 34.0 / 183 & 23.9 / 586 + q1318@xmath20113 & 13 21 09.38 & @xmath2011 39 32.0 & 2.306 & 2.80 & 2006 dec 28 & 47.7 / 853 & 46.9 / 876 & 33.2 / 2621 + sbs1425@xmath21606 & 14 26 56.18 & @xmath2160 25 50.9 & 3.202 & 1.58 & 2006 nov 12 & 20.6 / 232 & 20.9 / 255 & 7.0 / 304 we have performed spectral imaging observations of four new sources from the shemmer et al . ( 2004 ) sample with ( jansen et al . 2001 ) ; a log of these observations is presented in table[tab_obs_log ] . these sources were selected for observations for being luminous , high - accretion rate sources , predicted to have high fluxes , and having relatively low galactic column densities . the data were processed using standard science analysis system v6.5.0 tasks . the event files of all the observations were filtered to remove periods of flaring activity in which the count rates of each mos ( pn ) detector exceeded 0.35 ( 1.0 ) counts s@xmath17 for events having @xmath22kev . the time lost due to flaring in each observation varied between of the entire observing time ; the net exposure times in table[tab_obs_log ] reflect the filtered data . the spectra of the quasars were extracted from the images of all three european photon imaging camera ( epic ) detectors using apertures with radii of 30 . local background regions were at least as large as the source regions . the spectra were grouped with a minimum of 20 counts per bin , except for the spectrum of q 1318@xmath20113 which was grouped with a minimum of 50 counts per bin . joint spectral fitting of the data from all three epic detectors for each source was performed with xspec v11.3.2 ( arnaud 1996 ) . we employed galactic - absorbed power - law models at rest - frame energies 2kev , corresponding to 0.50.6kev in the observed frame of the sources , where the underlying power - law hard- spectrum is less prone to contamination due to any potential soft excess emission or absorption . in each fit , the photon indices in the spectra of all three epic detectors were tied to a single value , while the normalizations were free to vary . the best - fit @xmath2 values , power - law normalizations , and @xmath23 values from these fits are given in columns ( 2 ) , ( 3 ) , and ( 4 ) of table[tab_properties ] , respectively , and the data , their joint , best - fit spectra , and residuals appear in fig.[fig_spectra ] . we also searched for intrinsic absorption in each source by jointly fitting the spectra with a galactic - absorbed power law model including an intrinsic ( redshifted ) neutral - absorption component with solar abundances in the same energy range quoted above . no significant intrinsic absorption was detected in any of the sources ; upper limits on intrinsic @xmath24 values appear in column ( 5 ) of table[tab_properties ] . each panel of fig.[fig_spectra ] includes a confidence - contour plot from this fitting for each source . by applying @xmath25-tests between the models including intrinsic absorption and those that exclude it , we found that none of the spectra require an intrinsic absorption component . the remarkably flat hard- spectrum of lbqs 0109@xmath210213 ( with @xmath26 ; see table [ tab_properties ] ) motivated an alternative modeling , searching for an indication of partial covering . we found that when the spectrum is fitted with a redshifted partial - covering fraction absorber ( zpcfabs model in xspec ) and a galactic - absorbed power law , the covering fraction is consistent with zero and the fit is not improved with respect to a galactic - absorbed power law model . therefore , the unusually flat hard- spectrum of lbqs 0109@xmath210213 can not be explained as being due to partial covering . he 0926@xmath200201 is a highly luminous rq agn at @xmath27 for which near - infrared spectroscopy of the h@xmath9 region is presented in sulentic et al . this source is serendipitously detected in an 8.8ks _ chandra_/acis - s observation from 2002 february 20 ( i d 3139 ) with 126 source counts in the 0.58 kev observed - frame band . we analyzed the _ chandra_observation using standard ciao v3.2 routines . the spectrum was extracted using psextract from a circular aperture with a radius of eight pixels , corresponding to @xmath74@xmath28 , and the events were grouped to have a minimum of 10 counts per bin ( the extremely low _ chandra _ background was determined from a source - free annular region with inner and outer radii of 10@xmath28 and 25@xmath28 , respectively , centered on the source ) . the spectrum was fitted over the @xmath29kev rest - frame energy band using xspec with a model including a power law and a galactic - absorption component with a column density of @xmath24=3.17@xmath30 cm@xmath12 . the data were also fitted with an additional intrinsic - absorption component , similar to the procedure described above for the observations , and we found that the data do not warrant any additional absorption . the best - fit spectral parameters as well as the upper limit on the intrinsic absorption are included in table [ tab_properties ] , and the _ chandra _ spectrum and best - fit model appear in fig . [ fig_he0926_chandra_spec ] . basic optical spectroscopic properties of the new high - redshift sample are given in table[tab_properties ] . the monochromatic luminosity at a rest - frame wavelength of 5100 [ @xmath32 is given in column ( 6 ) , and fwhm(h@xmath9 ) is given in column ( 7 ) ; except for he0926@xmath200201 , for which the optical data were obtained from sulentic et al . ( 2006 ) , these data were obtained from netzer et al . ( 2007 ) , who recently presented new and improved spectroscopic measurements for all the shemmer et al . ( 2004 ) sources . the @xmath4 and @xmath1 values in columns ( 8) and ( 9 ) , respectively , were determined as in s06 , using the @xmath33 and fwhm(h@xmath9 ) values given in columns ( 6 ) and ( 7 ) , respectively , and based on the recent reverberation - mapping results of peterson et al . ( 2004 ) and the kaspi et al . ( 2005 ; hereafter k05 ) belr size - luminosity relation ( see also kaspi et al . 2000 ) ; the general expression we use for @xmath1 is of the form @xmath1@xmath34 . we note that the k05 relation relies on a sample of agns having luminosities up to , and extrapolating it to higher luminosities , such as those of ten of the sources presented here , is somewhat uncertain ; a reverberation - mapping effort is underway to test the validity of such extrapolations ( see e.g. , kaspi et al . in addition , bentz et al . ( 2006 , 2007 ) have recently suggested that subtraction of host - galaxy starlight from the agn optical continuum may result in a somewhat flatter slope ( e.g. , @xmath35 compared with @xmath36 in k05 ) for the belr size - luminosity relation across the entire k05 luminosity range . we test the possibility of using a flatter slope below , but we note that the bentz et al . measurements were performed for sources with @xmath3710@xmath38ergss@xmath17 , while all the sources studied here have @xmath3910@xmath38ergss@xmath17 ( with a median luminosity of @xmath4010@xmath41ergss@xmath17 ; fig . [ fig_hist ] ) . we also expect a much smaller fractional amount of host contamination in the most luminous sources in our sample . in light of all this , we have computed @xmath4 values using the k05 slope across our entire luminosity range . moreover , since in this work we employ nonparametric statistical ranking tests , our main results are not significantly sensitive to the precise value of the slope of the belr size - luminosity relation in the range @xmath42 . to derive the normalized accretion rates , we have employed the luminosity - dependent bolometric correction method of marconi et al . ( 2004 ) ; the bolometric correction factor for the monochromatic luminosity at 5100 [ i.e. , @xmath10/@xmath43(5100 ) ] is for the moderate - luminosity @xmath44 ( high - luminosity @xmath0 ) sources in this work ( see s06 for more details ) . the optical spectral slopes ( @xmath5 ) in column ( 10 ) of table [ tab_properties ] are defined as + @xmath5@xmath45 , where @xmath46 and @xmath47 are the flux densities at 2kev and 2500 , respectively . the @xmath5 values were derived using the photon indices and fluxes in columns ( 2 ) and ( 3 ) , respectively , and the optical luminosities in column(6 ) , assuming a uv continuum of the form @xmath48 ( vanden berk et al . the @xmath5 values for all of our sources are consistent with the expected values , given their optical luminosities ( e.g. , steffen et al . 2006 ) . the relatively high redshifts of our sources allowed us to search for a compton - reflection continuum as well as emission in their spectra . a compton - reflection continuum may be observed in agns within the rest - frame @xmath49kev energy range , peaking at rest - frame @xmath730kev , and it presumably originates from reflection of hard- photons off the relatively colder outskirts of the accretion disk and/or the torus . we also searched for signatures of a neutral narrow emission line at rest - frame 6.4kev , as this is expected to appear in conjunction with a compton - reflection continuum . the search for continua reflected from neutral material was carried out by fitting all our spectra in the 2kev rest - frame energy range with xspec , employing a galactic - absorbed power - law and a compton - reflection continuum model ( i.e. , the pexrav model in xspec ; magdziarz & zdziarski 1995 ) simultaneously with a redshifted gaussian emission line model ( using the zgauss model in xspec ) ; the gaussian rest - frame energy and width were fixed at @xmath50kev and @xmath51kev , respectively . we also included in the analysis the five luminous , high - redshift sources from s06 that have spectra and h@xmath9spectroscopy , namely pg1247@xmath21267 , q1346@xmath20036 , pg1630@xmath21377 , pg1634@xmath21706 , and he2217@xmath202818 ; the luminosities and accretion rates of these sources are comparable to those of the sources presented in this work . table [ tab_compton ] lists the best - fit parameters from these fits . column ( 3 ) gives the rest - frame equivalent width ( ew ; throughout the paper , ews refer to rest - frame values ) of the emission line and column ( 4 ) gives the relative - reflection component ( @xmath52 ) of the compton - reflection continuum expressed as @xmath53 , where @xmath54 is the solid angle subtended by the continuum source . except for the cases of pg1247@xmath21267 and pg1634@xmath21706 , @xmath25-tests carried out between the results of the pexrav+zgauss fits and the galactic - absorbed power - law fits ( [ sec_xmm ] ) have indicated that the new ( reflection ) model did not improve the fits with respect to the corresponding galactic - absorbed power - law fits . table [ tab_compton ] shows that , except for pg1630@xmath21377 , we have not detected any neutral emission in any of the sources , and instead , we have placed upper limits on the ews of such emission . although the ew ( ) for pg1630@xmath21377 has a nonzero lower limit , and the ew value we found is consistent with the one found for this source by jimnez - bailn et al . ( 2005 ) , we argue that the data for pg1630@xmath21377 do not warrant the existence of a reflection spectrum based on the @xmath25-test described above . we have also tested whether the spectra of our sources exhibit hard- emission reflected from ionized material . this was performed by employing the same xspec fitting routine described above for the case of reflection from neutral material , except that the pexrav model was replaced with the pexriv model ( magdziarz & zdziarski 1995 ) , and the energy of the line was free to vary in the 6.76.97 rest - frame energy range to account for an ionized iron line . the results of these fits , i.e. , @xmath52 parameters and ew ( ) values , are consistent with those presented in table [ tab_compton ] for the case of reflection from neutral material ( even though the energies of the lines were required to be somewhat higher ) . the non - detections of emission lines in the highly luminous sources under study are in accord with previous studies claiming that the strength of such emission lines decreases with increasing luminosity ( aka the ` baldwin effect ' ; e.g. , iwasawa & taniguchi 1993 ; page et al . 2004a ; zhou & wang 2005 ; bianchi et al . 2007 ) however , the relatively high upper limits we obtain for our ew ( ) values can not rule out the possibility that such an ` baldwin effect ' is very weak or does not even exist ( e.g. , jimnez - bailn et al . 2005 ; jiang et al . 2006 ) . given the high luminosities of our sources , the expected ew ( ) values from such an ew-@xmath55 relation are of order @xmath750ev , while in most cases , our upper limits are considerably higher ( see table [ tab_compton ] and e.g. , bianchi et al . 2007 ) . we detect statistically significant compton - reflection continua only in pg1247@xmath21267 and pg1634@xmath21706 ( table [ tab_compton ] ) . the @xmath52 value we find for the first of these sources is consistent with the one found by page et al . ( 2004b ) for the same spectrum . however , in contrast with our results , page et al . ( 2004b ) have not detected a significant compton - reflection feature in the same spectrum of pg1634@xmath21706 that we have analyzed , while the upper limit we obtain on the ew ( ) for the source is consistent with their finding ; we also note that nandra et al . ( 1995 ) have not detected any reflection features in an _ asca _ spectrum of the source . in addition , page et al . ( 2004b ) _ have _ detected an line in pg1247@xmath21267 while we have not . interestingly , inspection of fig . 1 of page et al . ( 2004b ) suggests the existence of a reflection component in both sources . an alternative spectral fitting , including a thermal component and a power - law spectrum for each source , was recently employed by ruiz et al . ( 2007 ) ; their study finds a soft excess and no reflection component in both sources . these partial discrepancies may be a consequence of the different modelings and different energy ranges used in the different studies . the apparently non - physical @xmath52 values we obtain for pg1247@xmath21267 and pg1634@xmath21706 ( i.e. , @xmath56 steradians ) may be due to the effects of general relativistic light bending ( e.g. , fabian et al . 2002 ; fabian & vaughan 2003 ) . to improve our ability to detect or constrain the strengths of any reflection features , we jointly fitted the spectra of our nine high - redshift sources from table [ tab_compton ] . in the joint fitting , we used the pexrav+zgauss models as before and tied all the @xmath52 values and normalizations to a single value ( the flux level of each source was controlled by assigning to it a scaling factor that was allowed to vary freely , and the photon index for each source was also free to vary ) . we ran the joint fitting process four times . in the first run , all nine sources were considered . in the second run , we removed the two sources from table [ tab_compton ] in which a significant compton - reflection continuum was detected . in the third run , we removed all sources in which a non - zero @xmath52 value was detected , and in the fourth run we removed pg1630@xmath21377 in which an line was detected . the best - fit values of ew ( ) and @xmath52 from the joint - fitting process appear in table [ tab_compton_joint_fit ] where we also provide the total number of counts used in the fit and the mean redshift of the sources considered in each run . by comparing the results in table [ tab_compton_joint_fit ] with those in table [ tab_compton ] , one can see that when all sources are considered ( runi ) , a large value of @xmath52 is detected , while a tight constraint is placed on the mean ew ( ) ; this result is dominated by the properties of pg1247@xmath21267 and pg1634@xmath21706 that comprise @xmath760% of the counts . in runii , one can see that despite the removal of all sources in which a compton - reflection continuum was detected , a mean reflection component is detected , although with a relatively small value of @xmath52 . in runiii , the detection is probably due to the fact that pg1630@xmath21377 comprises @xmath730% of the total counts ; no significant compton reflection is detected in this case . in the final run , no significant mean reflection emission is detected ; the constraint on @xmath52 is relatively tight while the constraint on ew ( ) is rather weak . lccccccccc lbqs0109@xmath210213 & @xmath57 & @xmath58 & 70/53 & @xmath59 & 46.8 & 7959 & 10.4 & 0.1 & @xmath201.89 + 2qzj023805.8@xmath20274337 & @xmath60 & @xmath61 & 47/39 & @xmath62 & 46.6 & 3403 & 9.5 & 0.5 & @xmath201.66 + q1318@xmath20113 & @xmath63 & @xmath64 & 70/62 & @xmath65 & 46.9 & 4665 & 10.0 & 0.3 & @xmath201.64 + sbs1425@xmath21606 & @xmath66 & @xmath67 & 38/39 & @xmath68 & 47.4 & 4964 & 10.4 & 0.4 & @xmath201.82 + he0926@xmath200201 & @xmath69 & @xmath70 & 2/10 & @xmath71 & 47.0 & 5100 & 10.1 & 0.3 & @xmath201.80 these results do not provide a clear picture for the dependence of the reflection spectrum on luminosity or the accretion rate . most of our high - redshift sources do not exhibit significant reflection components , as might be expected given their very high luminosities . only two of these sources show a significant compton ` hump ' , and a strong line is detected in another . in addition , significant detections of compton humps are not accompanied by corresponding detections of lines . the results of the joint - fitting process also portray a mixed picture for the average reflection spectrum of luminous , high - accretion rate rq agns as a class . the relatively weak constraints on the mean reflection properties of this class do not allow us to either confirm or rule out the existence of an ` baldwin effect ' , regardless of whether the luminosity or the accretion rate drives the anticorrelation with ew ( ) ( e.g. , zhou & wang 2005 ; bianchi et al . 2007 ) . by extending the spectral fitting to the entire epic energy range ( 0.210kev ) , we checked whether any of our nine high - redshift sources with spectra ( including the five high - redshift sources from s06 ) shows evidence for excess soft- emission , frequently observed in lower - redshift agns ( e.g. , comastri et al . 1992 ; reeves & turner 2000 ; piconcelli et al . while the physical nature of the soft excess is uncertain its presence is more pronounced among local high - accretion rate agns , i.e. , narrow - line seyfert 1 ( nls1 ) galaxies ( e.g. , vaughan et al . 1999a ; boller et al . 2002 ; czerny et al . 2003 ; vignali et al . 2004 ) ; hence it is of interest to search for its existence in our luminous high - accretion rate sources at high redshift . we extrapolated the best - fit galactic - absorbed power - law model obtained for rest - frame energies 2kev ( see [ sec_xmm ] and table [ tab_properties ] ) to the 0.210 kev observed - frame energy range and no signature of soft excess emission was detected in any of the sources ( i.e. , no systematic residuals were present for the extrapolated fits ) . this result is not unexpected given the relatively high redshifts of our sources and the low - energy cutoff ( 0.2kev ) of the epic cameras . for example , porquet et al . ( 2004 ) have found that the effective temperatures of the soft excess - components in a sample of moderate - luminosity sources at @xmath44 are of the order of @xmath72 kev . for our ten sources with @xmath73 , such temperatures shift to observed - frame values of @xmath74 kev . furthermore , since the soft - excess component in most type 1 agns is typically observed below a rest - frame energy of @xmath75 kev ( except for some high @xmath1 nls1s , in which this component may extend up to 1.5 kev ; e.g. , vaughan et al . 1999b ) , the detection of such a component with is challenging even for sources with @xmath76 . nevertheless , we have placed upper limits on the strength of a potential soft - excess component in two of our ten luminous sources that have the lowest redshifts , namely pg1630@xmath21377 and pg1634@xmath21706 ( with @xmath77 and @xmath78 , respectively ) . in these two cases , the extension of the energy range to 0.2kev has increased the number of photons in the fit by @xmath79% . the constraint on the data - to - model ratio in the 2kev rest - frame band , as a result of the extrapolation of the best - fit slope to the entire epic energy range , is 2.0 and 1.5 , respectively . finally , we note that the non - detection of soft - excess emission in pg1247@xmath21267 , pg1630@xmath21377 , and pg1634@xmath21706 is consistent with previous studies of their spectra ( page et al . 2004b ; piconcelli et al . 2005 ) . two of our sources , sbs1425@xmath21606 and lbqs0109@xmath210213 , were previously detected by the _ rosat _ pspc and hri , respectively . for the first of these , we measured an unabsorbed flux of @xmath80ergscm@xmath12s@xmath17 in the 0.52.0kev observed - frame band from our data , which is somewhat higher than , but consistent ( within the errors ) with the _ rosat _ flux ( see also just et al . similarly , we measured for the second source an unabsorbed flux of @xmath81ergscm@xmath12s@xmath17 in the 0.52.0kev observed - frame band , which is a factor of @xmath82 lower than the value calculated from the _ rosat _ count rate . in both cases , we assumed that the photon index we measured from the observations at rest - frame energies of @xmath29kev also extends to the _ rosat _ bandpass ( i.e. , down to energies of 0.1kev in the observed frame ) , based on the lack of detectable soft excesses in these sources ( [ sec_soft_excess ] ) . although highly luminous rq agns such as those presented here are not expected to exhibit pronounced variations even on timescales of several years ( e.g. , lawrence & papadakis 1993 ) , the case of lbqs0109@xmath210213 is not unique , since several other luminous sources at high redshifts have displayed variations with similar amplitudes ( e.g. , paolillo et al . 2004 ; shemmer et al . 2005 ) . lccc lbqs0109@xmath210213 & 2.349 & @xmath83 & @xmath84 + 2qzj023805.8@xmath20274337 & 2.471 & @xmath85 & @xmath86 + pg1247@xmath21267 & 2.038 & @xmath87 & @xmath88 + q1318@xmath20113 & 2.306 & @xmath89 & @xmath90 + q1346@xmath20036 & 2.370 & @xmath91 & @xmath92 + sbs1425@xmath21606 & 3.202 & @xmath93 & @xmath94 + pg1630@xmath21377 & 1.476 & @xmath95 & @xmath96 + pg1634@xmath21706 & 1.334 & @xmath97 & @xmath98 + he2217@xmath202818 & 2.414 & @xmath99 & @xmath100 we also searched for rapid ( on timescales of @xmath71hr in the rest frame ) variations in the and _ chandra _ data of our five new sources by applying kolmogorov - smirnov tests to the lists of photon arrival times from the event files , but no significant variations were detected . together with the results of s06 , this suggests that high accretion rates do not necessarily lead to faster and higher - amplitude flux variations , as has been previously expected based on variability properties of some nls1s , hence spoiling the analogy between nls1s and luminous , high - redshift agns ( e.g. , grupe et al . . on the other hand , our results are consistent with the idea that more massive bhs lead to longer timescales and smaller amplitudes of flux variations ( e.g. , oneill et al . 2005 ) . in fig.[fig_ge ] we plot @xmath2 versus fwhm(h@xmath9 ) and @xmath1for our core sample of 35 agns . twenty five of these sources are palomar green ( pg ) quasars ( schmidt & green 1983 ) at @xmath44 with @xmath101 { \raisebox{-.5ex}{$\;\stackrel{<}{\sim}\;$}}46 $ ] , and ten are at @xmath0 with @xmath102 { \raisebox{-.5ex}{$\;\stackrel{<}{\sim}\;$}}48 $ ] . in s06 , it was shown that both fwhm(h@xmath9 ) and @xmath1 are significantly correlated with @xmath2 when only the 25 moderate - luminosity ( @xmath44 ) sources are considered ( in agreement with porquet et al . 2004 ; piconcelli et al . 2005 ; see also table [ tab_corr_coeff ] ) . this fwhm(h@xmath9)-@xmath1 degeneracy emerges as a consequence of considering only sources with a relatively narrow luminosity range , i.e. , @xmath101 { \raisebox{-.5ex}{$\;\stackrel{<}{\sim}\;$}}46 $ ] ( see also [ sec_l5100_hb ] and s06 ) . when the ten highly luminous agns at are introduced to the analysis , both fwhm(h@xmath9 ) and @xmath1 remain significantly correlated with @xmath2 ( with @xmath103% confidence ) ; however , the significance ( in terms of chance probability ) of the @xmath2-fwhm(h@xmath9 ) correlation drops considerably while the significance of the @xmath2-@xmath1 correlation increases ( table [ tab_corr_coeff ] ) . the results of the @xmath2-@xmath1correlation do not significantly change when a constant bolometric correction factor of 7 ( which is the average correction factor for the luminosity range of our sample ; see [ sec_l5100_hb ] ) is used to derive @xmath1 values ; in this case , the significance of the correlation increases with the chance probability dropping from @xmath104 to @xmath105 . as suggested in s06 , the extension of the luminosity range of the sample to high luminosities has allowed us to break the degeneracy between the @xmath2-fwhm(h@xmath9 ) and @xmath2-@xmath1 correlations and show that @xmath1 drives the correlation with @xmath2 . lccccc i & 9 & 39,173 & 2.22 & @xmath106 & @xmath107 + ii & 7 & 15,307 & 2.37 & @xmath108 & @xmath109 + iii & 5 & 9,814 & 2.39 & @xmath110 & @xmath86 + iv & 4 & 7,112 & 2.61 & @xmath111 & @xmath112 we have also repeated the mann - whitney ( mw ) nonparametric rank test performed in s06 on all sources with fwhm(h@xmath9 ) values that lie in the range 3400@xmath113fwhm(h@xmath9)@xmath1138000 ^ -1 km s@xmath17 [ which is the fwhm(h@xmath9 ) interval of the ten luminous sources in our sample ] . we find a significant deviation ( with @xmath114% confidence ) between the @xmath2 values of the ten luminous sources and the nine moderate - luminosity sources in this range . in contrast , when the mw test is performed in the [email protected] range , the @xmath2 values of the two groups of agns are not significantly different ( as can be seen clearly from fig.[fig_ge]b ) . these results reinforce the s06 argument that @xmath2 depends primarily on @xmath1 . we note that in the case of two of our luminous sources , lbqs 0109@xmath210213 and sbs 1425@xmath21606 , the @xmath2 values are consistent with the expected values of both fwhm(h@xmath9 ) and @xmath1 , and hence they do not assist in removing the fwhm(h@xmath9)-@xmath1degeneracy . the first of these sources has a remarkably flat hard- spectral slope ( with @xmath26 ; see [ sec_xmm ] and table [ tab_properties ] ) , which was not expected given its original @xmath1 determination from shemmer et al . ( 2004 ; see their table 2 ) . owing to the new and improved spectroscopic measurements of netzer et al . ( 2007 ) for lbqs 0109@xmath210213 ( sbs 1425@xmath21606 ) , fwhm(h@xmath9 ) has increased from 5781 ^ -1 km s@xmath17 to 7959 ^ -1 km s@xmath17 ( 3144 ^ -1 km s@xmath17 to 4964 ^ -1 km s@xmath17 ) , and hence @xmath1 has decreased from 0.2 to 0.1 ( 0.9 to 0.4 ) . for the other shemmer et al . ( 2004 ) sources used in this work , the netzer et al . ( 2007 ) fwhm(h@xmath9 ) measurements are consistent with those of shemmer et al . the optical data presented in table [ tab_properties ] are based on the new netzer et al . ( 2007 ) measurements . our result that @xmath2 depends more strongly on @xmath1 , rather than on fwhm(h@xmath9 ) , may also serve as a partial vindication for the use of the reverberation - mapping based method to determine @xmath1 ( see [ sec_l5100_hb ] ) . from a pure mathematical point of view , in the tests described above , we essentially multiplied one observational parameter , fwhm(h@xmath9 ) , with another , @xmath115 , and yet the dependence of @xmath2 on the product of the two parameters ( i.e. , @xmath1 ) turned out to be stronger than the dependence on the single parameter . if the product of the two parameters had no physical significance , then there would be no apparent reason to expect this result . we use the significant correlation we find between @xmath2 and @xmath1 to derive a linear relationship between the two parameters employing a variety of statistical methods . hereafter , we take the error bars on @xmath2 at the 1@xmath116 level . a standard @xmath23 minimization method weighted by the errors on @xmath2 yields the following relation ( with 1 @xmath116 errors ) : @xmath117 the best - fit coefficients of this relation are consistent with those obtained by wang et al . ( 2004 ) and kelly ( 2007 ) who find a similar correlation in low moderate luminosity sources . however , the @xmath118 value obtained by this minimization suggests that either the data are not well represented by a linear model , the error bars on @xmath2 are underestimated , or there is additional intrinsic scatter in the data . following the methods outlined in tremaine et al . ( 2002 ) and k05 , and by assuming a @xmath119 ( i.e. , a reduced @xmath120 ) , we obtain an estimate of the additional potential scatter of @xmath121 in the dependent parameter ; this scatter is larger than the typical measurement errors on @xmath2 . lccc @xmath2-fwhm(h@xmath9 ) & @xmath122 & @xmath123 & 25 + @xmath2-fwhm(h@xmath9 ) & @xmath124 & @xmath125 & 35 + @xmath2-@xmath1 & @xmath126 & @xmath127 & 25 + @xmath2-@xmath1 & @xmath128 & @xmath129 & 35 the observed intrinsic scatter may be induced by the uncertainty in the bh - mass estimate , and/or additional unknown physical properties , such as the optical depth in the corona , orientation , and bh spin , that vary from source to source . an additional potential source for this scatter may be attributed to variability ; for example , variability may seem to have a significant contribution to the scatter since the data and the rest - frame optical data ( used for obtaining @xmath1 ) are not contemporaneous . however , as discussed in s06 , variability ( i.e. , changes in both @xmath2 and @xmath1 ) is not expected to dominate the observed scatter in this correlation ( see also k05 ; steffen et al . 2006 ; wilhite et al . 2007 ) ; its effects are expected to be even less significant in this case , since our sources are mostly luminous and thus have smaller amplitudes of and optical flux variations ( e.g. , lawrence & papadakis 1993 ; kaspi et al . 2007).10 or more ) and corresponding spectral slope variations ( with @xmath130 ) that are often due to changing spectral contributions from reflection and/or absorption ( e.g. , taylor et al . 2003 ; krongold et al . 2007 ; grupe et al . such changes have not been observed for moderate high luminosity sources such as those under study in this work . ] to test whether @xmath2 can serve as an accretion - rate indicator we switched the roles of @xmath2 and @xmath1 , allowing the first to serve as the independent variable . to account for the scatter in the correlation , we performed a linear - regression analysis using the bivariate correlated errors and scatter method ( bces ; akritas & bershady 1996 ) on the data . for this purpose , we assumed that typical uncertainties on the determination of @xmath4 are a factor of @xmath131 ( e.g. , kaspi et al . 2000 ) , and therefore assigned homoscedastic ( i.e. , uniform variance ) , @xmath132 errors of 0.3 dex on @xmath133(@xmath1 ) . the best - fit linear relation using the bces bisector result ( with 1 @xmath116 errors ) is @xmath134 a linear regression based on @xmath23 minimization using the fitexy routine ( press et al . 1992 ) gives consistent results to those obtained with bces with @xmath135 . in addition , we also performed a linear - regression analysis on the data using the maximum - likelihood estimate ( mle ) of kelly ( 2007 ) . the results of the different linear fits are presented in table [ tab_linear_fits ] , and are shown in fig . [ fig_ge_yx ] . in all three linear - regression methods outlined above , the average scatter on the predicted value of @xmath133(@xmath1 ) is @xmath136 dex . this is only slightly higher than the typical uncertainty , @xmath137 dex , associated with @xmath133(@xmath1 ) determinations using the reverberation - mapping based @xmath4 determinations . we have also computed @xmath1 values using the bentz et al . ( 2007 ) slope for the belr size - luminosity relation ( see [ sec_l5100_hb ] ) . this flatter slope ( @xmath35 ) results in @xmath4 values that are smaller by a factor of @xmath75 , on average , and hence @xmath1 values that are larger , on average , by the same factor . this has no noticeable effect on the strength of the correlation ( as expected for a nonparametric correlation ) or on its slope ; only the intercept ( i.e. , @xmath138 values in table [ tab_linear_fits ] ) is affected , and it increases slightly within the current uncertainty on its value . we conclude that the @xmath2-@xmath1 relation is not significantly affected by small deviations from the k05 slope . cccccc @xmath139 & @xmath140 & @xmath141 & @xmath142 & @xmath143 & @xmath144 based on our high - quality sample of 35 sources , we conclude that the hard- power - law spectral slope can predict the value of the normalized accretion rate in rq agns , across four orders of magnitude in agn luminosity , with an acceptable uncertainty level of a factor of 3 . this may offer a useful new tool to probe the history of bh growth , based almost exclusively on the availability of high - quality hard- spectra of rq agns ; we discuss this in more detail in [ sec_bhgrowth ] below . although not unexpected , and previously predicted by several studies ( see [ sec_introduction ] ) , our result is by far the most reliable indication that the shape of the hard- power - law spectrum is largely controlled by @xmath1 . as mentioned in [ sec_introduction ] , a possible explanation is that the corona acts as a ` thermostat ' by cooling more efficiently when the disk emission increases , manifested by a steepening of the hard- spectrum . in this context , it is also interesting to note that our result may readily explain the narrow ranges observed for the values of both @xmath2 and @xmath1 in optically selected moderate high luminosity , type 1 rq agns . while @xmath1 is relatively narrowly distributed around a value of @xmath70.3 ( 0.28 in our sample of 35 sources ) , with a typical dispersion of a factor of @xmath75 around that value ( e.g. , mclure & dunlop 2004 ; kollmeier et al . 2006 ; netzer et al . 2007 ; shen et al . 2008 ) , @xmath2 values for such sources typically lie in the narrow range of ( 1.22.6 for our sample of 35 sources ; see also vignali et al . these ranges in @xmath1 and @xmath2 may be different for lower luminosity sources . in particular , for sources with @xmath145(5100)@xmath146ergss@xmath17 , a possible difference may , in part , be attributed to a difference in their accretion mode as compared with the moderate high luminosity sources studied in this work . finally , we point out that a possible dependence of the bolometric correction for @xmath147 on @xmath1 , recently reported by vasudevan & fabian ( 2007 ) , may , in part , be reflected by our results . the @xmath5 parameter is known to have a strong anticorrelation with optical uv luminosity at 2500 [ hereafter @xmath148 , and it shows no significant dependence on redshift ( e.g. , vignali et al . 2003 ; strateva et al . 2005 ; steffen et al . 2006 ; just et al . 2007 ; but see also kelly et al . 2007 ) . this is equivalent to a non - linear relation between and optical uv luminosity of the form , where @xmath149 ( see eq.7 of just et al . 2007 who find ) . this relation still lacks a sound physical interpretation and , in particular , it is not clear what mechanism controls the proportion of reprocessed hard- emission from the corona with respect to the uv emission from the disk . in order to test whether @xmath5 depends on @xmath1 , we selected 81 sources from the steffen et al . ( 2006 ) sample with available fwhm(h@xmath9 ) measurements ( all these have @xmath150 and they include the 25 sources with @xmath44 from our core sample ; [ sec_degeneracy ] ) , and complemented this sample with the ten luminous sources at @xmath0 from s06 and this work . for about half of the steffen et al . ( 2006 ) sub - sample , fwhm(h@xmath9 ) values were obtained from boroson & green ( 1992 ) , and for the rest we used fwhm(h@xmath9 ) measurements from netzer & trakhtenbrot ( 2007 ) , kindly provided by b. trakhtenbrot , 2007 , private communication . for each source we determined the value of @xmath1 following the procedure outlined in [ sec_properties ] . following the arguments made in [ sec_ge_corr ] , the non - simultaneous and optical observations are not expected to affect our analysis considerably ( see also 3 of steffen et al . 2006 ) . we find that @xmath5 is significantly correlated with both @xmath151 and @xmath1 ; however , the correlation with luminosity is considerably tighter ( see fig . [ fig_aox ] ) . even when a constant bolometric correction factor of 7 is used to determine @xmath1 values ( as performed in [ sec_degeneracy ] ) , the correlation is considerably tighter than the correlation , although the significance of the latter correlation has increased ( the chance probability has dropped from to ) . by investigating this further , we find that the @xmath5-@xmath1correlation disappears when the correlation is taken into account ; i.e. , the difference between the observed @xmath5 and the expected @xmath5 [ based on the @xmath152 values of the sources and the most recent relation from just et al . 2007 ] is not correlated with @xmath1 . in another test , we divided the data into sources that have @xmath1values either above or below the median value of @xmath1@xmath153 . in both sub - samples we detect a strong correlation ( and no correlation ) . these tests suggest that the correlation is persistent and that @xmath5 does not depend primarily on @xmath1 . in this scenario , the ( relatively weak ) correlation may be a consequence of the inherent dependence of @xmath1 on @xmath151 , and the relative weakness of this correlation compared with the strong correlation may be due to the additional uncertainty introduced by determining @xmath1 ( see , for example , the typical error bars in fig . [ fig_aox ] ) . subsequently , if @xmath5-@xmath151 is the fundamental correlation , then the inclusion of the fwhm(h@xmath9 ) parameter to produce @xmath1 adds substantial scatter resulting in a rather weak @xmath5-@xmath1 correlation . it is also of interest to point out that we find no correlation between @xmath5 and @xmath2 ( that depends on @xmath1 ) in our core sample of 35 sources discussed in [ sec_ge_corr ] . we also checked whether the @xmath5-@xmath152 correlation is induced by luminosity - dependent obscuration ( e.g. , lawrence 1991 ; gaskell et al . 2004 ) . in a first test , we found a strong correlation between @xmath5 and @xmath154 for our sample , with similar coefficients as in the @xmath5-@xmath151 correlation . we note that the fluxes at 5100were obtained independently from the fluxes at 2500 , i.e. , no extrapolations have been made between the flux densities in these two wavelengths . in a second test , we redefined @xmath5 by changing the optical uv continuum threshold from 2500 to 5100 ; the correlations between this modified @xmath5 and both @xmath155 and @xmath152 for our sample returned similar correlation coefficients as in the @xmath5-@xmath151 case . we conclude that , within the limits of our sample , it is unlikely that the @xmath5-@xmath152 correlation is induced by potential reddening effects . the fact that @xmath152 appears as the underlying parameter determining the slope of the sed instead of @xmath1 is puzzling , since @xmath1 indicates the relative agn power and significantly affects the sed of the disk . one possibility is that our empirical expression for @xmath1 ( see s06 ) is inaccurate or more complicated . this , however , is not supported by the fact that @xmath2 _ does _ depend on @xmath1 as defined here and not on @xmath33 ( see [ sec_introduction ] and [ sec_degeneracy ] ; s06 ) . we have also considered the possibility that the correlation is induced by selection effects ( e.g. , green et al . 2006 ) , but this is highly unlikely given the fact that the correlation spans over five orders of magnitude in luminosity ( see , e.g. , just et al . 2007 ) . indeed , future correlations between @xmath5 , @xmath152 , and @xmath1 should involve @xmath1determinations for -selected sources ( see [ sec_bhgrowth ] below ) . an alternative approach to interpreting the non - dependence of @xmath5 on @xmath1 stems from the recent results of vasudevan & fabian ( 2007 ) . these authors confirm the strong dependence of @xmath5 on @xmath152 in their agn sample , but they do not find an @xmath5-@xmath1 correlation . utilizing far - uv , as well as optical , near - uv , and spectra for a sample of agn with known @xmath4 and @xmath1 , they argue that the reason for the non - dependence of @xmath5 on @xmath1 lies in the choice of the optical uv and continuum thresholds ( i.e. , 2500 and 2kev ) for the @xmath5definition . while they find dramatic differences between the optical sed of sources with different @xmath1 values , the ratio between the flux densities at 2500 and 2kev remains almost constant as @xmath1 changes . the most pronounced differences between the optical seds of different sources are expected to be concentrated in the @xmath156kev spectral region ( i.e. , the ` big blue bump ' ) , which can not be traced effectively using the current optical uv and thresholds of @xmath5 . in this context , the @xmath5 parameter , given its current definition , while providing useful information on the @xmath157-@xmath3 connection , can not be used as an accretion - rate indicator . by inspection of fig.13 of vasudevan & fabian ( 2007 ) , one may expect a strong dependence of @xmath5 on @xmath1 if , for instance , the optical uv threshold is shifted from 2500 to as close as possible to @xmath158 , although this is observationally challenging and may not be practical . when using modified definitions for @xmath5 , there also remains the question of which correlation , @xmath5-@xmath1 or @xmath5-@xmath157 , is stronger than the other . in general , the choice of optical uv and thresholds should likely depend upon the scientific question of interest . in [ sec_ge_corr ] we have shown that @xmath2 may serve as an @xmath1 indicator in unabsorbed , moderate high luminosity rq agns with an acceptable mean uncertainty level on @xmath1 of a factor of 3 . therefore , in principle , given a high - quality spectrum in the 2kev rest - frame band of a rq agn , one can empirically estimate @xmath1 and @xmath4 for the source . accurate measurements of @xmath2 and @xmath3 ( in the 210kev rest - frame band , for example ) may provide @xmath1 and @xmath10 using our eq.[eq_lledd_gamma ] and eq.21 of marconi et al . ( 2004 ) , into @xmath10 depends on @xmath1 , while marconi et al . ( 2004 ) have used the vignali et al . ( 2003 ) @xmath5-@xmath159 correlation ( which is not significantly different from the most recent correlation of this type given by just et al . 2007 ) to derive bolometric corrections that depend on @xmath10 . accurate measurements of @xmath2 ( that provides @xmath1 ) and @xmath3 will enable comparisons of @xmath10 values obtained from these two methods . ] respectively , and thus provide an estimate of @xmath4 . we note that the marconi et al . ( 2004 ) determination of @xmath10 ( whether from @xmath160 or @xmath3 ) relies , in part , on the very strong and non - linear @xmath3-@xmath157 dependence ( see [ sec_sed ] ) . the observed ( luminosity - dependent ) rms errors on that relation are given in table5 of steffen et al . ( 2006 ) ; these represent typical deviations of up to @xmath161 from the mean @xmath5 [ i.e. , a factor of @xmath162 uncertainty on @xmath163 for our luminosity range . this scatter is inherent in the @xmath1-@xmath2 correlation as it reflects uncertainties in determining @xmath10 from @xmath164 and potential -optical variability . for example , when @xmath4 values for our core sample of 35 sources are recovered from @xmath1 using the bolometric corrections for @xmath165 , the ratio between these @xmath4 values and the original @xmath4 values obtained in [ sec_l5100_hb ] is 1 , on average , with a dispersion of a factor of 2 . in an additional test , we used the @xmath2 and @xmath147 values of six nearby sources from our sample to recover their @xmath4values and compare them with the most recent , direct ( reverberation - mapping based ) @xmath4 measurements from peterson et al . the six sources , namely pg 0804@xmath21761 , pg 0844@xmath21349 , pg 0953@xmath21414 , pg 1211@xmath21143 , pg 1307@xmath21085 , and pg 1613@xmath21658 , comprise @xmath166% of all the rq agns with @xmath145(5100)@xmath167ergss@xmath17 in the kaspi et al . ( 2000 ) sample ( i.e. , the pg quasar sample ) that have direct @xmath4measurements . we find that the ratio between the recovered -based masses and the corresponding @xmath4 measurements from peterson et al . ( 2004 ) is 0.8 , on average , with a dispersion of a factor of 1.9 ; we also find that the mass recovered for each individual source is consistent with the measured value , given the uncertainties from eq . [ eq_lledd_gamma ] and those from the reverberation - mapping measurements . high - quality spectra may thus be useful for estimating the accretion rates and bh masses for moderate high luminosity type 1 , rq agns and , in particular , for -selected sources . this method may allow tracing the history of bh growth in the universe by utilizing large agn datasets ( see e.g. , brandt & hasinger 2005 ) in which bh growth can not be determined effectively using existing methods [ i.e. , reverberation - mapping based methods ( aka single - epoch methods ) , e.g. , k05 , and host - agn type relations , e.g. , marconi & hunt ( 2003 ) ] , either due to the faintness of the sources in the optical near - ir bands , or that the required spectroscopic features are either not accessible or can not be modeled reliably . one advantage of the -based method is the ability to obtain @xmath1 and @xmath4 by measuring broad spectroscopic properties ( i.e. , @xmath2 ) as opposed to detailed spectroscopy for obtaining the width of an emission line ( e.g. , h@xmath9 or ; this involves , for example , careful decontamination of emission features from the uv optical spectra ) . moreover , measurements of the hard- power - law spectral slope are typically not limited to specific redshift ranges , such as those dictated by atmospheric transmission bands ( and detector bandpass ) for ground - based spectroscopy . these advantages offer a way to obtaining many @xmath1 and @xmath4 estimates , economically , in contrast with more complicated ( and redshift - restricted ) spectroscopic measurements for individual sources done with existing methods . nevertheless , we caution that the measurements of @xmath2 and @xmath3 should be done carefully to account for potential complex absorption and compton reflection . the -based method for estimating @xmath1 and @xmath4 may prove to be even more rewarding in cases where a source is either optically faint and/or the broad - emission lines are too weak to measure if , for example , these lines are overwhelmed by host - galaxy continuum ( e.g. , moran et al . 2002 ; provided the source is not radio loud and that the redshift can be determined ) . in particular , this method may be the best way to determine @xmath1 and @xmath4 directly in obscured ( i.e. , optical type2 ) , moderate high luminosity agns . provided the absorption column density is not too high ( @xmath24@xmath168cm@xmath12 ) and can be modeled accurately , the intrinsic hard- power - law spectrum can , in principle , provide @xmath1 and @xmath4 as outlined above . examples of such sources with high - quality spectra are given in e.g. , civano et al . ( 2005 ) , mateos et al . ( 2005 ) , and mainieri et al . ( 2007 ) . since compton - thin type 2 agns comprise a significant fraction of the agn population , tracing the growth of the supermassive bhs in their centers using the -based method is crucial for testing models of the evolution of bh growth and accretion luminosity in the universe in an unbiased way . we present spectroscopy for five highly luminous rq agns at @xmath0 , with accurate fwhm(h@xmath9 ) measurements that allow determinations of their @xmath4 and @xmath1 values . analysis of the spectra provided measurements of the hard- photon index in the rest - frame 2kev band and @xmath5 . we have combined these data with the s06 sample of 30 moderate high luminosity sources with similar properties , while doubling the number of highly luminous sources in their sample . our main goal was to test the s06 claim that @xmath2 can serve as an accretion - rate indicator in rq agns . we have also tested whether any additional properties of our highly luminous sources depend on @xmath1 . our main results are summarized as follows : 1 . our new highly luminous sources with fwhm(h@xmath9 ) measurements have allowed us to break the degeneracy between the dependence of @xmath2 on fwhm(h@xmath9 ) and on @xmath1 , suggesting that the accretion rate largely determines the hard- spectral slope across four orders of magnitude in agn luminosity ( i.e. , @xmath169 { \raisebox{-.5ex}{$\;\stackrel{<}{\sim}\;$}}48 $ ] ) . 2 . we found a significant correlation between @xmath1 and @xmath2 with a best - fit line of the form @xmath170 , and an acceptable uncertainty of a factor of 3 on a predicted value of @xmath1 . 3 . utilizing a sample of 91 sources from steffen et al . ( 2006 ) and this work , we find that @xmath5 depends strongly on optical uv luminosity and only weakly on @xmath1 ; the ( weak ) correlation with @xmath1 is probably due to the strong dependence . we discuss possible explanations for this result including the possibility that @xmath5 can not be used as an accretion - rate indicator based on its current definition . we find a significant compton - reflection feature in two of our sources , and the mean relative reflection for seven other sources is @xmath171 . by setting rather loose constraints on the strengths of emission lines in our highly luminous sources , we can neither confirm nor rule out a suggested anticorrelation between ew ( ) and either luminosity or @xmath1 ; the upper limit on the mean rest - frame ew ( ) for seven of these sources that do not show compton - reflection features is 105ev . we have not detected any signature of a soft - excess component in any of our highly luminous sources , including two sources at @xmath172 where our rest - frame coverage extends to @xmath75kev , suggesting that the soft excess does not depend strongly on the accretion rate . although one of our highly luminous sources , lbqs0109@xmath210213 , exhibits long - term ( i.e. , on timescales of years ) variations , rapid variations on timescales of @xmath71hr have not been detected in any of our highly luminous ( and high-@xmath4 ) sources , supporting the idea that variability timescale depends inversely on @xmath4 and does not depend on @xmath1 . the strong correlation between @xmath2 and @xmath1 may serve as a useful probe for tracing the history of bh growth in the universe . it may provide @xmath1 and @xmath4 estimates for -selected agns , with the possibility of estimating these properties for compton - thin type2 agns for the first time . this work is based on observations obtained with , an esa science mission with instruments and contributions directly funded by esa member states and the usa ( nasa ) . we thank an anonymous referee for a helpful report that assisted in improving the presentation of this work . we are also grateful to franz bauer , george chartas , brandon kelly , and aaron steffen for useful comments and fruitful discussions . we gratefully acknowledge the financial support of nasa grants and ( o.s , w.n.b ) , nasa ltsa grant ( o.s , w.n.b ) , and the zeff fellowship at the technion ( s.k ) . this work is supported by the israel science foundation grant 232/03 .

we prove that the asymptotic of the bulk local statistics in models of random lozenge tilings is universal in the vicinity of straight boundaries of the tiled domains . the result applies to uniformly random lozenge tilings of large polygonal domains on triangular lattice and to the probability measures describing the decomposition in gelfand tsetlin bases of tensor products of representations of unitary groups . in a weaker form our theorem also applies to random domino tilings .

1603.02707

this article is about random _ lozenge tilings _ , which are tilings of domains on the regular triangular grid by rhombuses of three types ( see figures [ fig_hex ] , [ fig_trap ] , [ fig_covered_domains ] , [ fig_not_covered_domain ] , [ fig_gt_tiling ] ) , and which can be identified with dimers , discrete stepped surfaces , @xmath0 young diagrams , see e.g. @xcite , ( * ? ? ? * section 2 ) . in more details , we are interested in the asymptotic behavior of _ gibbs measures _ on tilings , which means that tilings in each subdomain are always conditionally uniformly distributed given boundary conditions , i.e. positions of lozenges surrounding this domain . there are several ways to produce such gibbs measures . from the point of view of statistical mechanics , the most natural one is to fix a very large planar domain and consider the uniform measure on _ all _ ( finitely many ) lozenge tilings of this domain . asymptotic representation theory suggests two more ways , related to decompositions of tensor products of representations of classical lie groups and restrictions of characters of the infinite - dimensional counterparts of these groups , see e.g. @xcite , @xcite . finally , one can also _ grow _ tilings by means of interacting particle systems , see e.g. @xcite , @xcite . the rigorous asymptotic results available in the literature for various random lozenge tilings models all share the same features . let us describe such features on the example of the uniformly random lozenge tilings of an @xmath1 hexagon as @xmath2 ( in such a way that @xmath3 and @xmath4 converge to constants ) , see figure [ fig_hex ] . note that such tilings can be linked both to decompositions of certain irreducible representations ( cf . @xcite ) and to growth models ( see @xcite ) . the following asymptotic features of the tilings of hexagons are known : * the random lozenge tilings exhibit the _ law of large numbers _ section [ section_lln ] . one way to phrase it is that in each macroscopic sub - region of the hexagon the asymptotic proportions of three types of lozenges converge in probability to deterministic numbers , described by integrals over this sub - domain of three functions @xmath5 , @xmath6 , and @xmath7 . in particular , outside the inscribed ellipse one observes the _ frozen region _ where asymptotically only one type of lozenges remains present , see @xcite , @xcite , @xcite . the region where all three types of lozenges are asymptotically present is called _ liquid . _ * the global fluctuations of the _ height function _ of tiling ( see section [ section_lln ] for the detailed definition ) are asymptotically normal and can be described via the pullback of the gaussian free field with dirichlet boundary conditions , see @xcite , @xcite , @xcite . * locally , near each point inside the inscribed ellipse in the limit one observes a _ translation invariant ergodic gibbs measure _ on lozenge tilings of the plane . its _ slope _ is determined by the law of large numbers ( such measure is unique for each slope , see @xcite , and its correlation functions admit known closed expressions , cf . section [ section_main_proof ] ) , see @xcite , @xcite , @xcite . * the one point fluctuations of the boundary of the frozen region at generic point after proper rescaling converge to the tracy widom distribution @xmath8 , and multi point fluctuations are described by the airy line ensemble , see @xcite , @xcite . * at a point where the frozen boundary is tangent to a side of the hexagon , the fluctuations are described by the gue corners process , see @xcite , @xcite , @xcite , @xcite . the _ universality belief _ predicts that all these features should be very robust along different models of lozenge tilings , and should not depend on details . it means , that while the limit shape ( i.e. the asymptotic proportions @xmath5 , @xmath6 , @xmath7 ) and the exact shape of the boundaries of the frozen regions differ from system to system , but they should always exist . the fluctuations of the height function should always be given by a pullback of the gaussian free field , and only the map with respect to which this pullback is taken , might change . bulk local limits should be always described by translation invariant gibbs measures and only the slope of such measure might vary from system to system . finally , the fluctuations of the boundaries of frozen regions should be always governed by the tracy widom distribution , airy line ensemble and gue corners process . while this conjectural universality is confirmed by numerous examples , but by now only the law of large numbers has been proven in sufficient generality , cf.@xcite , @xcite , @xcite . in the present article we address the third feature about local ( also called `` bulk '' ) scaling limits . in simple words , we show that whenever for a random lozenge tiling the law of large numbers holds , if the tiled domain has a straight boundary , then near this boundary the _ universality _ for the local limits is valid . we proceed to a more detailed formulation of our main result . consider a trapezoid drawn on the triangular grid as shown in figure [ fig_trap ] . such domain is parameterized by the length of the left vertical side @xmath9 and the width @xmath10 and we denote it @xmath11 . consider the set of all tilings of @xmath11 , in which we allow horizontal lozenges to _ stick out _ of the right boundary . the combinatorial constraints imply that there are precisely @xmath10 horizontal lozenges sticking out . let @xmath12 be a probability measure on tilings of @xmath11 which satisfies the _ gibbs property _ , which means that given the positions of the @xmath10 horizontal lozenges on the boundary the conditional distribution of the tilings becomes uniform . further , given a tiling @xmath13 of @xmath11 with coordinates of the sticking out horizontal lozenges @xmath14 , let @xmath15 be a probability measure encoding them . then the pushforward of @xmath12 with respect to the map @xmath16 is a _ random _ probability measure @xmath17 . due to the gibbs property , @xmath12 is uniquely reconstructed by @xmath17 . note that each tiling of @xmath11 can be viewed also as a tiling of @xmath18 , by adding a row of lozenges on top . for example , the tiling @xmath19 in the right panel of figure [ fig_trap ] is simultaneously a tiling of @xmath20 with added row on top . because of this correspondence , if we fix the coordinates of horizontal lozenges along the right boundary , then the exact value of @xmath9 in @xmath11 becomes not important ( as long as it is large enough ) . in the same way , in order to reconstruct a @xmath21random tiling by @xmath17 we actually do not need to know the value of @xmath9 . [ theorem_main_intro ] let @xmath22 , @xmath23 , be a sequence of trapezoids equipped with probability measures @xmath24 on their lozenge tilings . suppose that : * @xmath25 stays bounded as @xmath26 . * the random probability measures @xmath27 converge ( weakly , in probability ) to a _ deterministic _ probability measure @xmath28 . then 1 . the @xmath24random lozenge tilings of @xmath22 exhibit the law of large numbers as @xmath26 : there are three deterministic densities @xmath29 , @xmath30 , and @xmath31 , @xmath32 , @xmath33 , such that for any subdomain @xmath34 with smooth boundary , the normalized by @xmath10 ( random ) numbers of lozenges of types @xmath35 inside @xmath36 converge in probability to the vector @xmath37 2 . take a point @xmath38 , such that all three asymptotic densities at this point are continuous and non - zero . assume @xmath32 . if @xmath39 , @xmath40 are two sequences of integers , such that @xmath41 , @xmath42 , then the point process of lozenges near the point @xmath43 weakly converges as @xmath26 to the ( unique ) translation invariant ergodic gibbs measure on lozenge tilings of slope @xmath44 . a more detailed formulation and the proof is below in theorems [ theorem_main_bulk ] , [ theorem_lln_bulk ] . of course , for the domains obtained from @xmath11 by rotating it by @xmath45 degrees , @xmath46 , the exact analogue of theorem [ theorem_main_intro ] is also valid . the conditions @xmath47 and @xmath48 of theorem [ theorem_main_intro ] are known to hold for many models of lozenge tilings , which implies that the local convergence to the translation invariant gibbs measures is true for them . let us provide some examples : [ corollary_tilings ] let @xmath49 be any simply - connected polygonal domain on the triangular grid , and define @xmath50 , @xmath51 to be the domain obtained by multiplying all the side lengths of @xmath49 by @xmath52 . for any part of @xmath50 covered by a trapezoid ( three sides of such trapezoid should belong to the same lines as sides of @xmath50 , cf . figure [ fig_covered_domains ] ) , near any point in the liquid region in this part , the uniformly random lozenge tilings of @xmath50 converge locally to the ergodic translation invariant gibbs measure of the corresponding slope . many polygonal domains can be _ completely covered _ by trapezoids , which implies the convergence to the ergodic translation invariant gibbs measures _ everywhere _ in the liquid region , cf . figure [ fig_covered_domains ] for examples . yet more complicated domains are covered only partially , cf . figure [ fig_not_covered_domain ] . for specific polygons , which are covered by a _ trapezoid , an analogue of corollary [ corollary_tilings ] was previously proven in @xcite . for a class of ( non - polygonal ) domains , such that the limit shape has _ no frozen regions _ an analogue of corollary [ corollary_tilings ] was previously proven in @xcite . a general conjecture that the bulk asymptotic behavior of corollary [ corollary_tilings ] should hold _ universally _ for tilings of finite planar domains dates back to @xcite . the tiled domains do not have to be polygonal , see theorem [ theorem_tilings ] for the detailed formulation . the assumption of being simply connected is also probably not essential , however , most law of large numbers type theorems in the literature stick to this assumption for simplicity , and so we have to use it here as well . there are examples of domains with holes for which the law of large numbers is explicitly known ( see e.g. ( * ? ? ? * section 9.2 ) ) , and for them corollary [ corollary_tilings ] holds . another example comes from the representation theory . recall that the irreducible representations of the @xmath10dimensional unitary group @xmath53 are parameterized by @xmath10tuples of integers @xmath54 called _ signatures _ , see e.g. @xcite . it is convenient to shift the coordinates by introducing strictly ordered coordinates @xmath55 , @xmath56 . we denote the corresponding irreducible representation through @xmath57 . such representation has a distinguished _ gelfand tsetlin basis _ , parameterized by the _ tsetlin patterns _ , which are in bijection with lozenge tilings of trapezoidal domains as above . here the coordinates of the horizontal lozenges sticking out of the right boundary of the domain are precisely the label @xmath14 of the representation , see @xcite and also section [ section_gt_assy ] for the details . in particular , the dimension @xmath58 equals the total number of the tilings of trapezoidal domain with fixed @xmath10 lozenges along the right boundary . if we take any _ reducible _ ( finite dimensional ) representation @xmath59 of @xmath53 , then we can decompose it into irreducible components @xmath60 and further consider the gelfand tsetlin basis in each of them . recalling the bijection with tilings , we thus determine for each lozenge tiling of trapezoidal domain a non - negative number , which is equal to @xmath61 with @xmath62 encoding the horizontal lozenges along the right vertical boundary of the domain . all these numbers sum up to @xmath63 , so that after dividing by @xmath63 they define a probability measure @xmath64 on lozenge tilings . varying @xmath59 in this construction one arrives at several intriguing probability distributions . the first example is given by the tensor product of two irreducible representations . [ corollary_tensor_products ] let @xmath65 , @xmath66 be two sequences of signatures ( in the notation with strictly increasing coordinates ) such that * the numbers @xmath67 , @xmath68 , @xmath69 are uniformly bounded . * there exist two monotonous functions @xmath70 , @xmath71 with finitely many points of discontinuity and such that @xmath72 consider the _ tensor product _ @xmath73 and the corresponding measure on lozenge tilings @xmath74 . then the conclusion of theorem [ theorem_main_intro ] is valid for @xmath74 as @xmath26 . the proof of corollary [ corollary_tensor_products ] is a combination of theorem [ theorem_main_intro ] with ( * ? ? ? * theorem 1.1 ) describing the law of large numbers for the tensor products . we remark that since theorem [ theorem_main_intro ] is restricted to @xmath32 , we do not get any information about @xmath75 , i.e. the behavior of the lozenges at the right boundary of the domain . _ conjecturally _ this behavior should be similar , cf . @xcite for the recent results in this direction for sums of random matrices , which are continuous analogues of tensor products ( see e.g. ( * ? ? ? * section 1.3 ) ) . another celebrated representation of @xmath53 is @xmath76 , which was intensively studied in the context of the _ schur weyl duality _ , cf . @xcite . [ corollary_schur - weyl ] consider the representation of @xmath53 in @xmath76 by natural action in each component of the tensor product . suppose that as @xmath26 , @xmath77 varies in such a way that @xmath78 . let @xmath79 denote the probability measure on lozenge tilings corresponding to the decomposition of this representation . then the conclusion of theorem [ theorem_main_intro ] is valid for @xmath80 as @xmath26 . the proof of corollary [ corollary_schur - weyl ] is a combination of theorem [ theorem_main_intro ] with the law of large numbers for the decomposition of @xmath81 obtained in @xcite , see also ( * ? ? ? * theorem 5.1 ) corollaries [ corollary_tilings ] , [ corollary_tensor_products ] , [ corollary_schur - weyl ] do no exhaust the list of possible applications of theorem [ theorem_main_intro ] and we refer e.g.to @xcite , @xcite , @xcite for some further examples of the situations where this theorem holds . further , theorem [ theorem_main_intro ] also has consequences for _ domino _ tilings on square grid , as in @xcite . in more details , it implies that in the particle process corresponding to the rectangular parts of domains tiled with dominos , the 1d bulk scaling limit along a section is universally governed by the discrete sine process , see section [ section_main_proof ] for the definition of the discrete sine process and @xcite for more details on rectangular domains and corresponding particle configurations . one way to interpret theorem [ theorem_main_intro ] is that straight boundaries lead to a _ smoothing effect_. indeed , we do not know much about the local structure of random measures @xmath27 , and it can be anything , yet as soon as we move macroscopic distance towards the straight boundary , the local measures become universal . this interpretation can be put into a wider context by observing that the stochastic process of @xmath82 horizontal lozenges on @xmath83th ( from the right ) vertical line of a trapezoid @xmath11 is a markov chain in time variable @xmath84 , as follows from the gibbs property . therefore , we see that this markov chain has a _ homogenization _ property , i.e. its local statistics become universal . in a parallel work @xcite we study a similar homogenization for the families of non - intersecting paths on very short time scales . in the continuous setting there is a close analogy with the homogenization properties of _ dyson brownian motion_. such property was first observed in @xcite , and used there for proving the universality of the local statistics for certain ensembles of wigner random matrices . more recently , such homogenization was developed much further and has led to many exciting universality results , see e.g. @xcite , @xcite , @xcite and references therein . it is natural to ask whether an analogue of theorem [ theorem_main_intro ] can hold for other conjectural universal features of the lozenge tilings , such as the fluctuations of the frozen boundary or global fluctuations of the height function . regrettably , the answer is _ no _ , the knowledge of the law of large numbers at the right boundary of the trapezoid domain is not enough for proving other universal features . a simple way to see that is to take a bernoully random variable @xmath85 and to add @xmath86 to the coordinates of the lozenges at the right boundary . this addition clearly does not change the law of large numbers ( the assumption of theorem [ theorem_main_intro ] ) , yet it will lead to the same shift everywhere in the tiling , and therefore will influence all the conjectural limiting behaviors , _ except _ for the local bulk limits that we study here . we also refer to ( * ? ? ? * section 1.9 ) for a related discussion . the proof of theorem [ theorem_main_intro ] is based on several ingredients . the first one is the double contour integral expression of @xcite for the correlation kernel of the determinantal point process describing the uniformly random lozenge tilings of trapezoids with _ fixed _ positions of horizontal lozenges on the right boundary . our asymptotic analysis of this kernel reveals a new fact : the bulk asymptotic behavior depends only on the global scale law of large numbers for the right boundary . this paves a way to pass from fixed to random horizontal lozenges on the right boundary . we also need to make a link between the slope of the local limiting gibbs measure and the slope in the global law of large numbers , for which we make use of the description of @xcite of the limit shape through the quantization of the voiculescu @xmath87transform . the author would like to thank alexei borodin , alexey bufetov and leonid petrov for helpful discussion . this work was partially supported by the nsf grant dms-1407562 . for a fixed @xmath23 let @xmath88 be an @xmath10 tuple of integers . a _ gelfand tsetlin pattern _ with top row @xmath89 is an array of @xmath90 integers @xmath91 , @xmath92 , which satisfy the interlacing condition @xmath93 and the top row condition @xmath94 , @xmath56 . such patterns are in bijection with lozenge tilings of certain specific domains , shown in figure [ fig_gt_tiling ] ; the bijection is given by positions of horizontal lozenges . is used . we stick to the notations of @xcite , as we use many ideas from that paper . ] in this section we investigate the asymptotic behavior of _ uniformly random _ gelfand tsetlin patterns with fixed top row as @xmath26 . we adapt important ideas from @xcite ; yet the results of that article are not enough for our purposes , so we need to generalize them and supplement with new considerations . the following theorem of @xcite is our starting point , see also @xcite . [ theorem_leo ] fix an @xmath10tuple of integers @xmath95 , and let @xmath96 , @xmath97 be uniformly random gelfand tsetlin pattern with top row @xmath89 . then for any @xmath98 and any collection of distinct pairs of integers @xmath99 with @xmath100 , @xmath101 , we have @xmath102=\det_{i , j=1}^k \left [ k(x(i),n(i);x(j),n(j))\right],\ ] ] where @xmath103 the contour @xmath104 encloses points @xmath105 and no other singularities of the integrand , and the contour @xmath106 is a very large circle ; both contours have positive orientation . the content of this section is the asymptotic analysis of the correlation kernel . although we use a somewhat standard steepest descent approach to such analysis ( cf . @xcite , @xcite , @xcite ) , but the technical details are delicate , as we need to deal with arbitrary @xmath10tuples @xmath89 . we start by rewriting the integrand in the double integral of as @xmath107 with @xmath108 let us analyze the zeros of the derivative of @xmath109 , i.e. the solutions to @xmath110 [ lemma_number_of_roots ] the equation has either @xmath111 or @xmath112 non - real roots . in the latter case the non - real roots are complex conjugate to each other . set @xmath113 and @xmath114 . then can be written in the form @xmath115 } \frac{1}{z - d } - \sum_{d'\in { \overline{\mathbf t}}\bigcap [ x_\kappa - n+ n_\kappa , x_\kappa-1 ] } \frac{1}{z - d'}=0\ ] ] let @xmath116 denote the total number of terms in . then after clearing the denominators , the equation becomes a polynomial equation of degree @xmath117 . since all the coefficients of this equation are real , all its non - real roots split into complex conjugate pairs . we will now show that has at least @xmath118 real roots , which then implies that there is at most one complex conjugate pair . let @xmath119 be those elements of @xmath120 which are smaller than @xmath121 and let @xmath122 be those which are greater than @xmath123 . also let @xmath124 be elements of @xmath125 $ ] . clearly , @xmath126 . further , for each @xmath127 the function continuously changes from @xmath128 to @xmath129 on the real interval @xmath130 and therefore has a zero on this interval . similarly , there is a zero on each interval @xmath131 , @xmath132 and on each interval @xmath133 , @xmath134 . summing up , we found @xmath135 distinct real roots of . we will further stick to the case when has a pair of non - real roots ( the asymptotic analysis in the case when all roots are real is more delicate and is left out of the scope of the present paper ) and denote these roots @xmath136 and @xmath137 with convention @xmath138 . let us state the main result of this section . [ theorem_kernel_asympt ] fix an arbitrary real parameter @xmath139 . assume that the pairs @xmath140 and @xmath141 are such that the complex critical points @xmath142 and @xmath143 of @xmath144 and @xmath145 , respectively , satisfy @xmath146 and @xmath147 . further , assume @xmath148 and that the points @xmath149 satisfy @xmath150 . then the kernel satisfies @xmath151 where @xmath152 is the _ uniformly small _ remainder as @xmath26 , and the integration contour crosses the real axis to the right from @xmath153 when @xmath154 , and on the interval @xmath155 when @xmath156 . the change of variables @xmath157 , i.e. @xmath158 transforms the leading asymptotic of into @xmath159 where the integration contour intersects the real line inside @xmath160 when @xmath154 and inside @xmath161 otherwise ; @xmath162 the form is known as the incomplete beta - kernel . in the rest of this section we prove theorem [ theorem_kernel_asympt ] . the level lines of functions @xmath163 and @xmath164 passing through @xmath136 , @xmath165 , are important for what follows , they are schematically sketched in figure [ fig_real_part ] . let us explain the key features of figure [ fig_real_part ] . since @xmath136 and @xmath166 are simple critical points of @xmath109 ( i.e. @xmath167 , @xmath167 ) there are four branches of @xmath168 and four branches of @xmath169 going out of each critical point and these two kinds of branches interlace . since @xmath170 as @xmath171 the level lines of @xmath172 are closed curves . these level lines can not intersect in a non - real point except at @xmath136 , @xmath137 , as such an intersection would have been another critical point for @xmath109 . further , since @xmath163 is harmonic everywhere outside the singularities on the real axis , each of its closed level lines must have one of these singularities inside . finally , @xmath173 and therefore , the level lines are symmetric . combination of these properties implies that there are four non - intersecting curves of constant @xmath163 joining @xmath136 with @xmath137 , as in figure [ fig_real_part ] . let us now in addition draw all other level lines @xmath174 ; there might be no others in addition to the 4 we just drawn , but also there might be some additional loops near the real axis . we would like to distinguish four regions on the plane bounded by level lines of @xmath163 ; these are 4 regions adjacent to @xmath136 . ( note that , in principle , these four regions do not have to cover the whole plane , but we know that each of them contains some points of the real axis ) . one of these regions is unbounded , while the other two are bounded and we call them left , middle and right ( according to the position of their intersections with the real axis ) due to the maximum principle for the harmonic functions and the behavior at the infinity of @xmath175 in the exterior ( infinite ) region we have @xmath176 . therefore , since @xmath136 is a simple critical point for @xmath109 , also @xmath177 in the middle bounded region , and @xmath178 in both left and right bounded regions . we will further distinguish four contours . @xmath179 is the curve @xmath180 which starts at @xmath136 and continues in the middle bounded region until it reaches the real axis . after that we continue the curve symmetrically to @xmath137 . note that @xmath181 is monotonous along @xmath179 except , perhaps , at the real axis , since its local extremum would necessary be a critical point . therefore , @xmath181 monotonously decreases as we move away from @xmath136 in the upper half plane and increases as we move towards @xmath137 in the lower halfplane . also observe that the choice of the branch of the logarithm in is not important here as the curve @xmath179 is constructed locally and therefore does not depend on this choice . further , @xmath182 is the curve @xmath180 which starts at @xmath136 and continues in the right bounded region until it reaches the real axis . after that we continue the curve symmetrically to @xmath137 . along this curve @xmath181 monotonously increases as we move away from @xmath136 in the upper half plane and decreases as we move towards @xmath137 in the lower halfplane . the contour @xmath183 is defined in the same way , but inside the left bounded region . the contour @xmath184 is the curve @xmath180 which starts at @xmath136 and continues in the infinite region . this curve will never go back to the real axis ( this is established below by counting the points on the real axis where @xmath180 ) and thus goes to infinity in the upper half plane . we then return symmetrically from the infinity to @xmath185 in the lower half plane . [ lemma_contour_positioning ] the curve @xmath179 intersects the real axis inside the interval @xmath186 , and the curve @xmath184 does not intersect the real axis . the curve @xmath183 intersects the real axis to the left from @xmath187 , and the curve @xmath182 intersects the real axis to the right from @xmath123 . choose a small positive parameter @xmath188 and let @xmath189 be the curve which is the real axis , except near the singularities of @xmath190 ( i.e. the points @xmath191 and @xmath192 in ) . @xmath189 walks around each singularity by a half circle in the upper - halfplane centered at this singularity and of the radius @xmath193 . we remarked above that the curves @xmath179 , @xmath182 , @xmath183 , and @xmath184 do not depend on the choice of branches of the logarithm in . but we need to choose _ some _ branch and so we use the branch of @xmath194 with a cut along the negative _ imaginary _ axis in terms of @xmath195 and such that the value at positive real @xmath195 is real . with this notation in mind , let us trace the values of @xmath196 along the curve @xmath189 moving in the direction of growing @xmath197 . clearly , if we choose @xmath193 small enough , then @xmath196 is constant on straight segments between the singularities , increases along the half circles corresponding to singularities @xmath191 in and decreases along the half circles corresponding to singularities @xmath192 in . at @xmath128 the value of this function is @xmath111 . this is schematically illustrated in figure [ fig_imaginary_part ] . at a point @xmath198 where one of our curves intersects the real axis we should have @xmath199 . we claim that there can not be two intersections on the same horizontal segment of the graph of @xmath200 along @xmath201 . indeed , each such intersection is a critical point of @xmath109 ( since the real axis itself is also a level line @xmath202 ) , but the analysis of critical points in the proof of lemma [ lemma_number_of_roots ] shows that there are never more than one critical points between two real singularities of @xmath203 . also these curves can not intersect the real axis on the segments @xmath204 and @xmath205 because we know that there are no critical points of @xmath206 on these segments . therefore , our curves can not have more than @xmath207 intersections with real axis , and , thus , they have exactly three , in particular , @xmath208 does not intersect the real axis . the middle intersection then belongs to @xmath209 and should be in the interval @xmath210\subset ( x_\kappa+n_\kappa - n , x_\kappa-1)$ ] . the right intersection belongs to @xmath211 and should be in the interval @xmath212 . finally , the left intersection belongs to @xmath213 and should be in the interval @xmath214\subset ( -\infty , x_\kappa+n_\kappa - n)$ ] . next , we need analogues of lemmas [ lemma_number_of_roots ] , [ lemma_contour_positioning ] , in which the point configuration @xmath89 is replaced by a probability measure . take a probability density @xmath215 , @xmath216 such that @xmath217 for all @xmath197 , and two numbers , @xmath33 , @xmath32 . given this data , define @xmath218 [ lemma_number_of_roots_cont ] the equation @xmath219 has either @xmath111 or @xmath112 non - real roots . in the latter case the non - real roots are complex conjugate to each other . the function @xmath190 outside the real axis can be ( uniformly on the compact subsets of @xmath220 ) approximated by functions of the form @xmath221 of . it remains to combine lemma [ lemma_number_of_roots ] with rouche s theorem . suppose that @xmath219 has two complex roots , and let @xmath201 denote the root in the upper half - plane . define four curves @xmath222 , @xmath223 , @xmath224 , @xmath225 in the same way as @xmath179 , @xmath183 , @xmath182 , @xmath184 , i.e. they are parts of the lines @xmath180 in the upper half plane continued symmetrically to the lower half plane . [ lemma_contour_positioning_cont ] the curve @xmath222 intersects the real axis inside the interval @xmath226 , and the curve @xmath225 does not intersect the real axis . the curve @xmath223 intersects the real axis to the left from @xmath227 , and the curve @xmath224 intersects the real axis to the right from @xmath228 . the proof follows the same lines as that of lemma [ lemma_contour_positioning ] . in more details , we trace the imaginary part of @xmath190 along the curve @xmath229 . the last technical ingredient needed for the proof of theorem [ theorem_kernel_asympt ] is the following statement . [ lemma_bound_derivative ] for @xmath23 , set @xmath230 then @xmath231 integrating @xmath232 termwise we get @xmath233 where @xmath234 is the inverse functions to @xmath235 , i.e. @xmath236 split the sum in into two : the first one has @xmath98 such that @xmath237 and the second one is the rest . the first sum is bounded from above by @xmath238 for the second sum we use the inequality @xmath239 which gives the bound @xmath240 the stirling s formula yields @xmath241 for large @xmath197 , and therefore , we further bound by @xmath242 for a constant @xmath243 . summing and , dividing by @xmath10 , sending @xmath26 and then @xmath244 we get the desired claim . we denote through @xmath245 the signed measure on @xmath246 describing the scaled by @xmath10 points @xmath191 and @xmath192 from : @xmath247 } \delta_{d / n } - \frac{1}{n } \sum_{d'\in { \overline{\mathbf t}}\bigcap [ x_1-n+ n_1 , x_1 - 1 ] } \delta_{d'/n}.\ ] ] the assumption @xmath150 implies that the measures @xmath245 are tight , and therefore by the standard compactness arguments we can ( and will ) assume that the measures @xmath245 weakly converge as @xmath26 to a signed measure @xmath28 . note that @xmath245mass of any interval @xmath248 $ ] is satisfies the bounds @xmath249)\le b - a+\frac{1}{n},\ ] ] therefore @xmath28 is an absolutely continuous signed measure with density @xmath250 bounded between @xmath251 and @xmath252 . recall the functions @xmath253 of , and let us introduce their normalized versions through : @xmath254 } \ln(u - d / n ) - \frac{1}{n } \sum_{d'\in { \overline{\mathbf t}}\bigcap [ x_1-n+ n_1 , x_1 - 1 ] } \ln(u - d'/n)\\= \int_{\mathbb r } \ln(z - x ) \mu_n(dx),\end{gathered}\ ] ] and similarly for @xmath255 . also set @xmath256 which is well - defined due to the assumption @xmath150 , and note that the same assumption implies @xmath257 uniformly over @xmath198 in compact subsets of @xmath258 . since the differences @xmath259 and @xmath260 stay finite as @xmath26 , we also have @xmath261 we now introduce two integration contours . for the @xmath262 contour we start from the union of @xmath263 and @xmath264 oriented from top to bottom . from the technical point of view it is convenient to modify this contour near the real axis namely , we choose @xmath188 to be fixed later and assume that @xmath265 . as soon as @xmath262 contour ( recall that be are tracing it from the bottom ) reached the level @xmath266 , it immediately has a horizontal segment so that to turn @xmath267 into a half integer ( i.e. number of the form @xmath268 , @xmath269 ) , and then vertical segment from @xmath266 to @xmath270 . in the upper halfplane @xmath271 we do the same modification . similarly , and @xmath272 is defined as the union of @xmath273 and @xmath274 oriented counter clockwise and modified to a vertical line with half integer real part for @xmath275 . we refer to figure [ fig_contour_modification ] for an illustration . note that due to convergence @xmath276 , the contours do not oscillate as @xmath26 , but smoothly approximate similar contours constructed using @xmath277 . we next deform the @xmath198 and @xmath278 contours in into @xmath262 and @xmath272 , respectively . lemmas [ lemma_contour_positioning ] , [ lemma_contour_positioning_cont ] imply that the only residues which we collect in this deformation are those coming from the @xmath279 pole . let us first deform the @xmath198 contour ( there is no residue coming from @xmath280 at this stage , as @xmath278 is very large ) , and then proceed to the @xmath278contour . therefore , the result of the deformation is the @xmath198integral over a contour @xmath281 , of the @xmath278residue of the integrand in the double integral in at point @xmath280 , i.e.@xmath282 note that @xmath283 is passed from bottom to top and it intersects the real line to the right from @xmath284 . the change of variables @xmath285 , @xmath286 transforms the kernel @xmath287 into the form @xmath288 where @xmath289 and @xmath290 contours are the contours of rescaled by @xmath10 . note that since the functions @xmath291 and @xmath292 uniformly converge to @xmath293 outside @xmath193neighborhood of the real line , and the contours are the level lines of the imaginary part of the former functions , they converge to similar level lines for @xmath293 . in particular , the lengths of all the involved contours are bounded as @xmath26 , and @xmath294 might intersect with @xmath295 only in a neighborhood of @xmath296 . let us analyze the @xmath26 behavior of each term in . the first line does not depend on @xmath10 . for the second line observe that for @xmath297 , @xmath298 we have @xmath299 due to lemmas [ lemma_contour_positioning ] , [ lemma_contour_positioning_cont ] the integration contours are bounded away from the points @xmath300 and @xmath301 . therefore , uniformly converges on the integration contours and thus the second line in is as @xmath26 @xmath302 when @xmath303 or @xmath304 the computation is the same . since the point of the intersection of the contours @xmath294 and @xmath295 approaches @xmath296 as @xmath26 , the final asymptotic for the second line of is @xmath305 where the integration contour crosses the real axis to the right from @xmath306 . now we turn to the third line in . fix any @xmath307 . we claim that outside the @xmath308neighborhood of the points @xmath296 , @xmath309 the integrand is exponentially ( in @xmath10 ) small . indeed , by the construction of the contours , @xmath310 ( strictly ) decreases as we move away from the point @xmath311 ( similarly with @xmath312 ) as long as we do not get into the @xmath193neighborhood of the real axis . lemma [ lemma_bound_derivative ] gives a uniform bound for the derivative of @xmath255 near the real axis , which shows , that @xmath310 can not grow much near the real axis . in the same way @xmath313 increases as we move away from @xmath296 and does not grow much near the real axis , where monotonicity no longer holds . on the other hand for small values of @xmath308 , we can taylor expand @xmath314 and @xmath255 near the points @xmath296 , @xmath311 , respectively . since @xmath315 uniformly converges to @xmath316 , we essentially deal with the taylor expansion of the latter function . since @xmath296 , @xmath311 are critical points of the corresponding functions , we have @xmath317 since the difference @xmath318 is ( uniformly ) of order @xmath319 as @xmath26 , the factor @xmath320 stays bounded as @xmath26 . turning to the second line in , note that by the definition , the tangent to the @xmath294 contour at @xmath296 is such that on this tangent @xmath321 is negative real and the tangent to the @xmath295 contour at @xmath311 is such that @xmath322 is positive real . therefore , the integral ( over @xmath308neighborhood of @xmath296 , along our contours ) in the third line of decays as @xmath26 ( in fact , it behaves as @xmath323 ) . summing up , behaves when @xmath26 as @xmath324 where the integration contour crosses the real axis to the right from @xmath306 . we claim that when @xmath156 , @xmath325 , then the first term in is precisely the minus residue of the second term at @xmath300 . indeed , from one side @xmath326 on the other side , @xmath327\\ = \frac{1}{(x_1-x_2)!}\left(\frac{\partial}{\partial u}\right)^{x_1-x_2 } \left(u-\frac{x_1}{n}+1-\frac{n_1}{n}\right)^{n_1-n_2+x_1-x_2 - 1 } \biggr|_{u = x_1/n } \\ = \frac{(n_1-n_2+x_1-x_2 - 1)!}{(x_1-x_2)!(n_1-n_2 - 1 ) ! } \left(1-\frac{n_1}{n}\right)^{n_1-n_2 - 1}.\end{gathered}\ ] ] we conclude that @xmath328 where the integration contour crosses the real axis to the right from @xmath153 when @xmath329 and inside the interval @xmath330 otherwise . take a lozenge tiling of an arbitrary _ simply - connected _ domain on a regular triangular grid . we aim to define for each vertex @xmath331 of the grid inside the domain the value of _ height function _ @xmath332 . for that we choose two types of lozenges out of three ( we have chosen _ non - horizontal _ ones , see figure [ fig_paths_height ] ) and draw their middle lines , thus arriving at a family of non - intersecting paths corresponding to the tiling . the value of the height function increases by @xmath252 when we cross such a path ( from bottom to top ) , this condition defines the values of @xmath333 up to an addition of an arbitrary constant . we fix this constant by a convention that @xmath334 vanishes at the bottom point of the left - most vertical of the domain , see figure [ fig_paths_height ] . observe that along the boundary of the domain , the values of the height function do not depend on the choice of lozenge tiling . indeed , the height function changes linearly along the vertical segments of the boundary and is constant along two other types of boundary segments . let @xmath335 , @xmath51 , be a sequence of simply connected domains on the triangular grid such that each @xmath335 has at least one lozenge tiling . further , let @xmath49 be a simply connected domain with piecewise - smooth boundary @xmath336 , and let @xmath337 be a continuous real function on @xmath338 . for two sets @xmath339 we say that they are within @xmath193distance from each other , if for each @xmath340 there exists @xmath341 such that @xmath342 , and for each @xmath341 there exists @xmath340 , such that @xmath342 . here @xmath343 is the euclidian distance . we say that @xmath335 approximates @xmath344 as @xmath345 , if @xmath346 for each @xmath52 , and for each @xmath188 the set @xmath347 becomes within @xmath193distance of @xmath348 as @xmath345 . here @xmath349 is the height function of lozenge tilings of @xmath335 on @xmath350 . [ theorem_lln ] suppose that the domains @xmath335 approximate @xmath344 as @xmath345 . then the rescaled height function @xmath351 of uniformly random lozenge tiling of @xmath335 converges in uniform norm , in probability to a non - random function @xmath352 on @xmath49 , which coincides with @xmath337 on @xmath338 . the proof of theorem [ theorem_lln ] is given in @xcite , see also @xcite , @xcite . the function @xmath352 is identified there with a solution to a certain variational problem and therefore depends only on @xmath49 and @xmath337 . more direct descriptions for restrictive classes of domains @xmath49 were given in @xcite and @xcite . the function @xmath352 is always lipshitz , but its derivatives might have discontinuities as e.g. at the points where the inscribed circle is tangent to the hexagon in figure [ fig_hex ] . theorem [ theorem_lln ] can be reinterpreted as the law of large numbers for the average proportions of lozenges of three types . the definition of the height function implies that the derivatives of @xmath352 in up left and up right grid directions correspond to average densities of two types of lozenges : @xmath353 and @xmath354 , respectively , see figure [ fig_hd ] . the third density is then also reconstructed from directional derivatives , e.g. using , @xmath355 . therefore , the limit shape @xmath356 can be encoded by @xmath207 functions @xmath357 , @xmath358 , and @xmath359 , which sum up to identical @xmath252 . then theorem [ theorem_lln ] yields that for any subdomain @xmath360 the numbers of lozenges ( inside @xmath361 in uniformly random lozenge tiling of @xmath335 ) of three types divided by @xmath52 converges as @xmath345 , in probability , to the integrals over @xmath362 of these three functions , as in theorem [ theorem_main_intro ] . in this section we prove the main results announced in the introduction . theorem [ theorem_main_intro ] is a combination of theorems [ theorem_main_bulk ] and [ theorem_lln_bulk ] . theorem [ theorem_tilings ] is a refinement of corollary [ corollary_tilings ] . for a probability measure @xmath363 on @xmath246 with bounded by @xmath252 density @xmath215 , and two parameters @xmath33 , @xmath32 we define a function @xmath364 through the formula . it is an analytic function on @xmath365 with derivative given by @xmath366 according to lemma [ lemma_number_of_roots_cont ] , the equation @xmath367 has at most @xmath252 solution @xmath368 in the upper half plane . denote @xmath369 and note that when @xmath201 is in the upper half plane , then so is @xmath370 . we will say that @xmath371 is _ well - defined _ , if the equation @xmath372 has a non - real solution . given a complex number @xmath370 with positive imaginary part the _ incomplete beta kernel _ ( see @xcite , @xcite ) is defined through @xmath373 where the integration contour crosses the real line inside @xmath160 for @xmath329 and inside @xmath161 for @xmath156 . further , we define @xmath377 as a probability measure on lozenge tilings of the plane with correlation functions of _ horizontal _ lozenges given for each @xmath378 by : @xmath379 = \det_{i , j=1}^k [ k_\xi(x_i , n_i;x_j , n_j)].\ ] ] it is known that @xmath377 is a translation invariant ergodic gibbs measure . we call @xmath370 the _ complex slope_. the @xmath377average proportions @xmath353 , @xmath354 , and @xmath380 of three types of lozenges can be reconstructed through the following geometric procedure ( see @xcite , @xcite ) : they are the angles of the triangle on @xmath381 with vertices @xmath111 , @xmath252 and @xmath370 normalized to sum up to @xmath252 , see figure [ fig_triangle ] . in particular , @xmath382 , which matches at @xmath383 . the triplet ( @xmath384 is the ( geometric ) slope of @xmath377 . @xcite shows that @xmath377 is a _ unique _ translation invariant ergodic measure of such slope . take a sequence @xmath24 of probability measures on trapezoids , such that for each @xmath388 the vector @xmath389 is @xmath27distributed . take any point @xmath390 with @xmath32 , such that the complex number @xmath391 is well - defined . if @xmath39 , @xmath40 are two sequences of integers , such that @xmath392 , @xmath393 , then the point process of @xmath24distributed lozenges near the point @xmath394 weakly converges as @xmath26 to @xmath395 [ remark_convergence_meaning ] in more details , the last claim says that if @xmath396 is the @xmath98th correlation function computing the probability that there is a horizontal lozenge at each of the positions @xmath397 , @xmath398 , in the @xmath24random lozenge tiling , then for each @xmath399 and each @xmath10-dependent collection of integers @xmath400 , such that the differences @xmath401 and @xmath402 do not depend on @xmath10 , the limit @xmath403 exists and is given by . we fix @xmath98 and @xmath404 , and aim to compute @xmath405 . for @xmath406 , let @xmath407 denote the @xmath98th correlation function of horizontal lozenges in tilings corresponding to uniformly random gelfand tsetlin schemes with _ fixed _ top row @xmath195 . then we can write @xmath408.\ ] ] theorem [ theorem_leo ] expresses @xmath409 as a determinant involving correlation kernel @xmath410 given by . the next step is to apply theorem [ theorem_kernel_asympt ] and we need to check that its assumptions are satisfied . choose @xmath307 such that @xmath308neighborhood of @xmath411 is bounded away from the real axis . the weak convergence of @xmath245 towards @xmath28 and tightness of @xmath412 imply that for @xmath253 defined through ( cf . ) @xmath413 we have @xmath414 in probability , and therefore the rescaled by @xmath10 critical points of @xmath415 are in @xmath308neighborhood of @xmath416 with probability tending to @xmath252 as @xmath26 . we conclude that with probability tending to @xmath252 the assumptions of theorem [ theorem_kernel_asympt ] are valid and we can use it to conclude that @xmath417 where @xmath152 is a remainder which tends to @xmath111 in probability as @xmath26 , and @xmath418 with @xmath419 being the critical point of @xmath415 in the upper half plane . as @xmath26 , @xmath420 converges in probability to @xmath421 since the random variable under expectation in is between @xmath111 and @xmath252 , the convergence in probability implies the convergence of expectations and therefore @xmath422 . \qedhere\ ] ] take a sequence @xmath24 of probability measures on trapezoids , such that for each @xmath388 the vector @xmath389 is @xmath27distributed . then the rescaled height function ( cf . theorem [ theorem_lln ] ) converges as @xmath26 uniformly , in probability to a non - random limit shape @xmath424 , @xmath425 , @xmath426 , in the coordinate system of figure [ fig_gt_tiling ] ; this limit shape is encoded by three proportions of lozenges @xmath427 . take any point @xmath390 with @xmath32 , such that the proportions of lozenges are continuous and non - zero at this point . then the complex number @xmath391 is well - defined , and moreover @xmath427 are precisely the normalized angles of the triangle @xmath428 . and vice - versa , if the complex number @xmath429 is well - defined , then the proportions of lozenges at @xmath430 are continuous and given by the angles of the @xmath431 triangle . the assumption @xmath423 of theorem [ theorem_lln_bulk ] is much more restrictive than the tightness assumption in theorem [ theorem_main_bulk ] . probably , @xmath423 can be weakened , but we will not go into this direction . the first part of the theorem , i.e. the convergence of the height function to a non - random limit shape @xmath432 is a direct corollary of theorem [ theorem_lln ] , as the weak convergence of measures @xmath245 implies the convergence of the height functions ( which are their distribution functions ) along the right boundary . given a probability measure @xmath28 of bounded by @xmath252 density and compact support , define its exponential stieltjes transform through @xmath433 @xmath434 is an analytic function outside the support of @xmath28 . set @xmath435 where @xmath436 is the functional inverse of @xmath437 , and @xmath198 is taken to be close to @xmath252 in in order for this inverse to be uniquely defined . @xmath438 is a close relative of the voiculescu @xmath87transform of @xcite , but is slightly different , see @xcite for the details . for @xmath439 $ ] , let @xmath440 $ ] be a measure on @xmath246 with density at a point @xmath197 equal to @xmath441 . note that @xmath440 $ ] is a probability measure due to combinatorial constraint on the number of horizontal lozenges on a vertical line . then the results of ( * ? ? ? * section 3.2 ) is piecewise continuous . however , one easily sees that this condition is not important for the proofs . another way to remove this condition is to use the known continuity of dependence of the limit shape on boundary conditions , see ( * ? ? ? * proposition 20 ) for the corresponding statement and proof for the domino tilings ; the proof for lozenge tilings is the same . ] yield that @xmath442}(v)=\frac{1}{\boldsymbol{\eta } } r_{\mu[1]}(v).\ ] ] let us rewrite the equation in a different equivalent form . let @xmath443 denote the function @xmath444}^{(-1)}$ ] , then plugging @xmath445}(z)$ ] into we get @xmath446}(z ) ) + \frac{e_{\mu[\boldsymbol{\eta}]}(z)}{e_{\mu[\boldsymbol{\eta}]}(z)-1 } ( 1-\boldsymbol{\eta}).\ ] ] we also have @xmath447}(h(e_{\mu[\boldsymbol{\eta}]}(z))= e_{\mu[\boldsymbol{\eta}]}(z)\ ] ] express @xmath448}(z))$ ] from and plug it into to get @xmath449}(z)}{e_{\mu[\boldsymbol{\eta}]}(z)-1 } ( 1-\boldsymbol{\eta } ) \right)= e_{\mu[\boldsymbol{\eta}]}(z).\ ] ] further , let @xmath450}(z)}{e_{\mu[\boldsymbol{\eta}]}(z)-1 } ( 1-\boldsymbol{\eta } ) , \ ] ] i.e.@xmath451}(z ) , \ ] ] therefore , if the latter has only real roots , then @xmath289 is necessarily real , which implies that @xmath456}(v ) \label{eq_exp_stieltjes_limit}\ ] ] is also real . but since the density of @xmath440 $ ] is always between @xmath111 and @xmath252 , this is possible only when this density at @xmath457 ( which is @xmath31 by the definition ) is either @xmath111 or @xmath252 . now , since upon the change of variables @xmath458 , @xmath454 is precisely @xmath459 , we conclude that @xmath460}(v)=\xi(\mu,\mathbf x,\boldsymbol{\eta}),\ ] ] where the limit is taken from the _ lower _ half plane . on the other hand , the definition of @xmath461}$ ] implies @xmath462}(v))=\lim_{{\varepsilon}\to 0 } \int_{-\infty}^{\infty } \frac{{\varepsilon}}{{\varepsilon}^2+t^2 } p^{{\scalebox{0.15}{\includegraphics{lozenge_hor.pdf}}}}(\mathbf{x}+ t,\boldsymbol{\eta } ) dt,\ ] ] where the first limit is taken in the _ lower _ half plane . since @xmath463 approaches the delta - function as @xmath244 ( and using also the continuity of @xmath429 under small perturbations of @xmath228 ) , we conclude that @xmath464 which is precisely one of the angles of the @xmath465 triangle . in order to identify another angle , define @xmath466 for @xmath198 in the lower half plane as an analytic continuation of @xmath391 , satisfying @xmath467}\left(\frac{z}{\boldsymbol{\eta}}+1\right).\ ] ] plugging in and then differentiating with respect to @xmath198 and @xmath468 , one observes an identity ( for all @xmath198 in the lower half plane ) @xmath469 which is the complex inviscid burgers equation , cf . @xcite . since the proportions of lozenges are identified with derivatives of the limiting height function , we can further write ( here @xmath52 is an arbitrary _ very large _ positive number ) @xmath470 we implicitly use in the last computation that @xmath471 is non - zero due to , and that it is not equal to @xmath252 , since its argument ( computed as in the right hand side of ) is strictly between @xmath111 and @xmath472 . for large positive @xmath52 , @xmath473 is a real number greater than @xmath252 due to . therefore , the imaginary part of the first term in the right hand side of vanishes . we conclude that @xmath474 which is precisely the formula for the angle of the @xmath475 triangle adjacent to the upper vertex . [ theorem_tilings ] consider a sequence of domains @xmath335 , @xmath476 which approximate @xmath344 as in theorem [ theorem_lln ] . in addition suppose that @xmath477 is a sequence of trapezoids , such that the symmetric difference @xmath478 is a union of triangles inside the trapezoid and adjacent to its base , cf . figure [ fig_covered_domains ] . take a sequence of points @xmath479 inside @xmath480 , such that @xmath481 . if the proportions of lozenges @xmath482 encoding the limit shape @xmath352 of theorem [ theorem_lln ] are continuous and non - zero at point @xmath390 , then the point process of lozenges near @xmath479 in uniformly random lozenge tiling of @xmath335 converges as @xmath345 to a translation invariant ergodic gibbs measure of geometric slope @xmath483 j. baik , t. kriecherbauer , k. t. -r . mclaughlin , p. d. miller , discrete orthogonal polynomials : asymptotics and applications . annals of mathematics studies , princeton university press , 2007 . arxiv : math.ca/0310278 k. johansson , universality of the local spacing distribution in certain ensembles of hermitian wigner matrices . communications in mathematical physics , 215 , no . 3 ( 2001 ) , 683705 . arxiv : math - ph/0006020 a. okounkov , n. reshetikhin , correlation functions of schur process with application to local geometry of a random 3-dimensional young diagram . journal of american mathematical society , 16 ( 2003 ) , 581603 . arxiv : math.co/0107056

in the double detonation scenario for type ia supernovae it is suggested that a detonation initiates in a shell of helium - rich material accreted from a companion star by a sub - chandrasekhar - mass white dwarf . this shell detonation drives a shock front into the carbon - oxygen white dwarf that triggers a secondary detonation in the core . the core detonation results in a complete disruption of the white dwarf . earlier studies concluded that this scenario has difficulties in accounting for the observed properties of type ia supernovae since the explosion ejecta are surrounded by the products of explosive helium burning in the shell . recently , however , it was proposed that detonations might be possible for much less massive helium shells than previously assumed . moreover , it was shown that even detonations of these minimum helium shell masses robustly trigger detonations of the carbon - oxygen core . therefore it is possible that the impact of the helium layer on observables is less than previously thought . here , we present time - dependent multi - wavelength radiative transfer calculations for models with minimum helium shell mass and derive synthetic observables for both the optical and-ray spectral regions . these differ strongly from those found in earlier simulations of sub - chandrasekhar - mass explosions in which more massive helium shells were considered . our models predict light curves which cover both the range of brightnesses and the rise and decline times of observed type ia supernovae . however , their colours and spectra do not match the observations . in particular , their colours are generally too red . we show that this discrepancy is mainly due to the composition of the burning products of the helium shell of the models which contain significant amounts of titanium and chromium . using a toy model , we also show that the burning products of the helium shell depend crucially on its initial composition . this leads us to conclude that good agreement between sub - chandrasekhar - mass explosions and observed type ia supernovae may still be feasible but further study of the shell properties is required .

1006.4489

although the standard single - degenerate ( sd ) chandrasekhar - mass scenario ( see @xcite for a review ) is capable of explaining most of the observed diversity of type ia supernovae ( sne ia ) @xcite via the delayed - detonation @xcite model , it suffers from severe problems in explaining the observed rate of sne ia . in particular , binary evolution population synthesis calculations predict rates which are an order of magnitude too low compared to the observed rate of sne ia ( @xcite , but see @xcite ) . in addition , recent observational studies suggest that chandrasekhar - mass explosions of hydrogen - accreting carbon - oxygen ( c / o ) white dwarfs ( wds ) in sd binary systems can not account for all sne ia : @xcite found that the x - ray flux of nearby elliptical galaxies is significantly weaker than expected for a population of wds accreting hydrogen towards the chandrasekhar - mass needed to explain the observed supernova rate in elliptical galaxies ( see also @xcite ) . moreover , there is growing observational evidence that there are different populations of sne ia @xcite . this has led to a revived interest in alternative explosion mechanisms . here we consider the double detonation scenario applied to sub - chandrasekhar - mass c / o wds . in that scenario a helium - accreting c / o wd explodes below @xmath2 due to a detonation in the accreted helium shell which triggers a secondary core detonation by compressional heating @xcite . this model has some very appealing features . depending on the initial mass of the wd , a wide range of explosion strengths can be realized ( e.g. @xcite ) . moreover , population synthesis studies @xcite predict rates comparable to the observed galactic supernova rate . however , earlier work @xcite found light curves and spectra of such models to be in conflict with the observed spectra and light curves of sne ia . the differences were mainly attributed to the composition of the outer layers . due to the initial helium detonation in the outer shell , the ejecta of sub - chandrasekhar - mass models are surrounded by a layer of helium and its burning products ( which can include iron - peak nuclei ) . this , however , is in apparent contradiction to the layered composition structure of observed sne ia , where the composition changes from iron - group elements in the core to lower mass elements in the outer layers . in a preceding paper @xcite have shown that artificial explosions of `` naked '' sub - chandrasekhar - mass wds can reproduce the observed diversity of sne ia . thus it is natural to ask if somewhat modified properties in the initial helium shells of realistic sub - chandrasekhar - mass models can reduce the negative post - explosion effect of this shell on the observables . in particular , @xcite recently presented new calculations , indicating that detonations might occur for much less massive helium shells than previously thought . @xcite adopted the minimum helium shell masses of @xcite and investigated whether such low - mass helium detonations are capable of triggering a secondary detonation in the c / o core of the wd . in that study , they concluded that as soon as a detonation in the helium shell initiates , a subsequent core detonation is virtually inevitable . for example , they found that even a helium shell mass as low as @xmath3 is sufficient to detonate a c / o wd of @xmath4 . here , we focus on the observable properties of the models presented in @xcite and their comparison to real sne ia . in particular , we investigate whether the low helium shell masses of these models help to alleviate the problems encountered previously when comparing double detonation sub - chandrasekhar - mass models to observed spectra and light curves of sne ia @xcite . the outline of the paper is as follows : in section [ sec : models ] we give a short summary of the models of @xcite before briefly describing details of our radiative transfer simulations in section [ sec : rt ] . in section [ sec : oo ] we present synthetic observables for the @xcite models and compare them to the observed properties of sne ia . the results of this comparison and implications for future work on sub - chandrasekhar - mass double detonations are discussed in section [ sec : discussion ] . finally , we draw conclusions in section [ sec : conclusions ] . adopting the minimum helium shell masses required to initiate a helium detonation in the shell according to @xcite , @xcite investigated the double detonation scenario for six models representing a range of different c / o core masses ( the model parameters are summarized in table [ tab : modelparas ] ) . in all their models , they ignite an initial helium detonation in a single point at the base of the helium shell located on the positive @xmath5-axis ( hereafter referred to as the `` north - pole '' of the wd ) . from there the helium detonation wave sweeps around the wd core until it converges at the south pole . at the same time a shock wave propagates into the core and converges at a point off - centre . finding conditions that might be sufficient to initiate a core detonation in a finite volume around this point , @xcite then trigger a secondary core detonation at that point . this secondary detonation disrupts the entire wd and yields ejecta with a characteristic abundance distribution which is shown for model 3 of @xcite in figure [ fig : m03_composition ] . lcccccccc @xmath6 & 0.936 & 1.004 & 1.080 & 1.164 & 1.293 & 1.3885 & 1.025 & 1.071 + @xmath7 & 0.810 & 0.920 & 1.025 & 1.125 & 1.280 & 1.3850 & 1.025 & 1.025 + @xmath8 & @xmath9 & @xmath10 & @xmath11 & @xmath12 & @xmath13 & @xmath14 & @xmath11 & @xmath15 + @xmath16 & @xmath17 & @xmath18 & @xmath19 & @xmath20 & @xmath21 & @xmath22 & @xmath19 & @xmath23 + @xmath24 & @xmath25 & @xmath26 & @xmath27 & @xmath28 & @xmath29 & @xmath30 & @xmath27 & @xmath31 + @xmath32 & @xmath33 & @xmath34 & @xmath35 & @xmath36 & @xmath36 & @xmath37 & @xmath35 & @xmath38 + @xmath39 & @xmath40 & @xmath41 & @xmath42 & @xmath43 & @xmath44 & @xmath45 & @xmath42 & @xmath42 + @xmath46 & @xmath47 & @xmath47 & @xmath48 & @xmath49 & @xmath50 & @xmath51 & @xmath48 & @xmath48 + @xmath52 & @xmath53 & @xmath54 & @xmath55 & @xmath56 & @xmath57 & @xmath58 & @xmath55 & @xmath55 + @xmath59 & @xmath60 & @xmath61 & @xmath62 & @xmath63 & @xmath64 & @xmath49 & @xmath62 & @xmath65 + @xmath66 & @xmath67 & @xmath68 & @xmath69 & @xmath70 & @xmath71 & @xmath72 & @xmath69 & @xmath47 + @xmath73 & @xmath63 & @xmath74 & @xmath75 & @xmath76 & @xmath77 & @xmath78 & @xmath75 & @xmath79 + @xmath80 & @xmath81 & @xmath82 & @xmath83 & @xmath84 & @xmath85 & @xmath86 & @xmath83 & @xmath87 + @xmath88 & 0.126 & 0.084 & 0.055 & 0.039 & 0.013 & 0.0035 & & 0.046 + @xmath89 & @xmath90 & @xmath91 & @xmath51 & @xmath82 & @xmath92 & @xmath93 & & @xmath94 + @xmath95 & @xmath17 & @xmath96 & @xmath97 & @xmath98 & @xmath99 & @xmath100 & & @xmath101 + @xmath102 & @xmath70 & @xmath103 & @xmath82 & @xmath104 & @xmath105 & @xmath106 & & @xmath107 + @xmath108 & @xmath20 & @xmath109 & @xmath110 & @xmath45 & @xmath111 & @xmath112 & & @xmath113 + @xmath114 & @xmath115 & @xmath116 & @xmath104 & @xmath104 & @xmath105 & @xmath117 & & @xmath98 + @xmath118 & @xmath119 & @xmath116 & @xmath120 & @xmath121 & @xmath122 & @xmath123 & & @xmath124 + @xmath125 & @xmath104 & @xmath99 & @xmath126 & @xmath127 & @xmath128 & @xmath129 & & @xmath43 + @xmath130 & @xmath131 & @xmath132 & @xmath133 & @xmath134 & @xmath135 & @xmath123 & & @xmath136 + @xmath137 & @xmath138 & @xmath139 & @xmath140 & @xmath141 & @xmath142 & @xmath143 & & @xmath144 + @xmath145 & @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath151 & & @xmath152 + @xmath153 & @xmath99 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & & @xmath159 + @xmath160 & @xmath161 & @xmath162 & @xmath57 & @xmath40 & @xmath44 & @xmath51 & & @xmath163 + @xmath164 / mag & 0.88 & 1.25 & 1.74 & 1.77 & 1.23 & 1.29 & 1.63 & 1.37 + @xmath165 / days & 17.0 & 17.0 & 17.7 & 16.4 & 15.2 & 14.3 & 18.2 & 18.1 + @xmath166 / mag & -15.9 & -17.3 & -18.4 & -19.3 & -19.8 & -19.9 & -19.2 & -19.1 + @xmath167 / mag & -17.6 & -18.8 & -19.6 & -19.9 & -20.1 & -20.1 & -19.4 & -19.4 + @xmath168 / mag & -18.4 & -19.1 & -19.4 & -19.4 & -19.4 & -19.2 & -19.0 & -19.1 + @xmath169 / mag & -18.9 & -19.2 & -19.2 & -19.4 & -19.6 & -19.7 & -18.9 & -19.0 + @xmath170 / mag & 0.58 & 0.41 & 0.50 & 0.19 & 0.08 & -0.26 & -0.04 & 0.35 + @xmath171 / mag & 1.67 & 1.51 & 1.13 & 0.59 & 0.28 & 0.08 & 0.11 & 0.30 + @xmath172 / mag & 0.74 & 0.36 & -0.13 & -0.46 & -0.67 & -0.82 & -0.30 & -0.28 + @xmath173 / mag & 1.26 & 0.39 & -0.47 & -1.05 & -1.48 & -1.47 & -0.64 & -0.49 + in the initial helium shell the burning does not reach nuclear statistical equilibrium due to the low densities . thus it mainly produces iron - group elements lighter than @xmath174ni such as titanium , chromium and iron including some amount of the radioactive isotopes @xmath175 and @xmath176 . aside from a small mass of calcium , no significant amounts of intermediate - mass elements are produced in the shell . however , a large fraction of helium remains unburned . the core detonation yields both iron - group and intermediate - mass elements , the relative amounts of which depend crucially on the core density and thus wd mass . models with more massive wds produce more iron - group material and less intermediate - mass elements ( similar to the explosions of the naked sub - chandrasekhar - mass wds studied in @xcite ) . the most massive model 6 of the sequence of @xcite produces almost only iron - group material and hardly any intermediate - mass elements ( the most important burning products in the shell and core of the different models are listed in table [ tab : modelparas ] ) . due to the single - point ignition the models show strong ejecta asymmetries which can be divided into two main categories . first , the helium shell contains more iron - group material on the northern hemisphere as discussed by @xcite . second , the ignition point of the secondary core detonation is offset from the centre - of - mass of the model due to the off - centre convergence of the shock waves from the helium detonation . to derive synthetic observables for the models , we performed radiative transfer simulations with the time - dependent multi - dimensional monte carlo radiative transfer code artis described by @xcite and @xcite . since the models produce significant amounts of @xmath175 and @xmath176 , we extended artis to take into account the energy released by the decay sequences @xmath177 and @xmath178 in addition to the radioactive decays of @xmath179 and @xmath180 which form the primary energy source of sne ia ( @xcite , @xcite ) . for other radioactive nuclei which are synthesized during the explosion in a non - negligible amount ( e.g. @xmath181 in the shell ) the life times are much longer than those of the @xmath174ni decay - sequence . thus they can be neglected at early times when the decays of @xmath174ni and @xmath174co power the light curves . the total @xmath0-ray energy @xmath182 that will be emitted from @xmath183 to @xmath184 in the decay chains we consider , will be @xmath185 here @xmath186 , ( @xmath187 , @xmath188 , @xmath189 , @xmath190 , @xmath191 ) is the mean energy emitted per decay of @xmath174ni , ( @xmath174co , @xmath192fe , @xmath192mn , @xmath193cr , @xmath193v ) . similarly , @xmath194 ( @xmath195 , @xmath196 ) is the initial mass of @xmath174ni ( @xmath197 , @xmath198 ) synthesized in the explosion and @xmath199 ( @xmath200 , @xmath201 ) the mass of the @xmath174ni ( @xmath197 , @xmath198 ) atom . following @xcite , this energy is quantized into @xmath202 identical indivisible energy `` pellets '' of initial co - moving frame ( cmf ) energy @xmath203 . the pellets are first assigned to one of the decay sequences in proportion to the amount of energy deposited in the different decay sequences ( terms on the right - hand - side of equation [ eq : decays ] ) . then they are distributed on the grid according to the initial distribution of @xmath174ni , @xmath192fe and @xmath193cr , as appropriate , and follow the homologous expansion until they decay . decay times are sampled in a two step process . if the pellet was assigned to the @xmath174ni chain , for example , we first choose whether it belongs to a decay of the parent nucleus @xmath174ni or the daughter nucleus @xmath174co by sampling the probabilities @xmath204 and @xmath205 , respectively . finally an appropriate decay time is sampled @xmath206 @xmath207 from the mean life times of the @xmath174ni [ @xmath208 and @xmath174co [ @xmath209 nuclei . @xmath210 are random numbers between 0 and 1 . the @xmath197 [ @xmath211 , @xmath212 and @xmath198 [ @xmath213 , @xmath214 chains are treated in the same way . upon decay , a pellet transforms to a single @xmath0-packet representing a bundle of monochromatic - radiation of cmf energy @xmath215 . the cmf photon energy of the @xmath0-packets is randomly sampled from the relative probabilities of the @xmath0-lines in the appropriate decay of the selected decay sequence including annihilation lines due to positron emission . following @xcite , we assume that positrons released by radioactive decays annihilate in situ , giving rise to the emission of two 511 kev @xmath0-ray photons . in doing so , we neglect the kinetic energy released by stopping the positrons , any positron escape and possible positronium formation which gives rise to the emission of continuum photons ( see discussion by @xcite ) . thus , our prediction of the 511 kev line flux should be considered as an upper limit . the @xmath0-packets are then propagated through the ejecta as described by @xcite . the @xmath0-line data is taken from table 1 of @xcite for the @xmath174ni decay - sequence and from @xcite for the @xmath193cr decay - sequence , respectively . owing to the comparatively short life times of @xmath192fe and @xmath192mn , these nuclei have mostly already decayed to their daughter nuclei when we start the radiative transfer simulation at @xmath216 day . since the ejecta at these early times are almost opaque to @xmath0-rays , the @xmath0-packets released by the @xmath192fe decay chain will be thermalized rapidly . therefore we do not follow the propagation of the @xmath0-packets released by the @xmath192fe decay chain , but immediately convert their energy to thermal kinetic energy ( @xmath217-packets in the framework of artis ) . the input models ( composition , density , velocities ) for the radiative transfer simulations were derived using the tracer particles from the hydrodynamics simulations ( see @xcite for details on the hydro setup and use of tracer particles for nucleosynthesis ) . since the tracer particles are lagrangian , we use an approach similar to the reconstruction of the density field in smooth particle hydrodynamics ( sph ) simulations to construct the input model from the tracers . for the models described here , the density field was obtained using the sph method described by @xcite ( specifically , their equations 1 3 ) adopting @xmath218 for the sph smoothing - length normalisation factor . for the centre ( @xmath219 ) of each grid cell ( @xmath220 ) in the model , the mass fractions ( @xmath221 ) of the elements considered ( @xmath222 to 30 ) were reconstructed using @xmath223 where @xmath224 is the reconstructed mass density , @xmath225 is the mass - fraction of element @xmath226 for tracer particle @xmath227 ( which lies at position @xmath228 ) , @xmath229 is the sph particle smoothing length and @xmath230 the sph kernel function ( defined via equations 3 and 1 of @xcite , respectively ) . the mass fractions of the important radioactive isotopes ( @xmath174ni , @xmath174co , @xmath192fe and @xmath193cr ) were reconstructed from the tracer particle yields in exactly the same manner . the reconstruction was performed on an @xmath231 @xmath232-grid using the final state of the tracer particles at the end of the simulations ( which were run up to the phase of homologous expansion ) . for the radiative transfer simulation this 2d model is re - mapped to a 3d cartesian grid of size @xmath233 which co - expands with the ejecta . we then follow the radiative transfer from 2 to 120 days after the explosion , discretized into 111 logarithmic time steps each of duration @xmath234 . using our detailed ionization treatment and the cd23_gf-5 atomic data set @xcite , we simulated the propagation of @xmath235 packets . to speed up the initial phase of the calculation we made use of our initial grey approximation for the first 30 time steps ( adopting a parameterised grey opacity for the highly optically thick cells in the centre of our simulation volume ) . the first ten time steps were treated in local thermodynamic equilibrium ( lte ) . in this section we present synthetic observables for the models of @xcite . first we consider angle - averaged ultraviolet , optical and infrared light curves and explore the diversity within this set of models and compare it to that of observed sne ia ( section [ sub : oo_lcs ] ) . then we investigate the colour evolution and spectra of these models ( sections [ sub : oo_colours ] and [ sub : oo_spectra ] ) . in section [ sub : oo_los ] we study the effects of the asymmetric ejecta composition . finally , in section [ sub : oo_gamma ] , we briefly discuss the @xmath0-ray emission from these models . in figure [ fig : lightcurves_all ] we show the angle - averaged ultraviolet - optical - infrared ( @xmath236 ) bolometric and @xmath237 , @xmath238 , @xmath239 , @xmath240 , @xmath241 , @xmath242 , @xmath243 , and @xmath244 band - limited light curves for the model sequence of @xcite as obtained from our radiative transfer simulations . for comparison to observations we included photometric data of sne 2005cf @xcite , 2004eo @xcite , 2001el @xcite and 1991bg @xcite in the figure . while sn 1991bg , the prototypical event of the subluminous 1991bg - like objects , marks the faint end of observed sne ia , sne 2005cf , 2004eo and 2001el are representative of the spectroscopically normal objects . since our models form a sequence of increasing @xmath245-mass , the peak brightness of the synthetic light curves increases from model 1 to model 6 : for @xmath238-band maximum , for example , we find a range from @xmath246 for model 1 to @xmath247 for model 6 ( values for the intermediate models and other bands are given in table [ tab : modelparas ] ) . this covers almost the full range of peak brightnesses observed among sne ia ( e.g. * ? ? ? * ) , excluding only the very brightest events like sn 2003fg @xcite which have been suggested to derive from super - chandrasekhar - mass progenitor systems ( see however @xcite ) . despite their lower total mass , and thus lower overall opacity , our model light curves do not show particularly fast evolution compared to chandrasekhar - mass models . in contrast , @xcite reported a fast initial rise and broad peaks for their helium detonation models . for the rise times from explosion to @xmath238-band maximum [ @xmath165 ] we find values between 17.0 and 17.7 days for our least massive models ( 1 to 3 , cf . table [ tab : modelparas ] ) . this is in good agreement with observational findings ( e.g. @xcite find rise times to @xmath238-band maximum between 13 and 23 days with an average value of @xmath249 ) . in contrast , the more massive models ( 4 to 6 ) show shorter @xmath238-band rise times with increasing mass ( see table [ tab : modelparas ] ) : for model 6 , @xmath165 is only 14.3 days . the peak time is mainly set by the diffusion time for photons to leak out from the @xmath174ni - rich inner core . owing to the larger densities in the more massive models , nuclear burning produces @xmath174ni out to much higher velocities than in the lower mass models ( compare figure 3 of @xcite ) . moreover , the helium shell masses in our models decrease for the heavier wds . thus , the mass ( and therefore opacity ) on top of the @xmath174ni - rich inner core decreases with increasing wd mass and photons start to escape earlier for the more massive models making their light curves peak faster ( compare also the rise times to bolometric maximum in @xcite ) . in addition , we note that the radioactive nuclides which are produced in the helium shell burning have some influence on the initial rise phase of the light curves ( see figure 8 of @xcite ) . in the post - maximum decline phase , characterized by the decline parameter @xmath164 which gives the change in @xmath238-band magnitude between maximum and 15 days thereafter , our models show a peculiar characteristic . observationally , brighter sne ia show a trend of slower declining light curves ( e.g. @xcite ) . in contrast , the model light curves decline faster along our sequence from model 1 to 4 despite increasing brightness ( cf . table [ tab : modelparas ] ) . specifically , we find @xmath250 for model 1 , which according to peak brightness would be classified as a subluminous explosion . however , observationally these events are characterized by a fast decline [ @xmath251 , e.g. @xcite ] . in contrast , model 4 , which has a @xmath238-band peak magnitude typical for a spectroscopically normal sn ia , yields @xmath252 which is much faster than typically observed [ @xmath253 , e.g. @xcite ] . although @xmath164 is lower for the brighter models 5 and 6 , they still decline too fast compared to observed objects of corresponding brightness . a similar trend for more rapidly declining light curves from more massive models is also seen in the bolometric light curves of model 2 to 6 ( cf . table 4 of @xcite ) . in contrast to earlier work by @xcite , who studied sub - chandrasekhar - mass models with significantly more massive shells , we do not find particularly fast evolution in the post - maximum decline of our light curves compared to those obtained for other types of models . applying our radiative transfer code artis for example to the well - known w7 model @xcite , which is widely regarded as a good standard for sne ia , we find @xmath254 when using the same atomic data set as adopted here . this value is comparable to our fastest declining sub - chandrasekhar - mass model and also too fast compared to normal sne ia . in addition , we note that @xmath164 is also quite sensitive to the details of the radiative transfer treatment . using an atomic data set with @xmath255 lines ( the big_gf-4 data set of @xcite ) , for example , yields @xmath256 for w7 . in contrast , a simulation using the atomic data set adopted in this study but applying a pure lte treatment for the excitation / ionization state of the plasma yields @xmath257 for w7 . this shows that systematic uncertainties in the radiative transfer treatment can affect @xmath164 by several tenths of a magnitude ( see also the comparison of different radiative transfer codes in figure 7 of @xcite ) . thus , we argue that there is no evidence that our sub - chandrasekhar - mass models fade too fast compared to other explosion models . the most striking difference between our light curves and those of the comparison objects in figure [ fig : lightcurves_all ] concerns their colour . to highlight this we show the angle - averaged time - evolution of the @xmath1 , @xmath258 and @xmath259 colours for all our models in figure [ fig : colours_all ] and compare them again to our fiducial sne 2005cf , 2004eo , 2001el and 1991bg . in @xmath1 all our models show positive colour indices for the whole simulation period and a red peak at @xmath260 days after explosion . contrary to observed sne ia @xcite , however , we find no convergence of the different models at epochs after the red peak . instead , our models at all times form a sequence of increasingly redder @xmath1 colour towards the fainter explosions . with the exception of model 6 , all our models are generally too red compared to spectroscopically normal sne ia . at maximum light , for example , we find @xmath261-values of 1.67 and 0.28 for model 1 and model 5 , respectively ( values for the other models are given in table [ tab : modelparas ] ) . in contrast , spectroscopically normal sne ia are characterized by @xmath262 . subluminous 1991bg - like objects show a redder @xmath1 colour before the red peak , reaching @xmath263 at @xmath238-band maximum @xcite . although model 4 can reproduce this , it is not a good fit to 1991bg - like objects , since it is considerably too bright ( compare figure [ fig : lightcurves_all ] ) . our subluminous models 1 and 2 , on the other hand , are significantly redder than observed ( @xmath1 colours of 1.67 and 1.51 at maximum , respectively ) . in contrast , @xcite and @xcite found too blue colours at maximum light for their subluminous sub - chandrasekhar - mass models compared to 1991bg - like objects . the origin of our red colours traces back to the extended layer of titanium and chromium which is present in the helium shell ejecta of our models . this will be discussed in more detail in section [ sub : oo_spectra ] . while the trend of increasingly redder colours for fainter models persists in the @xmath258 index , our models are not systematically too red here . model 2 and 3 populate about the right range for 1991bg - like and spectroscopically normal sne ia , respectively . however , the details of their @xmath258 evolution , especially the initial decline present in all the models , do not match the observations . a similar behaviour is found for the @xmath259 colour where again model 2 and 3 lie closest to the observed colours of 1991bg - like and spectroscopically normal sne ia , respectively . however , again the agreement is imperfect . for model 2 this is most obvious at the latest epochs , where 1991bg - like objects start to become bluer while the model colour maintains a redward evolution . moreover , the model shows a secondary blue minimum , due to the post - maximum plateau of the @xmath239-band light curve which is not observed in 1991bg - like objects . in model 3 the initial decline and the rise to the red peak are significantly different from the observed behaviour of spectroscopically normal sne ia . the properties of our model light curves can be understood from consideration of the synthetic spectra . figure [ fig : maxspectra_all ] shows the angle - averaged spectrum of all our models at three days before @xmath238-band maximum . for comparison , we also show the spectrum of sn 2004eo @xcite at the corresponding epoch . the colour coding below the synthetic spectrum indicates the fraction of escaping quanta in each wavelength bin which were last emitted in bound - bound transitions of a particular element . similarly , the coloured regions above the spectra indicate which elements were last responsible for removing quanta from a wavelength bin by bound - bound absorption . this coding allows us to both identify individual spectral features and track the effect of fluorescence on the spectrum formation directly . the contributions of bound - free and free - free emissions to the total flux are so small that they are not discernable in the figure . from this it is immediately obvious that the colours of our models are due to fluorescence in titanium and chromium lines . having a wealth of strong lines in the uv and blue , even small amounts of these elements effectively block radiation in the uv and blue and redistribute it to redder wavelengths where the optical depths are smaller and the radiation can escape . compared to other explosion models , this effect is particularly strong in the @xcite models , since they produce relatively large amounts of titanium and chromium in the outer layers during the initial helium detonation ( cf . table [ tab : modelparas ] and figure [ fig : m03_composition ] ; typical titanium and chromium yields for other explosion models are on the order of the yields from the core detonation ) . this also explains the trend for redder colours in the fainter models : according to table [ tab : modelparas ] , the production of titanium and chromium in the shell increases continuously from model 6 to model 1 . since this titanium and chromium layer is located at higher velocities than most of the intermediate mass elements ( cf . figure [ fig : m03_composition ] ) , redistribution by titanium and chromium also dilutes the absorption features of intermediate - mass elements like silicon and sulphur by reprocessing flux into the relevant wavelength regions . this can be clearly seen in figure [ fig : maxspectra_all ] . although models 1 and 2 produce the largest amounts of silicon of all our models ( 0.27 and 0.25 solar masses in the core , respectively ) , they show only a ( very ) weak siii @xmath264 feature since this wavelength region is strongly polluted by flux redistribution from titanium and chromium . the only feature of intermediate - mass elements which is clearly visible in these models is the caii near - infrared ( nir ) triplet at @xmath265 . this feature remains unburied since calcium exists co - spatially with titanium and chromium in the outer shell ( figure [ fig : m03_composition ] ) . of all our models , model 3 [ @xmath266 produces the strongest siii @xmath264 feature . for more massive models the silicon yields from the core detonation drop dramatically since , owing to the higher densities , a larger fraction of the core is burned to iron - group elements ( see table [ tab : modelparas ] and figure 3 of @xcite ) . as expected from studies of pure detonations of chandrasekhar - mass wds @xcite , the extreme case of model 6 ( almost the chandrasekhar mass ) produces no significant amounts of intermediate - mass elements . thus the spectrum is totally dominated by iron - group elements . since it shows no indication of any siii @xmath264 feature , this model would not be classified as a sn ia . since none of our models give a good match to the colours and line strengths of observed sne ia , we will not discuss line velocities in detail . we note , however , that along the sequence from model 1 to 5 there is a trend for higher line velocities of intermediate mass elements . this is obvious in the siii @xmath264 feature of our models which is apparent in figure [ fig : maxspectra_all ] . along the model sequence ( 1 to 5 ) , this line moves to shorter wavelengths compared to the observed siii absorption feature of sn 2004eo . this arises because the inner boundary of the region rich in intermediate - mass elements moves to higher velocities with increasing mass due to the more complete burning in the inner regions ( compare figure 3 of @xcite ) . finally , we note that none of our model spectra show any indication of helium lines , despite our models having up to @xmath267 of helium in their outer layers ( cf . table [ tab : modelparas ] ) . since helium has rather highly excited levels , this might simply be a consequence of our approximate treatment of the plasma state which neglects non - thermal excitation and ionization . we note , however , that despite using the phoenix code which does include a treatment of non - thermal processes , @xcite also found no strong evidence of helium lines for models with even more ( @xmath268 ) helium in the outer shell . as discussed in section [ sec : models ] , our models show strong ejecta asymmetries due to their ignition in a single - point at the north pole of the wd ( cf . figure [ fig : m03_composition ] ) . these asymmetries are expected to have some influence on the observables along different lines - of - sight . since they are characteristically the same for all our models , we first use model 3 as an example to give a detailed discussion of line - of - sight dependent spectra and light curves of this model ( sections [ sub : oo_los_spectra ] and [ sub : oo_los_lightcurves ] ) . in section [ sub : oo_los_all ] we then present the variations in peak magnitudes and colours due to line - of - sight effects for all our models . to obtain line - of - sight dependent observables we bin the escaping photons into a grid of ten equal solid - angle bins in @xmath269 with @xmath270 being the angle between the line - of - sight and the @xmath5-axis of the model . in figure [ fig : m03_losspectra ] we show the maximum - light spectra of model 3 for three different directions , corresponding to lines - of - sight close to the southern polar axis ( @xmath271 ) , the northern polar axis ( @xmath272 ) and equator - on ( average of the @xmath273 and 0.1 bins ) . while the equator - on spectrum looks similar to the angle - averaged spectrum , the spectra seen from the polar directions are significantly different from each other as well as from the angle - averaged spectrum . for @xmath274 , an observer is looking through the extended layer of iron - group elements in the outer shell of the northern - hemisphere ( cf . figure [ fig : m03_composition ] ) . this blocks almost all the flux in the uv and blue wavelength range and the redistribution of this flux leads to a very red spectrum . in contrast , this layer of iron - group elements in the outer shell is far less extended on the southern hemisphere , where the shell burning is less efficient . thus , an observer with @xmath271 sees a bluer spectrum than equator - on . equator - on the extension of the layer of iron - group elements is somewhere between these two extremes and the spectrum resembles the angle - averaged case . a secondary effect results from the off - centre ignition of the core . since the core has been compressed less strongly on the northern hemisphere , the core detonation yields more intermediate - mass elements on the northern than on the southern hemisphere ( cf . figure [ fig : m03_composition ] and the discussion in section 4.3 of @xcite ) . this leads to stronger features of intermediate - mass elements in the spectrum seen along the northern polar - axis and becomes particularly visible in the strength of the siii @xmath275 feature and the caii nir - triplet which become weaker from @xmath276 . figure [ fig : m03_lightcurves ] shows band - limited synthetic light curves from our radiative transfer simulations for model 3 as seen equator - on and from the two polar directions . as already noted for the maximum - light spectra , our model shows a trend for increasingly red colours from @xmath277 . similarly , we find a clear dependence of the light curve rise and decline times on the line - of - sight . while the rise times increase along the sequence @xmath277 ( for @xmath238 band , for example , we find rise times between 16.8 and 19.7 days for @xmath271 and @xmath274 , respectively ) , the light curve declines more slowly [ @xmath278 and @xmath279 for @xmath271 and @xmath274 , respectively ] . moreover , we find a clear trend for a weaker dependence of the light curves on the line - of - sight at lower photon energies . thus , while @xmath237 and @xmath238 band show a variation of @xmath280 and @xmath281 magnitudes at maximum light , respectively , the variation in @xmath239 band is already less than half a magnitude and in the nir bands there is virtually no viewing - angle effect . all these effects are due to the asymmetry of the layer of iron - group elements in the outer shell . since this layer is more extended on the northern hemisphere it causes additional line blocking and thus enhanced fluorescence and photon trapping for inclinations close to @xmath282 . this explains the redder colours as well as the increasing rise and decreasing decline times for @xmath277 . for lower photon energies , the asymmetry of the outer shell is less important , since the optical depths are smaller and photons typically escape from deeper layers of the ejecta . in the nir bands the entire ejecta contribute to the emitted photons and the viewing - angle dependence of the light curves is very small . this is illustrated in figure [ fig : m03_lastscat_snapshots ] which shows where photons of selected bands were last emitted before they escaped from the supernova ejecta at different times . while @xmath237-band photons leak out predominantly from the regions on the southern hemisphere , where the layer of iron - group elements is least extended , @xmath241-band photons show no strong preference for a particular region . in fact , even before maximum light the whole ejecta contribute to @xmath241-band emission . @xmath239-band photons , in contrast , show some preference for leaking out from the southern hemisphere before and immediately after maximum light . from about 10 days after maximum light this preference disappears . this is directly reflected in the @xmath239-band light curve in figure [ fig : m03_lightcurves ] which becomes viewing - angle independent at about that time . finally , we note that the strong line - of - sight dependence of our synthetic observables poses an additional problem for the @xcite models . even if some particular line - of - sight might compare more favourably to observed sne ia than others ( as does for example the southern line - of - sight for the light curves in figure [ fig : m03_lightcurves ] ) , all lines - of - sight should occur in nature including the most peculiar ones . moreover , the large dispersion in brightness at @xmath238-band maximum which we find in our model is in conflict with observations . since the other models in the @xcite sequence have the same characteristic asymmetries , their light curves and spectra show a similar viewing - angle dependence . however , the strength of this viewing - angle dependence varies between the models , due to their different helium shell masses . to demonstrate this , figure [ fig : los - dependence_all ] shows selected light curve properties of all models for the ten different viewing directions within our uniform grid in @xmath283 . -band maximum [ @xmath165 ] , the decline in @xmath238-band between maximum light and 15 days thereafter [ @xmath164 ] , the @xmath238-band peak magnitude ( @xmath166 ) , the @xmath1 colour at @xmath238-band maximum [ @xmath171 ] and the @xmath258 colour at @xmath238-band maximum [ @xmath172 ] , respectively . the lines connecting the different data points are just to guide the eye.,width=321 ] as expected , the viewing - angle dependence decreases for smaller helium shell mass . thus , we find a scatter of more than 2 magnitudes for the brightness at @xmath238-band maximum between the different lines - of - sight in model 1 and 2 , while model 6 shows only a scatter of @xmath287 magnitudes . similar trends can be observed for the @xmath1 and @xmath258 colours at @xmath238-band maximum . while model 6 shows almost no dependence on the line - of - sight , the less massive models show a very clear trend for redder colours towards @xmath274 ( as discussed for model 3 in detail above ) . for the rise times the situation is similar : while models 1 , 2 and 3 ( relatively massive shells ) show a clear trend for increasing @xmath238-band rise times from @xmath271 to @xmath274 due to the enhanced photon trapping by optically thick lines of iron - group elements , we do not observe a clear trend for models 4 , 5 , and 6 ( less massive shells ) . recall , that the average rise times increase from model 1 to 3 , while models 4 to 6 form a sequence of decreasing rise times . this was already discussed in section [ sub : oo_lcs ] . the enhanced trapping of photons by the iron - group layer of the northern hemisphere also affects the post - maximum decline rate in @xmath238 band . thus , we find decreasing values of @xmath164 from @xmath271 to @xmath274 . this trend persists for all our models , although it is weaker for the models with the least massive shells . due to their peculiar composition including a mixture of the radioactive isotopes @xmath174ni , @xmath192fe and @xmath193cr close to the surface , @xmath0-observations might provide an additional discriminant between our sub - chandrasekhar - mass models and more standard explosion models which do not show radioactive isotopes close to the surface . to investigate this , figure [ fig : gamma ] shows @xmath0-ray light curves and spectra for models 1 , 3 and 6 of our sequence . line identifications are given for some important features in table [ tab : gammalineidentifiers ] . -ray light curves ( top panel ) and spectra ( lower panels ) for models 1 , 3 and 6 of @xcite as indicated by the colour coding . the spectra shown are for two different epochs which correspond roughly to maximum light in @xmath238 band ( 20 days , middle panel ) and in @xmath0-rays ( 60 days , bottom panel ) . line features discussed in the text are identified by labels ( a f ) in the middle panel ( for their identification , see table [ tab : gammalineidentifiers ] ) . in addition , the @xmath174ni , @xmath174co and @xmath193v line systems are indicated . the line systems of @xmath193cr , @xmath192fe and @xmath192mn are omitted , since their life times are sufficiently short that these nuclei have already decayed at the shown epochs . note that at 20 days the line features in the spectra are generally offset towards higher energies compared to the line identifications , due to blue - shifted emission . at 60 days the emission comes from regions of lower velocity and the offset disappears . , width=321 ] ccl a & 0.158 & @xmath174ni + b & 0.270 & @xmath174ni + c & 0.511 & annihilation of positrons from @xmath174co and @xmath193v + d & 0.750 & @xmath174ni + e & 0.812 & @xmath174ni + f & 1.562 & @xmath174ni + broadly speaking , our @xmath0-ray spectra are not dramatically different from those obtained by @xcite for parameterised chandrasekhar - mass models : they are dominated by strong emission lines , mainly due to @xmath174co , and a continuum which results from compton scattering of line photons . models 1 and 3 have characteristically similar @xmath0-ray light curves . after an initially fast rise ( lasting for about 5 and 10 days after the explosion for model 3 and 1 , respectively ) their light curves have a small plateau before passing to a second rise to maximum light at about 60 days . in contrast , model 6 shows no early rise / plateau but reaches maximum at about 60 days in one continuous rise . these differences result from the significantly different masses of the helium shells of the models . model 1 has a rather massive shell of @xmath288 with @xmath289 of radioactive isotopes . the fast initial rise of the light curve of model 1 is caused by @xmath0-photons which originate from this outer shell . due to the small optical depth in the outer shell , those @xmath0-photons start to stream freely at about 10 days and the emerging flux from the outer shell decreases . at the same time , however , @xmath0-photons from the c / o core start to escape and keep the light curve rising until the core also becomes transparent to @xmath0-rays at about 60 days after which the @xmath0-photons start to stream freely . for model 3 , we observe the same effect . however , due to its lower helium shell mass ( @xmath290 ) @xmath0-photons from the shell escape even earlier ( at about 5 days ) . since the detonation of the c / o core of this model produces much more @xmath174ni , the @xmath174ni - rich region is far more extended than in model 1 . this means that @xmath174ni nuclei from the c / o core are present in regions where the compton optical depth is relatively low , allowing their @xmath0-photons to more easily escape . thus , photons from the core dominate the rise earlier than in model 1 . in the extreme limit of model 6 , the shell ( @xmath291 ) is completely negligible and the massive @xmath174ni region in the c / o core extends so close to the surface that the rise of the light curve is totally dominated by @xmath0-photons escaping from the c / o core . a similar effect shows up in the early - time ( 20 days ) @xmath0-ray spectra of our models . while model 6 shows a clear indication of the 0.158 and 0.270 mev lines of @xmath174ni , those are invisible for model 1 . since the compton cross - section increases with decreasing photon energy , soft - energy lines are most easily buried by photons being compton down - scattered from higher energies and can only be observed if @xmath174ni is present at low optical depths @xcite . thus the presence of the 0.158 and 0.270 mev lines of @xmath174ni in model 6 is a direct consequence of the large extension of the @xmath174ni bubble in this model , compared to model 1 . the harder @xmath174ni lines at 0.750 , 0.812 and 1.562 mev , in contrast , are also visible in model 1 . however , they are weaker due to the smaller mass of @xmath174ni synthesized in model 1 . note that despite containing @xmath193cr or @xmath193v close to the surface , our models show no clear features of these radioactive isotopes in their @xmath0-ray spectra . an interesting effect concerns the 511 kev annihilation line : for model 3 and 6 the strength of this line increases from 20 to 60 days significantly , but it does not for model 1 . in model 1 , the 511 kev line is dominated by positrons from @xmath193v at 20 days . being located in the outer layers , the annihilation photons escape easily making the line strong . at 60 days most of the @xmath193v has already decayed . then , the 511 kev line results from the annihilation of @xmath174co positrons from the c / o core . for model 3 and 6 , in contrast , the line is always dominated by annihilation photons originating from @xmath174co positrons in the c / o core . due to the longer life time of @xmath174co and the longer time it takes for photons to escape from the core , the strength of the 511 kev line increases from 20 to 60 days . in the last section we presented synthetic observables for the sub - chandrasekhar - mass double detonation models with minimum helium shell mass of @xcite . compared to observed sne ia , these models show some promising features : 1 . they predict a wide range of brightnesses that covers the whole range of observed sne ia . 2 . their light curve rise and decline times are for most bands in reasonable agreement with those observed . but despite these positive features the @xcite models can not account for all the properties of observed sne ia since they have peculiar light curves and spectra : 1 . the colours of the models are too red compared to observed sne ia . this is particularly obvious in the evolution of the @xmath1 colour , where all the models are redder than spectroscopically normal sne ia for all epochs . 2 . the model spectra can not reproduce the strong features of intermediate - mass elements typical of sne ia at maximum light . in detail , there are further problems concerning the exact light curve shapes and decline rates as well as a strong viewing - angle dependence which is caused by the point - like ignition of our models . we have argued that all these problems are mainly due to the burning products of the helium shell . moreover , @xcite , have recently shown that detonations of centrally ignited spherically symmetric naked sub - chandrasekhar - mass c / o wds are capable of reproducing the observed diversity of light curves and spectra of sne ia at least to a similar level of agreement as models of the more standard chandrasekhar - mass delayed detonations . this naturally leads us to speculate on whether the helium shell in double detonation models could be altered in some way to reduce its negative impact on the synthetic observables . we first explicitly investigate the extent to which the shortcomings and viewing - angle dependence of our models can indeed be attributed to the helium shell . for that purpose we constructed ( for model 3 ) a toy model which contains only the burning products of the detonation in the initial c / o core but not those of the initial helium shell . for the models of @xcite this can be done in a straightforward manner : since the models use two different sets of tracer particles to simulate the nucleosynthetic yields of core and shell burning respectively ( see section 3.3 of @xcite ) , we obtain such a model by restricting our sph - like reconstruction algorithm ( section [ sec : rt ] ) to the core tracers . properties of this `` core - only '' model ( hereafter 3c ) are listed in table [ tab : modelparas ] . figure [ fig : m03_nohe_lightcurves ] shows band - limited synthetic light curves obtained from our radiative transfer simulations for this model as seen equator - on and from the two polar directions . in contrast to the light curves of model 3 in figure [ fig : m03_lightcurves ] , which are strongly dependent on the viewing angle , the light curves of model 3c show only a moderate line - of - sight dependence . thus , at maximum light we find now only a variation of @xmath292 and @xmath293 magnitudes for @xmath237 and @xmath238 band , respectively . for model 3 these values were significantly larger ( @xmath280 and @xmath281 magnitudes , respectively ) . redder bands show no significant line - of - sight dependence in model 3c . moreover , the light curves of model 3c give an excellent representation of sn 2004eo in the @xmath238 , @xmath239 and @xmath240 bands . @xmath237 and @xmath241 are not in perfect agreement , but still reasonable compared to the agreement between other first principles explosion models and observed sne ia ( e.g. @xcite ) . in particular , the colours of this toy model are now in good agreement with observed sne ia and not too red as it is the case for model 3 . in the nir bands , in contrast , the agreement is no better . however , the nir light curves are much more difficult to model accurately since they require simulations with an extensive atomic data set to properly simulate flux redistribution by fluorescence which strongly affects these bands @xcite . here , however , we have restricted ourselves for computational reasons to a simplified atomic data set ( cd23_gf-5 of @xcite ) with only @xmath294 lines . this has been shown to give reliable results in the optical bands which , for our purposes , are the most important since they are the most different between our toy model 3c and model 3 of @xcite . in the nir bands , in contrast , model 3c and model 3 give rather similar results ( see figure [ fig : compare_m03_lightcurves ] ) . the good agreement between our toy model 3c and observational data does not only hold for band - limited light curves but also for individual spectral features as can be seen from figure [ fig : m03_nohe_spectrum ] which shows a spectrum of model 3c at 3 days before @xmath238-band maximum . compared to sn 2004eo , our toy model succeeds in reproducing the characteristic spectral features of intermediate - mass elements in sne ia . this is highlighted by our colour coding . moreover , it shows an overall flux distribution which is in almost perfect agreement with the observed spectrum of sn 2004eo and we see no strong flux redistribution by titanium ( compare with model 3 in figure [ fig : maxspectra_all ] ) . this confirms our conclusion from section [ sub : oo_los ] that the peculiarities of our model spectra with respect to the observations and their strong viewing - angle dependence are mainly due to the shell material and its compositional asymmetries . it also shows that the off - centre ignition of the secondary detonation in the c / o core causes only a minor viewing - angle dependence which is on the order of the observed variation between sne ia . in light of the discussion above , we are motivated to speculate on how the influence of the helium shell might be reduced . in the sub - chandrasekhar - mass double detonation scenario , the helium shell can not be removed entirely since it is required to trigger the detonation . also the helium shell mass adopted in the @xcite models is already the minimum that might be expected to detonate @xcite . in section [ sec : oo ] , however , we have argued that the differences between our model spectra and observations are not a consequence of the helium itself but of its particular burning products , namely titanium and chromium produced in the outer layers . the yields of these elements are affected by details of the nucleosynthesis in the shell . the degree of burning in the shell material ( and thus its final composition ) can be affected by the initial abundance of heavy nuclei ( e.g. @xmath295c ) which in turn depend strongly on triple-@xmath296 reactions during previous hydrostatic burning and dredge - up phases from the core @xcite . since the time - scale for @xmath296-captures behind the detonation shock front is significantly shorter than that of triple-@xmath296 reactions , such seed - nuclei can limit the @xmath296-chain before reaching nuclear statistical equilibrium . if , for example , in a shell consisting of a mixture of @xmath297he and @xmath295c the number ratio of free @xmath296-particles to @xmath295c - nuclei on average is less than 6 ( corresponding to a mass ratio of 2 ) , the @xmath296-chain will end at @xmath298ar . thus , it is possible that more intermediate - mass elements and less titanium and chromium may be produced . therefore it is interesting to consider how the burning of the helium might be different from that found by @xcite for different initial compositions of the shell . a full study of this goes beyond the scope of this work and will be published in a follow - up study . here , we illustrate the possibility of obtaining better agreement with data for just one example . in the @xcite models it was initially assumed that the shell consisted of pure helium . to demonstrate the sensitivity to the initial composition of the shell , we set up another toy model . for this `` modified '' model ( hereafter 3 m ) , we homogeneously polluted the shell of model 3 with 34% ( by mass ) of @xmath295c and repeated the hydrodynamics and nucleosynthesis calculation ( in the same way as described by @xcite ) ( compared to @xmath299 in model 3 ) to suppress further triple-@xmath296 burning in the shell . as a consequence the density and also the shell mass changes slightly compared to the original model 3 ( cf . table [ tab : modelparas ] ) . ] . we found that a core detonation was still triggered but the different shell burning led to a substantial reduction of the mass of @xmath300ti , @xmath193cr and @xmath192fe in the shell ( nucleosynthetic yields for core and shell of the model are given in table [ tab : modelparas ] ) . since the detonation tables of @xcite are only valid for pure helium , model 3 m is not fully self - consistent . nevertheless , it is a useful toy model to explore the basic effect of a modified shell composition . figure [ fig : compare_m03_lightcurves ] compares the angle - averaged band - limited light curves of this modified model ( 3 m ) , to those of model 3 of @xcite and our core - only toy model ( 3c ) . as can be seen , the modified model 3 m produces light curves very similar to those of model 3c despite having about the same shell mass as model 3 . the most obvious difference between our modified and core - only models occurs in the @xmath237 band : the titanium in the outer layer of model 3 m causes some line blocking , leading generally to a dimmer @xmath237-band magnitude than for model 3c which has no outer layer . compared to model 3 , with its large titanium mass in the shell , however , this effect is much weaker . the @xmath238 band , which is strongly affected in model 3 , shows no significant titanium absorption for model 3 m . another slight difference between model 3 m and 3c occurs after the first peak in the @xmath241 band . comparing the light curves of model 3 m to sn 2004eo , we find qualitatively similar agreement as for model 3c . this generally also holds for the angle - averaged spectrum at 3 days before @xmath238-band maximum , shown in figure [ fig : m03_modified_spectrum ] . compared to model 3c , where the agreement was almost perfect , there are some minor shortcomings . but the model is dramatically improved compared to model 3 of @xcite . the small differences between model 3c and 3 m are again mostly due to the titanium in the outer layers which leads to pronounced absorption troughs bluewards of the caii h and k lines and redwards of @xmath301 . this suggests that model 3 m still over - produced titanium in the shell . interestingly , the enhanced calcium abundance in the outer layers ( cf . table [ tab : modelparas ] ) leads to a stronger caii nir triplet , bringing the model in better agreement with the spectrum of sn 2004eo at the corresponding epoch . therefore some _ calcium _ in the outer shell is an _ improvement _ over model 3c . this suggests that a slight further reduction in the degree of burning so that titanium is further suppressed in favour of calcium ( just one step down the @xmath296-chain ) could lead to very good agreement . in summary , polluting the initial helium shell of model 3 m with @xmath295c significantly improved the agreement between our synthetic spectra and light curves and those observed for sne ia making this model a promising progenitor candidate for sne ia . we stress again that this improvement results only from the change in the composition of the burning products of the helium shell which contains much less titanium and chromium for this model ( the total shell mass stays about the same ) . given that the initial composition of the helium shell depends on several processes including details of the accretion physics and hydrostatic burning phases that might precede the detonation or possible dredge - up of material from the c / o core , this leaves some scope to find sub - chandrasekhar - mass double detonation models which give reasonable agreement with observed sne ia . this , however , must be investigated by future follow - up studies that more fully explore the influence of the initial composition of the helium shell on the burning products and link the initial composition of the helium shell directly to the evolution of progenitor models . moreover , we note that different ignition geometries might also lead to better agreement with observational data . in particular , more symmetric ignition geometries , e.g. ignition in an equatorial ring or simultaneous ignition in multiple points ( as studied by @xcite for the case of more massive helium shells ) , are likely to alleviate the strong viewing - angle dependence found for the point - like ignition of the @xcite models . our results also highlight the strong sensitivity of the radiative transfer to particular elements / ions ( in our case titanium and chromium which represent only a tiny fraction of the ejecta mass yet dominate our conclusions ) . this emphasizes the need for a better description of nuclear reaction rates and continued study of the radiative transfer processes ( and atomic data ) in order to quantify more fully the systematic uncertainties which arise due to the complexity of spectrum formation in supernovae . in particular , we note that almost all the flux redistribution done by titanium and chromium in our models is due to their singly ionized states . since the current ionization treatment of artis neglects non - thermal processes ( see @xcite for more details ) , we can not exclude that the actual ionization state in the helium shell ejecta would be higher due to non - thermal ionization from the radioactive isotopes produced during the helium burning . this could also significantly improve the agreement between our models and observational data , as a numerical experiment with an artificially enforced higher ionization state for titanium and chromium has shown . in this paper we presented synthetic observables for the sub - chandrasekhar - mass double detonation models of @xcite . we found that these models predict light curves which rise and fade on time - scales typical of sne ia . moreover , they produce a large range of brightnesses which covers the whole range of observed sne ia . however , they do not account for all the properties of observed sne ia since they have peculiar spectra and light curves . in particular , their @xmath1 colours are generally too red compared to observed sne ia . this is in contrast to the results of earlier work on models with more massive helium shells @xcite . in addition , our model light curves and spectra show an unreasonably strong viewing - angle dependence due to the point - like ignition of the @xcite models and the resulting ejecta asymmetries . detonation of a pure helium shell leads to a layer containing iron - group elements like titanium and chromium around the core ejecta . these elements have a vast number of optically thick lines in the uv and blue part of the spectrum making them very effective in blocking the flux in these wavelength regions and redistributing it to the red . we used a toy model to show that this layer of titanium and chromium causes the peculiar red colours of our light curves and also the peculiar spectral features . moreover , we found that this toy model reproduces the observed properties of sne ia remarkably well . the toy model also showed that the strong viewing - angle dependence of our models results from the compositional asymmetry in the helium shell ejecta and not from the off - centre ignition of the c / o core . we stress , that the additional energy release in the shell , due to the production of radioactive nuclides during the helium burning , is relatively inconsequential for our models even at @xmath0-ray energies the signatures of the surface @xmath193cr and @xmath192fe are not apparent . instead , in the optical / uv the shell has a strong signature but this is primarily due to the additional opacity in the outer layers which affects the transport of energy from the core to the surface . we conclude that , if the double detonation sub - chandrasekhar - mass model valid for normal sne ia , the properties of the post - burning helium shell material need to be different from those in the @xcite models . since @xcite considered the limit of the least massive helium shells which might ignite a detonation in the helium shell , their models represent the most optimistic case for reducing the influence of the shell material by simply reducing the shell mass . however , we argue that the mass of the helium shell ejecta is not the main problem but rather the peculiar composition including comparably large masses of titanium and chromium . we illustrated this using a second toy model , where the initial composition of the helium shell was polluted with 34% ( by mass ) of @xmath295c . by providing additional seed - nuclei for @xmath296-captures , this leads to burning products with lower atomic number ( i.e. intermediate - mass elements rather than mainly iron - group elements ) . spectra and light curves of this model which has about the same shell mass as model 3 of @xcite show comparably good agreement to observed sne ia as the shell - less toy model . taking into account all these results , we argue that these systems might yet be promising candidates for sn ia progenitors . much more work will be needed to properly investigate this possibility . besides a more detailed description of the excitation / ionization state in the radiative transfer modelling which includes non - thermal effects , we need a better understanding of the initial composition of the helium shell and the incomplete burning processes which take place in this material to reach reliable predictions of the burning products of the helium shell . although only a tiny fraction of the mass , the post - burning composition of the shell material is critical to assessing the viability of the sub - chandrasekhar - mass double detonation scenario . we thank r. pakmor and s. taubenberger for many helpful comments and our referee , david branch , for a supportive report . the simulations presented here were carried out at the computer center of the max planck society in garching , germany . this work was supported by the deutsche forschungsgemeinschaft via the transregional collaborative research center trr 33 `` the dark universe '' , the excellence cluster exc153 `` origin and structure of the universe '' and the emmy noether program ( ro 3676/1 - 1 ) .

far - ultraviolet ( fuv ) , h , and hi observations of dwarf galaxy holmberg ii are used to investigate the means by which star formation propagates in galaxies lacking global internal triggering mechanisms such as spiral density waves . the observations trace the interaction between sites of massive star formation and the neutral and ionized components of the surrounding ism in this intrinsically simple system . both local and large - scale triggering mechanisms related to massive star formation are seen , suggesting that feedback from massive stars is a microscopic process operating in all galaxies to a certain degree . the data emphasize the importance of local conditions in regulating star formation from evidence such as massive stars inside ionized shells , compact hii regions surrounding aging clusters , and stars formed in chains of progressing age . surface brightness profiles show current activity correlates with the time averaged level of past star formation at a given radius demonstrating a reliance on local conditions . large - scale triggering by hi shells is supported by observations of progenitor populations as well as secondary sites of star formation associated with their dense rims . analysis of the energy available from massive stars inside hi shells indicates energy deposited into the ism from supernovae and stellar winds is sufficient to account for the hi morphology . ages of individual star forming regions are derived using b , h , and fuv photometry and show both older , diffuse fuv regions and younger , compact hii regions . the distribution of ages is reconciled with the hi morphology , showing a clear preference of young regions for areas of dense hi and old regions for hi voids . global kinematical properties may also play a role in the star formation process since differences in the rotation characteristics of the neutral gas disk correlate with differences in triggering mechanisms . large - scale feedback from massive stars is shown to operate in regions that lack differential shear in the gas disk .

astro-ph9910098

the pervading questions surrounding our current view of large - scale star formation are primarily associated with the need to understand the mechanism for triggering and globally maintaining star formation . the first step in forming a fundamental theory to describe star formation in all environments is understanding the process in intrinsically simple systems . this study utilizes a unique observational dataset to explore the processes which may influence new star formation and its relation to the ism within a relatively uncomplicated dynamical environment . far - ultraviolet ( fuv ) , h@xmath0 , and hi images of dwarf galaxy holmberg ii ( hoii ) are used to gain new insight into the star formation process by studying the interconnection between massive star formation and the ionized and neutral components of the galaxy s ism . irregular galaxies lack a dominant internal global triggering mechanism for cloud compression and star formation such as density waves , yet exhibit a wide range of star formation activity . non - interacting low surface brightness late - type galaxies , including magellanic - type irregulars ( i m ) , are ideal for this study due to the limited number of global internal processes influencing star formation while at the same time having star formation rates per unit area similar to much more massive spiral galaxies ( hunter 1997 ) . due to their small size , the amount of energy input per a given star formation event has a pronounced effect on the global ism . instabilities in the ism introduced by massive star formation may play an important role in the star formation process in these systems . they display relatively uncomplicated internal gas dynamics , generally characterized as slow solid body rotation . these characteristics create an environment of reduced shear allowing features in the ism to be longer lived . they often possess huge gas envelopes extending well beyond the optical regime of the galaxy . large vertical scale heights allow features in the ism to grow to large sizes before breaking out of the hi layer . in addition , these systems allow for the test of star formation models at very low metallicities . for these reasons , late - type low surface brightness galaxies are powerful tools for understanding the star formation process and galaxy evolution . the apparently non - interacting im - type galaxy hoii ( ddo 50 , ugc 4305 ) is a member of the m81-ngc 2403 group of galaxies . the galaxy is classified as a david dunlap observatory ( ddo ) dwarf galaxy due to its low surface brightness ( van den bergh 1959 ) . although it is slightly too bright to fit the magnitude criterion of a `` true dwarf '' galaxy of m@xmath116 ( nilson 1973 ) , it is usually referred to as a dwarf galaxy in the literature . iras observations of hoii suggest a minimal presence of dust , so little hidden star formation is taking place ( hunter _ et al . the galaxy has a low metal abundance , z / [email protected] ( hunter & gallagher 1985 ) , similar to that of the large magellanic cloud ( lmc ) . et al . _ ( 1994 ) find that the sizes of hii regions as well as their size distribution are statistically normal in comparison to galaxies of the same type , distance , and luminosity . hoii has a hi envelope extending to 1.5 d@xmath3 , where the holmberg diameter ( d@xmath3 ) is the diameter defined by the b isophote at 26.6 mag arcsec@xmath4 . the hi morphology of hoii is intriguing due to the various shells and holes present throughout the galaxy ( puche _ et al . _ et al . _ ( 1998 ) estimate a distance of 3.05 mpc using 28 cepheid candidates . this is different than the distance estimate of 3.63 mpc determined by freedman _ et al . _ ( 1994 ) for m81 using hubble space telescope observations of cepheid variables . since hoii is likely the closest member of the m81-ngc 2403 group , 3.05 mpc is adopted for the distance of hoii . table 1 summarizes the parameters assumed for the galaxy in this study . lllr parameter & & value & reference + optical center & @xmath5 & 8@xmath6 13@xmath7 535 & dressel & condon ( 1976 ) + & @xmath8 & 705213 & + dynamical center & @xmath5 & 8@xmath6 13@xmath7 53@xmath9 & westpfahl ( 1997 ) + & @xmath8 & 7052473 & + distance & & 3.05 mpc & hoessel , saha , & danielson ( 1998 ) + m@xmath10 & & @xmath1116.52 & rc3 , de vaucouleurs _ et al . _ ( 1991 ) + inclination & & 41 & puche _ et al . _ ( 1992 ) + position angle & & 177 & puche _ et al . _ ( 1992 ) + axial ratio , q = b / a & & 0.74 & huchtmeier & richter ( 1988 ) + a@xmath10 & & 0.10 & rc3 , de vaucouleurs _ et al . _ ( 1991 ) + a@xmath12 & & 0.20 & this study + observations in the vacuum ultraviolet ( uv , @xmath13 912 - 3200 ) isolate hot sources from the underlying optical cool star background in galaxies . ultraviolet observations allow direct detection of young massive stars , which are responsible for the majority of photoionization , photodisassociation , elemental synthesis , and kinetic energy input in galaxies . the uv observations used in this study are in the fuv window ( @xmath13912 - 2000 ) and are particularly useful in characterizing high - mass star formation since the dominant luminosity sources at 1500 are ob stars with mass m @xmath14 5 m@xmath15(fanelli _ et al . _ 1992 ) . understanding the uv morphology and its relation to the star formation process has broad implications considering that the rest frame fuv is detectable at high redshift , z@xmath143 , with optical and ir instruments . oconnell ( 1997 ) stresses that determining a `` morphological k - correction '' for galaxies is needed to assess the morphology of distant objects . therefore , understanding the processes responsible for the uv emission in nearby systems will play an important role in advancing our understanding of galaxy evolution . this study addresses several issues which hunter ( 1997 ) stresses are vital to an improved understanding of star formation processes within irregular galaxies . these key issues are tied to the need for a better understanding of the regulation mechanism for star formation on global and local scales , including a quantitative description of both the feedback processes operating in irregular galaxies and the impact which huge stellar complexes have on the surrounding ism . morphological comparisons between the fuv flux and other radiation linked to star formation provide valuable information regarding the interaction between a star forming site and its surrounding medium . for example , the ratio of the fuv and h@xmath0 emission from a young star forming region is indicative of the cluster age . the suggested link between massive star formation and hi shells can be studied via the relative fuv and hi emission morphologies . since the fuv is able to isolate the massive stellar component over timescales similar to the kinematic ages of hi shells , this comparison could identify a progenitor population . understanding if characteristic differences in the ism are related to differences in star formation mechanisms is another important issue . the fuv image can be used to probe the possible relationship between massive star formation and the kinematics of its hi disk . the intrinsic dynamical reduction of shear in the gas disks compared to spiral galaxies most likely influences the overall star formation rate . despite the wealth of information provided by uv images , they are seldomly used in star formation studies because spatially - resolved uv images are rare . the fuv images used in this study are provided by the ultraviolet imaging telescope ( uit ) which flew as a member of the astro observatory on two shuttle missions in december , 1990 and march , 1995 . together , a total of @xmath16 fuv images of galaxies exhibiting recent star formation were observed . the @xmath17 spatial resolution of the uit was a 5@xmath1120 times improvement over previously available uv imagery . besides the apollo 16 uv image of the lmc in 1972 ( page & carruthers 1981 ) , the only other uv image of a galaxy similar in type to hoii prior to the astro-2 mission was the dwarf galaxy holmberg ix . it was serendipitously observed during the astro-1 mission due to its close proximity to m81 . using the uv data , et al . _ ( 1993 ) characterized the most recent star formation event , confirming that no recent star formation was occurring in holmberg ix . during the second flight of the uit , a dwarf galaxy program was initiated . the results of this program were first presented by gessner _ et al . _ detailed analysis of all of the uit dwarf galaxy targets is presented by stewart ( 1998 ) . global photometry of the galaxy is given in 3 and the radial surface brightness profiles in 4 . photometry of individual star forming complexes along with their derived ages and internal extinctions are presented in 5 . comparisons between the relative morphologies of the fuv and h@xmath0 emission are given in 6 and that of the fuv and hi in 7 . the principal results are summarized in 8 . holmberg ii was observed by the uit during the astro-2 mission in march , 1995 . table 2 summerizes the observations of the galaxy used in this study . the targets were observed using the broad - band uit b1 filter , which has @xmath18 and @xmath19 = 354 . the images were recorded on iiao film and digitized with a pds microdensitometer at goddard space flight center , producing 2048@xmath202048 pixel images with scale 114 pixel@xmath21 . the images were calibrated using iue spectrophotometry of stars observed by the uit . the fuv magnitudes are computed from fuv flux using m@xmath22= @xmath23 log@xmath24(fuv)@xmath1121.1 , where fuv is the incident flux in erg ( @xmath25 s ) @xmath21 . the chief photometric uncertainties are in the form of low - level nonuniformities introduced via the development and digitization processes . uncertainty in the absolute calibration can lead to uncertainties of up to 10 @xmath11 15 % in the fuv flux . a detailed discussion of the reduction of the uit data to flux - calibrated arrays is given by stecher _ et al . _ ( 1992 , 1997 ) . lccl telescope & filter & exposure & date + & & time ( s ) & + uit & fuv ( b1 ) & 1310 & mar 11 , 1995 + mlo & u & 1200 & nov 2 , 1995 + mlo & b & 300 & feb 24 , 1995 + mlo & r & 120 & feb 24 , 1995 + mlo & h@xmath0 & 600 & feb 24 , 1995 + the fuv background level ( @xmath26 25 mag arcsec@xmath4 ) is checked by taking the mean of 20 , 20@xmath2020 pixel boxes in areas void of any galaxy or stellar flux . the fuv background level is minimal , 0.0@xmath27 analog data units ( adu s ) . the uit has limited accuracy at the lowest light levels . the error in the background level reflects this intrinsic uncertainty . ccd observations of hoii were obtained in the johnson u and b bands , and the kron - cousins r and h@xmath0 bands with san diego state university s 1.0 m telescope at the mount laguna observatory ( mlo ) . the optical images are median combined with at least two other images having the same exposure time to remove cosmic rays and other defects . astrometry and photometry are implemented using standard idl procedures for data reduction . background values are determined for each image by taking the mean of pixels in 20 , 10@xmath2010 pixel boxes . the hoii broad - band images are calibrated using unpublished photoelectric photometry of three bright stars in the galaxy ( corwin 1997 ) . total magnitudes given in the rc3 agree with our measured values within reasonable limits providing a check of the calibration constants . the h@xmath0 filter at mlo has a centroid wavelength 6573 and width 61 which includes the [ n ii ] lines . a correction for this contribution to the h@xmath0 flux is made by assuming f ( h@xmath0 ) / ( f ( h@xmath0 ) + f ( [ n ii ] ) ) @xmath28 , based on spectral observations of h ii regions by hunter & gallagher ( 1985 ) . this correction is applied using the calibration technique of waller ( 1990 ) and observations of the h@xmath0 standard star bd + 82015 ( stone 1977 ) . attempting to reproduce the irregular apertures defined by hodge _ et al . _ ( 1994 ) for h ii regions in hoii provides a check for the h@xmath0 calibration constant . our photometry of these regions agree within @xmath29 % . deviations may be caused by failure to reproduce the irregular apertures exactly . an h@xmath0 emission image is produced by scaling stars in the h@xmath0 image with those in the r - band image and subtracting the stellar continuum component . the 21 cm images were obtained with the vla b , c , and d arrays . the uniformly weighted column density map has a spatial resolution of 4@xmath30 , corresponding to a spatial resolution of 60 pc at the distance of hoii . this image is used in the puche _ et al . _ ( 1992 ) study of hoii where a complete description of its observational characteristics is given . in figure 1a the 100 @xmath20 100 fuv image of hoii is presented . the fuv morphology is quite patchy with the brightest features contained in a central star forming arc . there is a region of diffuse fuv emission extending to the southwest and three separate patches of fuv emission to the north of the central arc . since galaxies at high redshift are seen in the rest frame fuv with optical instruments , the fuv image is compared to the r - band image in figure 1b to illustrate the importance of understanding the fuv morphology of local galaxies . the r - band image of the galaxy exhibits a relatively smooth distribution of light compared to the irregular distribution shown by the fuv image . the true optical morphology of a high redshift galaxy could be quite different than that observed in the optical bands . + integrated magnitudes of hoii are obtained using elliptical apertures whose shape and orientation are based upon the derived hi structural parameters given in table 1 . the dynamical center is adopted as the aperture center since it has more physical significance than the optical center , especially when comparing integrated parameters from observations ranging from the r - band to the fuv . the optical magnitudes are derived using images with the foreground stars masked , the brightest of which are given the average flux from the surrounding medium . therefore , the derived magnitudes may be slightly less than published catalog values . since the fuv image is free from foreground star contamination , foreground stars are masked from optical images in order to make a true comparison with the fuv band . the total integrated magnitudes , corrected for galactic extinction , a@xmath31 , are given for each bandpass in table 3 . galactic extinction in the fuv is derived assuming the galactic reddening curve of savage and mathis ( 1979 ) integrated over the uit b1 bandpass . these are related by a@xmath12 / e(b@xmath11v ) = 8.33 ( hill _ et al . the errors account for the photometric uncertainty and @xmath327% uncertainty in the distance estimate . corrections for internal extinction will be discussed later ( 5.2 ) . lccl filter & apparent & absolute & a@xmath31 + & magnitude & magnitude & + fuv & [email protected] & @xmath34&0.20 + u & [email protected] & @xmath35&0.12 + b & [email protected] & @xmath36&0.10 + r & [email protected] & @xmath37&0.06 + the fuv magnitude is at the faint end of the range of absolute magnitudes observed by the uit . in a sample of 35 galaxies exhibiting recent massive star formation , the range of observed absolute magnitudes spans from @xmath1117 to @xmath1122 ( fanelli _ et al . _ the corresponding fuv@xmath11b = @[email protected] color falls into the range observed by faust ( a balloon - born uv camera ) for a sample of im - type galaxies ( deharveng _ et al . _ 1994 ) . the integrated fuv flux as a function of radius provides information regarding the concentration of recent massive star formation . the fuv growth curve , corrected for galactic extinction , is shown in figure 2 . it illustrates that recent massive star formation is not particularly centrally concentrated in hoii despite the presence of the prominent central star forming arc . the aperture containing this feature in its entirety ( 17 ) only contains @xmath3243% of the total fuv flux . therefore , massive star formation extends well beyond the central star forming arc . the intrinsic emission of a young stellar population can be translated into a global star formation rate ( sfr ) by comparing integrated fluxes with predictions from stellar population synthesis models . the timescale characterized by a given sfr depends upon the bandpass of the observation since the luminosity of a single generation of stars decays roughly as a power law ( oconnell 1997 ) . the lyman continuum ( balmer emission lines ) describes the sfr over a timescale @xmath38 5 myr while the uv continuum characterizes the sfr over a 100 myr timescale . since massive stars ( m@xmath1410m@xmath15 ) are the ionizing sources for the lyman continuum , observed here via the h@xmath0 recombination line , and the uv emission is dominated by stars @xmath14 5m@xmath15 , the sfr derived using these bandpasses describes the formation history of massive stars . late - type irregular galaxies are dominated by short - lived massive stars and are therefore excellent sources to provide an estimate of the sfr using either fuv or h@xmath0 observations . the global sfr for massive stars in hoii is derived using both the integrated fuv and h@xmath0 flux by applying formulae which convert the integrated light from each bandpass to an estimate of the global sfr . these formulae are taken from kennicutt ( 1998 ) who provides a self - consistent set of formulae for various regimes based on the calibration of madau _ et al . _ ( 1998 ) for a fixed salpeter ( 1955 ) imf with mass limits 0.1 to 100 m@xmath15 . the observed ( not corrected for internal extinction ) fuv and h@xmath0 luminosity and global sfr are listed in table 4 . the h@xmath0 luminosity and sfr@xmath39 agree reasonably well with the values derived by miller & hodge ( 1994 ) after differences in distances are removed . since internal extinction is not extreme low metallicity systems , the comparison between sfr@xmath39 and sfr@xmath12 is valid to the first order in estimating the recent to current formation history of massive stars . for example , differences in internal extinction effects between the fuv and h@xmath0 observations for a given e(b - v ) are likely on the order of a@xmath12/a@xmath404 , assuming a lmc extinction curve for internal reddening ( this ratio doubles for the smc extinction curve ) . this difference translates into a higher sfr@xmath12 than sfr@xmath39 in a given system by underestimating a@xmath12 to a greater degree than a@xmath39 . the results indicate the `` recent '' ( traced by the fuv ) and `` current '' ( traced by h@xmath0 ) massive star formation rates are comparable . the ratio sfr@xmath12 to sfr@xmath39 indicates that the fuv flux traces a slightly larger fraction of massive star formation than the h@xmath0 flux , perhaps suggesting hoii is in a post - burst phase . ccccc log(l @xmath41 ) & log(l@xmath12)&sfr@xmath39 & sfr@xmath12 & sfr@xmath12/sfr@xmath39 + & & m@xmath15yr@xmath21 & m@xmath15yr@xmath21 & + 39.55 & 38.67 & 0.028 & 0.049 & 1.75 + surface brightness profiles are derived using magnitudes corrected for galactic foreground extinction in elliptical annuli 16@xmath42 wide . the optical profiles are created using images with the foreground star component masked ; the brightest of these stars is given the average value from their surroundings . to effectively compare the behavior of surface brightness profiles of irregular - type galaxies , the choice of aperture center is important . as with the integrated magnitudes , the dynamical center is chosen as the aperture center instead of the optical center or brightest feature in order to make a uniform physical comparison with other galaxies . the patchy distribution of fuv light shown in the uit image is represented in the surface brightness profiles by pronounced inflections at various radii . the fuv surface brightness profile , shown in figure 3 , has a double peaked shape with a central depression . at the aperture center , the average surface brightness is @xmath43=21.6 mag arcsec@xmath44 while at the first peak , 0@xmath45 from the center , the average value is @xmath43=21.2 mag arcsec@xmath46 . surface brightness maps of the galaxy indicate that the highest levels of activity lie in the region responsible for this peak ; the brightest knot in the center of the star forming arc has the highest levels , @xmath4718.9 mag arcsec@xmath4 in a resolution element ( 5@xmath42 ) . following the peak , the profile exhibits somewhat exponential behavior until r@xmath322 , where it takes a constant value of @xmath4722.4 mag arcsec@xmath4 . the flattening of the profile is due to flux from the group of regions on the eastern side of the central star forming arc . a second , yet relatively small peak occurs at [email protected] , after which it falls off sharply . the second peak can be accounted for by the diffuse patches of emission in the northern part of the galaxy . the sample of fuv surface brightness profiles presented by stewart ( 1998 ) offers little uniformity with regard to identifying characteristics for late - type low surface brightness systems . it is a relatively heterogeneous group which does not follow either a characteristic r@xmath48 law or a pure exponential . however , some general trends do emerge which could have important implications in interpreting optical observations of high redshift galaxies where the observed light is the redshifted uv continuum . these characteristics are also exhibited by the hoii profile : generally flat with inflections associated with prominent star forming complexes , healthy activity far from the galaxy center , and non - centralized organization of star formation . as compared to fuv profiles of other galaxies observed by the uit , the hoii profile has characteristics somewhat similar to those of ibm - type galaxy ngc 4449 ( fanelli _ et al . _ ngc 4449 has an exponential fuv surface brightness profile with inflections due to blue knots ( hill _ et al . the profile does not have a central light depression , but this may be due to the choice of the aperture center . although steeper than the hoii profile , the ngc 4449 profile shows similar characteristic inflections due to star formation at large radii from the center . in contrast to the lack of fuv profiles , there have been numerous studies characterizing the optical light profiles of low surface brightness galaxies ( karachentseva _ et al . _ 1996 ; patterson & thuan 1996 ; vader & chaboyer 1994 ) . the b - band profile of hoii , shown in figure 4 , can be characterized as exponential with a central depression . this behavior is typical of late - type low surface brightness galaxies . the underlying exponential component indicates the presence of an optical disk superimposed on bright star forming knots while the central trough demonstrates that star formation is not centrally organized around a nucleus . a rough estimate of the central surface brightness of the hoii b - band profile is made by fitting it with a pure exponential which is extrapolated to the center of the galaxy . corrected for galactic foreground extinction , @xmath49(0)=22.3@xmath50 mag arcsec@xmath4 the range in central surface brightness of the galaxies in the karachentseva _ et al . _ ( 1996 ) sample is 22.2 @xmath51(0)@xmath38 24 mag arcsec@xmath4 compared to the canonical value of 21.65 @xmath52 mag arcsec@xmath4 for disk galaxies ( freeman 1970 ) . holmberg ii falls at the brighter end of the observed range for low surface brightness irregular galaxies . other studies report an average @xmath53(0)@xmath54=23.5@xmath55 ( rnnback and bergvall 1994 ) and @xmath53(0)@xmath54=23.5@xmath56 mag arcsec@xmath4 ( vader & chaboyer 1994 ) . the general lack of correlation between star formation properties and other global properties such as abundance or global gas parameters in irregular galaxies suggests that local rather than global conditions are important in regulating star formation . for example , hunter , elmegreen , & baker ( 1998 ) do not find a correlation between the star formation and the ratio of the hi surface density to the critical density for gas instabilities in a sample of dwarf irregular galaxies , while the stellar surface brightness does correlate with this ratio . these results suggest that stellar energy provides feedback for star formation to a certain degree . studies of the radial dependence of the optical and h@xmath0 emission also suggest that local regulation of star formation may be important in both irregular ( hunter & gallagher 1985 ) and spiral galaxies ( ryder & dopita 1994 ) . the fuv bandpass gives some insight into the radial dependence of the recent star formation activity , yet over longer timescales than the h@xmath0 emission allows . figure 4 displays the fuv , u , b , and r - band profiles of hoii on the same linear scale . beginning with the r - band profile and moving counterclockwise around the figure , the profiles become less smooth as the wavelength of the profile decreases . this is expected due to the contribution from recent star formation , while at increasingly longer wavelengths , the light is more complicated measure of the entire history of star formation . the general shape of each profile is same and the radii at which increased star formation activity occurs remains relatively constant over the timescale which these profiles track star formation activity , roughly a few gyrs . figure 4 suggests that hoii is currently forming stars at the same radii as it has in the past , supporting the supposition that star formation in irregular galaxies is a local process and that local conditions such as gas distributions and feedback from massive stars may play a significant role in the star formation process . star forming properties are examined in further detail by characterizing the individual star forming complexes . star formation for individual regions is quantified by means of photometry in the observed bandpassses , used to derive a characteristic age and internal extinction for each region . the physical property upon which the age derivation is based is the different timescale of decay of the h@xmath0 and fuv flux from an evolving cluster . the h@xmath0 emission from an evolving cluster peaks at roughly 1 - 2 myr and falls off sharply . the fuv emission coming directly from the young massive stars in an evolving cluster peaks at roughly 5 myr and decays on a much longer timescale than the h@xmath0 emission ( see oconnell 1997 , fig . star formation regions are defined by selecting regions of associated fuv and h@xmath0 flux . since both the fuv and h@xmath0 emission define a star forming region in this context , the ages provide a snapshot of the pattern of star formation over a 100 myr timescale . this is an unconventional approach in defining the spatial distribution of young star formation which is normally described by h@xmath0 observations alone . the registered fuv and continuum subtracted h@xmath0 images of hoii are compared in figure 5 to highlight the differences between the spatial patterns of current and recent star formation activity . the choice of region boundary is facilitated by creating a fuv and h@xmath0 difference image , which reveals areas of associated emission . circular apertures are drawn so as to include hii regions defined in previous studies ( hodge _ et al . _ 1994 ) and approximately all fuv emission above a limiting surface brightness of 21.5 mag arcsec@xmath4 . using these criteria , 45 regions are identified . the apertures are drawn on a fuv , h@xmath0 difference image in figure 6 . some regions are defined by virtue of meeting this criteria in only one of the two bandpasses . the choice of aperture size is straightforward when associated emission is compact and coincident . ideally , an aperture would contain only a coeval association and its immediate hii regions . however , the resolution of the image limits the ability to select a single generation of star formation . the effect which sampling slightly different ages in a single aperture has on the subsequent derivations is discussed below . the fuv , b , and continuum subtracted h@xmath0 flux are measured for each region . a correction for the background light internal to the galaxy is made for the b - band photometry by averaging the values immediately surrounding a region . since this quantity may be variable for some regions , especially the larger ones , this estimate is the greatest source of error in the subsequent derivations . two time dependent quantities , the fuv@xmath11b color and log ( n@xmath57/l@xmath58 , are derived for each region after correction for galactic foreground extinction . the latter quantity is the logarithmic ratio of the number of lyman continuum photons , converted from h@xmath0 flux assuming case b recombination , to fuv flux . the observed fuv flux , fuv@xmath11b color , and log ( n@xmath57/l@xmath58 are listed in table 5 along with the aperture sizes . lcrcrcll 1 & 16.0 & 63.8 & @xmath113.18 & 24 & 1 & 0.07 & 0.63 2 & 16.0 & 41.4 & @xmath113.22 & 11 & 2 & 0.03 & 0.27 3 & 16.0 & 31.3 & @xmath113.45 & 7 & 2 & 0.0 & 0.0 4 & 25.6 & 168.2 & @xmath113.27 & 50 & 2 & 0.04 & 0.36 5 & 9.6 & 30.3 & @xmath113.34 & 10 & 1 & 0.05 & 0.45 6 & 22.4 & 122.9 & @xmath113.19 & 47 & 1 & 0.07 & 0.63 7 & 16.0 & 62.7 & @xmath112.97 & 28 & 1 & 0.09 & 0.81 8 & 28.8 & 67.5 & @xmath113.25 & 17 & 2 & 0.02 & 0.18 9 & 51.2 & 414.4 & @xmath113.15 & 172 & 1 & 0.08 & 0.72 10 & 19.2 & 19.9 & @xmath112.72 & 11 & 2 & 0.12 & 1.08 11 & 12.8 & 9.1 & @xmath113.04 & 2 & 3 & 0.02 & 0.18 12 & 22.4 & 37.3 & @xmath112.97 & 12 & 3 & 0.05 & 0.45 13 & 22.4 & 43.1 & @xmath112.53 & 18 & 4 & 0.08 & 0.72 14 & 22.4 & 67.9 & @xmath112.92 & 26 & 3 & 0.07 & 0.63 15 & 16.0 & 40.9 & @xmath113.09 & 9 & 3 & 0.01 & 0.09 16 & 16.0 & 22.0 & @xmath113.27 & 7 & 2 & 0.04 & 0.36 17 & 19.2 & 74.0 & @xmath113.04 & 22 & 2 & 0.04 & 0.36 18 & 19.2 & 67.8 & @xmath112.99 & 19 & 3 & 0.03 & 0.27 19 & 16.0 & 50.0 & @xmath113.08 & 11 & 3 & 0.0 & 0.0 20 & 16.0 & 33.4 & @xmath113.18 & 7 & 3 & 0.0 & 0.0 21 & 16.0 & 33.0 & @xmath112.82 & 8 & 4 & 0.01 & 0.09 22 & 16.0 & 18.9 & @xmath112.72 & 9 & 3 & 0.09 & 0.81 23 & 25.6 & 47.5 & @xmath113.27 & 11 & 2 & 0.01 & 0.09 24 & 15.6 & 10.1 & @xmath113.43 & 4 & 1 & 0.07 & 0.63 25 & 25.6 & 9.6 & @xmath112.28 & 10 & 2 & 0.19 & 1.71 26 & 35.2 & 25.6 & @xmath112.75 & 16 & 2 & 0.13 & 1.17 27 & 25.6 & 33.7 & @xmath113.18 & 7 & 2 & 0.0 & 0.0 28 & 32.0 & 61.2 & @xmath112.89 & 14 & 4 & 0.01 & 0.09 29 & 25.6 & 17.9 & @xmath112.73 & 8 & 3 & 0.09 & 0.81 30 & 54.4 & 174.6 & @xmath112.89 & 44 & 4 & 0.02 & 0.18 31 & 25.6 & 12.7 & @xmath112.55 & 8 & 3 & 0.13 & 1.17 32 & 19.2 & 4.6 & @xmath112.47 & 3 & 2 & 0.15 & 1.35 33 & 25.6 & 55.4 & @xmath113.04 & 29 & 1 & 0.11 & 0.99 34 & 22.4 & 22.2 & @xmath112.99 & 6 & 3 & 0.03 & 0.27 35 & 47.6 & 138.3 & @xmath112.60 & 49 & 4 & 0.06 & 0.54 36 & 32.0 & 69.6 & @xmath112.60 & 21 & 4 & 0.04 & 0.36 37 & 48.0 & 115.3 & @xmath112.92 & 25 & 4 & 0.0 & 0.0 38 & 32.0 & 8.7 & @xmath112.42 & 10 & 2 & 0.20 & 1.80 39 & 25.6 & 9.5 & @xmath112.62 & 10 & 1 & 0.19 & 1.71 40 & 25.6 & 16.2 & @xmath113.12 & 5 & 2 & 0.05 & 0.45 41 & 19.2 & 13.1 & @xmath113.33 & 3 & 2 & 0.0 & 0.0 42 & 19.2 & 6.6 & @xmath112.71 & 4 & 2 & 0.13 & 1.17 43 & 19.2 & 4.3 & @xmath113.16 & 2 & 1 & 0.06 & 0.54 44 & 16.0 & 2.5 & @xmath112.40 & 2 & 2 & 0.17 & 1.53 45 & 16.0 & 2.7 & @xmath113.13 & 1 & 1 & 0.09 & 0.81 the two time dependent parameters , fuv@xmath11b color and log ( n@xmath57/l@xmath58 , are compared to model values allowing a simultaneous determination of the characteristic age and internal extinction of each region . this method is used to estimate the ages of h ii regions in m81 ( hill _ et al . _ 1995 ) and ngc 4449 ( hill _ et al . _ 1998 ) where it is described in detail . a single generation instantaneous burst ( ib ) model , assuming the stellar evolutionary tracks of the geneva group ( schaerer _ et al . _ 1993 ) and the stellar atmosphere models of kurucz ( 1992 ) , is used . the model estimates l@xmath12 , n@xmath57 , and l@xmath10 for an evolving cluster over 100 myr with a 0.5 myr timestep . these parameters are then integrated over the each of the observed bandpasses . several assumptions are made to simulate the environment in hoii . a mass range of 1 @xmath11 100 m@xmath15 for model stars is assumed . under the assumption that regions are forming o - type stars , a relatively flat slope for the model imf is chosen . a slope of @xmath111.08 , derived by hill _ et al . _ ( 1994 ) for lucke - hodge regions in the vicinity of the 30 dor region in the lmc , is used . however , the subsequent age derivations are not very sensitive to differences in the imf . for example , using a model with a slope of @xmath111.0 and one with @xmath112.0 only results in small differences in the internal extinction estimates . at maximum , an underestimation of the extinction by e(b@xmath11v)=0.01 for stars 3.5 @xmath11 10 myr would result . however , differences in the assumed cluster metallicity show substantial differences in the shape of the model relationship . an lmc - like metallicity , z / z@xmath59 , is assumed . the observed values of fuv@xmath11b and log ( n@xmath57/l@xmath58 are corrected with an array of e(b@xmath11v ) s assuming a lmc curve , chosen under the assumption that the shape of the curve correlates with metallicity ( verter & rickard 1998 ) . in the fuv , a@xmath12/e(b@xmath11v)=9.04 for the lmc extinction curve ( hill _ et al . the array of corrected observations for each region are compared to the model plot . since the corrected data points are tied to e(b - v ) values and the model is age dependent , the point at which a corrected value agrees with the model curve gives both an age and internal extinction estimate . to illustrate this method , the lmc model for t= 0 @xmath11 50 myr is shown in figure 7 with several regions . the region number represents the uncorrected data point ; the dashed line connecting it to the model represents the array of corrected values . the length of the dashed line is a measure of the internal extinction and the point at which it crosses the model is the approximate age of the cluster . the age obtained using this method should be treated as the mean age of stars in the aperture due to the fact that a single generation model is used to interpret flux from what is potentially a mix of populations of slightly different ages . to compensate for this uncertainty , regions are classified into four age groups . these groups result from both the shape of the model and how the observations are naturally grouped with one another . this range is sufficient to establish the relative ages of regions for large scale comparisons and to identify star formation patterns . the age groups are shown in figure 7 by the alternating dotted and solid lines of the model : group 1 ( first dotted line ) = 0 @xmath11 3.5 myr , group 2 ( first solid line ) = 3.5 @xmath11 4.5 myr , group 3 ( next dotted line ) = 4.5 @xmath11 6.3 myr ; group 4 ( last solid line ) @xmath60 6.3 myr . when comparing regions , the actual age cutoffs for each group should be ignored and groups should be thought to consist of relatively very young , young , intermediate age , or older star forming regions for groups 1 @xmath11 4 , respectively . the derived age groups and internal extinctions are listed for each region in table 5 . the ages of star formation regions are plotted on the fuv image to reveal the spatial distribution of star formation as a function of age in figure 8 . the ages are identified by a symbol , representing the mean age of stars within the circular aperture of like size . two locations in the galaxy generally contain the youngest regions , the eastern side of the central arc and the linear group at the southern part of the galaxy . using figure 6 to reference the region numbers , these are region numbers 38 @xmath11 45 and 1 @xmath11 9 , respectively . however , the young regions are not restricted to these two areas . the youngest derived age , @xmath61 myr , is located on the western side of the galaxy ( region 24 ) adjacent to the oldest group of regions . being a strong , compact h@xmath0 source and lacking fuv emission , the young age of this region is not surprising while its location seems somewhat peculiar . the older regions in the galaxy are mostly the larger , diffuse regions in areas of less concentrated star formation ( regions 30 , 35 @xmath11 37 ) . the oldest region , region 37 , has a derived age of @xmath62 myr . it is most likely much older than this , but difficulties in measuring the h@xmath0 flux at this level do not allow the limit to be constrained further . the origin of diffuse fuv emission found in a galaxy is sometimes hard to determine . possible sources include scattering from dust grains and light from a spatially unresolved population of hot stars . in a low metallicity galaxy with low global extinction , scattering of uv light is not problematic and allows direct imaging of high - mass star formation . therefore , the diffuse light in the larger apertures is most likely a spatially dispersed population of b stars formed over the past few 100 myr . the derived internal extinctions , a@xmath12 , in table 5 show several trends when compared to their respective age groups . age groups 1 @xmath11 3 show a full range in extinction , a@xmath63 while regions in group 4 have substantially lower extinctions , generally a@xmath64 . extinction estimates for individual star forming regions derived from h@xmath65/h@xmath66 line ratios for dwarf galaxies , including 4 regions in hoii , also fall in this range ( hunter & gallagher 1985 ) . extinction proves to be quite patchy when comparing the spatial distribution of extinction levels . this distribution can be reconciled with the shell - like hi morphology of the galaxy ( discussed in 7 ) . the highest extinction levels are found in the dense hi ridges and the lowest levels in the hi voids . there is often not a smooth transition between extinction levels of adjacent regions . the distribution of internal extinction levels in hoii illustrates the problematic nature of using a global correction in irregular galaxies , especially those with hi shell structures . to further quantify the fuv fluxes in individual star forming regions , the number of equivalent early - o type main sequence stars is derived from the internal extinction corrected fuv flux of each region and listed in table 5 . this exercise allows a comparison between the relative stellar content of regions to the first order . the derivation is based on the m@xmath12 for normal stars in the uit bandpasses from the fanelli _ et al . _ ( 1992 ) stellar library . to derive the equivalent number of stars per a given region , the expected flux of a single o5 v star is scaled to the galaxy s distance and compared to the corrected fuv flux . a single o5 v star is m@xmath12(b1)=@xmath67 ; its unextincted magnitude at the distance of hoii is m@xmath12(b1)=17.2 . an equivalent number of 172 early - o type stars is estimated for the most energetic complex in hoii , region 9 . this number is comparable to , but slightly less than , the number generally accepted to be present in the central ( 1@xmath68 ) 30 dor complex in the lmc , @xmath32 230 ( vacca _ et al . _ 1995 ) . using fuv data , et al . _ ( 1997 ) estimate that the brightest regions in sbc spiral m51 , rand 308 and 78 , have an equivalent number of 245 and 242 early - o type stars , respectively . holmberg ii is dominated by one large complex which is able to produce a prodigious amount of star formation comparable to that in much larger galaxies . even the large , fuv diffuse regions have a significant equivalent number of early - o type stars present . for example , the diffuse fuv flux in region 35 is equivalent to 49 o5 v stars . massive stars dump energy into their surrounding ism via ionizing photons , supernovae explosions , stellar winds , and outflows . combined , these processes have a profound affect on the surrounding region of gas . the ionizing photon bath alone , provided by uv photons from massive stars , can deposit prodigious amounts of kinetic energy into the ism . the `` champagne '' model ( tenorio - tagle 1979 ) , summarized by hunter ( 1992 ) , describes the process by which massive stars ionize their surrounding gas creating compact hii regions that evolve into extended objects . as the ionization front reaches the edge of the parental molecular cloud , ionization of the neutral medium commences , where by the stars lie in a half - open cavity in the dense cloud and the hii region emerges as a `` blister '' on the side of the cloud . numerous observationally - based studies support both the existence of champagne flows and the premise that expanding ionized structures are caused by localized energy sources ( lafon _ et al . _ 1983 ; heydari - malayeri & testor 1981 ) . a tedious approach to identifying massive stars inside ionized structures can be made through spectroscopy . optical observations of massive stellar populations , the less massive remnants of the population actually responsible for the expanding features , have been employed to compare the energy available from the population with the ionization and formation requirements of the expanding ionized structures ( hunter _ et al . in contrast , parker _ et al . _ ( 1998 ) use uit observations of the lmc and smc to identify the hottest ob stellar members , those responsible for the majority of the ionizing flux , and correlate the fuv photometry of these stars with the h@xmath0 flux of the associated hii regions . several examples illustrating the relationship between massive stars and ionized structures in various evolutionary stages are found in hoii . figure 9 shows registered h@xmath0 and fuv images of region 6 . the shell in the center of the h@xmath0 image encircles a group of massive stars present in the fuv image providing evidence of a causal relationship . the emission on the edges of the h@xmath0 shell itself is not symmetrical ; the emission on the left side of the shell is more prominent than on the right , consistent with the champagne model . region 9 presents similar evidence in figure 10 where a more complex system of shells is present , suggesting that a less spatially concentrated group of massive stars is responsible for the ionization . the fuv image identifies the distribution of these stars , a @xmath69 pc chain decreasing in intensity from the lower right to the upper left . the fuv luminosity of this region indicates the presence of an equivalent number of 172 o5 v stars ; an average age of @xmath70 myr is estimated for the entire complex . as illustrated by elmegreen & lada ( 1977 ) , the high pressures generated by stellar winds , ionized gas , and supernovae explosions can act as a triggering mechanism to convert stable configurations of gas to unstable ones . the ionization front emerging from a star formation complex is followed by a shock front generated by the supersonic flow of the ionized material . if formed on the edge of a molecular cloud , the successive passage of ionization and shock fronts into a neighboring cloud allows unstable configurations of dense neutral material to accumulate to form new stars . these stars generate new ionization fronts and the cycle continues ( blaauw 1991 ) . massive star formation has the ability to influence the formation of new sites of star formation creating linear sequences of stellar subgroups within ob associations , potentially allowing self - propagation throughout a region . the cumulative chain reactions initiated by these local events have been suggested as an overall mechanism for star formation in irregular galaxies . first proposed by gerola & seiden ( 1978 ) as a global mechanism for star formation , the stochastic self - propagating star formation model ( sspsf ) describes the propagation of star formation between neighboring regions by means of a `` percolation '' process . et al . _ ( 1980 ) adapt the sspsf to small galaxies with no differential rotation and find the model predicts characteristics which correlate nicely with those of dwarf galaxies . in particular , the model produces galaxies that form stars in a bursting mode of short - lived episodes and a sfr that varies drastically with time . while the average sfr s are low , the galaxies have instantaneous sfr s that range from bursting to quiescent , which could explain the diverse star formation characteristics observed in dwarf galaxies . however , a recent survey of 110 virgo dwarf irregulars indicates that star formation is generally preferential to the edges , mostly to one side of the galaxy , while the sspsf model does not predict patterns that predominately favor any particular location ( brosch , heller , & almoznino 1998 ) . sequential star formation has been documented observationally in galaxies of similar type as the present sample ( lortet & testor 1988 ; dopita _ et al . _ 1985 ) . on local scales ( within the same molecular cloud ) , sequential star formation can be inferred by observations of two spatially adjacent regions with slight age differences . comparisons of h@xmath0 and fuv emission morphologies can indicate the relative ages of adjacent young star formation sites due to the longer timescale of decay of the fuv emission in a given region . the relationship between the h@xmath0 and fuv flux as a function of time for a given cluster is shown in figure 11 illustrating the physical basis for this qualitative treatment of relative cluster ages . a relatively strong fuv source surrounded by compact hii regions is suggestive of sequential star formation using this method . region 23 , shown in figure 12 , exemplifies this scenario . the region , having an average age of @xmath71 myr , consists of two compact hii regions surrounding a relatively strong fuv source . the group of regions on the western side of the central star forming arc ( regions 15 - 22 ) , shown in figure 13 , illustrates the geometry of sequential star formation as suggested by the sspsf model . age estimates of the individual regions show a mix of age groups 2@xmath114 . a chain of young star formation seen on the h@xmath0 image is located to the right of a faint chain of fuv emission suggesting star formation is propagating from left to right . massive star formation can influence new star formation activity on larger scales as high pressures from supernovae push gas away from aging ob associations . in theory , when a star forming complex has sufficient numbers of main sequence o and b stars , the combined pressures from stellar winds and supernovae are able to drive an expanding ionized shell of gas and at the same time , create a void in the neutral gas ( tenorio - tagle & bodenheimer 1988 ) . sequential star formation typically acts on only one side of a molecular cloud where as large - scale triggering describes the process by which a centralized source destroys the remainder of its parental molecular cloud , creating a void in the gas and a dense ridge of swept up material carried to large distances ( weaver _ et al . _ 1977 ) . described by elmegreen ( 1992 ) as the `` collect and collapse '' method , dynamical instability of the accumulated gas collapses under self gravity to form new stars along the periphery of the shell . this scenario is a plausible explanation for the correlation between stellar age and proximity to the hi shell boundary in dwarf galaxy sextans a , where the youngest stars are primarily found in the interior of the hi shell ( van dyk , puche , & wong 1998 ) . the ridge surrounding a swept - up shell may fragment leaving chains or arcs of new star formation , allowing a means for star formation to propagate on large scales . efremov & elmegreen ( 1998 ) recently demonstrated that two arcs of clusters inside and on the rim of the superbubble lmc4 in the lmc , which surrounds a roughly 30 myr - old dispersed group of supergiants and cepheid variables , may have been formed by the self - gravitational collapse of gas in swept - up pieces of rings or shells . on smaller scales , feedback from massive star formation is also evident in hi data , such as the turbulent and fractal nature of the local ism shown by the hi mosaic of the lmc ( kim _ et al . _ 1998 ) . puche _ et al . _ ( 1992 ) point out that the observed radial expansions , energy requirements and kinematic ages of the holes in hoii indicate that they originate from internal , pressure driven events . radio continuum observations of hoii confirm that supernovae do lie within several hii regions ( tongue & westpfahl 1995 ) . further analysis indicates that the supernovae energy injection rate into the galaxy is sufficient to account for the global number of hi features . on the other hand , numerical simulations using hydrodynamic methods to simulate powerful local energy sources have failed to reproduce all the features in the neutral medium of hoii ( maschenko & silich 1995 ) . some of the hi holes in hoii , especially those located beyond the optical disk of the galaxy , are likely formed by alternate mechanisms such as gravitational instability via collisional dissipation of gas particles ( byrd & howard 1992 ) or infall of gas clouds ( van der hulst & sancisi 1988 ) . et al . _ ( 1999 ) identify a single high velocity cloud candidate in their hi data . however , the distribution of holes is too regular and present at all radii to be explained entirely by infall . other environmental factors , such as tidal perturbations , are not a factor due to the fact that hoii is fairly isolated . the closest neighbor to hoii is m81 dwarfa ( kar 52 ) . given that the kinematic hi mass of hoii is a factor of 100 greater than that of m81 dwarfa ( melisse & israel 1994 ) , it is unlikely to have affected hoii . moreover , m81 dwarfa is at a distance on the sky of roughly 27 kpc from hoii with a relative radial velocity of 44 km / s . the current tidal perturbation of hoii due to this neighbor will be very tiny compared to the internal gravitation . using the values of the known components of position and velocity along with the masses , a positive two body orbital energy is obtained . the calculated value is less than the actual positive value if the unknown components of position and velocity were included . thus , analysis of the orbital energy indicates that the companion is likely to be in a hyperbolic orbit . there is not a high probability of a close encounter , especially a prolonged one . the hyperbolic relative velocity with hoii indicates that an interaction might have happened @xmath32600 myr ago , roughly three rotation periods within 1.8 kpc ( the majority of the optical portion of the galaxy ) . this timescale is sufficient for the effects of a weak interaction to have dissipated via differential rotation . rhode _ et al . _ ( 1999 ) search for the lower - mass remnants of the population which in theory created the hi holes in several dwarf galaxies , including hoii . they estimate the expected number of a and f main sequence stars from the kinematic ages of the holes and the energy required for their creation . they do not find sufficient numbers of these stars in the hi holes in deep optical searches to support the stellar wind / supernovae scenario , although the expected number is dependant upon the choice of imf . in addition , massey _ et al . _ ( 1996 ) find that the uv - bright stars in m33 are evolved b supergiants indicating that the optically brightest stars are not necessarily the ionizing sources . since fuv observations isolate the massive stellar component on timescales similar to the kinematical ages of hi holes , they can provide an observational `` smoking gun '' to the theoretical concept of pressure driven shells . instead of searching for a requisite number of lower - mass neighbors , a search for the progenitor stars themselves can be made . in addition , fuv observations can show sites of secondary star formation in the dense ridges , allowing insight into the possibility that star formation triggering via giant hi shells is a viable mechanism for star formation . fuv observations have the potential to provide a snapshot of a full cycle of propagating star formation . the numerous voids in the hi morphology of hoii are evident on the 21 cm map presented in figure 14 ; the fuv surface brightness contours are overplotted on the image . the large spatial extent of the hi disk is also apparent , extending well beyond both the fuv and optical disks . the hi holes are distributed evenly across the entire hi distribution including regions extending beyond the stellar disk . the largest hole is a @xmath32 1.7 kpc diameter void with a dense ridge in the southern part of the galaxy . since the expansion velocities of the hi holes were likely greater in the past , estimates of the kinematic ages using present day expansion rates yield upper limits . _ et al . _ ( 1992 ) derive ages for the hi holes ranging from 26 @xmath11 136 myr assuming a constant expansion rate . given these ages , some measure of fuv emission should be spatially coincident with the voids if they are caused by massive star formation . even if only the most massive stars are responsible for the generation of the hi holes , their less massive neighbors ( early b - type stars ) would still be detectable on the fuv images . however , the existence of hi holes beyond the stellar disk is not consistent with the supernovae / stellar wind scenario for their creation unless a very unusual imf is present in the outer parts of the galaxy . it is likely that an alternate mechanism is responsible . however , this mechanism is not a viable feedback mechanism for new star formation due to the absence of dense rims and corresponding secondary sites on the periphery of these holes . the lack of fuv emission in the centers of hi holes in the outer parts of hoii does not disprove the generation of large shells by massive stars since several hi voids in the inner part of the gas distribution are coincident with fuv emission . in addition , density enhancements along the rims of hi shells correspond to sites of fuv emission suggestive of propagating star formation . the largest hi hole illustrates the connection between two generations of star formation . the fuv - bright central star forming arc is aligned with the northern rim of this hi hole , while diffuse patches of fuv emission lie within its boundary . other examples of fuv emission coincident with hi voids are the eastern two patches in the northern part of the galaxy . several of the second generation sites also show they are beginning to disrupt the surrounding gas . for example , the intense star formation on the eastern edge of the central arc has hi density enhancements located on either side . tongue & westpfahl ( 1995 ) confirm the existence of supernovae inside star forming regions in this area ( regions 1 , 4 , and 6 ) using radio continuum observations . these supernovae are likely responsible in part for reshaping the surrounding neutral medium . photodisassociation of h@xmath72 on the interior of surfaces of the molecular cloud may also be responsible for the hi density enhancements surrounding the intense young star forming regions to a certain degree . however , tacconi & young ( 1987 ) do not detect any @xmath73co emission above the 3@xmath74 level in hoii . lacking other observations to trace the molecular component , its morphology remains a mystery . in general , evolved star forming knots are fairly efficient in converting in converting h@xmath75 to hi ( allen _ et al . _ 1987 ) . the derived ages for individual star forming regions further illustrate the connection between massive star formation and the hi morphology . the age groups for individual star formation regions are displayed in figure 15 on the hi map . each of the seven regions in the oldest age group ( group 4 , @xmath76 myr ) , lies in a hi hole or local minimum in the hi density . the largest hole contains three large regions in age group 4 on the western side of its interior . the youngest regions ( group 1 , 2.5 @xmath11 3.5 myr ) are each spatially coincident with hi density enhancements suggesting the potential for new star formation is high on the dense rims of the hi holes . the majority of new star formation in the whole system is occurring on the periphery of the largest hole . the distribution of star formation ages in general can be interpreted by this comparison . the spatial proximity of a very old region and a very young one is reconciled by the hi density distribution . a quantitative approach to test the observational evidence suggesting that massive star formation triggers holes in the neutral medium is offered by comparing the energy delivered to the ism by massive stars and the required creation energy of the hi holes . fuv observations of the clusters interior to the hi holes provide an estimate of the total mechanical energy delivered to the ism , assuming the derived ages and internal extinctions . several of these clusters are spatially coincident with hi holes included in the puche _ et al . _ ( 1992 ) study where estimates for their creation energy , e@xmath77 , are given . these hole / cluster combinations are listed in table 6 . region numbers refer to the previously defined star formation regions and hole numbers are taken from puche _ et al . _ ( 1992 ) . for reference , hi images showing each hole used in this comparison are presented with the fuv contours in figures 16 - 18 . the observed fuv luminosity , l@xmath12 , for each cluster is estimated from the fuv flux and internal extinction correction given in table 5 . cluster mass is estimated by comparing the observed l@xmath12 with the model of l@xmath12 ( ergs s@xmath21 m@xmath15@xmath21 ) for an evolving cluster at the given cluster age . this model is used to derive the ages and internal extinctions for all star formation regions and described previously ( 5.2 ) . the mechanical energy deposited into the ism , l@xmath78 , over a timescale equal to the age of the cluster can be calculated assuming a rate which is a function of cluster mass . evolutionary synthesis models of populations of massive stars given by leitherer & heckman ( 1995 ) provide estimates of the deposition rates from stellar winds and supernovae for an instantaneous burst as a function of time and mass . from clusters of similar imf and metallicity , an average rate of l@xmath79 ergs ( s m@xmath15)@xmath21 is assumed . the net mechanical energy deposited into the ism by supernovae and stellar winds , e@xmath80 , is estimated assuming this rate , the total cluster mass , and the derived cluster age . lccccccc reg & hole & age & l@xmath12 & mass & e@xmath77 & e@xmath80 + # & # & myr & 10@xmath81 ergs s@xmath21&@xmath82m@xmath15 & 10@xmath83ergs & 10@xmath83ergs + 28 & 48 & 8.5 & 7.4 & 7.57 & 470.3 & 608.8 + 13 & 23 & 8.5 & 9.3 & 9.53 & 41.1 & 766.4 + 36,37&21 & 10.0&23.6 & 32.10&2025 & 3036.9 + the derived properties of the clusters spatially coincident to hi holes are given in table 6 . in each case , e@xmath80 @xmath84 e@xmath77 indicating that to the first order , the energy provided by massive stars is sufficient to create the hi holes that encircle them . the largest hole ( # 21 ) actually contains two star forming regions which are added together to estimate the energy input . all the star forming regions interior to hi holes are in the oldest age group ( group 4 , @xmath84 6.3 myr ) . the derived ages in table 5 for these regions are upper limits in the sense that these regions are likely much older . ( the technique to derive the ages depends upon the presence of h@xmath0 emission and is not reliable for absolute ages @xmath846.3 myr . ) for these regions , e@xmath80 is a lower limit and would increase linearly with the true cluster age . region 13 has significantly more energy than the creation energy of its associated hi hole . it is possible that this region is also responsible for an adjacent hole ( # 22 ) , which would increase e@xmath77 by 35 @xmath85 ergs . these comparisons suggest that massive stars do provide sufficient mechanical energy to account for the observed properties of associated hi holes in hoii . the question remains : if massive star formation can provide sufficient energy to generate holes in the surrounding neutral medium , why are there not hi holes around every ob association ? the answer may be related to global properties of the ism . characteristic differences in the kinematical properties of the ism can be traced to differences in the star formation mechanisms . solid body rotation of the gas disk is a general property of dwarf galaxies which makes them appealing for star formation studies . the lack of shear in the absence of differential rotation allows features in the hi gas to be longer - lived . the thicker gas disks and lack of shear in irregulars allow the growth of larger gas clouds ( hunter 1997 ) . this combination provides an excellent environment for stars to form in giant complexes . since larger complexes impact the surrounding ism to a greater degree and hi features survive longer in regions of reduced shear , hi holes and associated sites of secondary star formation are more likely to be found in areas of the gas disk exhibiting solid body rotation . the rotation curve of hoii , taken from puche _ et al . _ ( 1992 ) and displayed in figure 19 , exhibits solid body rotation to a radius of r@xmath86 after which it flattens and exhibits differential rotation . an ellipse with a major axis of 2 , reduced to the plane of hoii , is plotted on the fuv image of the galaxy in figure 20 . the majority of the fuv emission is contained within the ellipse . this is also the region where examples of large - scale star formation triggering via hi shells are found . this simple illustration demonstrates that massive star formation in hoii can operate more easily as a feedback mechanism in regions with uncomplicated global kinematical properties . this study uses fuv , hi , and h@xmath0 observations of dwarf irregular galaxy hoii to trace the interaction between sites of massive star formation and the neutral and ionized components of the surrounding ism and to identify the means by which star formation propagates in this intrinsically simple system . both local and large - scale triggering mechanisms related to massive stars are identified suggesting that feedback from massive stars is a microscopic process operating in all galaxies to a certain degree . \(1 ) in comparison to its optical images , the fuv morphology of hoii is patchy and irregular . massive star formation does not have a smooth spatial distribution ; it is concentrated in several dominate complexes . the fact that massive star formation is not centrally organized has important implications for interpreting images of high redshift galaxies ( z @xmath14 3 ) where the restframe fuv is detected using optical and ir instruments . \(2 ) surface brightness profiles are used to characterize the radial dependence of the azimuthally averaged star formation activity . the fuv profile is generally flat with inflections due to prominent complexes . the b - band profile displays an exponential shape with a central light depression , characteristic of dwarf galaxies . the derived central surface brightness , @xmath49(0 ) , lies in the expected range for dwarf galaxies . comparisons of profiles in the r , b , u and fuv bands indicate that profiles retain the same general shape while inflections due to star formation increase as the wavelength of the profile decreases . the radial continuity of star formation sites demonstrates that the time averaged level of activity in the past correlates with the current activity at the same radius , supporting the theory that star formation in irregular galaxies is dominated by local processes . \(3 ) star formation rates derived from h@xmath0 and fuv photometry are used to characterize the current ( traced by h@xmath0 ) and recent ( traced by fuv ) global star formation activity . comparisons of the two rates allow a rough estimate of the recent star formation history and indicate that the galaxy is sustaining a somewhat recent star formation event . \(4 ) ages for individual regions are derived using b , h@xmath0 , and fuv photometry and show both older , diffuse fuv regions and younger , compact hii regions . the distribution of ages is reconciled with the hi morphology , showing a clear preference of young regions for areas of dense hi and older regions for hi voids . \(5 ) local star formation triggering mechanisms are identified via the relative h@xmath0 and fuv morphologies . observational evidence of the influence that massive stars have on their surroundings includes massive stars inside ionized shells and compact hii regions surrounding aging clusters . causal relationships between massive star formation and secondary sites of star formation are shown . star formation progression patterns are consistent with the sspsf model in that progressively older chains of star formation are seen in clusters . \(6 ) large - scale star formation triggering by hi shells generated by massive stars is seen by comparing the fuv and hi morphologies . given the timescale over which fuv observations trace massive star formation , both secondary sites of star formation and progenitor populations are identified . analysis of the energy available from massive stars inside hi voids indicates that energy deposited into the ism from supernovae and stellar winds are sufficient to account for the observed properties of the hi holes . \(7 ) global kinematical properties may also play a role in the star formation process since differences in the rotation characteristics of the neutral gas disk are coupled with differences in triggering mechanisms . large - scale feedback from massive star formation is shown to operate in regions that lack differential shear in the gas disk . s.g.s . wishes to acknowledge the funding of this study through the alabama space grant consortium project and to thank the university of alabama , nasa / goddard space flight center , and the u. s. naval observatory for their support . funding for the uit project has been through the spacelab office at nasa headquarters under project number 440 - 51 .

we revise and extend the stochastic approach to cumulative weak lensing ( hereafter the sgl method ) first introduced in ref . . here we include a realistic halo mass function and density profiles to model the distribution of mass between and within galaxies , galaxy groups and galaxy clusters . we also introduce a modeling of the filamentary large - scale structures and a method to embed halos into these structures . we show that the sgl method naturally reproduces the weak lensing results for the millennium simulation . the strength of the sgl method is that a numerical code based on it can compute the lensing probability distribution function for a given inhomogeneous model universe in a few seconds . this makes it a useful tool to study how lensing depends on cosmological parameters and its impact on observations . the method can also be used to simulate the effect of a wide array of systematic biases on the observable pdf . as an example we show how simple selection effects may reduce the variance of observed pdf , which could possibly mask opposite effects from very large scale structures . we also show how a jdem - like survey could constrain the lensing pdf relative to a given cosmological model . the updated code is available at .

1011.0732

inhomogeneities in the large - scale matter distribution can in many ways affect the light signals coming from very distant objects . these effects need to be understood well if we want to map the expansion history and determine the composition of the universe to a high precision from cosmolgical observations . in particular the evidence for dark energy in the current cosmological concordance model is heavily based on the analysis of the apparent magnitudes of distant type ia supernovae ( sne ) @xcite . inhomogeneities can affect the observed sne magnitude - redshift relation for example through gravitational lensing , in a way which essentially depends on size and composition of the structures through which light passes on its way from source to observer . the fundamental quantity describing this statistical magnification is the lensing probability distribution function ( pdf ) . it is not currently possible to extract the lensing pdf from the observational data and we have to resort to theoretical models . two possible alternatives have been followed in the literature . a first approach ( e.g. ref . @xcite ) relates a `` universal '' form of the lensing pdf to the variance of the convergence , which in turn is fixed by the amplitude of the power spectrum , @xmath0 . moreover the coefficients of the proposed pdf are trained on some specific n - body simulations . a second approach ( e.g. ref . @xcite ) is to build _ ab - initio _ a model for the inhomogeneous universe and directly compute the relative lensing pdf , usually through time - consuming ray - tracing techniques . the flexibility of this method is therefore penalized by the increased computational time . in ref . @xcite we introduced a stochastic approach to cumulative weak lensing ( hereafter sgl method ) which combines the flexibility in modeling with a fast performance in obtaining the lensing pdf . the speed gain is actually a sine - qua - non for likelihood approaches , in which one needs to scan many thousands different models ( see ref . the sgl method is based on the weak lensing approximation and generating stochastic configurations of inhomogeneities along the line of sight . the major improvements introduced here are the use of a realistic halo mass function to determine the halo mass spectrum and the incorporation of large - scale structures in the form of filaments . the improved modeling together with the flexibility to include a wide array of systematic biases and selection effects makes the sgl method a powerful and comprehensive tool to study the impact of lensing on observations . we show in particular that the sgl method , endowed with the new array of inhomogeneities , naturally and accurately reproduces the lensing pdf of the millenium simulation @xcite . we also study a simple selection effect model and show that selection biases can reduce the variance of the observable pdf . such reduction could at least partly cancel the opposite effect coming from large scale inhomogeneities , masking their effect on the observable pdf . we also show how a jdem - like survey could constrain the lensing pdf relative to a given cosmological model . along with this paper , we release an updated version of the package , which is a simple and very fast mathematica implementation of the sgl method @xcite . this paper is organized as follows . in section [ sup ] we introduce the cosmological background , the generic layout of inhomogeneities and review the basic formalism needed to compute the weak lensing convergence . in section [ sec : inhoprop ] we derive the halo mass function and the halo density profiles and define the precise modeling of filaments . in section [ hmf ] we present the revised and extended sgl method . the exact discretization of the model parameters , which is a crucial step in the sgl model building , is explained in section [ mbin ] and in section [ confi ] we explain how the realistic structures where halos are confined in filaments are modelled in the sgl method . section [ results ] contains our numerical results including the comparison with the cosmology of the millennium simulation @xcite and , finally , in section [ conco ] we will give our conclusions . we consider homogeneous and isotropic friedmann - lematre - robertson - walker ( flrw ) background solutions to einstein s equations , whose metric can be written as : @xmath1 \ , , \ ] ] where @xmath2 and @xmath3 & \ ; k<0 \\ \end{array}\right.\ ; , \ ] ] where @xmath4 is the spatial curvature of any @xmath5slice . in particular we will focus on @xmath6cdm models whose hubble expansion rate depends on redshift according to : @xmath7 where @xmath8 is given by : @xmath9 here @xmath10 could be taken to follow _ e.g. _ the parameterization @xcite : @xmath11 for which : @xmath12 for a constant equation of state @xmath13 , the latter reduces to @xmath14 . @xmath15 , @xmath16 and @xmath17 are the present - day density parameters of dark energy , matter and radiation and @xmath18 represents the spatial - curvature contribution to the friedmann equation . introducing the hubble radius @xmath19 and the spatial - curvature radius @xmath20 , we have at any time the relation @xmath21 . we will also need the matter and dark energy density parameters at a given time or redshift : @xmath22 throughout this paper , the subscript @xmath23 will denote the present - day values of the quantities . for example the critical density today is @xmath24 where @xmath25 km s@xmath26 mpc@xmath26 , while @xmath27 is the critical density at any time . substituting in eq . ( [ hhh ] ) @xmath28 and @xmath29 we obtain the equation we have to solve in order to find the time evolution of the scale factor @xmath30 , the only dynamical variable in an flrw model . we fix the radiation density parameter to @xmath31 . moreover @xmath32 , that is , dark and baryonic matter contribute together to the total matter density . the line - of - sight and transverse comoving distances are : @xmath33 from which we find the angular and luminosity distances together with the distance modulus : @xmath34 for scales smaller than the spatial - curvature radius @xmath35 , @xmath36 and we can use the euclidean geometry . the aim of this paper is to compute statistical weak lensing corrections to the measured light intensities , generated by the inhomogeneous matter distribution along the line of sight to a distant object . the basic quantity we need to model is the matter density contrast @xmath37 : @xmath38 where the lowercase @xmath39 indicates the local and inhomogeneous matter field while @xmath40 is the time dependent average mass density . the density contrast directly enters the expression for the weak lensing convergence @xcite : @xmath41 where @xmath42 is the co - moving position of the source and the integral is along an unperturbed light geodesic . the density @xmath43 is the constant matter density in a co - moving volume and we defined the auxiliary function @xmath44 which gives the optical weight of an inhomogeneity at the comoving radius @xmath45 . the convergence is related to the shift in the distance modulus by : @xmath46 where @xmath47 is the net magnification and the second - order contribution of the shear has been neglected @xcite . it is obvious that an accurate statistical modeling of the magnification pdf calls for a detailed description of the inhomogeneous mass distribution . in this paper we will significantly improve the modeling of the inhomogeneities from our previous work @xcite , where only single - mass spherical overdensities were considered . first of all , we improve the modeling of these halos " by using a realistic halo mass function @xmath48 , which gives the fraction of the total mass in halos of mass @xmath49 at the redshift @xmath50 . the function @xmath48 is related to the ( comoving ) number density @xmath51 by : @xmath52 where we defined @xmath53 as the number density of halos in the mass range @xmath54 . the halo function is by definition normalized to unity @xmath55 the idea is of course that halos describe large virialized mass concentrations such as large galaxies , galaxy clusters and superclusters . not all matter is confined into virialized halos however . moreover , only very large mass concentrations play a significant role in our weak lensing analysis : for example the stellar mass in galaxies affects the lensing pdf only at very large magnifications @xcite where the pdf is close to zero . therefore the fraction of mass @xmath56 concentrated in large virialized halos can be defined by introducing a lower limit to the integral ( [ eq : fmnorm ] ) : @xmath57 only mass concentrations with @xmath58 are treated as halos . the remaining mass is divided into a family of large mass , but low density contrast objects with a fraction @xmath59 and a uniform component with a fraction @xmath60 , such that @xmath61 the low contrast objects are introduced to account for the filamentary structures observed in the large scale structures of the universe . in our analysis they will be modeled by elongated objects with random positions and orientations . the mass in these objects can consist of a smooth unvirialized ( dark ) matter field and/or of a fine `` dust '' of small virialized objects with @xmath62 . in weak lensing this distinction does not matter because small halos act effectively as a mean field , with a sizeable contribution only at very large magnifications ( see comment above about stellar mass ) . for later use we define @xmath63 which gives the total mass fraction not in large virialized halos . if we only consider virialized masses larger than companion galaxies ( @xmath64 ) , then typical values for the concordance model are @xmath65 at @xmath66@xcite and @xmath67 at @xmath68 , with a weak dependence on the particular @xmath48 used . although @xmath69 is sensitive on @xmath70 , the lensing pdf depends only weakly on the cut mass . moreover , halo functions obtained from n - body simulations are valid above a mass value imposed by the numerical resolution of the simulation itself and so the use of @xmath70 is also necessary in this case @xcite . to obtain an as accurate modeling as possible , we will use realistic mass functions and spatial density profiles for the halo distribution . there is less theoretical and observational input to constrain the mass distribution of the filamentary structures or their internal density profiles . we will therefore parametrize the filaments with reasonable assumptions for their lengths and widths and by employing cylindrical nonuniform density profiles . mpc thick slice generated by our stochastic model . the shaded disks represent clusters and the shaded cylinders the filamentary dark matter structures . only large clusters are displayed.,width=8 ] our modeling allows treating the two families of inhomogeneities independently . both can be given random spatial distributions , or alternatively all halos can be confined to have random positions in the interiors of randomly distributed cylinders . the latter configuration more closely resembles the observed large scale structures , and an illustration created by a numerical simulation using the turbogl package is shown in fig . [ slice ] . in section [ discu ] we will discuss the power spectrum and the large - scale correlations of our model universe . we can now formally rewrite eq . ( [ contra1 ] ) for our model universe in which the density distribution is given by @xmath71 : @xmath72 where the index @xmath73 labels all the inhomogeneities , we used eq . ( [ eq : sumrule ] ) and we defined @xmath74 , so that both virialized halos and the unvirialized objects are described by a generic reduced density profile @xmath75 which satisfies the normalization @xmath76 . the lensing convergence we are interested in can now be written as @xmath77 where the positive contribution due to the inhomogeneities ( halos and low contrast objects ) is @xmath78 the also positive contribution due to the uniformly distributed matter is @xmath79 and the negative empty beam convergence is : @xmath80 a light ray that misses all the inhomogeneities @xmath81 will experience a negative total convergence @xmath82 , which gives the maximum possible demagnification in a given model universe . in an exactly homogeneous flrw model the two contributions @xmath83 and @xmath84 cancel and there is no net lensing . in an inhomogeneous universe , on the other hand , a light ray encounters positive or negative density contrasts , and its intensity will be magnified or demagnified , respectively . the essence of the sgl method is finding a simple statistical expression for the probability distribution of the quantity @xmath83 in the inhomogeneous universe described above . for a discussion about the validity of the weak lensing approximation within our setup see ref . @xcite , where it was shown that the error introduced is @xmath85% . we stress again that the lensing caused by stellar mass in galaxies is negligible in the weak lensing regime @xcite and so we can focus on just modeling the dark matter distribution in the universe . also , we will treat the inhomogeneities as perturbations over the background metric of eq . ( [ metric ] ) . in particular we will assume that redshifts can be related to comoving distances through the latter metric . see ref . @xcite for a discussion of redshift effects . in the next section we will give the accurate modeling of the inhomogeneities . we begin by introducing the halo mass function and the detailed halo profiles , and then move on to describe the precise modeling of the cylindrical filaments . we begin by explaining our dark matter halo modeling . the two main concepts here are the halo mass function @xmath48 giving the normalized distribution of the halos as a function of their mass and redshift , and the dark matter density profile within each individual halo . both quantities are essential for an accurate modeling of the weak lensing effects by inhomogeneities . we shall begin with the halo mass function introduced above in eq . ( [ eq : halomassf ] ) . the halo mass function acquires an approximate universality when expressed with respect to the variance of the mass fluctuations on a comoving scale @xmath45 at a given time or redshift , @xmath86 . relating the comoving scale @xmath45 to the mass scale by @xmath87 , we can define the variance in a given mass scale by @xmath88 . the variance @xmath86 can be computed from the power spectrum : @xmath89 where @xmath90 is the fourier transform of the chosen ( top - hat in this work ) window function and the dimensionless power spectrum extrapolated using linear theory to the redshift @xmath50 is : @xmath91 here @xmath92 is the spectral index and @xmath93 is the amplitude of perturbations on the horizon scale today , which we fix by requiring : @xmath94 where the value of @xmath0 is estimated by cluster abundance constraints @xcite . @xmath95 is the linear growth function which describes the growth speed of the linear perturbations in the universe . fit functions for @xmath96 could be found , _ e.g. _ , in ref . @xcite , but it can also be easily solved numerically from the equation : @xmath97 \nonumber \\ & -&{3 \over 2 } { d(z ) \over ( 1+z)^{2 } } \ , \omega_{m}(z)=0 \,,\end{aligned}\ ] ] where we have neglected the radiation . as usual , we normalize @xmath96 to unity at the present time , @xmath98 . finally , for the transfer function @xmath99 we use the fit provided by the equations ( 28 - 31 ) of ref . @xcite , which accurately reproduces the baryon - induced suppression on the intermediate scales but ignores the acoustic oscillations , which are not relevant for us here . with @xmath100 given , we can now define our halo mass function . several different mass functions have been introduced in the literature , but here we will consider the mass function given in eq . ( b3 ) of ref . @xcite : @xmath101 which is valid in the range @xmath102 . because of the change of variable , @xmath103 is related to our original definition of @xmath104 by : @xmath105 the mass function of eq . ( [ jenk ] ) is defined relative to a spherical - overdensity ( so ) halo finder , and the overdensity used to identify a halo of mass @xmath49 at redshift @xmath50 is @xmath106 with respect to the mean matter density @xmath107 . the so finder allows therefore a direct relation between halo mass @xmath49 and the radius @xmath108 : @xmath109 the subscript @xmath110 will denote the proper values of otherwise comoving quantities throughout this paper . for example , the comoving halo radius @xmath111 is related to the proper value by @xmath112 . a direct relation between the mass and the radius of a cluster is necessary for our sgl method . this is why we prefer mass functions based on so finders over mass functions based on friends - of - friends halo finders ; for the latter an appropriate @xmath113 to be used in eq . ( [ mdel ] ) is not directly available . moreover , as shown in ref . @xcite , the so(180 ) halo finder gives a good degree of universality to the mass function . another mass function which manifests approximate universality with the so(180 ) halo finder @xcite is provided by sheth & tormen in ref . @xcite . with the halo mass function given the only missing ingredient is the halo profile which , as said after eq . ( [ contra2 ] ) , we describe by means of the reduced halo profile @xmath114 . we stress that our halo of a given mass @xmath49 and redshift @xmath50 is an avererage representative of the total ensemble of halos , which in reality have some scatter in the density profiles . we will focus on the navarro - frenk - white ( nfw ) profile @xcite : @xmath115 here the scale radius is related to the halo radius by @xmath116 , where @xmath117 is the concentration parameter to be defined below . by integrating equation ( [ eq : nfwprofile ] ) one finds the total mass @xmath118 $ ] which , when combined with eq . ( [ mdel ] ) , gives : @xmath119 this relation fixes @xmath120 once the concentration parameter is given . the reduced halo profile @xmath114 in comoving coordinates is then : @xmath121^{-1 } , \ ] ] where the @xmath122-dependence ( as well as @xmath49-dependence ) arises through the redshift dependence of @xmath111 ( eq . ( [ mdel ] ) ) and @xmath117 ( eq . ( [ duffy ] ) ) . the last ingredient needed to fully specify the nfw profile is a relation between the concentration parameter and the halo mass at a given redshift . for the @xmath123cdm model we will use the following fit obtained from numerical simulations @xcite satisfying wmap5 cosmology @xcite : @xmath124 where the pivot mass scale is @xmath125 . apart from the nfw profile , alternative simple density profiles of interest include : @xmath126 \over ( 2 \pi)^{3/2 } ( r/3)^{3 } } \phantom{ggg } \textrm{gaussian , } \label{gau } \\ \varphi_{h } & = & { 1 \over 4 \pi r^{2 } r } \phantom{ggggggggggggggggggi } \textrm{sis , } \label{sis}\end{aligned}\ ] ] where @xmath111 is again given by eq . ( [ mdel ] ) . the normalization factor in ( [ gau ] ) is necessary to have the correct volume integral of unity and is due to the restriction of the profile between 0 and @xmath111 . in these cases there are no further parameters to be fixed . we will model dark matter filaments by cylindrical , non - uniform low density objects . let us stress that throughout this chapter by a filament " we refer only to the smoothly distributed dark matter component of a real filament ; we shall later explain how virialized halos can be embedded within these objects to form a more realistic model of the true filamentary structures in the full sgl framework . keeping the low contrast component and the halos as separate entities in our evaluation is convenient for the modeling and explicitly avoids double counting . using eqs . ( [ eq : halomassf ] ) and ( [ eq : sumrule ] ) we can obtain the following equation which relates the comoving filament density @xmath127 and the filament mass @xmath128 to the fraction @xmath69 of mass not confined in virialized halos : @xmath129 where the quantity @xmath130 defines the fraction of the total unvirialized mass which is confined in filaments : @xmath131 . because of the ongoing halo formation @xmath69 decreases with time and we shall simply assume that @xmath132 is a constant . we will assume that all filaments have the same mass , but this assumption can easily be relaxed later by introducing a filament mass function @xcite . we are also neglecting merging of smaller filaments into larger ones , whereby the comoving number density of filaments @xmath127 remains a constant . this assumption seems to be supported by numerical simulations @xcite at least at low redshifts . from eq ( [ eq : fila ] ) it then follows that the filament mass @xmath128 is not constant , but it gradually decreases as more and more halos virialize . ( the total mass of the full filamentary structure which includes the halos remains on average constant . ) now we need to specify the precise geometry , dimensions and time - dependent density profile @xmath133 for our filaments . first , we will relate the filament mass to its length and radius at present day according to : @xmath134 where @xmath135 and @xmath136 are the present - day filament proper radius and length and @xmath137 is the average overdensity of the filament with respect to the flrw matter density . consistent with our assumptions of neglecting filament mergers and using a constant comoving filament density , we will assume that filaments have a constant comoving length . this is again in accordance with the generic picture of a web of filaments of constant comoving length condensing with time seen in numerical simulations . finally , for simplicity , we will choose our filaments to have a constant proper radius . summarizing , we will use : @xmath138 the simplest assumption would be to take filaments to have uniform density . in this case the reduced density profile is simply @xmath139 . however , we will also consider a profile which is uniform along the length of the filament @xmath140 , but which has a gaussian profile in the radial direction . in this case @xmath141 \over l \ , 2 \pi ( r/3)^{2 } } \,,\ ] ] where @xmath142 and @xmath143 , and @xmath144 otherwise . finally , let us compute the mass of the `` dressed '' filaments which comprises the `` bare '' filaments until now discussed together with the halos ( to be embedded in section [ confi ] ) . it is easy to find that : @xmath145 \,,\ ] ] where also the contribution from the uniformly distributed matter is included . ( [ dressed ] ) is valid if all halos are embedded within the filaments and has to be slightly modified if different levels of confinement are considered . this completes our description of the _ objects _ in our model . next we shall have to understand how to simulate their effect on the weak lensing properties of the complete inhomogeneous model universe . in this section we will review and extend the stochastic gravitational lensing ( sgl ) method of ref . @xcite for computing the probability distribution function of the lens convergence @xmath146 in the presence of inhomogeneities . the basic quantity to evaluate is the inhomogeneity - induced part @xmath147 in eq . ( [ eq : kappaih ] ) : @xmath148 where @xmath149 . we wish to obtain a probabilistic prediction for this quantity along a line of sight to a source located at @xmath150 , through a random distribution of halos and filaments . the problem could be solved by constructing a large comoving volume of statistically distributed objects such as the one shown in fig . [ slice ] , and then computing the distribution of @xmath83 along random directions in this space . however , an equivalent and computationally much more efficient approach is to construct random realizations of object locations along a fixed geodesic and compute the lensing pdf from a large sample of such realizations . let @xmath151 refer to a particular _ realization _ of the integral in eq . ( [ eq : kappa3b ] ) and denote the resulting convergence by @xmath152 . because of the finite size of the halos and filaments , only a finite number @xmath153 of objects intercept the geodesic and contribute to the sum . moreover , because all objects are small compared to a typical comoving distance to the source @xmath154 , the function @xmath155 is , to a good approximation , a constant @xmath156 for each individual object . similarly , one can assume that @xmath157 with @xmath158 , and so one finds : @xmath159 where @xmath160 is the surface mass density of the @xmath73-th intercepting object in the realization . next divide the geodesic into @xmath161 subintervals with centers at @xmath162 and with ( possibly variable ) length @xmath163 , such that @xmath164 holds within each interval . in this way we can write @xmath165 where @xmath166 labels all objects encountered by the geodesic within the comoving length bin @xmath167 , such that @xmath168 . at this level eq . ( [ eq : realization2 ] ) still consists of a sum over all _ objects _ along a given path . next we categorize these objects into different classes depending on the parameters that define the surface density @xmath169 . for spherical halos of a given @xmath170-profile @xmath171 depends only on the mass of the halo and the impact parameter , and for filaments the relevant parameters are the angle between the main filament axis and the geodesic and the impact parameter in the cylindrical radius . now assume all these variables are discretized into finite - length bins . the precise way of the discretization will be discussed in detail in section [ mbin ] , but for now we simply let a generic index @xmath172 label the set of independent parameter cells created by the binning . if we now let @xmath173 denote the number of objects that fall into the cell labeled by indices @xmath174 and @xmath172 in the realization @xmath151 , we can replace the sum over the individual objects in eq . ( [ eq : realization2 ] ) by a weighted sum over the discrete parameter cells : @xmath175 where @xmath176 is the total number of independent parameter cells , and so the indexing @xmath172 and the number of different parameter cells @xmath177 could in fact depend on @xmath174 , but we suppress this notation for simplicity . ] used in the binning , @xmath178 and @xmath179 is the convergence due to one object in the bin @xmath180 : @xmath181 note that the quantity @xmath179 is a function of the distance and internal variables , it is universal for arbitrary realizations : all the information specific to a particular realization @xmath151 is contained in the set of integers @xmath182 . equation ( [ eq : realization3 ] ) is the starting point of the sgl analysis , because it can be easily turned into a probabilistic quantity . indeed , instead of thinking of realizations @xmath151 along arbitrary lines of sight through a pre - created model universe , we can define _ a statistical distribution of convergences _ through eq . ( [ eq : realization3 ] ) . indeed , we have shown in ref . @xcite that , for random initial and final points of the geodesics , the integers @xmath173 are distributed as poisson random variables : @xmath183 where the parameter @xmath184 is the _ expected _ number of objects in the bin volume @xmath185 : @xmath186 here @xmath187 is the comoving density of objects corresponding to the parameters in the bin @xmath188 and @xmath189 is the corresponding cross sectional area of these objects in co - moving units . we shall specify these quantities precisely in section [ mbin ] . physically , the statistical distribution of convergences is equivalent to our original set of realizations @xmath151 averaged over the position of the observer . this is a welcome feature because the statistical model explicitly incorporates the copernican principle . the original realization @xmath151 has now been replaced by _ a configuration _ of random integers @xmath190 and the convergence equation eq . ( [ eq : realization3 ] ) by a statistical convergence : @xmath191 where the probability for a particular configuration to occur is just @xmath192 it is easy to show that this probability distribution is correctly normalized . moreover , including the convergence @xmath193 due to the uniformly distributed matter and the empty space ( see eq . ( [ kem ] ) and below ) , one can show that the total convergence corresponding to the configuration @xmath190 becomes @xcite : @xmath194 for a configuration without any inhomogeneities we correctly recover @xmath195 , see section [ mbin ] . moreover , eq . ( [ eq : fullkappa ] ) shows explicitly that the _ expected _ convergence vanishes consistently with the photon conservation in weak lensing , because @xmath196 for a random variable following eq . ( [ eq : poisson - i ] ) . the final convergence pdf can now be formally written as @xmath197 where @xmath198 is a discrete probability distribution . note that the most likely configuration which maximises @xmath199 corresponds to the mode of the poisson distribution , which is the floor of its parameter : @xmath200 . moreover , for a large @xmath184 the poisson distribution approximates a gaussian with mean and variance equal to @xmath184 and the most likely configuration therefore approaches the mean : @xmath201 . when this is the case the mode of the lensing pdf vanishes even for a single observation and the pdf tends to a gaussian . one can similarly create the pdf in the shift in the distance modulus @xmath202 or in the magnification @xmath47 , using eq . ( [ eq : mukappa ] ) to compute these quantities for a given configuration and then replacing @xmath146 by the desired quantity in equations ( [ eq : pdf2a]-[eq : pdf1 ] ) . in practice , equations ( [ eq : pdf2a]-[eq : pdf1 ] ) are not useful for a direct evaluation of the pdf . indeed , the configuration spaces are infinite and a direct evaluation of the sum of configurations would not be feasible even for relatively roughly discretized systems . instead , we compute an approximation for @xmath203 by statistically creating a large set of random configurations @xmath190 from the probability distribution @xmath199 , and by forming a discrete normalized histogram out from the corresponding set of convergences ( or distance moduli or magnifications ) . that is , we define a histogram pdf using the function : @xmath204 where @xmath205 is the total number of realizations and @xmath206 is the number of convergences in the sample falling in the bin @xmath207 centered around the mean value @xmath208 . it is easy to check that eq . ( [ eq : pdf2 ] ) is correctly normalized to unity . moreover , the more configurations one creates in the simulation step , the more fine - detailed and accurate approximation for @xmath209 can be constructed . from the simulated data as a moving average weighted by some window function , but that would not give any advantage in the analysis with respect to using the simple histogram of eq . ( [ eq : pdf2 ] ) . ] to summarize , the sgl method consists of two steps : first the internal variables specifying the inhomogeneities are binned , reducing them to a finite number of different object classes labeled by the cell index @xmath172 , for which universal convergence functions @xmath179 can be computed . second the occupation numbers @xmath210 of these objects within a given co - moving distance bin are shown to be poisson distributed with parameter @xmath184 . the lensing pdf is then computed statistically , binning the convergence distribution obtained from a large set of random configurations @xmath190 on a given geodesic . computing the lensing pdf this way is very efficient . finding the pdf for a given model universe using the turbogl package @xcite typically takes a few seconds in an ordinary desktop computer . see the appendix [ perf ] for more details about its performance . let us point out that eq . ( [ eq : pdf2 ] ) treats all objects on the same footing . that is , all object families are independently randomly distributed throughout the space . confining the halos to within filaments will require slight modifications which we will explain in the section [ confi ] . for now , we shall study the effects that may cause the actual observed pdf deviate from the fundamental weak lensing pdf . the lensing pdf ( [ eq : pdf2 ] ) is the fundamental quantity inherent to a given background model and a set of inhomogeneities . however , in reality there are interferences that prevent us from observing @xmath203 directly . firstly , the intrinsic magnitudes of the sources are not known with arbitrary accuracy , and so the _ observed _ lensing pdf is at best a convolution : @xmath211 where @xmath212 describes the source magnitude dispersion for an imperfect standard candle . clearly the fundamental lensing pdf is recovered in the limit that @xmath212 becomes a delta function . if there were no other sources of errors , the distribution @xmath213 could be used to contrast observations against different cosmological background models , sets of inhomogeneities and the models for the source magnitude distributions . however , there can be several different types of selection effects caused by matter between the source and the observer or by the search strategies , that may bias the observed pdf from the one predicted by eq . ( [ eq : obspdf1 ] ) . one of the strengths of the sgl method is that it can be easily extended to model the effects of a wide range of selection biases on the observable pdf . let us begin with a simple class of biases that might affect the observed distance modulus beyond the weak lensing correction ; an obvious example would be an error made in the estimate of reddening . what makes such correction nontrivial , is that one would expect the error to be _ correlated _ with the types of mass concentrations encountered by the photon beam on its way from the source to the observer . the effects should thus be different along different lines of sight , affecting mostly those directions where light has to travel through volumes with higher mass densities . in the sgl method such correlated biases are very easily modelled , as one needs merely to replace @xmath214 where the first part @xmath215 is the usual weak lensing correction and the second part the object - specific correction . indeed , the bias function @xmath216 can be correlated with the redshifts @xmath217 and @xmath218 , and with all the internal characteristics of the objects , such as the halo and filament masses and impact parameters . a quantitative analysis of the possible magnitude of the function @xmath219 would require detailed astrophysical information and it is beyond the scope of this paper . however , once properly modelled , the sgl allows a very easy way of evaluating the outcome of biases on the observations . other types of _ selection _ biases might not directly affect the magnitudes , but rather the relative likelihood of observing particular events , in a way that is correlated with the mass density . this includes , for example , all sources leading to obscuration of the light beam , either alone or in combination with restrictions coming from imperfect search efficiencies and strategies . for example selection by extinction effects or by outlier rejection mainly relate to high - magnification events which are clearly correlated with having high intervening mass concentrations ( halos with large @xmath49 and/or small @xmath220 ) along the light geodesic . similarly , short duration events might be missed by search telescopes or be rejected from the data due to poor quality of the light curve , for example in cases where a supernova is not separable from the image of a bright foreground galaxy . probability of such events would obviously be correlated with the brightness of the source ( @xmath217 and source dependent parameters ) , with the density and redshift of the intervening matter and of course with the search efficiencies . the selection biases can be studied in the sgl framework by introducing a new , object - dependent _ survival probability function_. that is , we replace @xmath221 in the probability distribution for the magnification pdf ( [ eq : pdf1 ] ) . here @xmath222 is a normalization constant that makes sure that the final pdf is normalized to unity . the most generic form of the survival probability function is @xmath223 this allows the survival function to depend on the arbitrary local properties along the photon geodesic . the physical interpretation of @xmath224 is of course clear : it gives the relative probability that an event whose signal intercepts an object with characteristics given by internal parameters @xmath172 at redshift @xmath218 would make it to the accepted observational sample . given the form ( [ eq : piu ] ) the proper normalization is easily seen to be : @xmath225 where @xmath226 the correct expression for the convergence is still given by eq . ( [ eq : fullkappa ] ) , with the important difference that the random integers @xmath210 used to evaluate eq . ( [ eq : pdf2 ] ) are now drawn from the poisson distribution of eq . ( [ eq : survival2 ] ) with parameter the effective expected number of objects @xmath227 . as mentioned above , modeling @xmath224 quantitatively would need detailed input of the astrophysical properties of the intervening matter distributions and of the observational apparatus , which is beyond the scope of this work . see ref @xcite for an analytical example giving @xmath228 , where @xmath229 is the effective filling factor of a partially - filled beam @xcite ( see @xcite for an extension of the partially - filled beam formalism ) . we will discuss an illustrative numerical example in section [ results ] where we will consider as survival probability a simple step - function in the impact parameter . to summarise , let us note that it is easy to combine both types of observational biases discussed in this section . the true observational pdf can be defined as @xmath230 where @xmath212 again is the intrinsic source magnitude dispersion and the effective lensing probability distribution function @xmath231 is computed from eq . ( [ eq : pdf2 ] ) using the effective probability distribution of eq . ( [ eq : survival2 ] ) , and computing the effective magnitudes for configurations from equations ( [ eq : deltameff ] ) and ( [ eq : mukappa ] ) . this setting clearly allows modeling a very complex structure of observational biases . finally note that the generic trend among all the effects discussed above is to suppress events with higher magnifications . it could then be that a non - negligible overestimate of the observed magnitudes follows from neglecting the selection biases , even if the individual effects were relatively small . moreover , while the present data can not yet constrain the lensing pdf itself , it can put bounds on its variance . since the observational biases are suppressing the high magnification tail , they give @xmath232 a smaller effective variance which may open the possibility for a comparison with the experiments . we will further discuss these topics in section [ results ] . we conclude this section by deriving an effective distribution @xmath233 for a data sample @xmath234 , where the set @xmath234 refers to a given binning of the original data . @xmath233 could be computed directly from the fundamental pdf @xcite , but it is much faster to create it directly during the initial simulation , bypassing the calculation of the fundamental @xmath203 altogether . indeed , if we label the configurations within a given sample of observations by @xmath235 , the mean convergence after @xmath236 observations is : @xmath237 where in the second line we have used the fact that the quantities @xmath179 are independent of @xmath235 and in the last line we have used the property that the independent poisson variables @xmath238 ( with the same weight ) sum exactly ( in a statistical sense ) into the poisson variable @xmath239 of parameter given by the sum of the individual parameters , which is @xmath240 . @xmath233 is then given by an expression similar to eqs . ( [ eq : pdf2a]-[eq : pdf1 ] ) or eq . ( [ eq : pdf2 ] ) . note , however , that including the selection effects ( see section [ sbs ] ) in general does not commute with taking an average over the observations . in other words , if the @xmath241 measurements are correlated , we can not use eq . ( [ equation : kappa3 ] ) , but we have to start from the fundamental pdf . on the other hand , eq . ( [ equation : kappa3 ] ) displays explicitly the effect of the size of the data sample : even if @xmath146 had a skewed pdf and a nonzero mode for @xmath242 , for large @xmath241 the distribution approaches a gaussian and eventually converges to a @xmath243-function with a zero convergence , as it is clear from the properties of poisson variables . from the general expression eq . ( [ equation : kappa3 ] ) , including a nontrivial survival probability , it follows that the expected value and variance of the convergence are : @xmath244 and @xmath245 where we assumed that the survival probability for a light ray going through the uniformly distributed matter density @xmath246 equals unity and that the random variables @xmath210 are uncorrelated . ( [ averagekappa]-[varkappa ] ) can then be put back in integral form giving the following _ exact _ results : @xmath247 and @xmath248 where the integral limits for the last two integrals are implicitly defined and the explicit form of surface area element @xmath249 depends on the geometrical properties of the object of mass @xmath49 . for example , for spherical halos it is @xmath250 . we again stress that the quantity @xmath251 is a generic function that describes the probability that a light ray is observed when it hits a halo of mass @xmath49 at the redshift @xmath50 and impact parameter @xmath220 for a source at redshift @xmath252 . ( [ averagekappax]-[varkappax ] ) are a simple and direct prediction of the sgl method and allow to draw some general considerations . first , if @xmath253 , the expected value of the convergence is correctly zero , showing the `` benevolent '' nature of weak lensing corrections for unbiased observations . if however the survival probability is not trivial , selection biases persist even in very large datasets so that the average convergence over a large number of observations approaches a nonvanishing value . second , eq . ( [ varkappax ] ) is a product of positive quantities and so _ a nontrivial survival probability always reduces the observed variance . _ we would like to stress that it is easy to numerically evaluate the integrals of eqs . ( [ averagekappax]-[varkappax ] ) , and their predictions can be straightforwardly implemented in @xmath254 analyses based on gaussian likelihoods , which may be a reasonable approximation if the observable pdf is not strongly skewed . the key role played by the discretization of the model parameters in the sgl method is now quite obvious . of course , the final lensing pdf should not depend on the discretization . indeed , as the numerical value of a riemann integral does not depend on the binning chosen if the integrand does not significantly vary within the bins , the resulting pdf will not depend on the specific binning if the cosmological functions @xmath179 are approximately constant within the bin cells @xmath188 . once this condition is well met a further refinement of binning does not increase the accuracy . this statement can be proved formally as follows : imagine that we have a valid binning prescription @xmath255 and we want to move to a new prescription @xmath256 which is @xmath257 times finer . because @xmath258 is valid , the convergences @xmath259 are approximately constant within each @xmath255-bin . the relative poisson variables can then be resummed as done in eq . ( [ equation : kappa3 ] ) , and we obtain the same formal expression in terms of @xmath258-bins we started from . this theorem also explains why we do not need to bin any variable in which the surface mass density @xmath171 of a given object is constant , such as the rotational angle along the center of a spherical halo in the lens plane . as the convergences @xmath179 are constants within angular bins , their contributions can be resummed to give a model without any angular binning . that is , the fundamental halo objects are actually finite - width rings in the convergence plane . the first quantity to be binned is the co - moving distance to the source . here the basic quantity that needs to be accurately modelled is the function @xmath260 , which gives the optical weight of a given convergence plane . this is a rather smooth function with a shallow peak , and only slightly dependent on @xmath150 when expressed in the scaled variable @xmath261 . simple linear bins @xmath162 have been found to give accurate results , with only @xmath262 redshift slices . this turns out to be the typical number of bins needed for an accurate modeling of most of the internal variables as well . we begin by discretizing the halo parameters , which are the mass and the impact parameter in the lens plane . binning the halo masses involves two main considerations . first , the halo mass function only carries information about high - contrast virialized halos within a specific mass range ( see eq . ( [ jenk ] ) ) . moreover , as explained in section [ mati ] , only very large mass concentrations play a significant role in our weak lensing analysis . consequently , we introduced the lower cut @xmath70 to the mass distribution so that only mass concentrations with @xmath58 are treated as halos , while the remaining mass has been divided into low contrast objects and uniform matter density ( see eq . ( [ eq : sumrule ] ) ) . second , the fact that the concentration parameter of eq . ( [ duffy ] ) follows a power law suggests the use of a logarithmic binning in mass . the mass function drops quickly at very large masses , so that we can combine all masses above a certain limit to a single bin . we found it reasonable to define at any redshift . one could , for example , use bins such that @xmath69 has always the present - day value . see section [ mati ] for a discussion about @xmath69 . ] the cell boundaries by @xmath263 and the cell centers by @xmath264 , with @xmath265 . because @xmath266 to a very good approximation , it is sufficient to bin with an integer @xmath267 between 10 and 15 : @xmath268 the fraction of total mass in a bin is then @xmath269 which correctly gives the total halo mass fraction ( see eq . ( [ eq : def_of_fh ] ) ) when summed over all the halo bins : @xmath270 finally , the binned comoving halo densities are @xmath271 which also allows to relate the mass @xmath272 to the interhalo scale @xmath273 . the halo impact parameter @xmath274 in the lens plane is restricted to the range @xmath275 $ ] , where @xmath111 is the comoving halo radius . it should be noted that through eq . ( [ mdel ] ) the radius depends both on the halo mass and the redshift . see fig . [ sketch ] for an illustrative sketch . it is efficient to discretize @xmath220 into bins of constant integrated surface density ( constant equal mass ) : @xmath276 where @xmath277 is the number of @xmath220-bins used and the surface density is given by ( see fig . [ sketch ] ) : @xmath278 after the redshift and the halo mass is specified , the bin boundaries @xmath279 and the weighted centers of gravity @xmath280 can be computed using eq . ( [ binbounds ] ) . after these are defined , the corresponding area functions @xmath281 , needed for the poisson parameters @xmath282 and the binned surface densities @xmath283 , can be computed . overall , our halo model is described by the redshift index and two internal indices for the discretized mass and impact parameter . in terms of the generic parameter @xmath172 we can formally express this as : @xmath284 , where the label @xmath285 refers to the halo " family of the inhomogeneities . .,width=8 ] of eq . ( [ ltt ] ) as shown in fig . [ cylinder2].,width=8 ] the position and the orientation of a generic object in the three dimensional space is described by three position and three orientation degrees of freedom . since our filaments objects are confined to within a given @xmath286 slice , they generically have only five degrees of freedom . special symmetries of a sphere reduced the number of relevant parameters for halos to a sigle one , the impact parameter @xmath274 . for filaments the situation is slightly more complicated because a cylinder is invariant only for rotations around one symmetry axis . this leaves us with four degrees of freedom . the relevant parameters , however , are reduced by the symmetries to only two : the angle @xmath151 of the filament main axes with respect to the geodesic and the impact parameter @xmath287 . for illustration see figs . [ cylinder2]-[cylinder1 ] . this can be understood by the fact that all the possible filament configurations with the same @xmath151 and @xmath287 have the same surface density @xmath288 and so can be resummed using the theorem given above . in particular , the projected surface density of our filaments does not depend on the coordinate @xmath289 along the projected main axes we move " the triangular section as shown in fig . [ cylinder2 ] . strictly speaking our filaments do not have a cylindrical symmetry , but a @xmath151-dependent tilted " rotation symmetry , such that the projected surface densities follow equation ( [ gammat0 ] ) for all @xmath290 . this approximation should be good for long filaments with @xmath291 . ] allowing to resum the @xmath289-bins into a single effective bin of the total @xmath151-dependent length of the projection : @xmath292 the angle @xmath151 is physically important because objects seen with a smaller @xmath151 have a smaller cross section and a higher surface density . accounting for this geometrical effect leads to a more strongly skewed lensing pdf . the expression for the surface density is given by @xmath293 for the uniform filament profile this gives simply @xmath294 , where @xmath295 . for the gaussian profile ( [ filagau ] ) one instead finds @xmath296 \ , . \label{gammat}\ ] ] the angle @xmath151 can be restricted by symmetry to the interval @xmath297 $ ] , which we divide into @xmath298 uniform length bins around centers @xmath299 . the impact parameter is restricted to @xmath300 $ ] , and it will be discretized into @xmath301 bins using the equal mass criteria , as in the halo case defined in eq . ( [ binbounds ] ) . that is , our filaments will be represented by a family of bars " of different surface densities @xmath302 and length @xmath303 . the surface area of these bars to be used in the poisson parameters is @xmath304 . overall , our filament model is described by the redshift index , and two internal indices for the discretized angle and impact parameter . in terms of the generic parameter @xmath172 we can again formally express this as : @xmath305 , where the family label @xmath104 refers to the filament " family of the inhomogeneity . of course adding , _ e.g. _ , a mass distribution for filaments and/or an @xmath306-dependent density profile would increase the necessary number of indices in the discrete model . until now we have implicitly assumed that the halo and the filament families are independent . we will now show how a more realistic model with halos confined to the filaments can be set up in the sgl approach . the central observation is that the exact positions of halos and filaments within the co - moving distance bins are irrelevant in eq . ( [ eq : fullkappa ] ) for the convergence . it then does not matter if we really confine the halos to filaments ; for the same effect it is sufficient to merely confine them into the equivalent volume occupied by the filaments in a given bin . we can actually do even better than imposing a simple volume confinement . indeed , it is natural to assume that the halo distribution follows the smooth density profile that defines the filament , which can be accounted for by weighting the volume elements by their reduced density profiles . because only the projected matter density is relevant for lensing , this weighting is accounted for by the effective surface densities of the objects . we then introduce the effective co - moving thickness for the low density objects as : to refer to generic internal filamentary degrees of freedom . ] @xmath307 where @xmath308 is the average mass density of the filament with volume @xmath309 . the total effective length of a configuration of filaments @xmath310 along the line of sight is then : @xmath311 it is easy to show that the statistical average of @xmath312 over the configuration space is just the expected distance covered by filaments : @xmath313 where we defined the comoving average filament volume fraction : @xmath314 the confining simulation can now be performed in two steps : in the first step a random configuration of filaments is generated and eq . ( [ eq : master1 ] ) is used to compute the effective lengths @xmath315 at different co - moving slices . these lengths define then the poisson parameters for the halos through eq . ( [ deltan ] ) : @xmath316 note that the halo densities @xmath317 are corrected to account for the reduced volume available due to confinement : @xmath318 where @xmath319 are the usual halo densities of eq . ( [ nbins ] ) when the halos are uniformly distributed through all space . inserting eq . ( [ eq : haloweights ] ) back to ( [ deltanhalos ] ) and using eqs . ( [ eq : master1]-[eq : master2 ] ) we can rewrite eq . ( [ deltanhalos ] ) simply as @xmath320 where @xmath321 are the usual configuration independent poisson parameters computed without confinement . note that the central quantities needed in the evaluation of eq . ( [ deltanf ] ) are the filamentary weights @xmath322 and surface densities @xmath323 . with these given , all the information about the specific low contrast objects used is condensed into the single stochastic variable : @xmath324 which has an expectation value of unity and a mode smaller than unity . the lensing effects due to large - scale structures are thus tied to the probability distribution of @xmath325 : its added skewness is the effect coming from confining halos within filaments . this opens the possibility to investigate and compare different filamentary geometries by means of the @xmath325-pdf . we will develop these thoughts further in a forthcoming paper @xcite . we can further generalize the picture given in this section by considering different levels of confinement for halos of different masses . indeed , because halos of different mass are treated independently in eq . ( [ deltanf ] ) , one could have small halos populate also the voids and massive halos only the filamentary structure . let us finally point out that in practical computations the convergence pdf is obtained by creating a large number of independent halo configurations for each master " filament configuration . typically we use simulations with few hundred master configurations with a few hundred halo configurations each . in this section we will discuss the power spectrum of our model universe and the importance of large - scale correlations for the lensing pdf . we start by making a connection between the sgl method and the so - called `` halo model '' ( see , for example , @xcite and @xcite for a review ) , where inhomogeneities are approximated by a collection of different kinds of halos whose spatial distributions satisfies the linear power spectrum . the idea behind the halo model is that on small scales ( large wavenumbers @xmath326 ) the statistics of matter correlations are dominated by the internal halo density profiles , while on large scales the halos are assumed to cluster according to linear theory . the two components are then combined together . the sgl modeling of the inhomogeneities can be thought as a two - step halo model where we first create the random filamentary structures and then place the halos randomly within these structures . similarly to the halo model , one can then combine the linearly - evolved power spectrum with the nonlinear one coming from the filaments and the halos they contain . in this sense our filamentary structures extend the halo model by introducing correlations in the intermediate scales between the halo substructures and the cosmological scale controlled by the linear power spectrum . in the halo model the power spectrum can be computed analytically , but here the calculation has to be done numerically . we are currently extending the sgl method such that the simulation will produce also the power spectrum @xcite in addition to the lensing pdf . the power spectrum can indeed place useful constraints on the filament parameters which , as remarked in section [ mati ] , are not tightly constrained by observations . while the power spectrum is relevant for understanding the correlations at the largest observable scales , the lensing pdf depends mainly on the much stronger inhomogeneities at smaller scales . this can be understood from eq . ( [ eq : kappa1 ] ) which shows the direct dependence of the lensing convergence on the density contrast . the small lensing variance from the large - scale correlations induced by the linear power spectrum was numerically computed , for example , in @xcite . moreover , weak lensing and the power spectrum probe somewhat different aspects of the inhomogeneities . the web - like structures of filaments and voids that affect weak lensing are mainly described by higher order correlation terms beyond the power spectrum , and so special care has to be put in designing the filamentary structure . this is indeed the goal of the sgl method : to give an accurate and flexible modeling of the universe as far as its lensing properties are concerned . our main goal in this section is to compare the sgl approach with the convergence pdf computed from large scale simulations . the idea is that by achieving a good agreement with numerical simulations we are proving that the sgl method does provide a good and accurate description of the weak lensing phenomena . as we shall see , this is indeed the case , as we can naturally reproduce the lensing pdf of the millenium simulation . let us stress that while a comparison to simulations is a good benchmark test for the sgl approach , the simulations and underlying @xmath123cdm model do not necessarily describe nature . indeed observations do not yet provide strong constraints to the lensing pdf , which leaves room for very different types of large scale structures , examples of which have been studied for example in ref . @xcite . with the accuracy of the method tested , one can reliably compute the effects of selection biases using sgl . while we lack the necessary expertise to quantitatively estimate the possible bias parameters , we will study a simple toy model for the survival probability function @xmath224 to show qualitatively how selection effets might bias the observable pdf . finally , we will also show how a jdem - like survey could constrain the lensing pdf relative to a given cosmological model . we shall begin , however , by the comparison with the simulations . we shall now confront the sgl method with the cosmology described by the millennium simulation ( ms ) @xcite . accordingly , we will fix the cosmological parameters to @xmath327 , @xmath328 , @xmath329 , @xmath330 , @xmath331 @xmath332 and @xmath333 . moreover , the mass function of eq . ( [ jenk ] ) agrees with the ms results . these parameters completely fix the background cosmology and the halos . we are then left with the filament parameters to specify . first , we fix @xmath334 , which means that half of the unvirialized mass forms filaments , while the other half is uniformly distributed . this value determines the maximum possible demagnification @xmath335 in the lensing pdf , corresponding to light that misses all the inhomogeneities . for the background model described above this implies @xmath336 mag . for each @xmath337 the sgl method evaluates the corresponding occupational number from the poisson statistics . the program mathematica using one core of a cpu at @xmath338 ghz takes a time @xmath339 s to produce an array of @xmath340 poisson numbers . other programs will likely have similar performances . @xmath341 determines the statistics with which the lensing pdf is generated . using this information we can then estimate the performance of the sgl method by evaluating @xmath342 . the results are shown in table [ turbo ] where we fixed @xmath343 , @xmath344 and @xmath345 . confining halos within filaments typically makes calculations about ten times longer . p. valageas , astron . astrophys . * 356 * , 771 ( 2000 ) . d. munshi and b. jain , mon . not . soc . * 318 * , 109 ( 2000 ) . y. wang , d. e. holz and d. munshi , astrophys . j. * 572 * , l15 ( 2002 ) . s. das and j. p. ostriker , astrophys . j. * 645 * , 1 ( 2006 ) . l. amendola , k. kainulainen , v. marra and m. quartin , phys . * 105 * , 121302 ( 2010 ) . s. hilbert , s. d. m. white , j. hartlap and p. schneider , mon . not . * 382 * , 121 ( 2007 ) . s. hilbert , s. d. m. white , j. hartlap and p. schneider , mon . not . * 386 * , 1845 ( 2008 ) . v. springel _ et al . _ , nature * 435 * , 629 ( 2005 ) . m. chevallier and d. polarski , int . j. mod . phys . d * 10 * , 213 ( 2001 ) ; e. v. linder , phys . lett . * 90 * , 091301 ( 2003 ) . m. bartelmann and p. schneider , phys . rept . * 340 * , 291 ( 2001 ) . a. jenkins _ et al . _ , mon . not . soc . * 321 * , 372 ( 2001 ) . r. kantowski , astrophys . j. * 155 * , 89 ( 1969 ) . v. marra , e. w. kolb , s. matarrese and a. riotto , phys . d * 76 * , 123004 ( 2007 ) ; v. marra , e. w. kolb and s. matarrese , phys . d * 77 * , 023003 ( 2008 ) ; v. marra , arxiv:0803.3152 [ astro - ph ] ; e. w. kolb , v. marra and s. matarrese , gen . grav . * 42 * , 1399 ( 2010 ) . e. rozo _ et al . _ , astrophys . j. * 708 * , 645 ( 2010 ) . w. j. percival , astron . astrophys . * 443 * , 819 ( 2005 ) ; see also : s. basilakos , astrophys . j. * 590 * , 636 ( 2003 ) . d. j. eisenstein and w. hu , astrophys . j. * 496 * , 605 ( 1998 ) . m. j. . white , astrophys . j. suppl . * 143 * , 241 ( 2002 ) . r. k. sheth and g. tormen , mon . not . soc . * 308 * ( 1999 ) 119 . j. f. navarro , c. s. frenk and s. d. m. white , astrophys . j. * 462 * , 563 ( 1996 ) . a. r. duffy , j. schaye , s. t. kay and c. dalla vecchia , mon . not . 390 * , l64 ( 2008 ) . e. komatsu _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 180 * , 330 ( 2009 ) .

we present multi - wavelength observations of the unassociated-ray source 3fgl j2039.6 detected by the _ fermi _ large area telescope . the source-ray properties suggest that it is a pulsar , most likely a millisecond pulsar , for which neither radio nor-ray pulsations have been detected yet . we observed 3fgl j2039.6 with _ xmm - newton _ and discovered several candidate x - ray counterparts within / close to the-ray error box . the brightest of these x - ray sources is variable with a period of 0.2245.0081 d. its x - ray spectrum can be described by a power law with photon index , and hydrogen column density , which gives an unabsorbed 0.310 kev x - ray flux of erg s . observations with the gamma - ray burst optical / near - infrared detector ( grond ) discovered an optical counterpart to this x - ray source , with a time - averaged magnitude . the counterpart features a flux modulation with a period of 0.22748.00043 d that coincides , within the errors , with that of the x - ray source , confirming the association based on the positional coincidence . we interpret the observed x - ray / optical periodicity as the orbital period of a close binary system where one of the two members is a neutron star . the light curve profile of the companion star , with two asymmetric peaks , suggests that the optical emission comes from two regions at different temperatures on its tidally - distorted surface . based upon its x - ray and optical properties , we consider this source as the most likely x - ray counterpart to 3fgl j2039.6 , which we propose to be a new redback system .

1509.07474

the launch of the _ fermi _ gamma - ray space telescope in june 2008 marked a new era in @xmath0-ray astronomy , thanks to the unprecedented performance of its large area telescope ( lat ; atwood et al . the recently released third _ fermi_-lat @xmath0-ray source catalogue ( 3fgl ; acero et al . 2015 ) derived from the first 4 years of observations contains 3033 sources . about 70% of these sources have been either directly identified , e.g. either from the detection of @xmath0-ray pulsations ( pulsars ) or correlated @xmath0-ray and optical / radio variability ( active galactic nuclei , novae , x - ray binaries ) , or associated with objects that are either known or potential @xmath0-ray emitters . the remaining 30% of the 3fgl sources have not been associated with any object yet , hence they are referred to as _ unassociated _ , and their nature is unknown . being pulsars the largest family of @xmath0-ray sources identified in the galaxy ( @xmath9 and counting ) , a significant fraction of the unassociated _ fermi_-lat sources might be @xmath0-ray pulsars . some of them might have no or extremely faint radio emission , and escaped detection in all radio pulsar surveys so far . indeed , many of such radio - quiet ( rq ) , or radio faint ( rf ) , @xmath0-ray pulsars have been discovered through blind periodicity searches in the @xmath0-ray data ( e.g. , abdo et al . 2009 ) thanks to the use of novel search techniques ( e.g. , atwood et al . 2006 ; pletsch et al . 2013 ) . about 45% of the @xmath0-ray pulsars discovered by the _ fermi_-lat are milli - second pulsars ( msps ) . interestingly , the vast majority of these msps ( @xmath10 80% ) are in binary systems . some of them have an he white dwarf ( wd ) companion star of mass @xmath11 , whereas others have a usually non - degenerate companion ( a late main sequence star or a brown dwarf ) which is ablated by irradiation from the pulsar wind . two distinct families of binary msps are recognized depending on the degree of the ablation processes : the black widow ( bw ) msps , where the companion is a very low - mass star of @xmath12 almost fully ablated by the pulsar wind , and the redback ( rb ) msps , where the companion is only partially ablated and has an higher mass of @xmath13@xmath14 ( roberts 2013 ) . the use of automatic classification codes ( e.g. , ackermann et al . 2012 ; lee et al . 2012 ; mirabal et al . 2012 ) based on the @xmath0-ray characteristics is crucial to single out pulsar candidates among the many unassociated _ fermi_-lat sources and optimise a systematic search for new @xmath0-ray pulsars . since @xmath0-ray pulsars are also identified in the optical and x rays ( e.g. , abdo et al . 2013 ) , multi - wavelength follow - ups of unidentified _ fermi_-lat sources are still key to confirm the proposed pulsar classifications , though . in particular , optical observations are an important aid in the search for binary msps , for which blind periodicity searches in @xmath0 rays must account for the unknown orbital parameters , requiring a massive use of super - computing power facilities ( pletsch & clark 2014 ) . indeed , optical observations yielded the identification of the two _ fermi_-lat sources 2fgl j2339.7@xmath10531 and 2fgl j1311.7@xmath13429 as binary msps prior to the detection of radio or @xmath0-ray pulsations ( ray et al . in preparation ; pletsch et al . 2012 ; ray et al . 2013 ) through the discovery of orbital modulations in the flux of their companion stars ( romani & shaw 2011 ; kong et al . 2012 ; romani et al . 2012 ; kataoka et al . 2012 ) . in a similar way , new binary msp candidates have been identified for the two unassociated _ fermi_-lat sources 2fgl j1653.6@xmath10159 ( romani et al . 2014 ; kong et al . 2014 ) and 2fgl j0523.3@xmath12530 ( strader et al . 2014 ) . as a part of a pilot project aimed at identifying different classes of unassociated _ fermi_-lat sources , we studied 3fgl j2039.6@xmath15618 . this is a moderately bright @xmath0-ray source ( detection significance @xmath15 ) that was listed in both the first ( abdo et al . 2010 ) and second ( nolan et al . 2012 ) _ fermi_-lat @xmath0-ray source catalogues ( a.k.a . 1fgl j2039.4@xmath15621 and 2fgl j2039.8@xmath15620 , respectively ) , but has remained unassociated ever since . the 3fgl j2039.6@xmath15618 field was observed in x rays for the first time with the _ swift_/x - ray telescope ( xrt ) during snapshot observations ( 1 and 3.6 ks exposure times ) but no candidate x - ray counterpart was detected within the 2fgl @xmath0-ray source error circle ( takeuchi et al . 2013 ) . in radio , no potential counterpart was found in the sydney university molonglo sky survey ( summs ) source catalogue ( mauch et al . 2003 ) and in dedicated observations of unassociated 2fgl sources with the australia telescope compact array ( petrov et al . 2013 ) and the parkes radio telescope ( camilo et al . 2015 ) . at very high energies , 3fgl j2039.6@xmath15618 is not associated with any known tev source . based upon its @xmath0-ray characteristics , 3fgl j2039.6@xmath15618 was classified as a very likely pulsar candidate by mirabal et al . ( 2013 ) . our newly developed classification method ( salvetti et al . 2013 ; salvetti et al . , in preparation ) confirms that 3fgl j2039.6@xmath15618 is a very likely pulsar candidate and suggests that it is , most probably , an msp , either isolated or in a binary system ( see section [ ann ] ) . however , since @xmath16 of the msps detected by the _ fermi_-lat are in binary systems , one can expect that would be a binary msp . we investigated this scenario through a multi - wavelength observation campaign ( x rays , ultraviolet , optical , infrared ) of 3fgl j2039.6@xmath15618 carried out using both dedicated observations and exploiting data available in public archives . the observations and data reduction are described in section 2 and the results are presented in section 3 . discussion and conclusion follow in section 4 and 5 , respectively . recently , we developed an advanced classification code ( salvetti et al . 2013 ) that can recognise different classes of @xmath0-ray pulsars , e.g. young / middle - aged pulsars and msps . this code uses a statistical predictive method based on artificial neural network ( ann ) techniques to quantify the probability of a given source to be msp - like on the basis of its @xmath0-ray temporal and spectral characteristics . the method is based on an advanced hierarchical ann architecture consisting of simple neural networks applied in sequence to discriminate pulsar - like from agn - like objects in first place and disentangle msps from young / middle age pulsars in second place . such a method correctly classifies 84% of the identified msps , while the false positive fraction is lower than 10% ( salvetti et al . , in preparation ) . we then applied the optimized hierarchical neural network to all _ unassociated _ 3fgl sources , to rank them according to their msp probability function . as a result , 3fgl j2039.6@xmath15618 was classified as a msp - like object with a probability greater than 99% . therefore , it stands out as an obvious candidate for multi - wavelength investigations . we carried out an _ xmm - newton _ observation of the 3fgl j2039.6@xmath15618 @xmath0-ray error box ( programme i d : 0720750301 ) , which started on 2013 october 10 at 09:43:18 ut ( revolution 2534 ) and lasted 44.6 ks . the pn detector ( struder et al . 2001 ) of the european photon imaging camera ( epic ) instrument was operated in extended full frame mode , with a time resolution of 200 ms over a @xmath17 field - of - view ( fov ) , while the metal oxide semi - conductor ( mos ) detectors ( turner et al . 2001 ) were set in full frame mode ( 2.6 s time resolution on a 15@xmath18 radius fov ) . the thin optical filter was used for the pn while a medium one was used for the mos cameras . we retrieved the observation data files ( odf ) from the _ xmm - newton _ science archive and used the most recent release of the _ xmm - newton _ science analysis software ( sas ) v14.0 to analyze them . we performed a standard data processing , using the epproc and emproc tools , and screening for high particle background time intervals ( e.g. , de luca et al . our analysis revealed no significant contamination from soft protons . after the standard data processing , the good , dead - time corrected exposure time was 41.6 ks for the pn and 43.2 ks for the two mos detectors . in the optical / near - infrared ( near - ir ) , we observed the 3fgl j2039.6@xmath15618 field with the gamma - ray burst optical / near - infrared detector ( grond ; greiner et al . 2008 ) at the mpi / eso 2.2 m telescope on la silla ( chile ) . the field was repeatedly observed on august 16 , 17 , and 18 , 2014 in the g@xmath19 , r@xmath19 , i@xmath19 , z@xmath19 bands in the optical and in the j , h , k@xmath20 bands in the near - ir . the observations were split into sequences of 18 , 18 , and 17 exposures per day , each consisting of four 115 s dithered exposures in the optical and forty eight 10 s dithered exposures in the near - ir . the observations were executed in grey time with airmass between 1.12 and 1.28 and mean seeing of 1.0 . single dithered exposures were reduced ( bias subtraction , flat - fielding , distortion correction ) and stacked using standard iraf tasks implemented in the grond pipeline ( krhler et al . 2008 ; yoldas et al . the astrometry calibration was computed on single exposures against stars selected from the usno - b1.0 catalogue ( monet et al . 2003 ) in the optical bands and the 2mass catalogue ( skrutskie et al . 2006 ) in the near - ir bands , yielding an accuracy of 03 with respect to the chosen reference frame . the photometric calibration in the optical was computed against a close - by field from the sloan digital sky survey ( york et al . 2000 ) at @xmath21 observed in the first night under photometric conditions . from the calibrated images we extracted a grid of secondary photometric calibrators for direct on the frame calibration on the subsequent nights . in the near - ir , the photometric calibration was computed against 2mass stars identified in the grond field of view . the accuracy of the absolute photometry calibration was 0.02 magnitudes in the g@xmath19 , r@xmath19 , i@xmath19 , z@xmath19 bands , 0.03 magnitudes in j and h , and 0.05 in the k@xmath20 band . in addition to grond , we used serendipitous jhk@xmath22 images of the 3fgl j2039.6@xmath15618 field taken on july 16 and 25 2010 with the near - ir camera ( vircam ; dalton et al . 2006 ) of the visible and infrared survey telescope for astronomy ( vista ; emerson et al . the data set , processed and calibrated at the cambridge astronomical survey unit ( casu ) , consists of three sequences of 16 consecutive exposures of 15 s each in the j band and of 7.5 s each in both the h and k@xmath22 bands . in the optical and near - ultraviolet ( near - uv ) we used u , uvw1 ( @xmath23 ; @xmath24 ) and uvm2 ( @xmath25 ; @xmath26 ) images from the _ xmm - newton _ optical monitor ( om ; mason et al . 2001 ) obtained in parallel to our observations , with exposure times of 1900 , 2700 , and 3080 s , respectively . we also used archival uvw2 ( @xmath27 ; @xmath28 ) images from the _ swift _ ultraviolet and optical telescope ( uvot ; roming et al . 2005 ) performed on february 21 2011 ( obsid=00041479002 ) , consisting of five exposures for a total integration time of 3587 s. the om and uvot data were processed and calibrated using the sas tool omichain and the heasoft software package , respectively . to search for possible counterparts of 3fgl j2039.6@xmath15618 in our multi - wavelength observations , we used as a reference its recent 3fgl position ( acero et al . 2015 ) : @xmath29 and @xmath30 ( j2000 ) . its associated 95% confidence position error ellipse has semi - major and semi - minor axis of 2.6 and 2.4 , respectively , and a position angle of @xmath31 , measured east of north . we started from our _ xmm - newton _ observations to find potential x - ray counterparts to 3fgl j2039.6@xmath15618 . then , we used our multi - wavelength data base to single out those that are most likely associated with this @xmath0-ray source . in particular , since we expect that 3fgl j2039.6@xmath15618 is a binary msp , we focused our analysis on x - ray sources that show variability and/or have variable optical counterparts , possibly featuring periodic flux modulations . for our x - ray analysis , we selected only 0@xmath14 pattern events from the pn and 0@xmath112 from the two mos detectors with the default flag _ mask_. the source detection in the 0.310 kev energy range was run simultaneously on the event lists of each of the epic - pn and mos detectors using a maximum likelihood fitting with the sas task edetect_chain invoking other sas tools to produce background , sensitivity , and vignetting - corrected exposure maps . the final source list includes 90 x - ray sources from both the pn and mos detectors , with a combined pn+mos detection likelihood greater than 10 , corresponding to a significance above @xmath32 . figure [ xmm - image ] shows the 0.310 kev exposure - corrected _ xmm - newton _ fov obtained combining the images of the epic - pn and mos detectors . we focused our analysis on the 16 x - ray sources detected within , or close to , the 95% confidence position error ellipse of 3fgl j2039.6@xmath15618 . we summarized the positions , spectral parameters , fluxes , and variability indices of these sources in table [ tab1 ] . for each epic detector we extracted the source photons using an extraction radius of 20 , while we extracted background photons from source - free regions in the same ccd chip as the source , with radii of 50120 . for each detector , we used the sas task specgroup to rebin all the extracted spectra and have at least 25 counts for each background - subtracted spectral channel and generated ad hoc response matrices and ancillary files using the sas tasks rmfgen and arfgen . for each source , we fitted simultaneously the pn and mos spectra using xspec v12.8 , forcing the same set of parameters and considering three different spectral models : a _ power - law _ ( pl ) , well suited for both agn and pulsars , an _ apec _ ( ap ) for stellar coronae , and a _ black - body _ ( bb ) for the pulsar thermal component . in all cases , the hydrogen column density @xmath33 was left as a free parameter . for each emission model we computed the 90% confidence level error on the spectral parameters . when the best - fit @xmath33 values were comparable to zero , we assumed the measured uncertainties to determine the @xmath34 confidence level upper limit . spectra with very low counts were fitted after fixing the @xmath33 to the estimated value along the line of sight ( @xmath35 @xmath5 ; dickey & lockman 1990 ) , or fixing either the photon index or the temperature to typical values , i.e. @xmath36 and @xmath37 kev or @xmath38 kev in case of an ap and a bb model , respectively . as shown in table [ tab1 ] , only source 3 and 11 , the two brightest x - ray sources in our sample , were best fitted by a simple pl model , with a @xmath39=52.32 ( 42 degrees of freedom , d.o.f . ) for the former and @xmath39=16.06 ( 22 d.o.f . ) for the latter . for the remaining 13 sources , at least two different models were required to obtain an acceptable fit with null hypothesis probability @xmath40 0.01 . only for source 88 it was not possible to obtain acceptable results with any of the selected spectral models . indeed , this source is the faintest in our sample and is likely spurious . figure [ spec_src3 ] shows the binned spectrum of source 3 extracted simultaneously from each of the epic detectors , together with the best - fit pl model . since most msps are characterized by a significant thermal component to their x - ray spectra , we tried to fit the spectra of source 3 and source 11 with an absorbed bb plus pl model . this improved the accuracy of the fit only for source 3 , with a @xmath39=34.9 ( 40 d.o.f . ) . from the f - test ( bevington 1969 ) , we computed a 0.0003 probability that adding a thermal component to the model would produce a chance improvement to the fit . since this probability corresponds to a @xmath103.6@xmath41 significance only , hereafter we ignored the absorbed bb plus pl spectral model . since for most x - ray sources the measured counts are too few to clearly discriminate among different spectral models , we checked whether we could extract qualitative spectral information from an hardness ratio ( hr ) analysis ( marelli et al . however , we found that the observed hrs are compatible with different spectral models and , therefore , are not constraining . in order to detect possible time variability during the _ xmm - newton _ observation , we generated standard light curves from the pn - data for all the 16 x - ray sources in table 1 . starting from the source and background regions described in section [ spectra ] , we extracted source+background and background light curves , respectively , with the sas task evselect and combined them into a background - subtracted light curve with the task epiclccorr . this task also corrects the time series for vignetting , bad pixels , chip gaps , quantum efficiency , dead time , exposure and good time intervals . for each source we generated a light curve with time binning of 2500 s , or multiple of this value , to have at least 25 counts per bin and we ran a @xmath39 test to evaluate the variability significance . only source 3 is characterized by a significant variability during the observation ( @xmath42 , with 16 d.o.f . ) , with a chance probability of @xmath43 ( @xmath44 ) . after combining the data from all the three epic cameras the source 3 light curve shows a more apparent variability ( figure [ lc_src3 ] , left ) , with a chance probability of @xmath45 , whereas the other x - ray still show no evidence of significant variability . the source 3 light curve also hints at a possibly periodic modulation . we converted photon arrival times to the solar system barycentric dynamical time ( tbd ) with the sas task barycen and used the ftool task efsearch to find the best period in the light curve through a maximum @xmath39 test . we folded the light curve with periods ranging from 100 to 43000 s , the latter comparable to the length of the _ xmm - newton _ observation . we found the best - fit period at [email protected] d , where the period uncertainty was obtained following leahy ( 1987 ) . the corresponding @xmath39 of 104.5 ( 9 d.o.f . ) gives a chance probability of @xmath46 , accounting for the number of trials , and makes the periodicity statistically significant ( @xmath47 ) . the presence of a periodic signal at the corresponding frequency of @xmath48 hz was independently confirmed by the power spectrum produced with the ftool powspec . we note that the best - fit x - ray period of source 3 is comparable to about half the length of the _ xmm - newton _ observation , so that the observed periodicity might be spurious . we examined the light curves of other comparably bright x - ray sources detected in the whole _ xmm - newton _ fov and found that none of them showed evidence of periodicity at any time scales . nonetheless , the fact that the length of the _ xmm - newton _ observation only covers @xmath10 2.3 cycles , prevents us to firmly claim that source 3 is periodic . the x - ray light curve folded at the best - fit period ( figure [ lc_src3 ] , right ) is characterised by two peaks separated in phase by [email protected] . therefore , it can not be described by a simple sinusoidal or gaussian model , with a null hypothesis probability lower than @xmath49 . we checked whether the folded x - ray light curve of source 3 varied as a function of the energy and whether the x - ray spectrum changed as a function of the phase . in both cases , however , the available statistics is not sufficient to highlight significant differences in the energy - resolved light curves and the phase - resolved spectra . we looked for archival x - ray images of the 3fgl j2039.6@xmath15618 field to check for long - term variability of source 3 . the field was serendipitously observed with the x - ray imaging spectrometer ( xis ) of _ suzaku _ ( mitsuda et al . 2007 ) on october 28 , 2010 for a total exposure time of 21.5 ks ( obsid 705028010 ) . we extracted source counts from a circle of radius 13 around the best _ xmm - newton _ coordinates of source 3 and background counts from a nearby , source - free 2-radius circle using the heasoft(v.6.16 ) tool xselect and summed the spectra from all the xis cameras with the mathpha , addarf and addrmf tools . we obtained 350 counts , of which @xmath1050% are from the source . a fit with a pl gives a null hypothesis probability of 0.06 ( 4 d.o.f . ) , with @xmath50 @xmath5 ( 90% upper limit ) , photon index @[email protected] and an unabsorbed flux in the 0.310 kev energy range of 1.7@xmath52 @xmath53 erg @xmath5 s@xmath7 , fully compatible with the _ xmm - newton _ results . the _ xmm - newton _ count - rate and spectral parameters ( table [ tab1 ] ) are also compatible with the non - detection above the 3@xmath41 threshold ( @xmath10 1.05 @xmath53 erg @xmath5 s@xmath7 ) of source 3 in the short _ swift_/xrt images of takeuchi et al . ( 2013 ) , taken in 2010 and 2011 . therefore , we find no evidence of variability of source 3 on time scales of three years . we cross - matched the positions of all the 16 _ xmm - newton _ sources in table 1 with the source catalogues obtained from the grond observations . we performed the source detection on the single - band grond exposures using the starfind tool in iraf , matched the source catalogs over the different observations , and checked for variable sources against the median of all observations . object photometry was computed using the task daophot in iraf and the airmass correction was applied using the standard atmospheric extinction coefficients for the la silla observatory . for the cross - match we used a radius obtained by combining the statistical @xmath54 uncertainty on the x - ray source centroid plus the 90% confidence level systematic error associated with the absolute accuracy of the _ xmm - newton _ aspect solution , which is 15 per coordinate . the uncertainty on the absolute astrometry of the grond images ( 03 ) is much lower than the _ xmm - newton _ one , and is accounted for by our choice of the matching radius . there are only six _ xmm - newton _ sources ( source 3 , 13 , 22 , 24 , 40 , 76 ) with at least a candidate optical or near - ir counterpart in the grond data ( figure [ grond - image ] ) . we also cross - correlated the _ xmm - newton _ source list with the vista , om , and uvot source catalogues . the source detection and photometry in the vista images were carried out as a part of the casu pipeline ( sectn . [ opt_uv ] ) . for the om images , the source detection and photometry were carried out with the sas tasks omdetect and ommag , respectively , and for the uvot images with the heasoft taks uvotdetect and uvotsource . as before , the matching radius accounted for the uncertainty on the absolute astrometry of the used source catalogues , i.e. @xmath55 ( vista ; emerson et al . 2004 ) , 05 ( uvot ; breeveld et al . 2010 ) , and 07 ( om ; page et al . the cross - correlations with these catalogues provided additional near - ir , and near - uv magnitudes for some of the grond sources and unveiled candidate counterparts for source 11 , 19 , 29 , 60 , 88 . we found no counterparts in the om / uvm2 filter , whereas only source 3 was detected in the uvot / uvw2 images . the optical , near - uv , and near - ir magnitudes of the candidate counterparts to the _ xmm - newton _ sources are summarised in table 2 . only for the candidate counterparts to source 3 , 40 , and 76 we have an adequate spectral coverage in at least the optical and near - ir . among the six _ xmm - newton _ sources with a possible grond counterpart , only source 3 is associated with a clearly variable object ( @xmath56 , with 51 d.o.f . ) . figure [ grond - lc ] ( left ) shows the multi - band light curves of this object for the three nights spanned by the grond observations . as seen , the light curves seem modulated , with an amplitude of @xmath100.4 magnitudes in the g@xmath19 band . in particular , the modulation seems to be periodic and feature a double - peaked profile ( night 1 ) with the two peaks only partially seen in night 2 and 3 , likely owing to the different sampling of the light curve . this modulation is also seen in the r@xmath19 , i@xmath19 , and z@xmath19 light curves , with both shape and amplitudes similar to the g@xmath19 one , and is also recognised in the j and h - band light curves , and very marginally in the k - band one . this suggests that the observed modulation is real and not due , for instance , to possible problems with the photometry in a given filter . as a check , for all filters we extracted the light curves of several stars of comparable brightness detected in the grond fov but found no evidence of such a modulation this confirms that it is not due to random effects , such as variations in the sky conditions ( transparency , background , moon illumination ) , or systematic effects , such as variations in the encircled flux due to the fixed size of the photometry aperture with respect to the seeing disk . furthermore , the fact that the modulation seems to be periodic argues against the possibility that it is produced by any of such effects and implies that is associated with an intrinsic star variability . we computed the probability that the association between source 3 and its candidate grond counterpart is due to a chance coincidence . we computed the probability as @xmath57 , where @xmath58 is the matching radius used for source 3 ( 18 ) and @xmath59 is the density of stellar objects in the grond field , regardless of their brightness , measured in the co - addition of all g-band exposures . for a stellar density @xmath60 square degree@xmath7 we estimated a chance coincidence probability @xmath61 , which makes a chance coincidence unlikely . to confirm the existence of a periodic modulation of the source 3 candidate counterpart , we carried out a periodicity analysis based on the generalized lombe - scargle periodogram method ( lomb 1976 ; scargle 1982 ; zechmeister & kuerted 2009 ) . for cross - checking purpose , we also used the phase dispersion minimisation technique ( pdm ; stellingwerf 1978 ) using the pdm code in iraf and the period04 software package ( lenz & breger 2005 ) , which is especially dedicated to the statistical analysis of large astronomical time series containing gaps ( see figure [ grond - lc ] ) . all methods indicate the presence of a periodicity with a period of @xmath10 0.227 d. the analysis of the power spectrum of the optical time series shows a clear peak at the corresponding frequency , with a probability that is due to a chance noise fluctuation ( false - alarm probability ) of @xmath10 1.8@xmath6210@xmath63 ( @xmath10 8@xmath41 ) . the best period was found at [email protected] d in the optical bands and at [email protected] d in the near - ir bands , where we estimated the period uncertainty following gilliland et al . the optical and near - ir best - fit periods of the source 3 candidate counterpart are consistent within the uncertainties , which provides further evidence that the observed modulation is real . the period of the optical / ir modulations seen in the grond data for the source 3 candidate counterpart is consistent with that seen in the _ xmm - newton _ data for source 3 ( [email protected] d ; sect . [ x_var ] ) . therefore , the detection of virtually the same periodicity clearly indicates that the two objects are associated . therefore , based upon the x - ray and optical variability at the same period and the relatively low chance coincidence probability , we regard the association between source 3 and its grond candidate counterpart as robust . furthermore , since we based the selection of candidate x - ray counterparts to 3fgl j2039.6@xmath15618 on the search for variable sources , the optical / x - ray periodicity of source 3 makes it a very promising candidate . another possible candidate x - ray counterpart to 3fgl j2039.6@xmath15618 would be source 11 , the second brightest x - ray source detected in the 3fgl error circle , and the only other x - ray source with an obvious non - thermal spectrum ( table 1 ) . however , source 11 is not variable in x rays and is not associated with a periodic optical / near - ir grond counterpart . therefore , although it can not be firmly ruled out as a candidate x - ray counterpart to 3fgl j2039.6@xmath15618 , as of now , it is a by far a less likely candidate than source 3 . the same is true for all the other fainter x - ray sources in table 1 , whose poorly characterised x - ray spectra and sparse optical , near - uv , and near - ir flux measurements hamper a detailed multi - wavelength analysis of their properties and a non - ambiguous classification . the optical / near - ir light curves of the source 3 counterpart folded at the corresponding best - fit periods are shown in figure [ grond - lc ] ( right ) . a double - peak light curve is clearly visible , with a main peak and a secondary peak , separated in phase by @xmath64 . per each band , we determined the phase of the two peaks by fitting a gaussian function to their profiles to precisely compute their phase separations and errors . the peak phase separation remains constant in the optical ( @xmath65 ; g ) , while it seems to slightly increase in the near - ir ( @xmath66 ; j ) . to better recognise the light curve evolution , we defined four regions : the main peak ( @xmath67=0.20.5 ) , the secondary peak ( @xmath67=0.70.9 ) , the `` bridge '' ( @xmath67=0.50.7 ) , and the `` off - peak '' ( @xmath67=0.00.2 and @xmath67=0.91.0 ) . the shape of the light curve is similar in all bands , but the profile of the modulation changes from the optical to the near - ir ( figure [ grond - lc - diff ] , left ) . in particular , the primary peak becomes broader and its amplitude decreases , from @xmath10 0.4 magnitudes in the g@xmath19 band to @xmath10 0.2 magnitudes in the h band , and the difference between the primary and the secondary peaks decrease from 0.133 magnitudes in the g band to @xmath68 in the h band . similarly , the amplitude of the `` bridge '' decreases , becoming comparable to that of the `` off - peak '' region . there is also a possible evidence of a colour variation as a function of phase ( figure [ grond - lc - diff ] , right ) , which might indicate that we are observing regions of the star surface at different temperatures . in particular , the colours seem to be bluer in coincidence with the two peaks , and bluer in coincidence with the primary peak than with secondary one . the colours also seem bluer in the `` bridge '' than in the `` off - peak '' region . the colour variation seems less marked at longer wavelengths consistently with the fact that the light curve variations are smoother . unfortunately , the errors on the grond photometry calibration ( sectn . [ opt_uv ] ) make difficult to quantify the observed colour variations . we tried to align in phase the x - ray and optical light curves of source 3 to check the relative alignments between the x - ray and optical peaks . however , the difference between the epochs of the _ xmm - newton _ and grond observations ( mjd=5688556887 and mjd=56575 , respectively ) corresponds to a maximum time difference of 312 d. the accuracy on the best - fit optical period derived from the grond observations is 0.00043 d ( sectn . [ opt_var ] ) , which corresponds to a phase uncertainty of @xmath69 . thus , building a phase - coherent solution for the optical light curve backward to the epoch of the _ xmm - newton _ observation would bring an uncertainty of @xmath10 0.59 on the absolute phase determination , which is larger than the phase separation between the two peaks ( @xmath70 in the x rays and @xmath71 in the optical ) . we computed the time - average multi - band photometry of the source 3 counterpart from the grond data and obtained g@[email protected] , r@[email protected] , i@[email protected] , z@[email protected] , [email protected] , [email protected] , [email protected] , with all magnitudes in the ab system . the photometry errors account for both statistical errors and for the systematic uncertainty on the photometry calibration . we also identified the source 3 counterpart in the u and uvw1-filter exposures from the _ xmm - newton_/om with ab magnitudes [email protected] and m@[email protected] ( table 2 ) , whereas it is not detected in the uvm2 filter down to a 3 @xmath41 limiting magnitude of 21.99 and in the _ swift_/uvot uvw2 image down to a limiting magnitude of 23.34 ( ab ) . in the near - ir , we also identified the source 3 counterpart in the vista images , with ab magnitudes [email protected] , [email protected] , and k@[email protected] , converted from the native vega survey system . we found that the vista magnitudes ( epoch 2010.6 ) are all compatible with the grond ones ( epoch 2014.7 ) , after accounting for the difference between the k and k@xmath22 filters , which excludes long - term variability on year time scales from the source 3 counterpart . we used the time - averaged g@xmath19 , r@xmath19 , i@xmath19 , z@xmath19-band magnitudes of the source 3 counterpart as a reference for its classification by analysing its location in the observed ( i.e. not corrected for the reddening ) colour - magnitude ( cm ) and colour - colour ( cc ) diagrams of the field . the diagrams are shown in figure [ grond - cmd ] , where the location of the field stars is shown by the black filled circles and that of the source 3 counterpart as a red filled triangle . red and green triangles correspond to the location computed from the single - image photometry . in order to reject outliers and include only high - confidence measurements , we plotted only field stars for which at least 20 measurements per filter were available and with @xmath73 . we compared the observed cm and cc diagrams with simulated stellar sequences computed from the besanon models ( robin et al . 2004 ) for different stellar populations and distance values up to 15 kpc . the simulated sequences are shown in figure [ grond - cmd ] as the grey scale map . the dark grey regions in the cm diagrams correspond to a likely distance range ( @xmath74 pc ) for source 3 , whereas in the cc diagram the dark grey region corresponds to simulated magnitudes that are within @xmath2 0.05 the g@xmath19-band magnitude of the source 3 counterpart . the distance to source 3 is unknown a priori . the upper limit on the hydrogen column density derived from the fits to the x - ray spectrum of source 3 ( @xmath75 cm @xmath76 ; table [ tab1 ] ) indicates a distance lower than @xmath77 kpc , assuming the relation between distance and @xmath33 of he , ng & kaspi ( 2013 ) . without a parallax measurement , the distance to source 3 can not be precisely constrained from kinematic measurements of its optical counterpart . the nomad catalogue ( zacharias et al . 2005 ) gives a proper motion of @xmath78 mas yr@xmath7 and @xmath79 mas yr@xmath7 in right ascension and declination , respectively . this corresponds to a spatial velocity of @xmath80 @xmath62 ( d/1 kpc ) km s@xmath7 . if we equate it to the median of the transverse velocity distribution of msps ( @xmath10108 km s@xmath7 , with a standard deviation of @xmath1086 km s@xmath7 ) computed from the australia national telescope facility ( atnf ) pulsar catalogue ( manchester et al . 2005 ) we obtain a distance of @xmath107701500 pc , compatible with the estimate inferred from the @xmath33 . any determination of a lower limit on the distance is more uncertain . again , if source 3 were a binary msp , the distance distribution of known binary msps from the atnf pulsar catalogue gives a probability of @xmath10 0.006 to find one within a radius of 0.2 kpc . thus , the assumed distance range for source 3 ( @xmath74 pc ) is reasonable . we also used the upper limit on the @xmath33 to infer an interstellar extinction along the line of sight @xmath81 , after applying the relation of predehl & schmitt ( 1995 ) . then , we computed the extinction in the different filters using the extinction coefficients of fitzpatrick ( 1999 ) . the reddening vectors are shown in in figure [ grond - cmd ] for the limit case @xmath82 . since the field stars are affected by an unknown interstellar extinction , and the simulations based on the besanon models simply compute a reddening scaled proportionally to the assumed distance in a given direction , introducing a reddening correction in our simulations might bias a direct comparison between the observed and the simulated stellar sequences . therefore , for simplicity , in all cases we simulated the stellar sequences assuming a null reddening . then , we used the reddening vectors as a reference to trace the extinction - corrected locations of the observed points for the source 3 counterpart ( red points ) along the simulated stellar sequences . as seen , the location of the source 3 counterpart in the diagrams falls between the regions of the simulated ms and wd stellar sequences . only for distances as low as @xmath10 0.1 kpc the counterpart location in the diagrams could be compatible with a wd . however , if source 3 is an msp such possibility is unlikely ( see above ) . thus , we conclude that the star is , most likely , not a wd . we built the optical / near - uv / near - ir spectrum of the source 3 counterpart using the available multi - band photometry ( sect . [ cm ] ) . in all cases , we used as a reference the measured ab magnitudes to compute the spectral fluxes at the filter peak wavelengths . as a reference for the interstellar extinction correction we used a maximum extinction value of @xmath82 , derived from the upper limit on the hydrogen column density @xmath33 ( predehl & schmitt 1995 ) obtained with the fit with a pl spectral model ( sect . [ spectra ] ) . we fitted the spectrum with both a single and a double bb spectral model . however , we found that the optical / near - uv / near - ir data can not be simultaneously fitted by a single bb and that a double bb is required to fit the entire spectrum ( @xmath83 , 6 d.o.f . ) . the inferred temperatures are @xmath84 k for the hotter bb , which fits the optical / uv fluxes , and @xmath85 k , for the colder one , which fits the near - ir part of the spectrum . the two bb model is also consistent with the upper limits obtained in the om / uvw2 and uvot / uvm2 filters . the overall spectral shape and spectral parameters do not change significantly when adding the correction for the maximum estimated interstellar extinction ( @xmath86 k , @xmath85 k ) . while the temperature of the hot bb is compatible with the surface temperature of a mid - ms companion star , the temperature of the colder one is too low to be entirely ascribed to the emission from the star surface . although this can contribute to part of the observed near - ir emission , as indicated by the periodic modulation of the j , h and k - band light curves , an additional source external to the star is required to account for the low temperature of the bb that fits the near - ir part of the spectrum . this source might be associated with emission from cold intrabinary gas or dust , maybe the residual of an accretion disk left over by a past phase of matter accretion on the neutron star . the analysis of the counterpart colours as a function of phase ( figure [ grond - lc - diff ] , right ) suggests that its spectrum slightly changes along the @xmath10 0.2245 d period . to quantify this possible evolution we fitted the multi - band spectrum in the four phase intervals defined in sect . [ opt_lc_src3 ] . like in our phase - resolved colour analysis , we can not use the single - epoch flux measurements in the u and uvw1 bands obtained with the _ xmm - newton_/om . as done above , we fitted the four phase - resolved spectra using a two - bb model , considering both null and maximum interstellar extinction . however , we did not find evidence of a significant spectrum evolution across the different phase intervals . this is partially ascribed to the fact the spectra are less constrained at shorter wavelengths without the flux measurements in the u and uvw1 bands . the optical and x - ray emission of source @xmath87 , modulated at a common periodicity of @xmath88 is likely associated with the orbital motion of a tight binary system . the ir / optical / uv spectrum points at a late spectral type ( k or m ) for at least one of the stars in the system ( secn . [ opt_spec ] ) . the other object may either be another non - degenerate star or a compact object , likely an msp . in the first scenario , the x - ray activity , the orbital period and the shape of the optical light curves indicate the system could either be of the w uma or @xmath89 lyr type ( geske et al . . however , the system would be extremely peculiar even for these classes of binaries . the orbital period would be one of the shortest ever observed ; the spectral type one of the latest ; the asymmetry and separation of the peaks in the optical light curves hard to explain . the second scenario , sounds more plausible . moreover , when compared to the @xmath0-ray flux of 3fgl j2039.6@xmath15618 , @xmath90 erg @xmath5 s@xmath7 , the 0.310 kev unabsorbed x - ray flux of source 3 ( f@xmath91 erg @xmath5 s@xmath7 ) would give a @xmath0to x - ray flux ratio @xmath92 , which is consistent with that of msps ( abdo et al . 2013 ; marelli et al . 3fgl j2039.6@xmath15618 is also at a relatively high galactic latitude , with @xmath93 and @xmath94 , like most msps . we note that the nomad proper motion of source 3 in galactic coordinates , @xmath95 mas yr@xmath7 and @xmath96 mas yr@xmath7 , would suggest that it is moving towards the galactic centre from its present location . this , however , would not argue against an msp identification since several msps in the atnf catalogue have a negative proper motion in galactic latitude . being older than a gyr , msps are indeed expected to orbit in the galactic potential and periodically move away and towards the plane . interestingly , if ascribed to an orbital motion , the period of the observed optical flux modulation ( @xmath10 0.2245 d ) would be comparable to the orbital periods of bw and rb msps , which are usually less than one day ( see , e.g. roberts 2013 ) . since most binary msps detected as @xmath0-ray pulsars by the _ fermi_-lat are either bws or rbs , the possible identification of source 3 as a bw / rb system would , then , concur to make it the most likely x - ray counterpart to 3fgl j2039.6@xmath15618 . the x - ray emission model for bw and rb is generally described by a the combination of a thermal component , which originates from the neutron star surface , and a non - thermal component , which primarily originates from an intrabinary shock and from the neutron star magnetosphere ( gentile et al . the x - ray spectrum of source 3 is predominantly non thermal , with a photon index @xmath97 , and its x - ray luminosity in the 0.310 kev energy band is @xmath98 erg s@xmath7 @xmath99 , where @xmath100 is the distance in units of kpc . both the x - ray luminosity and photon index of source 3 are in general agreement with those of rb / bw msps ( gentile et al . 2014 ; roberts 2014 ) , although its relatively hard x - ray spectrum would point more at a rb than a bw . as we noted in sectn.[spectra ] , a thermal component might be present in the x - ray spectrum of source 3 . however , further observations are necessary to clearly discriminate between a purely non - thermal and a composite spectral model . the x - ray light curve of source 3 can be explained assuming a binary msp scenario , the emission from the intrabinary shock is expected to be modulated at the orbital period . recent studies suggest that the x - ray modulation may be due to synchrotron beaming , doppler boosting of the flow within the shock , or obscuration by the companion ( bogdanov et al . 2011 ; gentile et al . 2014 ; roberts 2014 ) . the shape of the x - ray light curve strongly depends on the geometrical and physical parameters of the binary system but , on average , it is characterized by an overall increase of a factor of @xmath1013 around inferior conjunction . the light curves of many bws and rbs show a double - peaked structure due to the obscuration of part of the shock by the companion around the superior conjunction . the inclination angle of the system , as well as the ratio of the companion radius to the intrabinary separation , characterize the phase separation and levels of the two peaks . the x - ray light curve of source 3 clearly points at a bw / rb scenario . its amplitude changes by a factor of @xmath102 during the orbit , the minima occur at orbital phases @xmath1030.1 and @xmath1040.0 , which corresponds to the superior conjunction , whereas the maxima occur at phase @xmath1050.7 , which corresponds to the inferior conjunction . unfortunately , we can not align in phase x - ray and optical light curves ( sectn [ opt_lc_src3 ] ) to confirm this scenario . the optical emission in bws and rbs is dominated by the non - degenerate companion and is characterized by significant luminosity and colour variations . the optical light curve profile can bring the signature of two different effects : the tidal distortion of the companion star surface , due to the gravitational pull of the msp , and the heating of the companion star surface , due to irradiation of relativistic photons and/or high energy particles from the msp ( breton et al . 2013 ) . most bws exhibit a single - peak optical light curve ( e.g. , stappers et al . 2001 ) , due to the irradiation of the nearly fully peeled companion star from the msp . half of the rbs also feature a single peak , whereas the rest feature a two - peaked light curve , with the peaks occurring at specific orbital phases due to the viewing geometry of the tidally - distorted , non - degenerate , companion star . in the latter case , the companion star nearly fills its roche lobes and tidal distortion effects dominate over the heating effects . examples of rbs with single - peak light curves are : psr j1023 + 0038 ( thorstensen & armstrong 2005 ) , j2215 + 5135 ( schroeder & halpern 2014 ) , j2339@xmath10533 ( romani & shaw 2011 ) , and j1227@xmath14853 ( bassa et al . 2014 ) , while examples of rbs with double - peak light curves are : psr j1628@xmath13205 ( li et al . 2014 ) , j1723@xmath12837 ( van staden et al . 2013 ) , j1816 + 4510 ( kaplan et al . 2012 ) , and j2129@xmath10428 ( hui et al . the double - peak light curves of the rbs look very similar to that of source 3 , displaying asymmetries in the relative phase of the two peaks . in particular , the optical light curve of psr j1628@xmath13205 ( li et al . 2014 ) shows the same remarkable asymmetry between the two peaks , with a main and a secondary peak , as observed in the source 3 counterpart . this similarity would , then , support the hypothesis that source 3 is a rb . the @xmath0-ray error ellipse of 3fgl j2039.6@xmath15618 has been observed four times at the csiro parkes telescope searching for pulsations up to a @xmath106 pc @xmath107 ( camilo et al . no pulsations were found , but this is not unexpected , because rbs are very elusive targets in radio . in fact , the intrabinary material ablated from the star causes strong and variable scattering and absorption of radio waves . the radio detection of the pulsations often requires several dedicated long observations ( ray et al . 2013 ; ray et al . , in preparation ) . standard models for the rb optical light curves based on tidal distortion and pulsar irradiation ( e.g. , thorstensen & armstrong 2005 ) can not fit well neither the asymmetric peaks nor the peak separation . therefore , we built a simple three - dimensional model including an additional component related to the asymmetric irradiation of the companion star to fit the light curve of the source 3 counterpart . in this process , we considered only the optical light curves because are those with the highest signal to noise . in order to probe the rb scenario and estimate its physical parameters we built a simple three - dimensional model of a rb binary system with very few free parameters and fit it to the observed optical light curves of source 3 . in this model , the shape of the companion star is approximated by a sphere and a tangent cone pointing to the neutron star . the cone is meant to account for the tidal deformation of the star as it approaches filling its roche lobe . by locking the star rotation to the orbital motion we assumed two different brightnesses for night and day , a characteristic commonly found in rb systems . the asymmetry in the two peaks of the optical light curve implies some asymmetry in the physical system that produces it . therefore , we allowed a tilt angle between the cone axis and the day / night separator line , as measured on the orbital plane . as for other rbs , we assumed a perfectly circular orbit , reducing the number of free parameters in the kepler s orbital parameter space . to summarize , our model accounts for four geometrical parameters : the star deformation ( distance of the cone tip from the star center in star radius units ) ; the night / day asymmetry ( angle between the cone axis and the line of sight to the star ) ; the orbital inclination ( angle between the line of sight and the orbital plane ) ; the epoch of quadrature ( when the projected axis of companion star orbit lies perpendicular to the line of sight ) . furthermore , for each band we have two free parameters : the light curve normalization and the brightness ratio between night and day . we fitted our model to the data , including both statiscal and systematic errors , and found good qualitative agreement for a narrow range of parameters . the overall goodness of the fit turns out to be 197.3 ( 196 d.o.f . ) . the configuration obtained from the best fit to the model implies a deformation of [email protected] for the companion star , indicating that it is subject to strong tidal effects , and a large value of the asymmetry , @xmath108 . the orbital inclination of the binary system obtained from the best fit is @xmath109 , whereas the epoch of quadrature is at mjd [email protected] . finally , the best - fit brightness ratios between the night and day sides in the four bands are : [email protected] [ g ] ; [email protected] [ r ] ; [email protected] [ i ] ; [email protected] [ z ] . the model assumes , as a first approximation , that the night and day sides of the star are in local thermodynamical equilibrium at different temperatures t@xmath110 and t@xmath111 , respectively . a brightness ratio implies a relation between these two temperatures , as a function of the temperature itself . this relation weakly depends on the interstellar extinction , which can only be constrained by our upper limits on the @xmath33 . we computed these relations for each band , adding in quadrature the uncertainty associated with the interstellar extinction correction . the values of temperature ratios are compatible in the four bands , at 1@xmath41 , only for a value of the day - side temperature t@xmath112 k. this constraint does not rely on the normalizations , hence on the accuracy of the photometric calibration . it is a side product of our model , consistent with the spectral results obtained in section 3.3.3 . on the other hand , the model implies a simple relation between the temperature ratio and their absolute value : @xmath113 the large value of the asymmetry , @xmath114 , is implied by the different levels of minima and maxima in the light curves . similar features are observed in other rbs , like psr j1628@xmath13205 ( li et al . 2014 ) , where asymmetry was also proposed as an explanation . in those cases , a measure of this effect was not possible , using the standard modelling tools , which motivated our alternative modelling . in our model , an asymmetry causes a phase shift between the sinusoid that represents the day / night variation and the peaks due to the bump . if the brighter temperature is induced by irradiation from the neutron star , this must be channeled through wind rather than photons . in fact , high - energy radiation travels straight lines , implying a symmetric heating , while particles may follow other trajectories . alternatively , the apparent difference in temperature could be due to very large spots on the surface of the low - mass star . this would not be totally unexpected given the purely convective nature of this kind of stars , and the perturbation induced by the pulsar . if this were the case , a future observation may reveal a different asymmetry , indicating migration of the spots on the surface . we carried out multi - wavelength observations of the unidentified _ fermi_-lat source 3fgl j2039.6@xmath15618 with _ xmm - newton _ and grond . we detected a likely x - ray counterpart ( source 3 ) within the @xmath0-ray error box of 3fgl j2039.6@xmath15618 , which is characterised by a pl x - ray spectrum ( @xmath3 ) that is indicative of strong magnetospheric emission . the upper limit on the hydrogen column density inferred from the x - ray spectral fit , @xmath4 @xmath5 , imply a distance probably lower than 1 kpc . the x - ray light curve of source 3 features a modulation with a period of [email protected] d and two peaks , separated in phase by @xmath70 . using the grond data , we found an optical counterpart to source 3 that features an asymmetric , double - peaked , light curve and a flux modulation with a period of [email protected] d , coincident with that measured in the x rays . if we interpret this periodicity as the orbital period of a compact binary system , source 3 would probably be a binary msp and , as such , a very likely counterpart to 3fgl j2039.6@xmath15618 . in particular , both the optical colors of its putative companion star and the optical light curve profile , with two asymmetric peaks separated in phase by @xmath64 , suggest that source 3/3fgl j2039.6@xmath15618 is a rb . this hypothesis is supported by the lack of apparent radio emission , which would be explained by eclipse of the radio signal as it propagates through the evaporated atmosphere of the irradiated msp companion , as observed in many rbs . optical spectroscopy observations of source 3 will be essential to obtain a more accurate classification of its companion star and measure its radial velocity curve , crucial to confirm the binary msp scenario and determine the orbital parameters of the binary system . a precise determination of the orbital parameters is necessary to fold @xmath0-ray photons and search for the msp pulsations . a timing solution , that can be extended back in time to the launch of _ fermi _ in 2008 , will provide even tighter constraints on the orbital parameters and their evolution . this has a potentiially huge scientific payoff in terms of fundamental physics ( romani et al . , 2012 ; pletsch & clark , 2015 ) the research leading to these results has received funding from the european commission seventh framework programme ( fp7/2007 - 2013 ) under grant agreement n. 267251 . this work was supported by the asi - inaf contract i/004/11/0 , art.22 l.240/2010 for the project `` studio di sorgenti di alta energia con swift '' . cd acknowledges support through extras , funded from the european union s seventh framework programme for research , technological development and demonstration under grant agreement no 607452 . part of the funding for grond ( both hardware as well as personnel ) was generously granted from the leibniz - prize to prof . g. hasinger ( dfg grant ha 1850/28 - 1 ) . cccccccc & j2000 coord . & counts rate & spectral & n@xmath115 & @xmath116 & flux@xmath117}$ ] & variability + & ra dec [ @xmath118 ( stat . & @xmath49 cts / s & model & @xmath119 @xmath5 & & @xmath120erg @xmath5 s@xmath7 & @xmath39 ( d.o.f . ) + 3 & 309.8956 -56.2861 ( 0.3 ) & [email protected] & pl & @xmath121 & [email protected] & 10.19@xmath122 & 66.18 ( 16 ) + 11 & 309.8439 -56.3292 ( 0.6 ) & [email protected] & pl & @xmath123 & 1.33@xmath124 & 5.73@xmath125 & 3.92 ( 8) + 13 & 309.9995 -56.3359 ( 0.5 ) & [email protected] & pl / ap & @xmath126 & 2.07@xmath127 & 2.66@xmath128 & 4.0 ( 8) + 19 & 309.8170 -56.2838 ( 0.9 ) & [email protected] & pl / ap & @xmath129 & 1.54@xmath130 & 2.98@xmath131 & 3.8 ( 8) + 22 & 309.8713 -56.3320 ( 0.7 ) & [email protected] & pl / ap & @xmath132 & 1.80@xmath133 & 3.37@xmath134 & 11.9 ( 8) + 24 & 309.9619 -56.2989 ( 0.8 ) & [email protected] & pl / ap & @xmath135 & 1.86@xmath136 & 1.24@xmath137 & 4.1 ( 8) + 29 & 309.8415 -56.3154 ( 1.0 ) & [email protected] & pl / ap / bb & @xmath138 & 2.37@xmath139 & 1.36@xmath140 & 6.2 ( 8) + 31 & 309.8172 -56.2948 ( 0.8 ) & [email protected] & pl / ap / bb & @xmath141 & 1.80@xmath142 & 3.31@xmath143 & 7.4 ( 8) + 40 & 309.9936 -56.3097 ( 1.1 ) & [email protected] & pl / ap & 0.05 & 2 & 0.63@xmath144 & 6.25 ( 6 ) + 44 & 309.8651 -56.2619 ( 1.5 ) & [email protected] & pl / ap / bb & 0.05 & 2 & 0.11@xmath145 & 5.5 ( 5 ) + 52 & 309.9451 -56.3099 ( 1.3 ) & [email protected] & pl / bb & @xmath146 & 1.11@xmath147 & 1.67@xmath148 & 2.6 ( 5 ) + 60 & 309.9159 -56.3583 ( 1.4 ) & [email protected] & pl / ap / bb & @xmath149 & 2.33@xmath150 & 0.49@xmath151 & 4.4 ( 4 ) + 72 & 309.8419 -56.3541 ( 1.6 ) & [email protected] & pl / ap / bb & 0.05 & 2 & 0.74@xmath152 & 0.7 ( 3 ) + 76 & 309.9548 -56.3170 ( 2.0 ) & [email protected] & pl / ap / bb & 0.05 & 2 & 0.68@xmath153 & 1.0 ( 3 ) + 85 & 309.8740 -56.3390 ( 1.6 ) & [email protected] & pl / ap & 0.05 & 2 & 0.74@xmath154 & 0.64 ( 3 ) + 88 & 309.9357 -56.2766 ( 1.7 ) & [email protected] & & & & & 8.89 ( 3 ) + c|ccccccc|ccc|ccc|cccccc i d & & & & + & g & r & i & z & j & h & k & j & h & k@xmath22 & u & uvw1 & uvm2 & uvw2 + 3 & 19.40 & 18.73 & 18.72 & 18.61 & 18.53 & 18.15 & 18.34 & 18.24 & 18.24 & 18.64 & & 21.26 & & 21.88 + 11 & & & & & & & & 20.33 & 20.02 & 20.73 & & & & + 13 & & & 22.61 & & & & & & & 20.70 & & 21.27 & & + 19 & & & & & & & & & & 20.64 & & & & + 22 & & & & & 19.37 & & 17.88 & & & & & & & + & & & & & 18.02 & 17.96 & & 17.84 & 17.93 & 17.97 & & & & + 24 & 24.21 & & & & & & & & & & & & & + 29 & & & & & & & & & & & 20.83 & 20.39 & & + 40 & 21.95&21.22 & 21.45 & 22.92 & & & & & 19.82 & 19.85 & 20.10 & 22.08 & & & + 72 & & & & & & & & & 19.66 & & & & & + 60 & & & & & & & & & & & & 19.33 & & & + 76 & 19.64 & 18.36 & 18.63 & 18.20 & 17.65&17.66&17.77 & 17.63 & 17.58 & 17.90 & & & & + & & & & 21.52 & & & & 19.36 & 18.84 & 18.64 & & & & + & & & & 23.97 & & & & & & & & & & + 88 & & & & & & & & 20.93 & 21.31 & & & & & + ) . the vertical dashed lines define the main peak ( @xmath67=0.20.5 ) , the secondary peak ( @xmath67=0.70.9 ) , the `` bridge '' ( @xmath67=0.50.7 ) , and the `` off - peak '' ( @xmath67=0.00.2 and @xmath67=0.91.0 ) regions . [ grond - lc],width=529 ] 5618 field obtained from the grond time - averaged photometry . _ bottom _ : observed cc diagram . in all panels , the location of field stars is indicated by the black filled circles , whereas that of the optical counterpart of the _ xmm - newton _ source 3 is indicated by the red filled triangle . the filled green triangles indicates the counterpart location computed from the photometry computed on the single image . stellar sequences simulated from the besanon models for different values of distance are shown in light and dark grey . in the cm diagrams the dark grey regions correspond to distance values @xmath155 pc , whereas in the cc diagram they correspond to magnitudes within @xmath2 0.05 the g-band magnitude of the source 3 counterpart . the ms and wd branches are labelled . [ grond - cmd],title="fig:",width=340 ] 5618 field obtained from the grond time - averaged photometry . _ bottom _ : observed cc diagram . in all panels , the location of field stars is indicated by the black filled circles , whereas that of the optical counterpart of the _ xmm - newton _ source 3 is indicated by the red filled triangle . the filled green triangles indicates the counterpart location computed from the photometry computed on the single image . stellar sequences simulated from the besanon models for different values of distance are shown in light and dark grey . in the cm diagrams the dark grey regions correspond to distance values @xmath155 pc , whereas in the cc diagram they correspond to magnitudes within @xmath2 0.05 the g-band magnitude of the source 3 counterpart . the ms and wd branches are labelled . [ grond - cmd],title="fig:",width=340 ] 5618 field obtained from the grond time - averaged photometry . _ bottom _ : observed cc diagram . in all panels , the location of field stars is indicated by the black filled circles , whereas that of the optical counterpart of the _ xmm - newton _ source 3 is indicated by the red filled triangle . the filled green triangles indicates the counterpart location computed from the photometry computed on the single image . stellar sequences simulated from the besanon models for different values of distance are shown in light and dark grey . in the cm diagrams the dark grey regions correspond to distance values @xmath155 pc , whereas in the cc diagram they correspond to magnitudes within @xmath2 0.05 the g-band magnitude of the source 3 counterpart . the ms and wd branches are labelled . [ grond - cmd],title="fig:",width=340 ] , r@xmath19 , i@xmath19 and z@xmath19 bands are marked by green , red , magenta and cyan circles . only statistical errors are plotted . the black lines display the best - fit light curve calculated using the model described in sec . [ opt_model ] . [ grond - lc - model],width=453 ]

we describe a number of recently developed cluster - flipping algorithms for the efficient simulation of classical spin models near their critical temperature . these include the algorithms of wolff , swendsen and wang , and niedermeyer , as well as the limited cluster algorithm , the multigrid methods of kandel and co - workers , and the invaded cluster algorithm . we describe the application of these algorithms to ising , potts , and continuous spin models . # 1([#1 ] ) # 1[#1 ] # 1[#1 ] # 1[#1 ] # 1#1 # 1 # 1to = # 1to # 1to # 1to # 1#2#3#4#5#6#7#8#9to to to # 1#2#3#4to # 1#2to monte carlo simulations of classical spin systems such as the ising model can usually be performed simply and very efficiently using local update algorithms such as the metropolis or heat - bath algorithms . however , in the vicinity of a continuous phase transition , these algorithms display dramatic critical slowing down , making them extremely poor tools for the study of critical phenomena . in the last ten years or so , a number of new monte carlo algorithms making use of non - local update moves have been developed to address this problem . in this chapter we examine a number of these algorithms in some detail , and discuss their strengths and weaknesses . the outline of the chapter is as follows . in section , we briefly discuss the problem of critical slowing down . in section we describe the wolff algorithm , which is probably the most successful of the algorithms developed so far . in section we discuss a number of other algorithms , such as the swendsen - wang algorithm , niedermayer s algorithm , and the invaded cluster algorithm . each of these algorithms is introduced first in the context of the ising model , but in section we discuss how they may be generalized to systems with many valued spins such as potts models or continuous valued spins such as the classical xy and heisenberg models . in section we give our conclusions .

cond-mat9703179

the divergence of the correlation length @xmath1 of a classical spin system as it approaches its critical temperature means that larger and larger volumes of spins must be updated coherently in order to sample statistically independent configurations . in monte carlo simulations this gives rise to a correlation time @xmath2 measured in monte carlo steps per site which diverges with @xmath1 as @xmath3 where @xmath4 is a dynamic exponent whose value depends on the update method . for a system of finite dimension @xmath5 close to the critical temperature , this means that the amount of cpu time required to generate an independent lattice configuration increases as @xmath6 , where @xmath7 is the dimensionality of the lattice . if we take the example of the ising model and make the assumption that the movement of domain walls under a dynamics employing a local update move is diffusive , then the number of monte carlo steps required to generate an independent spin configuration should scale as @xmath8 , and hence @xmath9 . the scaling arguments of bausch ( 1981 ) indicate that this is in fact a lower bound on the value of @xmath4 for all dimensions up to four , and numerical investigations have in general measured values slightly higher than this . the most accurate determination of @xmath4 of which we are aware is that of nightingale and blte ( 1996 ) , who measured a value of @xmath10 for the dynamic exponent of the two - dimensional ising model with a local update algorithm . thus the cpu time required to perform a simulation of this model with a given degree of accuracy increases slightly faster than @xmath11 with system size for temperatures close to the phase transition , which severely limits the system sizes which can be studied , even with the fastest computers . one solution to this problem is to try and find monte carlo algorithms which can update volumes on the order of @xmath12 in one step , rather than by the slow diffusion of domain walls . since @xmath1 can become arbitrarily large , this implies an algorithm with a true non - local update move one capable of updating an arbitrary number of spins in one time - step . in the first part of this chapter we will study one of the most widely used and successful such algorithms , the wolff algorithm , which has a measured critical exponent close to zero for the ising model , a great improvement over the metropolis case an elegant non - local monte carlo algorithm applicable to the ising model has been proposed by wolff ( 1989 ) , based on previous work by swendsen and wang ( 1987 ) . at each monte carlo step , this algorithm flips one contiguous cluster of similarly - oriented spins . clusters are built up by starting from a randomly - chosen seed spin and adding further spins if they are ( a ) adjacent to a spin which already belongs to the cluster and ( b ) pointing in the same direction as the spins in the cluster . spins which satisfy these criteria are added to the cluster with some probability @xmath13 . eventually the cluster will stop growing when all possible candidates for addition have been considered and at this point the cluster is flipped over with some acceptance probability @xmath14 . it is straightforward to calculate what value @xmath14 should take for a given choice of @xmath13 . consider two states of the system , @xmath15 and @xmath16 , illustrated in figure . they differ from one another by the flipping of a single cluster of similarly - oriented spins . the crucial thing to notice is the way the spins are oriented around the edge of the cluster ( which is indicated by the line in the figure ) . notice that in each of the two states , some of the spins just outside the cluster are pointing the same way as the spins in the cluster . the bonds between these spins and the ones in the cluster have to be broken when the cluster is flipped to reach the other state . inevitably , those bonds which are not broken in going from @xmath15 to @xmath16 must be broken if we make the reverse move from @xmath16 to @xmath15 . [ frames ] consider now a cluster monte carlo move which takes us from @xmath15 to @xmath16 . there are in fact many such moves we could choose any of the spins in the cluster as our seed spin , and then we could add the rest of the spins to it in a variety of orders . for the moment , however , let us just consider one particular move , starting with a particular seed spin and then adding others to it in a particular order . consider also the reverse move , which takes us back to @xmath15 from @xmath16 , starting with the same seed spin , and adding others to it in exactly the same order as the forward move . the probability of choosing the seed is the same in the two directions , as is the probability of adding each spin to the cluster . the only thing which changes between the two is the probability of `` breaking '' bonds around the edge of the cluster because the bonds which have to be broken are different in the two cases . suppose that , for the forward move , there are @xmath17 bonds which have to be broken in order to flip the cluster . these broken bonds represent correctly - oriented spins which were not added to the cluster by the algorithm . the probability of not adding such a spin is @xmath18 . thus the probability of not adding all of them , which is proportional to the selection probability @xmath19 for the forward move , is @xmath20 . if there are @xmath21 bonds which need to be broken in the reverse move then the probability of doing it will be @xmath22 , which is proportional to @xmath23 . the condition of detailed balance then tells us that @xmath24 where @xmath25 and @xmath26 are the acceptance ratios for the moves in the two directions . the change in energy @xmath27 between the two states also depends on the bonds which are broken . for each of the @xmath17 bonds which are broken in going from @xmath15 to @xmath16 , the energy changes by @xmath28 , where @xmath29 is the interaction strength between neighbouring spins . for each of the @xmath21 bonds which are made , the energy changes by @xmath30 . thus @xmath31 substituting into equation and rearranging we derive the following condition on the acceptance ratios : @xmath32^{n - m}. \label{wolffaccept}\ ] ] but now we notice a delightful fact : if we choose @xmath33 then the right - hand side of equation is just 1 , independent of any properties of the states @xmath15 and @xmath16 , or the temperature , or anything else at all . with this choice , we can make the acceptance ratios for both forward and backward moves unity , which is the best possible value they could take . every move we propose is accepted and the algorithm still satisfies detailed balance . the choice then defines the wolff cluster algorithm for the ising model , whose precise statement would go something like this : 1 . choose a seed spin at random from the lattice . 2 . look in turn at each of the neighbours of that spin . if they are pointing in the same direction as the seed spin , add them to the cluster with probability @xmath34 . 3 . for each spin that was added in the last step , examine each of _ its _ neighbours to find the ones which are pointing in the same direction and add each of them to the cluster with the same probability @xmath13 . ( notice that as the cluster becomes larger we may find that some of the neighbours are already members of the cluster , in which case obviously you do nt have to consider adding them again . also , some of the spins may have been considered for addition before , as neighbours of other spins in the cluster , but rejected . in this case , they get another chance to be added to the cluster on this step . ) this step is repeated as many times as necessary until there are no spins left in the cluster whose neighbours have not been considered for inclusion in the cluster . 4 . flip all the spins in the cluster . the flipping of these clusters is a much less laborious task than in the case of the metropolis algorithm as we shall see it takes a time proportional to the size of the cluster to grow it and then turn it over so we have every hope that the algorithm will indeed have a lower dynamic exponent and less critical slowing down . in section we show that this is indeed the case . first we look briefly at how the wolff algorithm is implemented . the standard way to implement the wolff algorithm is a very straightforward version of the list of steps given above . first , we choose at random one of the spins on the lattice to be our seed . we look at each of its neighbours , to see if it points in the same direction as the seed spin . if it does then with probability @xmath13 we add that spin to the cluster and we store its coordinates on a stack . when we have exhausted the neighbours of the spin we are looking at , we pull a spin off the stack , and we start checking _ its _ neighbours one by one . any new spins added to the cluster are again also added to the stack , and we go on pulling spins off the stack and looking at their neighbours , until the stack is empty . this algorithm guarantees that we consider the neighbours of every spin added to the cluster , as we should . furthermore , the amount of time spent on each spin in the cluster is , on average , the same . each one gets added to the stack , pulled off again at some later stage and has all of its neighbours considered as candidates for addition to the cluster . the time taken to flip over all the spins in the cluster also goes like the number of spins , so the entire time taken by one monte carlo move in the wolff algorithm is proportional to the number of spins in the cluster . in this section we look at how the wolff algorithm actually performs in practice . as we will see , it performs extremely well when we are near the critical temperature , but is actually a little slower than the metropolis algorithm at very high or low temperatures . [ wolffsnaps ] [ csize ] figure shows a series of states generated by the algorithm at each of three temperatures , one above @xmath35 , one close to @xmath35 , and one below it . up and down spins are represented by the black and white dots . consider first the middle set of states , the ones near @xmath35 . if you examine the four frames , it is not difficult to make out which cluster flipped at each step . clearly the algorithm is doing its job , flipping large areas of spins when we are in the critical region . in the @xmath36 case , it is much harder to make out the changes between one frame and the next . the reason for this is that in this temperature region @xmath13 is quite small ( see equation ) and this in turn makes the clusters small , so it is hard to see when they flip over . of course , this is exactly what the algorithm is supposed to do , since the correlation length here is small , and we do nt expect to get large regions of spins flipping over together . in figure , we have plotted the mean size of the clusters flipped by the wolff algorithm over a range of temperatures , and , as we can see , it does indeed become small at large temperatures . when the temperature gets sufficiently high ( around @xmath37 in two dimensions ) , the mean size of the clusters becomes hardly greater than a single spin . in other words , the single seed spin for each cluster is being flipped over with probability one at each step , but none of its neighbours are . however , this is just exactly what the single - spin flip metropolis algorithm does in this temperature regime . when @xmath38 is large , the metropolis acceptance ratio is 1 , or very close to it , for any transition between two states @xmath15 and @xmath16 . thus in the limit of high temperatures , the wolff algorithm and the metropolis algorithm become the same thing . but notice that the wolff algorithm will actually be the slower of the two in this case , because for each seed spin it has to go through the business of testing each of the neighbours for possible inclusion in the cluster , whereas the metropolis algorithm only has to decide whether to flip a single spin or not , a comparatively simple computational task . thus , even if the wolff algorithm is a good thing near to the phase transition ( which it is ) , there comes a point as the temperature increases where the metropolis algorithm becomes better . now let us turn to the simulation at low @xmath38 . looking at the bottom row of figure , the action of the algorithm is dramatically obvious in this case almost every spin on the lattice is being flipped at each step . the reason for this is clear . when we are well below the critical temperature the ising model develops a backbone of similarly - oriented spins which spans the entire lattice . when we choose our seed spin for the wolff algorithm , it is likely that we will land on one of the spins comprising this backbone . furthermore , the probability @xmath13 , equation , is large when @xmath38 is small , so the neighbours of the seed spin are not only likely to be aligned with it , but are also likely to be added to the growing cluster . the result is that the cluster grows ( usually ) to fill almost the entire backbone of spontaneously magnetized spins , and then they are all flipped over in one step . on the face of it , this seems like a very inefficient way to generate states for the ising model . after all , we know that what should really be happening in the ising model at low temperature is that most of the spins should be lined up with one another , except for a few excitation spins , which are pointing the other way ( see the figure ) . every so often , one of these excitation spins flips back over to join the majority pointing the other way , or perhaps one of the backbone spins gets flipped by a particularly enthusiastic thermal excitation and becomes a new excitation spin . this of course is exactly what the metropolis algorithm does in this regime . the wolff algorithm on the other hand is removing the single - spin excitations by the seemingly extravagant measure of _ flipping all the other spins in the entire lattice _ to point the same way as the single lonely excitation . in fact , however , it s actually not such a stupid thing to do . first , let us point out that , since the wolff algorithm never flips spins which are pointing in the opposite direction to the seed spin , all the excitation spins on the entire lattice end up pointing the same way as the backbone when the algorithm flips over a percolating cluster . therefore , the wolff algorithm gets rid of _ all _ the excitation spins on the lattice in a single step . second , since we know that the wolff algorithm generates states with the correct boltzmann probabilities , it must presumably create some new excitation spins at the same time as it gets rid of the old ones . and indeed it does do this , as is clear from the frames in the figure . each spin in the backbone gets a number of different chances to be added to the cluster ; for most of the spins this number is just the lattice coordination number @xmath4 , which is four in the case of the square lattice . thus the chance that a spin will not be added to the cluster at all is @xmath39 . this is a small number , but still non - zero . ( it is @xmath40 at the temperature @xmath41 used in the figure . ) thus there will be a small number of spins in the backbone which get left out of the percolating cluster and are not flipped along with the others . these spins become the new excitations on the lattice . the net result of all this is that after only one wolff monte carlo step , all the old excitations have vanished and a new set have appeared . this behaviour is clear in figure . at low temperatures , the wolff algorithm generates a complete new configuration of the model at every monte carlo step , though the payoff is that it has to flip very nearly every spin on every step too . in the metropolis algorithm at low temperatures it turns out that you have to do about one monte carlo step per site ( each one possibly flipping one spin ) to generate a new independent configuration of the lattice . to see this , we need only consider again the excitation spins . the average acceptance ratio for flipping these over will be close to 1 in the metropolis algorithm , because the energy of the system is usually lowered by flipping them . ( only on the rare occasions when several of them happen to be close together might this not be true . ) thus , it will on average just take @xmath42 steps , where @xmath42 is the number of sites on the lattice , before we find any particular such spin and flip it in other words , one monte carlo step per site . but we know that the algorithm correctly generates states with their boltzmann probability , so , in the same @xmath42 steps that it takes to find all the excitation spins and flip them over to join the backbone , the algorithm must also choose a roughly equal number of new spins to excite out of the backbone , just as the wolff algorithm also does . thus , it takes one sweep of the lattice to replace all the excitation spins with a new set , generating an independent state of the lattice . ( this is in fact the best possible performance that any monte carlo algorithm can have , since one has to allow at least one sweep of the lattice in order to generate a new configuration , so that each spin gets the chance to change its value . ) again then , we find that the wolff algorithm and the metropolis algorithm are roughly comparable , this time at low temperatures . both need to consider about @xmath42 spins for flipping in order to generate an independent state of the lattice . again , however , the additional complexity of the wolff algorithm is its downfall , and the extremely simple metropolis algorithm has the edge in speed . so , if the metropolis algorithm beats the wolff algorithm ( albeit only by a slim margin ) at both high and low temperatures , that leaves only the intermediate regime , close to @xmath35 in which the wolff algorithm might be worthwhile . this of course is the regime in which we designed the algorithm to work well , so we have every hope that it will beat the metropolis algorithm there , and indeed it does , very handily . before we measure the correlation time of the wolff algorithm , we need to consider exactly how it should be defined . if we are going to compare the correlation times of the wolff algorithm and the metropolis algorithm near to the phase transition as a way of deciding which is the better algorithm in this region , it clearly would not be fair to measure it for both algorithms in terms of number of monte carlo steps . a single monte carlo step in the wolff algorithm is a very complicated procedure , flipping maybe hundreds of spins and potentially taking quite a lot of cpu time , whereas the metropolis monte carlo step is a very simple , quick thing we can do a million of them in a second on a good computer though each one only flips at most one spin . in section we showed that the time taken to complete one step of the wolff algorithm is proportional to the number of spins @xmath43 in the cluster . such a cluster covers a fraction @xmath44 of the entire lattice , and so on average each monte carlo step will take an amount of cpu time comparable to @xmath45 sweeps of the lattice using the single - spin - flip metropolis algorithm . thus the correct way to define the correlation time is to write @xmath46 , where @xmath47 is the correlation time measured in steps ( i.e. , clusters flipped ) in the wolff algorithm . the conventional choice for the constant of proportionality is 1 . this makes the correlation times for the wolff and metropolis algorithms equal in the limits of low and high temperature , for the reasons discussed in the last section . this is not quite fair , since , as we pointed out earlier , the wolff algorithm is a little slower than the metropolis in these regimes because of its greater complexity . however , this difference is slight compared with the enormous difference in performance between the two algorithms near the phase transition which we will witness in a moment , so for all practical purposes we can write @xmath48 in our own simulations of the 2d ising model on a @xmath49 square lattice using the two algorithms , we measure the correlation time at @xmath35 of the wolff algorithm to be @xmath50 spin - flips per site , by contrast with the metropolis algorithm which has @xmath51 . a factor of a thousand certainly out - weighs any difference in the relative complexity of the two algorithms . it is this impressive performance on the part of the wolff algorithm which makes it a worthwhile one to use if we are interested in the behaviour of the model close to @xmath35 . [ taulw ] in figure we have plotted on logarithmic scales the correlation time of the wolff algorithm for the 2d ising model at the critical temperature , over a range of different system sizes . the slope of the line gives us an estimate of the dynamic exponent . our best fit , given the errors on the data points is @xmath52 . this was something of a rough calculation , though our result is competitive with other more thorough ones . the best available figure at the time of writing was that of coddington and baillie ( 1992 ) who measured @xmath53 . this figure is clearly much lower than the @xmath54 of the metropolis algorithm , and gives us a quantitative measure of how much better the wolff algorithm really is . in fact in studies of the wolff algorithm for the 2d ising model one does not usually bother to make use of equation to calculate @xmath2 . if we measure time in monte carlo steps ( that is , simple cluster flips ) , we can define the corresponding dynamic exponent @xmath55 in terms of the correlation time @xmath56 of equation thus : @xmath57 the exponent @xmath55 is related to the real dynamic exponent @xmath4 for the algorithm by @xmath58 where @xmath59 and @xmath16 are the critical exponents governing the divergences of the magnetic susceptibility and the correlation length . if we know the values of @xmath16 and @xmath59 , as we do in the case of the 2d ising model , then we can use equation to calculate @xmath4 without having to measure the mean cluster size in the algorithm , which eliminates one source of error in the measurement . directly , as we did in figure . ] the first step in demonstrating equation is to prove another useful result , about the magnetic susceptibility @xmath60 . it turns out that , for temperatures @xmath61 , the susceptibility is related to the mean size @xmath62 of the clusters flipped by the wolff algorithm thus : @xmath63 in many simulations using the wolff algorithm , the susceptibility is measured using the mean cluster size in this way . the demonstration of equation goes like this . instead of implementing the wolff algorithm in the way described in this chapter , imagine instead doing it a slightly different way . imagine that at each step we look at the whole lattice and for every pair of neighbouring spins which are pointing in the same direction , we make a `` link '' between them with probability @xmath34 . when we are done , we will have divided the whole lattice into many different clusters of spins , as shown in figure , each of which will be a correct wolff cluster ( since we have used the correct wolff probability @xmath13 to make the links ) . now we choose a single seed spin from the lattice at random , and flip the cluster to which it belongs . then we throw away all the links we have made and start again . the only difference between this algorithm and the wolff algorithm as we described it is that here we make the clusters first and choose the seed spin afterwards , rather than the other way around . this would not be a very efficient way of implementing the wolff algorithm , since it requires us to create clusters all over the lattice , almost all of which never get flipped , but it is a useful device for proving equation . well , we can write the total magnetization @xmath64 of the lattice as a sum over all the clusters on the lattice thus : @xmath65 here @xmath66 labels the different clusters , @xmath67 is their size ( a positive integer ) , and @xmath68 depending on whether the cluster is pointing up or down , thereby making either a positive or a negative contribution to @xmath64 . the mean square magnetization is then @xmath69 now , since the average of the magnetization is zero under the wolff algorithm ( this is even true below @xmath35 , since the algorithm flips most of the spins on the lattice at every step at low temperature ) , the first term in this expression is an average over a large number of quantities which are randomly either positive or negative and which will therefore tend to average out to zero . the second term on the other hand is an average over only positive quantities and therefore is not zero . noting that @xmath70 for all @xmath66 , we can then write @xmath71 or alternatively , in terms of the magnetization per spin @xmath17 : @xmath72 now consider the average @xmath62 of the size of the clusters which get flipped in the wolff algorithm . this is not quite the same thing as the average size @xmath73 of the clusters over the entire lattice , because when we choose our seed spin for the wolff algorithm it is chosen at random from the entire lattice , which means that the probability @xmath74 of it falling in a particular cluster @xmath66 , is proportional to the size of that cluster : @xmath75 the average cluster size in the wolff algorithm is then given by the average over the probability of the cluster being chosen times the size of that cluster : @xmath76 given that @xmath77 for @xmath61 , we now have our result for the magnetic susceptibility as promised : @xmath78 we can now use this equation to rewrite equation thus : @xmath79 which implies in turn that @xmath80 near the critical point , this implies that @xmath81 and thus @xmath82 as we suggested earlier on . [ tausteps ] in figure we have replotted the results from our wolff algorithm simulations of the 2d ising model using the correlation time measured in monte carlo steps ( i.e. , cluster flips ) . the best fit to the data gives a figure of @xmath83 . using the accepted values @xmath84 and @xmath85 for the 2d ising model ( onsager 1944 ) , and setting @xmath86 , equation then gives us @xmath53 , which is as good as the best published result for this exponent . we have looked in detail at the wolff algorithm for simulating the ising model . close to @xmath35 this algorithm is significantly more efficient than the metropolis one ; although it is more complex than the metropolis algorithm , the wolff algorithm has a very small dynamic exponent , which means that the time taken to perform a simulation scales roughly like the size of the system , which is the best that we can hope for in any algorithm . however , many other algorithms have been suggested for the simulation of the ising model . in this section we talk about a few of these alternative algorithms . after the metropolis and wolff algorithms , probably the most important other algorithm is the algorithm of swendsen and wang ( 1987 ) . in fact this algorithm is very similar to the wolff algorithm and wolff took the idea for his algorithm directly from it . ( swendsen and wang took the idea in turn from the work of fortuin and kasteleyn ( 1972 ) and sweeny ( 1983 ) . ) actually , we have seen the central idea behind the swendsen - wang algorithm already . in section we considered an alternative implementation of the wolff algorithm in which the entire lattice of spins is divided up into clusters by making `` links '' with probability @xmath87 between similarly - oriented neighbouring spins . we then imagined choosing a single spin from the lattice and flipping over the whole of the cluster to which it belongs . this procedure is clearly equivalent to the wolff algorithm , although in practice it would be an inefficient way of carrying it out . the swendsen - wang algorithm divides the entire lattice into clusters in exactly the same way , with this same probability @xmath13 of making a link . but then , instead of flipping just one cluster , each cluster is independently flipped with probability @xmath88 . we notice the following facts about this algorithm : 1 . the algorithm satisfies the condition of detailed balance . the proof of this fact is exactly the same as it was for the wolff algorithm . if the number of links broken and made in performing a move are @xmath17 and @xmath21 respectively ( and the reverse for the reverse move ) , then the energy change for the move is @xmath89 ( or @xmath90 for the reverse move ) . the selection probabilities for choosing a particular set of links differs between forward and reverse moves only at the places where bonds are made or broken , and so the ratio of the two selection probabilities is @xmath91 , just as it was before . by choosing @xmath92 , we then ensure , just as before , that the acceptance probability is independent of @xmath17 and @xmath21 and everything else , so any choice which makes it the same in each direction , such as flipping all clusters with probability @xmath88 will make the algorithm correct . notice however that other choices would also work . it does nt matter how we choose to flip the clusters , though the choice made here is good because it minimizes the correlation between the direction of a cluster before an after a move , the new direction being chosen completely at random , regardless of the old one . the algorithm updates the entire lattice on each move . in measuring correlation times for this algorithm , one should therefore measure them simply in numbers of monte carlo steps , and not steps per site as with the metropolis algorithm . ( in fact , on average , only half the spins get flipped on each move , but the number flipped scales like the size of the system , which is the important point . ) the swendsen - wang algorithm is essentially the same as the wolff algorithm for low temperatures . well below @xmath35 , one of the clusters chosen by the algorithm will be a percolating cluster , and the rest will correspond to the `` excitations '' discussed in section , which will be small . ignoring these small clusters then , the swendsen - wang algorithm will tend to turn over the percolating backbone of the lattice on average every two steps ( rather than every step as in the wolff algorithm see figure ) , but otherwise the two will behave almost identically . thus , as with the wolff algorithm , we can expect the performance of the swendsen - wang algorithm to be similar to that of the metropolis algorithm at low @xmath38 , though probably a little slower on average due to the complexity of the algorithm . 4 . at high temperatures , the swendsen - wang algorithm tends to divide the lattice into very small clusters because @xmath13 becomes small . as @xmath93 the clusters will just be one spin each , and the algorithm will just change all the spins to new random values on each move . this is also what the metropolis algorithm does at high temperatures in one sweep of the lattice , though again the metropolis algorithm can be expected to be a little more efficient in this regime , since it is a simpler algorithm which takes few operations on the computer to flip each spin . the combination of the last two points here implies that the only regime in which the swendsen - wang algorithm can be expected to out - perform the metropolis algorithm is the one close to the critical temperature . the best measurement of the dynamic exponent of the algorithm is that of coddington and baillie ( 1992 ) , who found @xmath53 in two dimensions , which is clearly much better than the metropolis algorithm , and is in fact exactly the same as the result for the wolff algorithm . so the swendsen - wang algorithm is a pretty good algorithm for investigating the 2d ising model close to its critical point . however , as table shows , for higher dimensions the swendsen - wang algorithm has a significantly higher dynamic exponent than the wolff algorithm , making it slower close to @xmath35 . the reason is that close to @xmath35 the properties of the ising model are dominated by the fluctuation of large clusters of spins . as the arguments of section showed , the wolff algorithm preferentially flips larger clusters because the chance of the seed spin belonging to any particular cluster is proportional to the size of that cluster . the swendsen - wang algorithm on the other hand treats all clusters equally , regardless of their size , and therefore wastes a considerable amount of effort on small clusters which make vanishingly little contribution to the macroscopic properties of the system for large system sizes . this , coupled with the fact that the swendsen - wang algorithm is slightly more complicated to program than the wolff algorithm , makes the wolff algorithm the algorithm of choice for most people . it is worth noting however , that the work of ferrenberg , landau , and wong ( 1992 ) appears to indicate that the wolff algorithm is unusually susceptible to imperfections in the random number generator used to implement the algorithm . although this result is probably highly sensitive to the way in which the algorithm is coded , it may be that the swendsen - wang algorithm would be a more sensible choice for those who are unsure of the quality of their random numbers . [ cols="^,^,^,^",options="header " , ] [ zval ] another variation on the general cluster algorithm theme was proposed by niedermayer ( 1988 ) . his suggestion is really just an extension of the ideas used in the wolff and swendsen - wang algorithms . in fact , niedermayer s methods are very general and can be applied to all sorts of models including glassy spin models . here we will just consider their application to the ordinary ising model . niedermayer pointed at that it is not necessary to constrain the `` links '' with which we make clusters to be only between spins which are pointing in the same direction . in general , we can define two different probabilities for putting links between sites one for parallel spins and one for anti - parallel ones . the way niedermayer expressed it , he considered the energy contribution @xmath94 that a pair of spins @xmath66 and @xmath95 makes to the hamiltonian . in the case of the ising model , for example , @xmath96 he then wrote the probability for making a link between two neighbouring spins as a function of this energy @xmath97 . in the ising model @xmath94 can only take two values @xmath98 , so the function @xmath99 only needs to be defined at these points , but for some of the more general models niedermayer considered it needs to be defined elsewhere as well . clearly , if for the ising model we make @xmath100 and @xmath101 , then we recover the wolff algorithm or the swendsen - wang algorithm , depending on whether we flip only a single cluster on each move , or many clusters over the entire lattice niedermayer s formalism is applicable in either case . to be concrete about things , let us look at the case of the single - cluster , wolff - type version of the algorithm . let us apply the condition of detailed balance to the algorithm . consider , as we did in the case of the wolff algorithm , two states of our system which differ by the flipping of a single cluster . ( you can look again at figure if you like , but bear in mind that , since we are now allowing links between anti - parallel spins , not all the spins in the cluster need be pointing in the same direction . ) as before , the probability of forming the cluster itself is exactly the same in the forward and reverse directions , except for the contributions which come from the borders . at the borders , there are some pairs or spins which are parallel and some which are anti - parallel . suppose that in the forward direction there are @xmath17 pairs of parallel spins at the border these correspond to bonds which have to be broken in flipping the cluster and @xmath21 pairs which are anti - parallel bonds which will be made when we flip . by definition no links are made between any of these border pairs , and the probability of that happening is @xmath102^m [ 1 - p_{\rm add}(-j)]^n$ ] . in the reverse direction the corresponding probability is @xmath102^n [ 1 - p_{\rm add}(-j)]^m$ ] . just as in the wolff case , the energy cost of flipping the cluster from state @xmath15 to state @xmath16 is @xmath31 thus , the appropriate generalization of the acceptance ratio relation , equation , is @xmath103^{n - m}. \label{eqniedaccept}\ ] ] any choice of acceptance ratios @xmath25 and @xmath26 which satisfies this relation will satisfy detailed balance . for the wolff choice of @xmath13 we get acceptance ratios which are always unity , but niedermayer pointed out that there are other ways to achieve this . in fact , all we need to do is choose @xmath13 to satisfy @xmath104 and we will get acceptance ratios which are always one . niedermayer s solution to this equation was @xmath105 $ ] where @xmath106 is a free parameter whose value we can choose as we like . notice however that this quantity is supposed to be a probability for making a bond , so it is not allowed to be less than zero . thus the best expression we can write for the probability @xmath107 of adding a link between sites @xmath66 and @xmath95 is @xmath108 and this defines niedermayer s algorithm . in figure , we have plotted @xmath13 as a function of @xmath109 for one particular choice of the constant @xmath106 . [ niedaccept ] notice the following things about this algorithm : 1 . as long as all @xmath94 on the lattice are greater than or equal to @xmath106 , the right - hand side of equation is always unity , so the two acceptance ratios can be chosen to be one for every move . since we are at liberty to choose @xmath106 however we like , we can always choose it to satisfy this condition by making it less than or equal to the smallest value that @xmath94 can take , which is @xmath110 in the case of the ising model . this gives us a whole spectrum of wolff - type algorithms for various values @xmath111 , which all have acceptance ratios of one . as @xmath106 gets more and more negative , the probabilities @xmath112 tend closer and closer to one , making the clusters formed larger and larger . this gives us a way of controlling the sizes of the clusters formed in our algorithm , all the way up to clusters which encompass ( almost ) every spin on the lattice at every move . if we choose @xmath106 be greater than the smallest possible value of @xmath94 ( which is @xmath110 in the ising case ) then the right - hand side of equation is no longer equal to one , and we can no longer choose the acceptance ratios to be unity . instead we have @xmath113^{n - m } = [ \e^{2\beta(e_0 + j)}]^{n - m}.\ ] ] just as with the metropolis algorithm , the optimal choice of the two acceptance ratios is then to make the larger of the two equal to one , and choose the smaller to satisfy this equation . if we do this , we again achieve detailed balance , and we now have an algorithm which , as @xmath106 is made larger and larger , produces smaller and smaller clusters , though it does so at the expense of an exponentially decreasing acceptance ratio for cluster moves which increase the energy of the system . if @xmath106 is chosen to be larger than the largest possible value of @xmath94 , which is @xmath114 , then @xmath115 for all pairs of spins @xmath66 , @xmath95 , so every cluster formed has only one spin in it and the acceptance ratios are given by @xmath116^{n - m}.\ ] ] bearing in mind that @xmath17 is the number of neighbours of this single spin which are pointing in the same direction as it , and @xmath21 is the number which point in the opposite direction , we can see that this is exactly the same as the acceptance ratio for the metropolis algorithm . thus we see that by varying the parameter @xmath106 , niedermayer s cluster algorithm includes as special cases both the metropolis algorithm and the wolff algorithm , and interpolates smoothly from one to the other and beyond , varying the average cluster size from one spin all the way up to the entire lattice . the trouble with the algorithm is that no one really knows what value one should choose for @xmath106 . niedermayer himself conjectured that the wolff choice @xmath117 might not give the optimal correlation time and that shorter ones could be achieved by making other choices . he gave preliminary evidence in his 1988 paper that , in some cases at least , a larger value of @xmath106 than this ( i.e. , a value which produces smaller clusters on average , at the expense of a lowered acceptance probability ) gives a shorter correlation time . however , to our knowledge no one has performed an extensive study of the dynamic exponent of the algorithm as a function of @xmath106 , so for the moment at least , the algorithm remains an interesting extension of the wolff idea which has yet to find use in any large - scale simulations . an alternative technique for varying the sizes of the clusters flipped by the wolff algorithm has been studied by barkema and marko ( 1993 ) , who proposed a general method for placing constraints on the clusters generated in the wolff algorithm . in their algorithm , cluster growth proceeds just as in the normal wolff algorithm , except when the addition of a site would violate some condition which we wish to preserve . for example , we might want to place a limit on the total number of spins in the cluster , or limit the radius of the cluster around the initial seed spin . if the addition of a certain spin would violate such a condition , we simply make the probability of adding it to the cluster zero , instead of the usual @xmath118 . on its own , this procedure would violate detailed balance since it alters the selection probability for the move @xmath19 away from its carefully chosen wolff value . however , the requirement of detailed balance only places a condition on the product of the selection probability for a move and the acceptance ratio . thus it is possible to compensate for the change in @xmath19 by making an opposing change to the acceptance ratio @xmath25 . by doing this , detailed balance is restored and the desired condition is met , albeit at the expense of changing the acceptance ratio from the optimal value of 1 which it has in the wolff algorithm . in detail the method works like this . consider the two states @xmath15 and @xmath16 , and suppose that for a move which takes us from @xmath15 to @xmath16 there are @xmath17 bonds which have to be broken in order to flip the cluster . these broken bonds represent correctly - oriented spins which are not added to the cluster by the algorithm . for most of these broken bonds , the probability of not adding such a spin is @xmath119 as in the usual wolff algorithm , but there will now be some number @xmath120 that were rejected because of the applied constraint and thus had no chance of being added . thus the probability of not adding all the broken bonds , which is proportional to the selection probability @xmath19 for the forward move , is @xmath121 . if there are @xmath21 bonds which are broken in making the reverse move from @xmath16 to @xmath15 , @xmath122 of which had no chance of being added since they violated our constraint , then the selection probability @xmath23 will be proportional to @xmath123 . the ratio of the forward and backward selection probabilities is thus the same as in the wolff algorithm , except for a factor of @xmath124 . in order to restore detailed balance , we compensate for this factor by introducing an acceptance ratio of @xmath125 what is the purpose of introducing constraints such as these ? in the context of the normal ising model , a constraint can be enforced to limit the cluster to a certain region . this allows us to divide the lattice into a number of regions in each of which cluster flips can be performed independently of the others . this leads to an efficient algorithm for simulating the model on a parallel ( distributed ) computer which shows near - linear speed - up in the number of processors employed ( barkema and macfarland 1994 ) . periodically , the division of the lattice has to be rearranged in order to obtain ergodicity . another application of the idea of cluster constraints can be found in the simulation of the conserved - order - parameter ising model : clusters of up- and down - pointing spins are grown with a hard constraint on their mass , and if both clusters reach the same mass , an exchange of the spins is proposed . size constraints have also proved useful in the simulation of the random - field ising model ( rfim ) . the rfim is normally simulated using a single - spin - flip metropolis algorithm . it is also possible to apply the wolff algorithm to the problem by growing wolff clusters ignoring the random fields and incorporating their energy contribution into an acceptance ratio for the cluster flip ( dotsenko 1991 ) . however in practice this algorithm works poorly in the critical region of the model because the rfim has a critical temperature lower than that of the normal ising model . this means that the clusters grown are nearly always large percolating clusters , which have little chance of being flipped over because the random fields pin them in one direction or the other . this problem can be overcome if the size of the cluster is limited to a radius @xmath126 , which is chosen at random on each step from a distribution as @xmath127 ( newman and barkema 1996 ) . also worthy of mention is a class of methods developed by kandel ( 1989 , kandel 1991 ) , which are referred to as multigrid methods . these methods are also aimed at reducing critical slowing down and accelerating simulations at or close to the critical temperature . multigrid methods have not been used to a very great extent in large - scale monte carlo simulations because they are considerably more complex to program than the cluster algorithms discussed above . however , they may yet prove useful in some contexts because they appear to be faster than cluster algorithms for simulating very large systems . the fundamental idea behind these methods is the observation that , with the divergence of the correlation length at the critical temperature , we expect to see fluctuating domains of spins of all sizes up to the size of the entire lattice . the multigrid methods therefore split the cpu time of the simulation up , spending varying amounts of time flipping blocks of spins of various sizes . this idea certainly has something in common with the ideas behind the wolff and swendsen - wang algorithms , but the multigrid methods are more deliberate about flipping blocks of certain sizes , rather than allowing the sizes to be determined by the temperature and configuration of the lattice . kandel and his co - workers gave a number of different , similar algorithms , which all fall under the umbrella of multigrid methods . here , we describe one example , which is probably the simplest and most efficient such algorithm for the ising model . the algorithm works by grouping the spins on the lattice into blocks and then treating the blocks as single spins and flipping them using a metropolis algorithm . in detail , what one does is this . in the swendsen - wang algorithm we made `` links '' between similarly - oriented spins , which effectively tied those spins together into a cluster , so that they flipped as one . any two spins which were not linked were free to flip separately there was not even a ferromagnetic interaction between them to encourage them to point in the same direction . in the present multigrid method , two adjacent spins can be in three different configurations : they can be linked as in the swendsen - wang case so that they must flip together , they can have no connection between them at all so that they can flip however they like , or they can have a normal ising interaction between them of strength @xmath29 which encourages them energetically to point in the same direction , but does not force them to as our links do . by choosing one of these three states for every pair of spins , the lattice is divided up into clusters of linked spins which either have interactions between them , or which are free to flip however they like . the algorithm is contrived so that only clusters of one or two spins are created . no clusters larger than two spins appear . the procedure for dividing up the lattice goes like this . 1 . we take a spin on the lattice , and examine each of its neighbours in turn . if a neighbour is pointing in the opposite direction to the spin , then we leave it alone . in kandel s terminology , we `` delete '' the bond between the two spins , so that they are free to assume the same or different directions with no energy cost . if a neighbour is pointing in the same direction as our spin , then we make a link between the two with the same probability @xmath34 as we used in the wolff and swendsen - wang algorithms . kandel calls this `` freezing '' the bond between the spins . since we only want to create clusters of at most two spins , we stop looking at neighbours once we have created a link to any one of them . in fact , what we do is to keep the normal ising interactions between the spin and all the remaining neighbours that we have not yet looked at . we do this regardless of whether they are pointing in the same direction as our first spin or not . kandel describes this as `` leaving the bonds active '' . now we move onto another spin and do the same thing , and in this way cover the entire lattice . notice that , if we come to a spin which is adjacent to one we have considered before , then some of the spin s bonds will already have been frozen , deleted , or marked as active . in this case we leave those bonds as they are , and only go to work on the others which have not yet been considered . notice also that if we come to a spin and it has already been linked ( `` frozen '' ) to another spin , then we know immediately that we need to leave the interactions on all the remaining bonds to that spin active . in this way , we decide the fate of all the spins on the lattice , dividing them into clusters of one or two , joined by bonds which may or may not have interactions associated with them . then we treat those clusters as single spins , and we carry out the metropolis algorithm on them , for a few sweeps of the lattice . but this is not the end . now we do the whole procedure again , treating the clusters as spins , and joining them into bigger clusters of either one or two elements each , using exactly the same rules as before . ( note that the lattice of clusters is not a regular lattice , as the original system was , but this does not stop us from carrying out the procedure just as before . ) then we do a few metropolis sweeps of this coarser lattice too . and we keep repeating the whole thing until the size of the blocks reaches the size of the whole lattice . in this way , we get to flip blocks of spins of all sizes from single spins right up to the size of the entire system . then we start taking the blocks apart again into the blocks that made them up , and so forth until we get back to the lattice of single spins again . in fact , kandel and co - workers used a scheme where at each level in the blocking procedure they either went towards bigger blocks ( `` coarsening '' ) or smaller ones ( `` uncoarsening '' ) according to the following rule . at any particular level of the procedure we look back and see what we did the previous times we got to this level . if we coarsened the lattice the previous two times we got to this point , then on the third time , we uncoarsen . this choice has the effect of biasing the algorithm towards working more at the long length scales ( bigger , coarser blocks ) . well , perhaps you can see why the complexity of this algorithm has put people off using it . the proof that the algorithm satisfies detailed balance is even more involved , and , since you re probably not dying to hear about it right now , we ll refer you to the original paper for the details ( kandel ( 1989 ) ) . in the same paper it is demonstrated that the dynamic exponent for the algorithm is in the region of @xmath128 for the two - dimensional ising model a value similar to that of the wolff algorithm . before we dismiss the multigrid method out of hand , however , let us point out that the simulations do indicate that its performance is superior to cluster algorithms for large systems . these days , with increasing computer power , people are pushing simulations towards larger and larger lattices , and there may well come a point at which using a multigrid method could win us a significant speed advantage . finally , in our round - up of monte carlo algorithms for the ising model , we come to an unusual algorithm proposed by jon machta and co - workers called the invaded cluster algorithm ( machta 1995 ) . the thing which sets this algorithm apart from the others we have looked at so far is that it is not a general purpose algorithm for simulating the ising model at any temperature . in fact , the invaded cluster algorithm can only be used to simulate the model at the critical point ; it does not work at any other temperature . what s more , for a system at the critical point in the thermodynamic limit the algorithm is just the same as the swendsen - wang cluster algorithm of section . so what s the point of the algorithm ? well , there are two points . first , the invaded cluster algorithm can find the critical point all on its own we do nt need to know what the critical temperature is beforehand in order to use the algorithm . starting with a system at any temperature the algorithm will adjust its simulation temperature until it finds the critical value @xmath35 . this makes the algorithm very useful for actually measuring the critical temperature , something which normally requires either finite - size scaling or monte carlo rg . second , the invaded cluster algorithm equilibrates extremely quickly . although the algorithm is equivalent to the swendsen - wang algorithm once it reaches equilibrium at @xmath35 , its behaviour whilst getting there is very different , and in a direct comparison of equilibration times between the two , say for systems starting at @xmath129 , there is really no contest . for large system sizes , the invaded cluster algorithm can reach the critical point as much as a hundred times faster than the swendsen - wang algorithm . ( a comparison with the wolff algorithm yields similar results the performance of the wolff and swendsen - wang algorithms is comparable . ) so , how does the invaded cluster algorithm work ? basically , it is just a variation of the swendsen - wang algorithm in which the temperature is continually adjusted to look for the critical point . the algorithm finds the fraction of links at which a percolating backbone of spins first forms across the lattice and uses this measurement to make successively better approximations to the critical temperature at each monte carlo step . in detail , here s how it goes : 1 . we choose some starting configuration of the spins . it does nt matter what choice we make , but it could , for example , be the @xmath129 state in which all spins are aligned . we add links at random between pairs of similarly - oriented nearest - neighbour sites until one of the clusters formed reaches percolation . once the links are made , we flip each cluster on the lattice separately with probability @xmath88 , just as we do in the normal swendsen - wang algorithm . then the whole procedure is repeated from step ( ii ) again . why does this algorithm work ? well , for any given configuration of the lattice , the maximum number of links which can be made between spins is equal to the number of pairs of similarly - oriented nearest - neighbour spins . let us call this number @xmath21 . the number of links @xmath17 needed to achieve the percolation of one cluster will normally be smaller than this maximum value . a given step of the algorithm is equivalent to a single step of the swendsen - wang algorithm with @xmath130 . ( the number of links @xmath17 required for percolation is fairly constant , and most of the variation in @xmath13 comes from @xmath21 . ) substituting this value into equation and rearranging , we can then calculate the effective temperature of the monte carlo step : @xmath131 consider what happens if the system is below the critical temperature . in this case , the spins are more likely to be aligned with their neighbours than they are at the critical temperature , and @xmath21 is therefore large . this gives us a value of @xmath132 which is higher than @xmath35 . in other words , when the system is below the critical temperature , the algorithm automatically chooses a temperature @xmath133 for its swendsen - wang procedure . conversely , if the system is at a temperature above the critical point , neighbouring spins are less likely to be aligned with one another than they are at @xmath35 , giving a lower value of @xmath21 and a correspondingly lower value of @xmath132 . so , for a state of the system above the critical temperature , the algorithm automatically chooses a temperature @xmath134 for its swendsen - wang procedure . the algorithm therefore has a kind of negative feedback built into it , which always drives the system towards the critical point , and when it finally reaches @xmath35 it will stay there , performing swendsen - wang monte carlo steps at the critical temperature for the rest of the simulation . note that we do not need to know what the critical temperature is for the algorithm to work . it finds @xmath35 all on its own , and for that reason the algorithm is a good way of measuring @xmath35 . the same feedback mechanism also drives the algorithm towards @xmath35 faster than simply performing a string of swendsen - wang monte carlo steps exactly at @xmath35 . this point is discussed in more detail below . before we review the result of simulations using the invaded cluster algorithm , let us briefly examine its implementation . by and large , the algorithm is straightforward to program , but some subtlety arises when we get to the part about testing for a percolating cluster . since we need to do this every time a link is added to the lattice , it is important that we find an efficient test . there are various criteria one can apply to test for percolation on a finite lattice . here we use one of the simplest : we assume a cluster to be percolating when it `` wraps around '' the periodic boundary conditions on the lattice . in other words the cluster percolates when there is a path from any site leading across the lattice and back to the same site which has an integrated displacement which is non - zero . this test can be efficiently implemented as follows . all sites belonging to the same cluster are organized into a tree structure in which there is one chosen `` reference site '' . all the other sites in the cluster are arranged in a hierarchy below this reference site . the immediate daughter sites of the reference site possess vectors which give their position relative to the reference site . other sites are arranged as grand - daughter sites , with vectors giving their position relative to one of the daughter sites , and so on until every site in the cluster is accounted for . the vector linking any site to the reference site can be calculated by simply ascending the tree from bottom to top , adding the vectors at each level until the reference site is reached . the particular way in which the sites in a cluster are arranged on the tree is dictated by the way in which the cluster grows . at the start of each monte carlo step no links exist and each site is a reference site . each time a link is added we must either join together two separate clusters to make one larger cluster , or join together two sites which already belong to the same cluster . since our criterion for percolation is the appearance of a cluster which wraps around the boundary conditions , it follows that percolation can only appear on moves of the latter type . in other words we need only check for percolation when we add a link joining two sites which are already members of the same cluster . the procedure is therefore as follows . each time we add a link , we use the tree structures to determine the reference site of each of the two sites involved . these reference sites tell us which clusters the sites belong to . when a link is made between two sites belonging to different clusters , we have to combine the two clusters into one , which we do by making one of the reference sites the daughter of the other . when we make a link between sites in the same cluster , we do nt need to combine two clusters , but we do need to check for percolation . to do this we take the lattice vectors from each of the two sites to the reference site of the cluster and subtract them . normally , this procedure should result in a difference vector of length one the distance between two nearest - neighbour spins . however , if the addition of a link between these two spins causes the cluster to wrap around the boundary conditions , the difference of the two vectors will instead be a unit vector _ plus _ a vector of length the dimension of the lattice . in this case we declare the cluster to the percolating , and stop adding links . [ ictc ] the results for the invaded cluster algorithm are impressive . machta ( 1995 ) found equilibration times 20 or more times faster than for the swendsen - wang algorithm at @xmath35 for the two- and three - dimensional ising systems they examined , and of course , the invaded cluster algorithm allowed them to measure the value of @xmath35 , which is not directly possible with the normal swendsen - wang algorithm . figure shows their results for the critical temperature of a set of 2d ising systems of different sizes . because there is always a finite chance of producing a percolating cluster on a finite lattice with any value of @xmath13 no matter how small ( as long as it s not zero ) , the average estimate of @xmath35 which the algorithm makes on a finite system tends to be a little high . however , we can easily extrapolate to the limit @xmath135 using finite - size scaling . in this case we do this by plotting the measured @xmath35 as a function of @xmath136 and extrapolating to the @xmath137 axis . the result is @xmath138 , which compares favourably with the known exact result of @xmath139 , especially given the small amount of cpu time taken by the simulation . ( the runs were @xmath140 monte carlo steps , for each system size . ) the invaded cluster algorithm can also be used to measure other quantities at the critical temperature magnetization for example , or internal energy . however , a word of caution is in order here . for lattices of finite size , which of course includes all the lattices in our monte carlo simulations , the invaded cluster algorithm does not sample the boltzmann distribution exactly . in particular , the fluctuations in quantities measured using the algorithm are different from those you would get in the boltzmann distribution . to see this , consider what happens once the algorithm has equilibrated to the critical temperature . at this point , as we argued before , it should stop changing the temperature and just become equivalent to the swendsen - wang algorithm at @xmath35 , which , as we know , certainly samples the boltzmann distribution correctly . however , in actual fact , because the lattice is finite , statistical variations in the order in which we generate the links on the lattice will give rise to variations in the temperature @xmath38 of successive steps in the simulation . the negative feedback effect that we described above will ensure that @xmath38 always remains close to @xmath35 , but the size of fluctuations is very sensitive to small changes in temperature near @xmath35 and as a result the measured fluctuations are not a good approximation to those of the true boltzmann distribution . thus the invaded cluster algorithm is not suitable for measuring , for instance , the magnetic susceptibility @xmath60 or the specific heat @xmath141 of the ising model at @xmath35 , both of which are determined by measuring fluctuations . on the other hand , one could use the algorithm to determine the value of @xmath35 , and then use the normal wolff or swendsen - wang algorithm to perform a simulation at that temperature to measure @xmath60 or @xmath141 . cluster monte carlo algorithms are not only applicable to the ising model . they also work for other types of spin models . in this section we consider how they can be modified to work with potts models and continuous spin models . potts models suffer from critical slowing down in just the same way as the ising model . all of the algorithms discussed in this chapter can be generalized to potts models . here we discuss the example of the wolff algorithm , whose appropriate generalization is as follows : 1 . choose a seed spin at random from the lattice . 2 . look in turn at each of the neighbours of that spin . if they have the same value as the seed spin , add them to the cluster with probability @xmath142 . ( note that the 2 has vanished from the exponent ; the 2-state potts model , for example , is equivalent to the ising model except for a factor two in the interaction constant @xmath29 . ) 3 . for each spin that was added in the last step , examine each of _ its _ neighbours to find which ones , if any , have the same value and add each of them to the cluster with the same probability @xmath13 . ( as with the ising model , we notice that some of the neighbours may already be members of the cluster , in which case you do nt have to consider adding them again . also , some of the spins may have been considered for addition before , as neighbours of other spins in the cluster , but rejected . in this case , they get another chance to be added to the cluster on this step . ) this step is repeated as many times as necessary until there are no spins left in the cluster whose neighbours have not been considered for inclusion in the cluster . 4 . choose at random a new value for the spins in the cluster , different from the present value , and set all the spins to that new value . the proof that this algorithm satisfies detailed balance is exactly the same as it was for the ising model . if we consider two states @xmath15 and @xmath16 of the system which differ by the changing of just one cluster , then the ratio @xmath143 of the selection probabilities for the moves between these states depends only the number of bonds broken @xmath17 and the number made @xmath21 around the edges of the cluster . this gives us an equation of detailed balance which reads @xmath144 just as in the ising case . the change in energy is also given by the same expression as before , except for a factor of two : @xmath145 and so the ratio of the acceptance ratios @xmath25 and @xmath26 for the two moves is @xmath146^{n - m}. \label{contacc}\ ] ] for the choice of @xmath13 given above , this is just equal to one . equation is then satisfied by making the acceptance ratios equal to 1 , and with this choice the algorithm satisfies detailed balance . as in the case of the ising model , the wolff algorithm gives an impressive improvement in performance near to the critical temperature . arguments similar to those of bausch ( 1981 ) indicate that the dynamic exponent of the metropolis algorithm for a potts model should have a lower bound of 2 . by contrast , baillie and coddington ( 1991 ) have measured a dynamic exponent of @xmath147 for the wolff algorithm in the 2d , @xmath148 case . as before , the single - spin - flip algorithms come into their own well away from the critical point , because critical slowing down ceases to be a problem and the relative simplicity of these algorithms over the wolff algorithm tends to give them the edge . ones choice of algorithm should therefore ( as always ) depend on exactly which properties of the model one wants to investigate , but the wolff algorithm is definitely a good choice for examining critical properties . the swendsen - wang algorithm can also be generalized for use with potts models in a very simple fashion , as can all of the other algorithms described in section . cluster algorithms can also be generalized to models with continuous spins . let us take the xy model as our example . a version of the wolff algorithm for this model was given by wolff in his original paper in 1989 . it also works for similar models in higher dimensions . the idea is a simple one : one chooses at random a seed spin to start a cluster and a direction vector @xmath149 . to represent the direction of the vector . in three dimensions we need to choose both a @xmath150 and a @xmath151 . @xmath150 is again uniformly distributed between 0 and @xmath152 , and @xmath151 is given by @xmath153 where @xmath126 is random real number uniformly distributed between zero and one . ] then we treat the components @xmath154 of the spins in that direction roughly in the same way as we did the spins in the ising model . a neighbour of the seed spin whose component in this direction has the same sign as that of the seed spin can be added to cluster by making a link between it and the seed spin with some probability @xmath13 . if the components point in opposite directions then the spin is not added to the cluster . when the complete cluster has been built , it is `` flipped '' by reflecting all the spins in the plane perpendicular to @xmath149 . the only complicating factor is that , in order to satisfy detailed balance , the expression for @xmath13 has to depend on the values of the spins which are joined by links thus : @xmath155.\ ] ] readers may like to demonstrate for themselves that , with this choice , the ratio of the selection probabilities @xmath19 and @xmath23 is equal to @xmath156 , where @xmath157 is the change in energy in going from a state @xmath15 to a state @xmath16 by flipping a single cluster . thus , detailed balance is obeyed as in the ising case by an algorithm for which the acceptance probability for the cluster flip is 1 . this algorithm has been used , for example , by gottlob and hasenbusch ( 1993 ) to perform extensive studies of the critical properties of the heisenberg model . similar generalizations to continuous spins are possible for all the algorithms discussed in section . we have discussed a number of recently developed cluster algorithms for the simulation of classical spin systems , including : the wolff algorithm , which flips a single cluster of spins at each update step and almost completely eliminates critical slowing down from the simulation of the ising model ; the swendsen - wang algorithm , which updates the entire lattice at each step and has performance similar to , though not quite as good as the wolff algorithm ; neidermayer s algorithm , a variation of the wolff algorithm which allows one to tune the size of clusters flipped ; the limited cluster algorithm , another technique for controlling the sizes of clusters ; multigrid methods , which may be even more efficient close to criticality than the wolff algorithm , although they achieve this at the expense of considerable programming complexity ; and the invaded cluster algorithm , which self - organizes to the critical point of the system studied and permits accurate measurements of @xmath35 to be made with little computational effort . we have also discussed briefly how algorithms such as these can be generalized to potts models and continuous spin models . the authors would like to thank eytan domany , jon machta and alan sokal for useful discussions about cluster algorithms , and daniel kandel for supplying a copy of his ph.d . thesis .

the period luminosity sequences and the multiple periods of luminous red giant stars are examined using the ogle iii catalogue of long - period variables in the large magellanic cloud . it is shown that the period ratios in individual multimode stars are systematically different from the ratios of the periods at a given luminosity of different period luminosity sequences . this leads to the conclusion that the masses of stars at the same luminosity on the different period luminosity sequences are different . an evolutionary scenario is used to show that the masses of stars on adjacent sequences differ by% at a given luminosity , with the shorter period sequence being more massive . the mass is also shown to vary across each sequence by a similar percentage , with the mass increasing to shorter periods . on one sequence , sequence b , the mass distribution is shown to be bimodal . it is shown that the small amplitude variables on sequences a , a and b pulsate in radial and nonradial modes of angular degree=0 , 1 and 2 , with the=1 mode being the most common . the stars on sequences c and c are predominantly radial pulsators (=0 ) . matching period ratios to pulsation models shows that the radial pulsation modes associated with sequences a , a , b , c and c are the 4th , 3rd , 2nd and 1st overtones and the fundamental mode , respectively . [ firstpage ] stars : agb and post - agb stars : oscillations stars : variables : general .

1502.03137

it is now well known that luminous red giants are variable and that the periods of the observed variability lie on discrete sequences in period luminosity diagrams . @xcite and @xcite used data from macho observations of stars in the large magellanic cloud ( lmc ) to define five sequences a , b , c , d and e in the ( @xmath3 , @xmath4 ) and ( @xmath5 , @xmath4 ) planes , respectively ( where @xmath3 is a reddening - free weisenheit index defined as @xmath61.38(@xmath7@xmath6 ) ) . the sequences a , b and c were identified as containing pulsating stars , while sequence e was found to consist of binary stars and sequence d was the sequence formed by the long secondary periods ( lsps ) that occur in 3050% of luminous pulsating red giants . in this paper , the pulsating stars will be discussed . these stars lie on sequences a , b and c and two other sequences a@xmath1 and c@xmath1 which are described below . further examination of the ellipsoidal variables on sequence e can be found in @xcite , @xcite , @xcite and @xcite while studies of the origins of the lsps can be found in @xcite , @xcite , @xcite , @xcite , @xcite and @xcite . following the discovery of sequences a , b and c , @xcite used extensive @xmath5 band photometry combined with macho periods to show that sequence b was in fact two sequences which were renamed b and c@xmath1 . the next refinement to the luminous sequences was by @xcite who used ogle data to show the existence of a new sequence on the short period side of sequence a. @xcite labelled this sequence a@xmath8 but in this paper the designation a@xmath1 will be used ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite also made the important discovery that sequence a consisted of three closely - spaced parallel sequences . @xcite claimed that sequences b and a@xmath1 also consisted of three closely - spaced parallel sequences ( hereinafter referred to as sub - sequences ) . more recently , @xcite have drawn attention to a probable faint sequence designated f between sequences c@xmath1 and c. @xcite assumed that all the variable luminous red giants found by the macho observations were asymptotic giant branch ( agb ) stars . however , @xcite and @xcite showed that both agb and red giant branch ( rgb ) stars were variable and that there was a slight offset in period between the sequences of rgb and agb stars . @xcite also noted this offset but they did not find evidence for two slightly offset sequences for sequence a@xmath1 which they suggested consisted of agb stars only . however , this result was challenged by @xcite who showed that an rgb component of sequence a@xmath1 does exist . various attempts have been made to identify the pulsation modes associated with the pulsation sequences c , c@xmath1 , b , a and a@xmath1 and the sub - sequences of b , a and a@xmath1 . @xcite ( see also @xcite ) showed that the radial fundamental mode could fit sequence c with reasonable assumptions about the mass of the star and that the first few radial overtone modes had periods similar to those of sequences b ( which now includes c@xmath1 ) and a , although the data available at the time was not extensive enough for a detailed matching of these latter sequences with modes . they also noted that the large - amplitude mira variables fell on sequence c , making them fundamental mode pulsators . additional evidence in favour of this modal assignment is provided by the fact that nonlinear fundamental mode radial pulsation models reproduce the light and velocity curves of local mira variables quite well ( e.g. * ? ? ? * ; * ? ? ? recently , using extensive observational data from ogle , @xcite showed that the period ratios in stars of sequences c and c@xmath1 were consistent with those expected for fundamental mode pulsation of sequence c and first overtone pulsation on sequence c@xmath1 . @xcite compared period ratios for ogle variables near the tip of the rgb with models and concluded that sequences c@xmath1 , b and a corresponded to the radial first , second and third overtones , consistent with the assignments by @xcite and @xcite . in contrast to the above results , @xcite concluded that the sequences a , b and c@xmath1 corresponded to the radial 2nd overtone , first overtone and fundamental modes , respectively , when they used used radial pulsation models to try fit the sequences in the period luminosity ( pl ) plane at rgb luminosities ( see also @xcite ) . this means that the longer period sequence c containing the miras has no explanation in terms of radial pulsation . this study was extended by @xcite who matched the a and b sequences at rgb luminosities to sequences of kepler radial mode pulsators in a pl plane where frequency data was used as a proxy for luminosity . this result produces a close match between sequences b and a and the radial first and second overtones , respectively , in agreement with @xcite but not with @xcite , @xcite and @xcite . there thus exists two extant assignments of the radial modes to the pl sequences a , b , c@xmath1 and c that differ by one radial order . in another study using kepler data , @xcite showed that in the luminous kepler red giants , triplets of modes with periods corresponding to spherical degrees @xmath2=0 , 2 and 1 ( in increasing order of period ) are excited for a given radial order . the mode with @xmath2=1 has the largest amplitude for radial orders @xmath9=3 or more ( the second or higher radial overtones ) . this suggests the ogle sequences b , a and a@xmath1 may be dominated by @xmath2=1 modes . in this paper , the period and luminosity data given in the ogle iii catalogue of long - period variables ( lpvs ) in the lmc @xcite is analysed in detail . purely empirical results related to the modes , masses and evolution of the stars on the sequences are derived . then the data are compared to pulsation models in order to throw further light on the modes of pulsation involved . , @xmath10 ) plane , where @xmath11 is the period of the mode with the largest amplitude and @xmath12 is a reddening - free measure of the luminosity . mira variables are shown as red points ( in the online version of the paper ) . the ellipsoidal variables belonging to sequence e are not shown . ] the sequences of luminous red giants in the lmc are shown in figure [ wjk - p ] . the period data for this figure come from the ogle iii catalogue of lpvs in the lmc @xcite . all the variables in the catalogue are included in the analysis in this paper - no selection is made according to the catalogue types mira , srv or osarg . in the catalogue , three periods are given for each star with the first ( or primary ) period ( p1 ) having the largest amplitude and the secondary periods p2 and p3 having progressively smaller amplitudes ( most luminous red giants exhibit multiple periods of variation ) . the luminosity in figure [ wjk - p ] is represented by @xmath12 where the @xmath13 and @xmath5 magnitudes are from the 2mass point source catalogue @xcite . the individual sequences are labelled in figure [ wjk - p ] although sequence e ( produced by the binary ellipsoidal variables ) is omitted as these stars are not included in the catalogue . the tip of the rgb is clearly seen at @xmath14 . the large - amplitude mira variables are shown as red points in figure [ wjk - p ] . they clearly belong to sequence c. as noted in the introduction , there are several lines of evidence which suggest that stars on sequence c are fundamental mode radial pulsators , although some authors have provided arguments against this modal assignment . in section [ comp_mods ] , we will present what we believe to be compelling evidence that the stars on sequence c are indeed fundamental mode pulsators . for these reasons , and to simplify some of the discussion , in this paper it will be assumed from here on that the sequence c stars , including the mira variables , are fundamental mode radial pulsators . it is worth considering at this point what causes the existence of the sequences in diagrams like figure [ wjk - p ] . firstly , remember that the period used in figure [ wjk - p ] is the primary or largest - amplitude period . at a given luminosity , a star will have possible modes of pulsation with periods that could fall on any of the sequences or ( as shall be shown below ) between the sequences . the actual position of a star in the figure is determined by the period of the mode of largest amplitude . for self - excited modes , this mode is the one that has the strongest driving - see @xcite for a demonstration of this effect for 1st overtone pulsation on sequence c@xmath1 . for modes that are excited stochastically by convective motions ( solar - like oscillations ) , there is also a preferred pulsation period which is related to the acoustic cut - off frequency ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? period luminosity ( pl ) laws such as those in figure [ wjk - p ] provide one type of diagram for the comparison of models and observations of variable stars . a more precise diagram for comparing observations and models is the petersen diagram wherein period ratios in multimode pulsators are plotted against the longer of the two periods in the ratio . in this section , a modified petersen diagram appropriate for lpvs is discussed . the upper panel of figure [ em_strips ] shows the same pl diagram as figure [ wjk - p ] but now coloured straight lines have been fitted to each of the pulsation sequences c , c@xmath1 , b , a and a@xmath1 . it has been assumed that the three sequences b , a and a@xmath1 have the same slope @xmath15 while the two sequences c and c@xmath1 have a slightly different slope @xmath16 . since many of the variable lpvs are multimode pulsators , they can be plotted in a petersen diagram . the lower panel of figure [ em_strips ] shows a variant of this diagram where the logarithmic period ratios @xmath17/@xmath11 and @xmath18/@xmath11 in each star are plotted against the variable @xmath10[@xmath19=12 ] . the latter quantity is the value of @xmath10 obtained by projecting the actual value of @xmath20 along a line parallel to sequence a until it meets the level @xmath21 . specifically , @xmath11[@xmath19=12 ] = @xmath22 . stars with primary periods on a given sequence are thus bounded in the same horizontal range in the lower panel of figure [ em_strips ] , with this range corresponding to the values of @xmath10 for this sequence when @xmath23 . for lpvs , this form of petersen diagram is much better at revealing preferred period ratios than the traditional petersen diagram used by @xcite and @xcite which plots @xmath24 against @xmath25 . a form of petersen diagram which used a quantity like @xmath11[@xmath19=12 ] on the horizontal axis was presented by @xcite . the coloured points and short lines in the lower panel of figure [ em_strips ] correspond to the ratios of the periods on the fit lines shown in the top panel i.e. they represent the ratios of the periods of the different period luminosity sequences . the horizontal position of the coloured point / lines is @xmath10[@xmath19=12 ] for the line which is in the denominator of the period ratio . the colour of the point / line is the colour of the line in the numerator . coloured points correspond to period ratios of the fits to sequences a@xmath1 , a and b ( these all have the same slope ) . for example , the red point at ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) @xmath27 ( 1.12 , 0.17 ) represents the ratio of @xmath10 at the middle of sequence a to @xmath10 at the middle of sequence a@xmath1 , the red point near ( 1.3 , 0.0 ) represents this ratio for sequence a to itself and the red point near ( 1.52 , -0.23 ) represents the ratio for sequence a to sequence b. since the lines fitted to sequences c and c have a slightly different slope to that of sequence a , the period ratios of fits involving sequences c or c lead to the short coloured lines . for example , the short red line at ( @xmath10[@xmath19=12 ] , @xmath28/@xmath11 ) @xmath27 ( 1.65 , -0.35 ) represents the ratio of @xmath10 at the middle of sequence a to @xmath10 at the middle of sequence c@xmath1 . we now examine the structure in the lower panel of figure [ em_strips ] in detail . the black points in the lower panel of figure [ em_strips ] show observed period ratios for individual stars . the clustering of these points represents ratios of specific modes present in the data . firstly , we start with the points in the vertical strip 1.2 @xmath29 @xmath10[@xmath19=12 ] @xmath29 1.4 which belong to stars with their primary period on sequence a. there is a large cluster of points near ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) = ( 1.3 , 0.14 ) which one would expect to belong to stars on sequence a that show a secondary period on sequence b. similarly , the cluster of points near ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) = ( 1.3 , 0.3 ) should correspond to stars on sequence a that show a secondary period on sequence c@xmath1 and a third cluster of points near ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) = ( 1.3 , -0.12 ) which should correspond to stars on sequence a that show a secondary period on sequence a@xmath1 . however , the period ratio of sequences b , c@xmath1 and a@xmath1 to sequence a , represented by the green , magenta and orange points at @xmath10[@xmath19=12]=1.3 , respectively , do _ not _ coincide with these clouds of black points . it can be seen that this problem exists for stars with primary periods on any of the sequences . we will explore this situation in detail in section [ chans ] . another interesting feature of figure [ em_strips ] is that there are clouds of points such as those near ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) = ( 1.45 , -0.15 ) for which the primary period lies _ between _ the sequences . one can see in the upper panel of figure [ em_strips ] that some stars do have their primary periods between the sequences although they are not as common as the stars on the sequences . in fact , it turns out that the existence of stars between the sequences is related to the problem noted in the previous paragraph and we come back to this in section [ chans ] . a prominent feature in the lower panel of figure [ em_strips ] is the set of sub - ridges first noted by @xcite and especially prominent for stars on sequence a where five almost - horizontal sub - ridges can be seen in the cluster of points near ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) = ( 1.3 , 0.0 ) . these five sub - ridges must be due to pulsation modes with very closely spaced periods . the spacing is sufficiently small that the individual modes can not be distinguished in a pl diagram such as the top panel of figure [ em_strips ] . the period separation of low - order radial modes in red giants is much larger than the period separation of the modes belonging to the five sub - ridges of sequence a so nonradial pulsation modes , as well as a radial pulsation mode , must be involved . the five sub - ridges of sequence a can be created by three closely - spaced modes each separated in @xmath4 by @xmath00.022 , provided that any of the three modes can be the primary mode in at least some stars . we assume here that this is the case ( if one mode was always dominant , five modes would be required ) . there is some evidence that there are three sub - ridges , corresponding to two close modes , associated with sequence b ( see the cluster of points around ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) = ( 1.53 , 0.0 ) ) . this evidence will be made more apparent in section [ ldep_pr ] . as noted above , the three modes associated with sequence a must include nonradial modes because of their close spacing . we assume that all stars on sequence a , and indeed all stars on any given sequence , have the same radial order of pulsation . nonradial pulsation models of luminous red giants ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) show that for modes of a given radial order but different angular degree @xmath30 , 1 or 2 ( the only ones likely to be observed in light curves ) , the @xmath31 mode has the longest period , the @xmath30 mode has the shortest period and the @xmath32 mode has an intermediate period . figure [ em_strips ] provides a method for determining which of the three modes within sequence a is dominant . among the stars with their primary period on sequence a@xmath1 ( the points with 1.05 @xmath29 @xmath10[@xmath19=12 ] @xmath29 1.2 ) , those that also have the closely - spaced modes associated with sequence a can be seen near ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) = ( 1.12 , 0.12 ) . there are three horizontal bands here with the same spacing as the sub - ridges associated with sequence a. the upper band is by far the most prominent . assuming that one mode dominates sequence a@xmath1 ( because there are no prominent sub - ridges near ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) = ( 1.12 , 0.0 ) ) , this band structure suggests that it is the longest period mode on sequence a that generally has the largest amplitude . as noted above , this is the @xmath31 mode . the dominance of the @xmath31 mode found on sequence a in the present high luminosity red giants is consistent with the findings of @xcite from kepler observations but different from that of @xcite who find the radial @xmath30 mode becomes dominant at high luminosities in their analysis of kepler stars . and @xmath31 modes of sequence a when these two modes are present together . see text for further details . ] .stars on the sub - sequences of sequence a [ cols="<,^,^",options="header " , ] [ n_subseqa ] estimates of the relative amplitudes of the three modes that make up the three sub - sequences of sequence a can be found by examining the populations of the five sub - ridges of sequence a in the petersen diagram . an exploded view of these sub - ridges is shown in the top panel of figure [ mode_amps ] . boxes have been drawn to enclose the sub - ridges and the sub - ridges have been named after the possible modes occurring in them . under our assumption that the three modes involved are the @xmath30 , 1 and 2 modes of a single radial order and that the period increases from @xmath30 to 2 to 1 , a star found in the top box must have @xmath30 for its primary period p1 and @xmath31 for the secondary period ( p2 or p3 ) . this box has been labelled 0 + 1 . similarly , the lowest box 1 + 0 must have @xmath31 for its primary period p1 and @xmath30 for its secondary period . the central box is made up of stars where the same mode is detected twice with slightly different periods and the mode involved could be any of the @xmath30 , 1 and 2 modes . the two intermediate boxes are labelled by the possible pairs of modes that could be present . note that the period separation @xmath33 of the @xmath30 and @xmath31 modes obtained from the positions of the sub - ridges agrees with the period separation of these modes given by the observations and the models of @xcite , giving added support for the mode identifications . the fact that the ridges in boxes 0 + 1 and 1 + 0 are not strictly parallel to the horizontal axis means that @xmath34 varies systematically by a small amount across sequence a. quantitatively , for stars on sequence a , @xmath34 decreases by @xmath00.0045 as @xmath10[@xmath19=12 ] increases by 0.1 . an explanation for this behaviour is given in section [ prat - m ] . the number of stars in each box is given in table [ n_subseqa ] . the total number @xmath35 of stars with their primary period p1 on sequence a is also given . this latter number is the total number of stars with @xmath10[@xmath19=12 ] lying between the values corresponding to the two vertical lines in figure [ mode_amps ] . stars that have both the @xmath31 and @xmath30 modes detected in their light curves lie in boxes 1 + 0 and 0 + 1 . it is clear that @xmath36 is much smaller than @xmath37 which shows that the @xmath31 mode is much more likely to be detected as the _ primary _ pulsation mode than the @xmath30 mode and this in turn suggests that the @xmath31 mode is likely to have a larger amplitude than the @xmath30 mode . this is shown directly in the bottom panel of figure [ mode_amps ] where histograms of modal amplitudes are shown ( the amplitudes come from the ogle iii catalogue ) . the top sub - panel shows the amplitudes of the @xmath30 and @xmath31 modes for stars in box 1 + 0 , the middle sub - panel shows the amplitudes for stars in box 0 + 1 and the bottom sub - panel shows the sum of the top and middle sub - panels . it is clear that the @xmath31 mode tends to have the larger amplitude in stars where the @xmath30 and @xmath31 modes are both present although the non - zero population in box 0 + 1 shows that the @xmath30 mode can sometimes have a larger amplitude than the @xmath31 mode . the amplitude distributions shown in figure [ mode_amps ] apply to stars with two modes on sequence a. however , about 80% of sequence a stars have only one detected mode on sequence a ( see the fractions in the last column of table [ n_subseqa ] , and note that stars on the middle sub - ridge have two detections of the same mode ) . assuming that the relative amplitude distributions shown in figure [ mode_amps ] are typical , the stars with only a single mode on sequence a will be mostly @xmath31 pulsators , in agreement with the result derived above from the stars that have their primary mode on sequence a@xmath1 and a secondary mode on sequence a. one factor that has not been mentioned above is the orientation of the pole of the pulsating star relative to the observer in the case of nonradial pulsation . different orientations will cause different apparent amplitudes of pulsation for a nonradial pulsator , with a zero apparent amplitude being possible for specific values of @xmath2 , @xmath38 and the orientation . as a result , the true numbers of @xmath31 and @xmath32 nonradial pulsators relative to the number of @xmath30 radial pulsators should be higher than indicated by the numbers in table [ n_subseqa ] . in section [ pr ] it was shown that period ratios in individual stars do not agree with the period ratios of the mean sequences . a vivid demonstration of this situation is shown in figure [ channel ] . looking at panel a , it can be seen that the secondary periods form three strips parallel to sequence a , but they are closer to sequence a than are sequences a@xmath1 , b or c@xmath1 . this situation , wherein the secondary modes are closer to the primary mode than are the adjacent sequences to the central sequence , exists for sequences a@xmath1 , a and c. it also exists for the modes shortward of the primary mode of sequence b and for the mode longward of the primary mode for sequence c@xmath1 . the only exception to this rule occurs for the mode longward of the primary mode of sequence b ( which is further from sequence b than is sequence c@xmath1 ) and for the mode shortward of the primary mode of sequence c@xmath1 ( which is further from sequence c@xmath1 than is sequence b ) . we come back to this exception later . we now seek to explain why the green strips in figure [ channel ] do not align with the sequences . firstly , it is assumed here that the pulsation modes that produce the green strips are the same pulsation modes as those that produce the corresponding , neighbouring sequences ( where the correspondence is obtained from the ordering in @xmath4 of the strips and sequences ) . this assumption will be justified in section [ evsec ] . in this situation , the period of the same mode must be different on a green strip and on its corresponding sequence . now , in general , the period of a pulsation mode can be written as @xmath39 where @xmath40 and @xmath41 are both positive and for red giants typically range from 1.52.3 and 0.51.9 , respectively ( e.g. * ? ? ? in addition , the position of the giant branch at a given luminosity means that @xmath42 . combining these two relations with the definition @xmath43 yields @xmath44 . thus if a green strip and its corresponding sequence have the same mode but a different @xmath45 at a given @xmath46 then the masses of the stars in the green strip and its corresponding sequence must be different at a given luminosity . finally in this section , we return to the exceptional cases of ( 1 ) the green strip originating from sequence b which lies on the long period side of sequence c@xmath1 and ( 2 ) the green strip originating from sequence c@xmath1 which lies on the short period side of sequence b. remember that in section [ pr1 ] , it was deduced that the @xmath31 mode was the dominant mode in sequence a and that this mode has a period which is @xmath00.043 longer in @xmath4 than the radial mode with @xmath30 . recall also that there appear to be at least two modes ( of the same radial order ) associated with sequence b. we will show later that these are most likely the @xmath31 and @xmath30 modes , with the former being most common . finally , as noted in section [ seqs ] , stars on sequence c appear to be pulsating in the radial fundamental mode . with these facts in mind , the exceptional cases noted here can be explained naturally if stars on sequences a@xmath1 , a and b are most commonly pulsating in the @xmath31 mode in both their primary pulsation mode and their secondary modes @xmath47 and @xmath48 , while stars on sequences c@xmath1 and c are most commonly pulsating in the radial @xmath30 mode in both their primary pulsation mode and their secondary modes . thus the stars with primary modes on sequence b will have @xmath4 for their secondary modes increased relative to @xmath4 of the radial mode pulsators on sequence c@xmath1 by @xmath49 , thereby causing the exceptions noted in points ( 1 ) and ( 2 ) above . note that the offset @xmath49 is derived from the populous rgb . since the slope of sequences c and c@xmath1 is steeper than the slope of sequences a@xmath1 , a and b , this offset will increase with luminosity and will be larger for agb stars . an alternative explanation for the exceptional cases would be for the stellar mass to decrease from sequence a@xmath1 to a to b , then to increase to sequence c@xmath1 then decrease again to c. this seems contrived and is not consistent with the evolutionary picture described in the next section . although generally not part of this study , in the last panel of figure [ channel ] the secondary periods of stars with their primary period on sequence d are shown as green points . this plot highlights the fact that the periods of pulsation associated with the lsps of sequence d lie mainly on or near sequence b. the faint sequence f between sequences c@xmath1 and c is marked in figure [ wjk - p ] . in panel c@xmath1 of figure [ channel ] , it can be seen that secondary periods of stars whose primary period lies on sequence c@xmath1 form a sequence close to sequence f. this suggests that the sequence f stars are possibly fundamental mode pulsators like the stars on sequence c. however , it is not clear why the sequence f stars should form a distinct sequence . one possible explanation is that they are high mass lpvs ( so that their periods are shorter than the periods typical of sequence c stars ) formed in a relatively recent burst of star formation . against the primary period @xmath11 for stars in the ogle iii catalogue ( black points ) . evolutionary channels cross each sequence from left to right and stars whose primary periods fall in these channels are plotted as orange points . the secondary periods @xmath47 and @xmath48 of the stars with @xmath11 in the orange channels are plotted as green points . see text for further details . the colours can be seen in the online version of the paper . ] for most of the stars in pl diagrams such as figure [ wjk - p ] , mass loss is small so that the evolutionary track of a star is represented by a line of constant mass and pulsation mode . an examination of pulsation models shows that in the ( @xmath50 ) plane , lines of constant mass and pulsation mode have a smaller slope than the sequences ( e.g. * ? ? ? * ) . in figure [ evol ] , channels similar to those in figure [ channel ] have been drawn but now the channels are roughly parallel to lines of constant mass and pulsation mode so they represent evolutionary tracks . the slopes of the tracks are estimated from the pulsation models described in section [ comp_mods ] . for convenience of plotting , at lower luminosities on sequences a@xmath1 , a and b , the orange channels have been drawn more tilted relative to the sequences than the models indicate but this does not qualitatively affect the empirical evolutionary scheme presented here . consider stars in figure [ evol ] whose primary period lies in the orange strip through sequence a. as stars in this channel evolve to higher luminosity , they slowly move from the short period side of sequence a to the long period side . at the same time , there is a declining prominence of the secondary period corresponding to the next highest radial order ( the green strip to the left of the orange channel in sequence a ) and an increasing prominence of the secondary period corresponding to the next lowest radial order ( the green strip to the right of the orange channel in sequence a ) . this means that the primary period most likely detected by observations will move to a lower radial order as the luminosity increases . thus the orange points representing the evolving stars in figure [ evol ] will jump from the long period side of sequence a to the short period side of sequence b because the highest amplitude mode has moved from that associated with sequence a to that associated with sequence b. evolution will then continue up and across sequence b before a jump to sequence c@xmath1 and so on . note that in the empirical scheme shown here , the point of transition of the orange channels from sequence to sequence is judged by eye since current pulsation models do not give reliable theoretical estimates of mode amplitudes . when a star reaches the rgb tip , it will transit to a lower - luminosity core - helium burning phase before rising again up the agb with a slightly shorter period than it had at the same luminosity on the rgb . this situation is shown in figure [ evol ] for the stars in the channel on sequence b. the agb evolution is shown only after the rgb tip luminosity ( @xmath14 ) is exceeded . the evolutionary scenario described in the previous sub - section has implications for the masses of stars on the sequences . note firstly that , on a given sequence , stars of different mass will follow essentially parallel tracks but the higher mass stars will be found at higher luminosities . this means that the mass increases up a sequence , it increases to shorter periods across a sequence at a given luminosity , and it also increases from sequence to sequence at a given luminosity as the radial order increases ( the sequence period decreases ) . note that a mass difference between sequences was invoked above in order to explain why the period ratios between the sequences did not match the period ratios in individual stars . the evolutionary scenario provides an understanding for this empirical finding . in panel a of figure [ channel ] . the horizontal position of each star is given by the value of its coordinate @xmath4[@xmath19=12 ] . the black curve is obtained by using the primary period @xmath11 of each star . the orange segment of this curve represents the sequence a stars in the orange channel through the middle of sequence a. the green curve is obtained by using the secondary periods @xmath47 and @xmath48 of those stars in the orange channel ( i.e. this is the histogram of the green points in panel a of figure [ channel ] ) . the colours can be seen in the online version of the paper . ] the periods of stars in the various sequences formed in pl diagrams by using the primary or secondary periods can be used to estimate the mass variation from sequence to sequence and the mass variation across sequences . figure [ hista ] shows a histogram of the number of stars in a horizontal band defined by @xmath51 in panel a of figure [ channel ] . here we are interested in exploring the mass variation from sequence b to sequence a and the mass variation across sequence a. the positions of sequences a and b are marked on figure [ hista ] . the peak in the green distribution at @xmath4[@xmath19=12 ] @xmath27 1.44 is formed by sequence a stars whose secondary periods are produced by the pulsation mode which produces sequence b. the difference @xmath52 in the positions of this peak and the sequence b peak is caused by the different masses of stars on sequences a and b. pulsation models for low radial order modes at the luminosities and masses of the stars involved here show that , at a given luminosity , @xmath53 ( this result comes from the pulsation models described in section [ comp_mods ] ) . thus the shift @xmath54 corresponds to the average mass of stars on sequence a being approximately 26% larger than the average mass of stars on sequence b. mass variations _ within _ a sequence are also required . these mass variations make a large contribution to the finite width in @xmath4 of the sequences , with the mass increasing to shorter periods across the sequences . let us consider sequence a. the total width of this sequence ( represented by the full - width at half maximum ) is @xmath55 . as well as the mass variation across the sequence , there is also a contribution to the sequence width from the period variability represented by the vertical width of the sub - ridges in the top panel of figure [ mode_amps ] . this contributes approximately @xmath56 to the observed width of sequence a , leaving the contribution from mass variation alone at @xmath57 . the sub - sequences of sequence a ( i.e. the multiple modes ) will also make a small contribution to the width but we ignore this . the width @xmath57 means that the mass increases by about 29% from the longer period to the shorter period side of sequence a. in the evolutionary scenario described above , the mass variation across a sequence should be equal to the mass variation between sequences . remember that stars jump between sequences from the long period edge of one to the short period edge of the other . thus the masses of stars on the long period side of a sequence should be the same as the masses of stars on the short period edge of the adjacent longer period sequence . consequently , the mass difference between the short period edges of two adjacent sequences will be the mass difference across the shorter period sequence . the results given above for sequences a and b show that the mass difference between these two sequences is indeed close to the mass difference across sequence a. the procedure applied above to determine the difference in the average masses of stars on sequences a and b can be applied to other pairs of sequences in which stars pulsate mostly in the same mode . we find that stars on sequence a@xmath1 are @xmath017% more massive than stars on sequence a at luminosities around @xmath21 . the difference in the average masses of stars on sequences c@xmath1 and c can not be readily determined at this luminosity since the mode of sequence c is rarely seen there . however , near @xmath58 , which is on the agb , many stars exhibit primary and secondary oscillations in both of the modes associated with sequences c@xmath1 and c. the period shifts indicate that the stars on sequence c@xmath1 are @xmath016% more massive than those on sequence c. to derive this result we used the formula @xmath59 rather than @xmath53 which is appropriate at lower luminosities on the rgb ( these formulae are derived from the pulsation models described in section [ comp_mods ] ) . finally , we consider sequences b and c@xmath1 . comparing the masses of stars on these two sequences is difficult at any luminosity because of the paucity of stars that exhibit both the sequence b and c@xmath1 modes , and because the stars with primary periods on sequence b appear to be @xmath31 pulsators while stars with primary periods on sequence c@xmath1 appear to be @xmath30 pulsators . we do not make any estimate of the mass difference between these two sequences . although figure [ em_strips ] demonstrates the existence of secondary periods in many lpvs , it does not indicate how the presence of secondary periods varies with luminosity . the luminosity dependence of the secondary periods can be seen in figure [ lum_dep ] . for each sequence , this figure plots @xmath60 against the secondary period relative to the primary period whose position is marked by a vertical blue line . the vertical strips of black points show where individual secondary modes are present in the stars . the radial order of the modes associated with these secondary modes are indicated by coloured numbers , with radial order 1 corresponding to the fundamental mode ( see section [ comp_mods ] for the justification of the assignment of radial orders ) . the positions of the individual sequences are indicated by coloured dots in each panel . the fact that the dots and numbers of the same colour do not align horizontally ( except for the reference mode ) is yet another demonstration of the fact that the ratio of the periods in individual stars do not match the ratio of the periods of the sequences . panel a@xmath1 of figure [ lum_dep ] shows the ratio of secondary to primary period for stars whose primary period lies on sequence a@xmath1 . looking at this panel , we see three closely - spaced vertical bands under and slightly left of the red 4 . these bands are caused by the three modes occurring on sequence a. the longest period mode ( the @xmath31 mode ) becomes dominant as the luminosity increases . this result is also seen for stars with their primary period on sequence a ( see panel a ) . there are five vertical bands associated with sequence a and these are caused by the the three modes present within this sequence . as discussed in section [ pr1 ] , the shortest period band is produced by stars whose primary mode has @xmath31 and whose secondary mode has @xmath30 while the longest period band is produced by stars whose primary mode has @xmath30 and whose secondary mode has @xmath31 . the shortest period band extends to higher luminosities than the longest period band which suggests that the @xmath31 mode is more prominent as the primary period ( i.e. the one with largest amplitude ) at higher luminosities . once again , we note that the dominance of the @xmath31 mode at higher luminosities on sequence a is consistent with the findings of @xcite from kepler observations but different from the findings of @xcite who suggest that the radial @xmath30 mode becomes dominant at high luminosities . the stars with their primary mode on sequence b show three of the outer bands of sequence b are cause by the 1-year alias of the primary period , so this feature is not real . a similar 1-year alias can be seen in the bands of sequence a near @xmath61 if the figure is expanded . ] vertical bands associated with the primary mode ( panel b of figure [ lum_dep ] ) . this structure indicates the presence of two close modes but they seem to be confined to rgb stars only ( @xmath62 ) . only one mode appears to be present on the agb . panel b in figure [ lum_dep ] reveals the presence of apparent extra modes not present in any other panel . in a simple sequential assignment of radial orders , there appear to be secondary modes present corresponding to radial orders 5 and 6 . however , looking at panel a@xmath1 immediately above panel b , it can be seen that the mode structure associated with the radial orders 4 and 5 in panel a@xmath1 is reproduced exactly in panel b. another point to notice is that the horizontal position of radial order 2 in panel a@xmath1 coincides exactly with the position of sequence b. we conclude from these facts that stars whose primary mode lies on sequence b are of two types : ( 1 ) stars pulsating in the third radial order with secondary periods in the second and fourth radial orders , and ( 2 ) stars pulsating in the second radial order with secondary periods in the third , fourth and fifth radial orders . the type 2 stars are more massive than type 1 stars and they have masses typical of stars on sequence a@xmath1 . these type 2 stars are really stars belonging to sequence a@xmath1 but the observed amplitude of the radial order 2 pulsations is larger than the observed amplitude of the radial order 5 pulsations so they end up on sequence b. the apparent extra radial orders in panel b have been labelled by their true radial orders using numbers in brackets . if we consider the stars on sequence b whose primary mode is of radial order 3 ( panel b of figure [ lum_dep ] ) , we see that the secondary modes pulsating in radial order 4 form two bands with the shorter period band being overwhelmingly dominant ( the apparent band on the short period side of the radial order 4 group belongs to radial order 3 of the type 2 stars examined in the last paragraph ) . given that there are two possible primary modes of radial order 3 , the dominance of the shorter period band in panel b suggests that the dominant primary mode is the longer period one . the separation @xmath63 of the two modes associated with radial order 3 ( derived from the separation of the vertical bands associated with this order in panel b of figure [ lum_dep ] ) is consistent with the separation of the @xmath64 and the @xmath65 modes of radial order 3 in the models of @xcite . these results therefore suggest that the @xmath64 and the @xmath65 modes are present on sequence b and that the @xmath31 mode is dominant . the direct finding that the @xmath31 mode is dominant on both sequences b and a is consistent with the suggestion in section [ chans ] that the @xmath31 mode is dominant on each of sequences b , a and a@xmath1 . in this section , period ratios observed in individual stars are compared to period ratios in radial pulsation models of red giants . the linear , non - adiabatic pulsation models were computed with the code described in @xcite using the same input parameters ( abundances , mixing - length , turbulent viscosity parameter , core mass ) as in that paper . evidence was presented in section [ chans ] that , in a given star , both the dominant primary pulsations and the secondary pulsations correspond to modes of the same spherical degree @xmath2 . thus , although the nonradial @xmath31 mode appears to dominate the primary pulsations in sequences a@xmath1 , a and b , provided the ratio of the @xmath31 to the @xmath30 periods is approximately the same for radial orders 3 to 5 associated with these sequences , the observed period ratios in individual stars should be the same as the period ratios in radial pulsation models . in figures [ lum_dep+mods ] and [ lum_dep+wrong_mods ] the observed period ratios in individual stars are compared to the period ratios in radial pulsation models . in figure [ lum_dep+mods ] , it is assumed that the radial mode associated with sequence c ( the mira sequence ) is the fundamental mode ( radial order 1 ) and that sequences c@xmath1 , b , a and a@xmath1 correspond to successively higher radial orders of pulsation . it can be seen that for each of the panels in figure [ lum_dep+mods ] , the model period ratios match the observed period ratios well for some mass in the range 0.8 to 2.4m@xmath66 except for the highest mass and luminosity models in panels c@xmath1 and c. this problem is discussed in section [ model - mass ] . note that in panel b , there is a mismatch between the models and those vertical bands that really belong to sequence a@xmath1 ( as described in section [ ldep_pr ] ) but this is expected because of the different masses of stars on sequences a@xmath1 and b. overall , these results provide strong evidence that the primary pulsation modes associated with sequences c , c@xmath1 , b , a and a@xmath1 belong to radial orders 1 , 2 , 3 , 4 and 5 . it also provides strong evidence that both primary and secondary pulsations in a given star tend to belong to the same @xmath2 value . in figure [ lum_dep+wrong_mods ] , it is assumed that the radial mode associated with sequence c@xmath1 is the fundamental mode ( radial order 1 ) and that sequences b , a and a@xmath1 correspond to successively higher radial orders . clearly , the mode marked by the cyan number 1 ( sequence c ) has no explanation in terms of radial modes with this assignment of radial orders . furthermore , even for higher radial orders , the ratios of mode periods in the models fail to match the observed ratios . we therefore reject this modal assignment . the finding here that sequences c , c@xmath1 , b , a and a@xmath1 correspond to radial orders 1 , 2 , 3 , 4 and 5 is consistent with most earlier findings ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? however , this assignment is inconsistent with the suggestion of @xcite and @xcite that sequence c@xmath1 corresponds to radial order 1 . in terms of nonradial modes , we find direct evidence for the presence of the @xmath65 mode on sequences a and b and for the @xmath32 mode on sequence a. in their analysis of ogle iii data , @xcite claim to find the presence of the two nonradial modes . firstly , they associate the @xmath67 mode of radial order three - mode @xmath68 corresponds to radial order 3 in this paper . ] with sequence b while the only nonradial mode we find has @xmath65 . secondly , they find the @xmath65 mode of radial order 5 associated with sequence a whereas we associate radial order 5 with sequence a@xmath1 and we find the @xmath65 mode of radial order 4 associated with sequence a. because @xcite only looked at lpvs on the rgb designated as osargs in the ogle iii catalogue , all lpvs with periods on sequences a@xmath1 and c were excluded ( by definition , osargs on the rgb do not have these modes ) so they had an incomplete picture of the modes involved . in addition , the petersen diagram they used meant that a number of modes overlapped on the figure causing confusion . in the lower panel of figure [ em_strips ] , the various clusters of points representing period ratios of specific pairs of modes tend to form ridges that are not horizontal . for example , we now know that the ridge at ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) @xmath27 ( 1.3 , 0.3 ) corresponds to stars whose primary pulsation mode belongs to radial order 4 and whose secondary pulsation mode belongs to radial order 2 . this group of points , and most of the other groups of points , form ridges such that the period ratio becomes smaller when moving from the long period side of the primary mode sequence to the short period side . it was shown in section [ mvarinseq ] that the mass of stars in a sequence increases when moving from the long period side to the short period side of the sequence . this suggests that the cause of the non - zero slope of the ridges in the lower panel of figure [ em_strips ] is mass variation across sequences . the models shown in figure [ lum_dep+mods ] confirm this . if we take the example given above for the ridge at ( @xmath10[@xmath19=12 ] , @xmath26/@xmath11 ) @xmath27 ( 1.3 , 0.3 ) , panel a of figure [ lum_dep+mods ] shows that the period ratio of radial order 2 to radial order 4 modes decreases as the model mass increases , just as suggested by the slope of the ridges . it was noted in section [ pr1 ] that the ratio of the periods of the @xmath31 and @xmath30 modes of radial order 4 increased slightly from the long period side to the short period side of sequence a ( the opposite slope to the cases considered in the previous paragraph ) . this result provides evidence for the mass dependence of the period ratio of the @xmath31 and @xmath30 modes , with the period ratio increasing with mass . combining the results for the variation of the period ratio across sequence a from section [ pr1 ] with the results for the variation of mass across sequence a from section [ mvarinseq ] , we find that @xmath69 for the stars on sequence a. this result provides a test for nonradial pulsation models of luminous red giant stars . against the primary period @xmath11 for stars in the ogle iii catalogue ( black points ) . the coloured lines show the positions of modes from the first five radial orders for stars of mass 0.8 , 1.6 and 2.4 m@xmath66 . the lines for each mode are plotted only in regions where they cross the sequence in which the mode is expected to dominate . the first two radial orders are shown as continuous lines and they correspond to the fundamental and first overtone radial pulsation modes . the lines for the next three radial orders are shown as dotted lines because the periods shown have been increased by @xmath70 from the period of the radial modes in an attempt to simulate the periods of the @xmath71 pulsation mode . the colours can be seen in the online version of the paper . ] until now in this paper , only period ratios have been compared to models because period ratios are less sensitive to model parameters than are the periods themselves . in addition , to compare model and observed periods in ( say ) the ( @xmath19 , @xmath4 ) plane , conversions from model @xmath72 and @xmath46 to the observed @xmath19 are required . these conversions are not particularly well known for cool stars such as lpvs and there are additional problem due to interstellar and circumstellar reddening and the distance modulus ( although the latter is now well - determined for the lmc ) . keeping these problems in mind , in this final sub - section , we do compare the models to the observations in the ( @xmath19 , @xmath4 ) plane . for each stellar model , @xmath72 and @xmath46 have been converted to @xmath13 and @xmath5 in the 2mass system using the tabulated bolometric corrections obtained from the model atmosphere calculation by @xcite and @xcite for a metal abundance @xmath73 } = -0.5 $ ] which is appropriate for the lmc and which is consistent with the metal abundance used in the model calculations . conversion of @xmath13 and @xmath5 in the system of @xcite and @xcite to the 2mass system is done using the conversions in @xcite . an lmc distance modulus of 18.54 and a reddening @xmath74 of 0.08 from @xcite are used , along with the reddening law of @xcite to get the reddening at @xmath13 and @xmath5 . figure [ masses ] is a plot in the ( @xmath19 , @xmath4 ) plane of the periods and luminosities of modes from the first five radial orders of stars with masses of 0.8 , 1.6 and 2.4 m@xmath66 . for each mode , only a short line segment is plotted centred on the position where it crosses the sequence in which the mode is assumed to be dominant . note that the models computed here only produce periods for radial modes . however , it was shown in sections [ chans ] and [ ldep_pr ] that the @xmath71 mode is dominant on sequences a@xmath1 , a and b , and in section [ comp_mods ] it was shown that these sequences correspond to radial orders 5 , 4 and 3 , respectively . furthermore , in sections [ pr1 ] and [ chans ] it was shown that the period of the @xmath65 mode is longer than the period of the @xmath64 mode by @xmath33 . hence , to compare the models with the observations , the computed periods for radial orders 3 , 4 and 5 have been increased by @xmath70 before being plotted in figure [ masses ] . this correction is made in order to simulate the @xmath31 mode periods . the resulting periods are shown in figure [ masses ] by dashed lines . we now make a detailed comparison of the models with the observations . looking at sequence c in figure [ masses ] , it can be seen that fundamental mode pulsation in stars with masses in the range 0.82.4 m@xmath66 explains most of these stars , with some stars outside this mass range required to explain the full length of the sequence ( note that the present - day population of red giant stars in the lmc peaks at masses around 1.6 m@xmath66 e.g. @xcite ) . stars in a similar mass range , but pulsating in the radial first overtone mode can explain the stars on sequence c@xmath1 . similarly , this mass range fits sequences @xmath75 , @xmath76 and @xmath77 but at lower luminosities , as expected from the empirical evolution scheme described in section [ evsec ] . the mass required to fit these three sequences at a given luminosity increases when moving from sequence @xmath75 to @xmath76 to @xmath77 , consistent with the results derived in section [ mvarinseq ] from the empirical evolution scheme . in addition , the width of sequence a requires a significant range of masses to be present at a given luminosity . this is also consistent with the results derived in section [ mvarinseq ] . given the various results presented in this paragraph , there appears to be reasonable agreement between the models and the observations in the ( @xmath60 , @xmath4 ) plane in spite of the potential difficulties mentioned above . there is one parameter on which the models and the empirical evolution scheme are in relatively poor agreement . this is the transition luminosity at which stars move from one sequence to the next , especially in the case of the transition from sequence c@xmath1 to sequence c. looking at the evolutionary track for a 1.6 m@xmath66 star ( i.e. the orange lines in figure [ masses ] ) , we see that the first overtone mode does not cross the long period edge of sequence c@xmath1 until @xmath78 . according to the empirical evolution scheme , the star should then jump to the short period edge of sequence c and start pulsating in the fundamental mode at @xmath78 . however , the period of the fundamental mode of the 1.6 m@xmath66 star at this stage puts the star at the _ long _ period edge of sequence c or beyond . this suggests that the ratio of the fundamental to first overtone periods in the models is too large . actually , the disagreement between the models and the observations is not as bad as just suggested . if we look at panel c@xmath1 of figure [ lum_dep+mods ] , we see that the logarithmic period difference @xmath34 between the fundamental mode to the first overtone mode is @xmath00.26 around @xmath79 . this is significantly larger than the difference @xmath80 between the long period edge of sequence c@xmath1 and the short period edge of sequence c. at @xmath81 , @xmath34 for the 1.6 m@xmath66 star is indeed @xmath82 but by @xmath83 @xmath34 is much too large at @xmath84 so that there is still a discrepancy between the model period ratios for the first two modes and the observed ratio of these two modes . it is worth noting that the discrepancy discussed here seems to be worst at high masses and luminosities . at @xmath85m@xmath66 , the models predict a transition from the long period edge of sequence c@xmath1 to the short period edge of sequence c which is as predicted by the empirical evolution scheme . finally , we take a look at the period difference between rgb and agb stars . the lines drawn in figure [ masses ] belong to both rgb and agb stars . the computed separation of the rgb and agb periods for each mode at a given luminosity is almost imperceptible on the plot ( the separation can just be seen for the 0.8 m@xmath66 models ) . it is certainly much less than the period shift of @xmath86 evident on sequence a when moving from the rgb at @xmath62 to the agb at @xmath87 . the observed period shift can not be explained by mass loss between the rgb tip and the agb since mass loss increases the period of pulsation whereas the period is observed to decrease . the models thus predict that there is a very small period change between the rgb to the agb for a given mode . a possible interpretation of the much larger period shift observed on sequence a when going from the rgb to the agb is that the agb is dominated by radial ( @xmath30 ) pulsators while the rgb is dominated by nonradial @xmath31 pulsators ( remember that @xmath49 for the period ratio of the @xmath31 and @xmath30 modes ) . however , this result conflicts with the results in section [ ldep_pr ] which suggested that the @xmath31 mode becomes more dominant at higher luminosity . the observational resolution of this conflict awaits larger data sets such as those that will come from ogle iv . multiple pulsation modes in stars in the ogle iii catalogue of lpvs in the lmc have been examined . it has been shown that the ratios of the multiple periods in individual stars do not agree with the ratios of the periods of the sequences of lpvs in the pl diagram . this leads to the conclusion that the average masses of stars on the different sequences vary , with the shorter period sequences containing more massive stars . an examation of the evolution of stars in the pl diagram shows that the mass variation between sequences is the same as the mass variation across individual sequences . as an example , it is shown that the average mass of a star on sequence a is @xmath026% higher than the average mass of a star on sequence b and that the mass variation across sequence a is of a similar magnitude , with the more massive stars on the shorter period side of the sequence . in general , the masses of stars at a fixed luminosity on adjacent sequences are found to increase by @xmath01626% towards the shorter period sequence . fitting pulsation models to the period ratios in multimode pulsators shows that the sequences c , c@xmath1 , b , a and a@xmath1 correspond to pulsation in radial orders 1 to 5 , respectively ( with radial order 1 corresponding to the fundamental mode ) . closely - spaced modes revealed by the period ratios in individual stars show the presence of nonradial modes . the observed period ratios indicate the presence of the @xmath30 , 1 and 2 modes on sequence a ( radial order 4 ) and the @xmath30 and 1 modes on sequence b ( radial order 3 ) . the relative populations and amplitudes of the modes of different @xmath2 show that the @xmath31 mode is the most common and of the largest amplitude in both sequences . the positions of secondary mode sequences relative to primary mode sequences in the pl diagram also suggest that the @xmath31 mode is the most prominent primary and secondary mode on sequences a@xmath1 , a and b and that the @xmath30 mode is the most prominent primary and secondary mode on sequences c@xmath1 and c. this result is supported by the good fit of the period ratios in radial pulsation models to the observed period ratios . i thank the anonymous referee whose constructive and detailed comments have considerably improved the readability of the paper . bessell m. s. , scholz m. , wood p. r. , 1996 , a&a , 307 , 481 cardelli j. a. , clayton g. c. , mathis j. s. , 1989 , apj , 345 , 245 carpenter j. m. , 2001 , aj , 121 , 2851 derekas a. , kiss l. l. , bedding t. r. , kjeldsen h. , lah p. , szab g. m. , 2006 , apj , 650 , l55 dziembowski w. a. , gough d. o. , houdek g. , sienkiewicz r. , 2001 , mnras , 328 , 601 dziembowski w. a. , soszyski i. , 2010 , a&a , 524 , aa88 hinkle k. h. , lebzelter t. , joyce r. r. , fekel f. c. , 2002 , aj , 123 , 1002 fox m. w. , wood p. r. , 1982 , apj , 259 , 198 houdashelt m. l. , bell r. a. , sweigart a. v. , wing r. f. , 2000 , aj , 119 , 1424 houdashelt m. l. , bell r. a. , sweigart a. v. , 2000 , aj , 119 , 1448 ireland m. j. , scholz m. , wood p. r. , 2008 , mnras , 391 , 1994 ita y. , et al . , 2002 , mnras , 337 , l31 ita y. , et al . , 2004 , mnras , 347 , 720 keller s. c. , wood p. r. , 2006 , apj , 642 , 834 kiss l. l. , bedding t. r. , 2003 , mnras , 343 , l79 mosser b. , et al . , 2013 , a&a , 559 , aa137 nicholls c. p. , wood p. r. , cioni m l. , soszyski i. , 2009 , mnras , 399 , 2063 nicholls c. p. , wood p. r. , cioni m .- r . l. , 2010 , mnras , 405 , 1770 nicholls c. p. , wood p. r. , 2012 , mnras , 421 , 2616 nie j. d. , wood p. r. , nicholls c. p. , 2012 , mnras , 423 , 2764 nie j. d. , wood p. r. , 2014 , aj , 148 , 118 olivier e. a. , wood p. r. , 2003 , apj , 584 , 1035 skrutskie m. f. , et al . , 2006 , aj , 131 , 1163 soszyski i. , udalski a. , kubiak m. , szymanski m. , pietrzynski g. , zebrun k. , szewczyk o. , wyrzykowski l. , 2004a , aca , 54 , 129 soszyski i. , et al . , 2004b , aca , 54 , 347 soszyski i. , et al . , 2007 , aca , 57 , 201 soszyski i. , et al . , 2009 , aca , 59 , 239 soszyski i. , wood p. r. , 2013 , apj , 763 , 103 soszyski i. , wood p. r. , udalski a. , 2013 , apj , 779 , 167 stello d. , et al . , 2014 , apj , 788 , ll10 stothers r. b. , 2010 , apj , 725 , 1170 tabur v. , bedding t. r. , kiss l. l. , giles t. , derekas a. , moon t. t. , 2010 , mnras , 409 , 777 takayama m. , saio h. , ita y. , 2013 , mnras , 431 , 3189 takayama m. , wood p. r. , ita y. , 2015 , mnras , in press wood p. r. , sebo k. m. , 1996 , mnras , 282 , 958 wood p. r. , et al . , 1999 , iaus , 191 , 151 wood p. r. , 2000 , pasa , 17 , 18 wood p. r. , olivier e. a. , kawaler s. d. , 2004 , apj , 604 , 800 wood p. r. , olivier e. a. , 2014 , mnras , 440 , 2576

the fixed - target mipp experiment , fermilab e907 , was designed to measure the production of hadrons from the collisions of hadrons of momenta ranging from 5 to 120 gev / c on a variety of nuclei . these data will generally improve the simulation of particle detectors and predictions of particle beam fluxes at accelerators . the spectrometer momentum resolution is between 3 and 4% , and particle identification is performed for particles ranging between 0.3 and 80 gev / c using , time - of - flight and cherenkov radiation measurements . mipp collected events of 120 gev main injector protons striking a target used in the numi facility at fermilab . the data have been analyzed and we present here charged pion yields per proton - on - target determined in bins of longitudinal and transverse momentum between 0.5 and 80 gev / c , with combined statistical and systematic relative uncertainties between 5 and 10% .

1404.5882

a growing number of neutrino experiments conducted at proton accelerators derive their neutrino beams from horn - focused beams of pions and kaons which result from proton - nucleus collisions in low - z materials . at the neutrinos at the main injector ( numi ) facility at fermilab , hadron production uncertainties in monte carlo ( mc ) simulations generally dominate the uncertainties of the neutrino flux predictions at the level of 1520% , and are a limiting factor in the neutrino and anti - neutrino cross - section measurements being done by many numi - based experiments @xcite . one of the goals of the main injector particle production ( mipp ) experiment was to measure the hadron production yield off of an actual numi target with 120 gev / c protons from the main injector ( mi ) to within 5% in order to verify and validate the monte carlo calculations of the numi flux . here we report a measurement of charged pion yield per 120 gev / c proton - on - target in approximately 120 bins of longitudinal and transverse momentum . the statistical and systematic uncertainties in most bins are between 5 and 10% . for the mipp experiment , an open geometry 25 m long spectrometer was assembled with two dipole magnets for momentum determination , a 1.5 m long time - projection chamber @xcite ( tpc ) located just downstream of the interaction region , and 3 drift chambers ( dcs ) and 2 multiwire proportional chambers ( mwpcs ) located further downstream for particle tracking . the tpc sits inside the most upstream dipole magnet , and targets are placed just upstream of the tpc . three wire chambers @xcite ( bcs ) positioned across 36 meters upstream of the target are used to track incident beam particles . a schematic of the spectrometer is shown in fig . [ fig : mippspect ] . the electric field used to drift ionization electrons produced in the mipp tpc is aligned the mipp was designed to provide particle identification ( pid ) with @xmath2 separation across the momentum range of a few hundred mev / c to greater than 80 gev / c using @xmath3information from the tpc ( 0.21.2 gev / c ) , a plastic scintillator - based time - of - flight ( tof ) detector ( 0.52.5 gev / c ) , a segmented gas cherenkov detector @xcite ( 220 gev / c ) and a gas ring imaging cherenkov @xcite ( rich ) detector ( 480 gev / c ) . electromagnetic and hadronic calorimeters are used to measure forward - going neutrons and photons @xcite . the high multiplicities present in this data set complicate the use of the cherenkov and tof detectors , and this analysis relies on measurements from the tpc and rich detectors . measurements from the tof detector are used to estimate backgrounds . the numi target used in this measurement was a spare target that was eventually used by the numi complex after the mipp data run . the target was designed for operation in the low - energy configuration of the numi beamline and consists of a 90-cm long , 3-cm diameter aluminum vacuum can encompassing 47 2-cm thick graphite slabs , adding up to two nuclear interaction lengths of material . the downstream end of the tube was positioned 8 cm away from the upstream end of the tpc active volume . the incident beam was 120 gev / c protons , slow - extracted directly from the mi . a pinhole collimator was used to reduce the incident mi proton beam flux by 8 orders of magnitude , such that the rate of incident beam pileup in the target over the 16 @xmath4 required to read out the tpc was reduced to a few percent . in order to ensure that the incident beam was centered on the face of the target , a numi trigger detector consisting of three thin ( @xmath5 ) overlapping pieces of plastic scintillator was mounted on the upstream face of the numi target . the light from each of the scintillator pieces was detected by a photomultiplier tube ( pmt ) . the middle and most downstream pieces of scintillator had circular holes 2 mm and 6 mm in diameter , respectively , drilled in the center . a `` 2-mm '' trigger was formed via a coincidence of a signal from the upstream beam counters with the signal from the upstream scintillator of the numi trigger and the absence of signals from the middle and downstream scintillators of the numi trigger . in a similar fashion , a second `` 6-mm '' trigger was defined by a signal from the upstream and middle scintillators and the absence of a signal from the downstream scintillator . this second trigger was pre - scaled such that the profile of the beam of protons striking the numi target in the mipp experiment matched the gaussian profile of the 1 mm beam width observed in the numi beam facility . incident beam pileup was reduced via a veto if two triggers fire within 1 @xmath6s of each other . shown in fig . [ fig : trgtprof ] is the measured probability that an incoming beam proton interacts with the target material as a function of the position of the proton at the upstream face of the target . the data used to generate fig . [ fig : trgtprof ] were collected with a minimum bias trigger during the numi target commissioning period and are not used in this analysis . during this commissioning period the beam was randomly swept across the upstream face of the target and some regions of the target were exposed to more beam than others . as a result , the outlines of the aluminum vacuum tube , water cooling pipe holes and graphite slabs is non - uniform . the reconstructed positions of the incident beam particles on the upstream face of the target are measured using data from the bcs . the reconstruction of the incident beam trajectory uses the charge and time information recorded in the 4 1-mm pitch wire planes of each bc , resulting in 0.1 mm track position reconstruction resolution at the upstream face of the target . the data recorded during the target scan involved random beam positioning , and some regions of the upstream face of the target received different levels of beam exposure . the color / shade in fig . [ fig : trgtprof ] is determined from the normalized difference between the number of events with greater than 3 secondary tracks observed in the downstream spectrometer and the number of events with a single beam proton in the downstream spectrometer . darker regions correspond to higher density materials , lighter regions to less dense materials . the graphite slabs located inside the numi target are clearly visible , with holes at the top and bottom where aluminum water cooling pipes , empty and exposed to air during data taking in the mipp experiment , run along the length of the target . the outer aluminum tube containing the graphite and water cooling pipes is also visible . the graphite slabs were actually found to be rotated @xmath7 about the longitudinal axis ; this rotation has been removed in fig . [ fig : trgtprof ] . the cross - hairs in fig . [ fig : trgtprof ] represent the width and height of the graphite slabs . the positions of the cross - hairs were determined by fitting the edges of the @xmath8- or @xmath9-projection of the plot . the measured width of the graphite slabs is 6.36 mm ; the technical specification of the width of the slabs in the numi target is 6.4 mm . the center of the 2-mm wide circle near the center of fig . [ fig : trgtprof ] represents the mean distribution of reconstructed incident beam positions on the face of the target for 2-mm triggered events . the beam center position is offset by 0.045 mm in the horizontal direction , and 0.174 mm in the vertical direction , from the center of the target . mc simulations are used to determine event and track selection criteria and acceptance , track reconstruction and particle identification efficiency corrections in the analysis . interactions of 120 gev / c protons striking the numi target in the mipp experiment are simulated using fluka ( v2005 ) @xcite for event generation ( e.g. , 120 gev / c proton interactions on the numi target ) and geant3 @xcite for particle trajectory tracking . the fluka simulation generates primary , secondary , tertiary , etc . interactions of particles within the target and housing , and has a detailed geometric description of the numi target , the same geometry employed by the minos experiment . the simulated beam profile at the target has a full - width at half - maximum ( fwhm ) of 0.26 mm in the horizontal direction and 0.29 mm in the vertical direction centered on the upstream face of the target , which are the measured widths of the beam spot size on the numi target in the numi facility , and is within 10% of of the measured beam spot size on the numi target in the mipp experiment . fluka tracks each particle produced in the target until it reaches the surface of the target , which is the outer edge of the aluminum pipe encasing the graphite slabs of the target . the next stage of the mc generation is geant3 , which uses the output of the fluka simulation as input , and tracks each particle taking into account multiple scattering , energy loss and decays through a detailed geometric description of the mipp spectrometer . geant `` hits '' are recorded in each detector volume until the particle either loses all energy or is well outside the volume of the mipp spectrometer . the last stage of the mc simulation converts the geant hits to simulated digital signals , tuned to match data recorded in the experiment . the mipp event reconstruction includes reconstruction of the trajectory of the primary beam particle using data from three wire chambers located upstream of the target , reconstruction of the secondary particles originating from the target , and matching the secondary particles to data recorded in specific channels in the tof , cherenkov , rich , emcal and hcal detectors . the secondary particle trajectories are reconstructed by merging reconstructed track segments from hits in the tpc detector with track segments formed from hits in the downstream dcs and pwcs . [ fig : evd ] shows an event display of numi target data recorded in the mipp experiment . in the analysis of the numi target data , monte carlo simulation studies indicate that the momentum resolution is @xmath10 , and the transverse momentum resolution is less than 20 mev / c for all momenta . the @xmath3is determined for every reconstructed track from tpc hits based on the charge recorded on 8 mm @xmath11 12 mm charge - sensitive pads in the readout - plane of the tpc . time - dependent corrections , relevant on the order of hours to days , are applied to the data to account for monitored changes in water - vapor and oxygen contamination in the tpc gas . tracks in this analysis had 2090 associated tpc hits providing @xmath3resolutions between 15 and 25% . reconstructed tracks are matched to hits recorded in the tof . temperature - dependent and cross - talk corrections are applied to the tof data , and improve the timing resolution of the detector from 1.2 ns to 0.4 ns . as a result , the tof data provide @xmath12 separation up to about 2 gev / c . the recorded flight time , @xmath13 , in the tof detector is converted to an invariant @xmath14 via @xmath15 where @xmath16 is the reconstructed momentum of the particle and @xmath17 is the reconstructed flight distance from the target to the tof detector . finite resolutions in @xmath13 and @xmath17 occasionally result in negative values of @xmath14 . due to the high multiplicities of secondaries in the numi data set , approximately 50% of all tof data recorded per event are a result of 2 or more particles passing through the same tof bar . it is therefore impossible to disentangle these particles in the tof data . the remaining 50% of the data from the tof detector are used in conjunction with the @xmath3measurements as described in section [ section : tpcdedxmeas ] . particles in the rich detector produce light cones which are reflected to form a ring of light on an array of approximately 2300 1/2-inch pmts . the high segmentation of the rich detector allows multiple rings to be clearly distinguished and matched to reconstructed tracks . the efficiency of matching a reconstructed track to a reconstructed rich ring is @xmath18 , and the high multiplicity of secondaries is not an issue for this detector . [ fig : richringvsmom ] is a scatter plot of the reconstructed rich ring radius versus matched reconstructed particle momentum . the predicted bands for different particle types are drawn with dashed curves . given a particle s momentum , @xmath16 , the matched rich ring radius is converted to an invariant @xmath14 assuming the small - angle approximation : @xmath19 where @xmath20 is the refractive index ( 1.00045 for the co@xmath21 used in the mipp rich detector ) , @xmath22 is the reconstructed rich ring radius and @xmath23 is the length of the rich radiator volume ( 990 cm ) . this analysis is a measurement of the pion yield off the numi target , @xmath24per proton - on - target ( pot ) . pot is defined as the number of reconstructed events that pass the event selection described in section [ sec : eventselection ] , and includes events where the proton does not interact in the target . yields are extracted from tpc @xmath3and rich @xmath14 distributions . corrections are applied to each measurement to account for spectrometer geometric acceptance , track reconstruction efficiency , pid detector geometric acceptance and pid detector efficiency : @xmath25 in general , unless otherwise noted , the measurement and calculations of corrections to be applied to the data are done for positive and negative particles separately . prompt muon production dominated by decays of charm mesons is known to be a @xmath26 effect , thus we treat all muons as arising from pion decay and pion mc pid distributions include contributions from muons . mc studies confirm that this contribution is negligible in all bins in the analysis . a small ( @xmath27 ) correction , based on a comparison between reconstructed and true momenta of mc tracks , is applied to the reconstructed momenta of tracks through the mipp spectrometer to account for energy loss and scattering , as well as biases introduced by the reconstruction algorithm . the overall momentum scale is calibrated using reconstructed primary beam protons that pass through the target and reconstructed k@xmath28 mass from pairs of oppositely charged tracks produced off the target . the primary beam momentum was found to agree with the expected 119.6 gev / c from the mi . the invariant mass distribution of combinations of oppositely charged tracks in data showed a peak near the k@xmath28 above a flat background . a gaussian fit to this peak had a central value @xmath29 lower than the pdg @xcite value . the momenta of tracks that contribute to the k@xmath28 mass peak is peaked around 1 gev / c . a linear interpolation between these two measurements , one at 1 gev / c ( 0.85% offset ) and the other at 120 gev / c ( 0% offset ) is used to correct for absolute momentum . event selection in this analysis is designed to reject events with multiple incident beam particles ( protons ) while requiring that the beam be centered on the numi target . we require exactly one reconstructed incident beam track from data recorded in the upstream beam wire chambers , a reconstructed beam track time that falls within the expected 18.9 ns - wide window from the accelerator rf bucket , and a reconstructed beam track position that falls within 0.648 cm of the center of the upstream face of the target . because the time to drift ionization from tracks out of the tpc volume is 16 @xmath4 and the drift is in the vertical direction , particles traversing the center of the tpc many @xmath4 before [ after ] the event trigger will have shorter [ longer ] recorded times and therefore appear to be well below [ above ] the center of the tpc . events with an excess of tpc tracks appearing at the top or bottom of the tpc were rejected , and mc studies , where no incident beam pileup is simulated , indicate the rejection of these events to have a negligible bias . track selection in this analysis , applied after event selection , is designed to reject tracks that are poorly reconstructed or do not originate from the target . we require all reconstructed tracks to be above a `` goodness - of - fit '' ( gof ) parameter threshold , derived from the trajectory fit residuals . the gof threshold was determined by a sharp cut - off in the gof distribution seen in both data and mc . all tracks must have at least 8 tpc hits , although in practice tracks with momenta in the analysis bins have @xmath30 tpc hits . finally , every track is required to have a trajectory that has a distance of closest approach ( doca ) to the target cylinder that is within 2 cm in @xmath22 and within 5 cm in @xmath31 . the resolution of the doca in @xmath22 is 0.7 cm and 2.8 cm in @xmath31 . no significant disagreement was found between the shapes of the distributions of the track selection criteria in data and mc . the yields of secondary pions produced in the target are measured in bins of @xmath32 and @xmath33 chosen to keep the statistical uncertainty in each bin less than @xmath34 while maintaining a bin width that exceeds the momentum resolution of the spectrometer by at least a factor of 4 . a total of 150 bins are defined in this analysis ; however due to limited statistics in some bins , positive [ negative ] pion yields are reported for 124 [ 119 ] bins covering 0.30 to 80 gev / c and 6 bins from 0 to 2 gev / c in @xmath33 . geometric acceptance and track reconstruction efficiency corrections are determined using mc simulations which have detailed descriptions of the target , spectrometer and detector geometries . the combined geometric acceptance of the spectrometer , track reconstruction efficiency and track selection efficiency is shown in fig . [ fig : recoeff ] as a function of . the geometric acceptance of the pid detectors is shown in fig . [ fig : pidaccept ] . the shade of the boxes indicates the scale of the effect , where 100% efficiency or acceptance is the darkest shade and the lightest shade represents 0% . both plots show results for negative particles ; positively charged particles have very similar efficiencies and acceptances . all tracks are required to have a reconstructed tpc track segment . the acceptance of the rich requires tracks to traverse the spectrometer , which is limited to high momentum . this is reflected in the 100% acceptance of the tpc @xmath3pid and the much smaller rich pid acceptance . the hashed bins have been excluded due to poor statistics . every reconstructed track has a corresponding measurement of tpc @xmath3 . in any given slice of total momentum , the distribution of log(@xmath3 ) for any particle type is nearly gaussian , and the @xmath35(@xmath3 ) distributions in narrow bins of are very nearly gaussian . we therefore fit the @xmath36log(@xmath3 ) distributions to a sum of six gaussians , 2 peaks each for @xmath37 and @xmath16 . the resulting distribution is centered about zero by construction . fitting the data in this way reduces the number of free gaussian parameters from 18 to 12 : 3 means , 3 widths , and 6 amplitudes . the fit function is rewritten as @xmath38 + \nonumber \\ & & a_{\pi^-}\left[f^{-}_{e\pi}g^-_e(y ) + g^-_\pi(y ) + f^-_{p\pi}g^-_p(y)\right ] \end{aligned}\ ] ] where @xmath39 is the gaussian function for each particle type , @xmath9 is the measured value of @xmath35@xmath3 , @xmath40 is the mean , @xmath41 is the width , @xmath42 is the fit amplitude of the pion peak , and @xmath43 ( @xmath44 ) is the amplitude ratio of the electron ( proton ) peak to the pion peak . one feature of the @xmath3distributions is that at higher momenta very large fractions of the proton peak fall under the pion peak and the fit to 6 peaks fails . however , in this range , the protons are clearly distinguished from pions and electrons in the tof . [ fig : tofvstpc ] shows the reconstructed @xmath45 versus the @xmath3of tracks where no other reconstructed trajectories traverse the tof scintillator bar . the protons are very clearly visible in the tof , whereas these protons fall under the pion and electron @xmath3peaks on the x - axis . the @xmath46 fraction is determined from these data by assuming all particles with @xmath45 between 0.5 and 1.2 gev@xmath47/c@xmath48 are protons , and fitting the @xmath3distributions for tracks with tof @xmath45 below 0.5 . the results of these fits are then used as a constraint to fits of the tpc @xmath3distribution for tracks with momentum greater than 0.88 gev / c where tof data may not be used because of multiple trajectories traversing the same tof scintillator bar . vs. tpc @xmath3for @xmath49 gev / c for tracks with isolated hits in the tof . note that the protons are clearly distinguished from pions and electrons in the tof , whereas they fall under the pion and electron peaks in the tpc @xmath3distribution.,scaledwidth=45.0% ] figs . [ fig : fitdedx7 ] and [ fig : fitdedx24 ] show two examples of fits to the @xmath3distributions for two bins , the former at lower momentum where the proton peak is clearly visible , and the latter at higher momentum where the proton peak falls mostly under the pion peak . in the latter case , the green curve is constrained from the tof data as described above . the initial pion yield in each bin is taken as the sum of the integrals of the fitted pion gaussian peaks for the two independent data sets where tof data are used in conjunction with tpc data , and where only tpc data are used . the uncertainty on the pion yield in each case is taken from the uncertainty in the fit parameters for the amplitude and width . however , it is clear that these fits are not perfect ; the bottoms of figs . [ fig : fitdedx7 ] and [ fig : fitdedx24 ] show the residuals of the fit ( data - fit ) . we take into account the imperfection of the fit by adding the sum of the residuals in the range @xmath50 $ ] ( the red lines in the figures ) , where @xmath51 is the fitted width of the pion gaussian peak , and the rms of the residuals in these regions is taken as the uncertainty on this correction to the pion yield . these corrections are typically on the order of 10% , and typical uncertainties of the corrections are on the order of 10% thereby contributing less than 1% to the total pion yield uncertainty . all uncertainties are added in quadrature . the tpc @xmath3data show no clear kaon peak , and so we do not attempt to fit for the kaon contribution in eq . [ eqn : dedx ] . a kaon contribution is estimated using mc to predict the @xmath52 fraction in each bin ; in most bins this contribution is well below 10% . the contribution is subtracted from the pion yield in each bin , and a 30% uncertainty , which is an estimate of the level of uncertainty in the mc prediction of the particle yields off the numi target , is attributed to the correction . the rich @xmath14 distributions are not well described by gaussians ; however , in general the @xmath53 and @xmath16 peaks in these distributions are quite well separated . therefore we take a simple cut - and - count approach , where we count the number of tracks that fall within a range in @xmath14 that contains pions . in practice , however , there is contamination from non - pions , mostly electrons and positrons on the low - side tail of the pions , as well as some pion signal that sits under the electron peak , both of which must be taken into account . we assume that the shapes of the @xmath14 distributions for each particle are properly modeled in the mc . we then define three ranges , one main signal range and two side - band ranges . the data in the sidebands are used to normalize the backgrounds predicted by the mc in the signal region . defining @xmath54 as the number of tracks within a range @xmath55 , @xmath56 as the number of tracks inside the other two ranges , @xmath57 as the mc background estimate ( number of non - pions ) in range @xmath55 and @xmath58 [ @xmath59 as the mc signal [ background ] estimate inside the other two ranges , the pion yield is then @xmath60 where @xmath61 and @xmath62 the uncertainty on the number of pions is @xmath63 where @xmath64 @xmath65 represents the relative amounts of pion to non - pion background in each range in the mc , and we assume a 30% conservative systematic uncertainty in this ratio ; the mc statistical uncertainty is negligible . the positions and widths of the signal and side - band ranges are set by hand , based on a visual scan of the @xmath14 distribution in each bin . the upper sideband where @xmath66 contains almost no background . in some cases , the mc predicts some small amount ( few percent ) of signal outside the ranges used to determine @xmath67 . the predicted fraction of missing signal is added back to the measure @xmath67 from data , and a 30% uncertainty is attributed to this correction . fig . [ fig : richm2a ] shows examples of data and mc rich @xmath14 distributions ; the window boundaries are defined by the vertical solid red lines , and the pion yield and uncertainty are displayed near the top of each in red . in most cases the rich pion yield has uncertainties below 5% . the pion yield measurements depend very weakly on the exact positions of the sidebands . the methods to determine the pion yields discussed above provide an uncertainty which combines statistical uncertainties and systematic uncertainties from backgrounds . the relative uncertainties as a function of are shown in fig . [ fig : statsyst ] for the @xmath68 yields ; the uncertainties are similar for the @xmath69 yields . in general , the combined uncertainty is a few percent for most bins of where a measurement is made ; the hashed bins are those excluded from the results because either no measurement was made or the uncertainty on the measurement is greater than 20% . the momentum corrections discussed above are known to better than 10% . the systematic uncertainties due to a 10% uncertainty on the momentum corrections were calculated for each bin independently . the effect is typically less than 1% , with a few bins as large as 2.5% . momentum resolution and reconstruction failures result in migrations of pions across bins of . since the bins used in the analysis are much larger than the momentum resolution , the net effect of pion migrations across bins is small . mc predictions imply this effect on the pion yield to be less than 4% across all bins , which we take as the systematic uncertainty . an additional systematic uncertainty of 1% is added to account for mis - modeling of noise in the detectors which could result in discrepancies between the measured and predicted yields . we note that these systematic effects do not cancel in the ratio of negative to positive pion yields . differences between the data and mc distributions of the gof and doca variables are used to determine the track selection cut systematic uncertainties . the gof distributions agree to within 14% in the relevant range of gof variable values . the gof cuts were effectively varied by this @xmath70 in both data and mc and the impact on the measured pion yields determined in each bin . the relative difference between the measured pion yields using these modified gof cuts and the nominal gof cut is typically less than 2% , with a few lower momentum bins as large as 6% . a similar approach was taken for the doca cuts . the data and mc doca distributions agree to within @xmath71 , and we find that modifying these cuts in both data and mc results in a systematic shift of the pion yields of less than @xmath72 in nearly all bins , with a few lower momentum bins as large as 7% . the uncertainty of the acceptance and efficiency corrections that are applied to the measured yields in each bin arises from mc statistics ( @xmath27 ) , imperfections in the geometry model of the spectrometer ( @xmath27 ) , imperfections in the modeling of the time - dependent performance of the tracking and pid detectors , and incorrect modeling of the particle yields in the mc . the same time - dependency of thresholds used in the data collection , reconstruction and analysis was also applied to the reconstruction and analysis of the mc simulation data . the time - dependent thresholds are known to within a few percent . we therefore assume a 2% uncertainty on all efficiencies in the pion yields due to imperfections in the modeling of the detectors in the mc . this uncertainty effectively cancels out in the ratio of the pion yields . incorrect modeling of the particle yields in the mc results in improper modeling of the effect of overlapping secondary particles ( pileup ) in the tracking and pid detectors . in high multiplicity events , the track reconstruction algorithm may either combine or confuse hits from different particles . pileup is the main cause of reconstruction and pid inefficiency ; this effect can be as large as 30% at low momenta , and is the main reason why results are not reported here for many bins at lower momenta . for bins where the pileup effect is below 20% , mc events are re - weighted such that the multiplicity distribution ( number of charged tracks coming off the surface of the target ) in mc matches that of data , and the efficiencies are re - calculated . we assume a conservative estimate of the uncertainty on the re - weighting factor to be 20% , based on hand - scan studies that indicate the uncertainty on the measurement of event multiplicity is much less than 20% . we therefore take 20% of the relative difference between the efficiencies determined from the nominal and the re - weighted mc as the systematic uncertainty due to this effect . typical relative uncertainties are a few percent , although some bins have uncertainties as high as 10% . the systematic uncertainties in the pion yield measurement described above are added in quadrature and displayed as a function of in fig . [ fig : totalsystematics ] for both positive and negative pions . nearly all bins have systematic uncertainties between 4 and 5% . the measured @xmath73/pot and @xmath74/pot per bin , along with the combined statistical and systematic errors , are shown in figs . [ fig : finalpiplusyields ] and [ fig : finalpiminusyields ] and table [ table : finalpionyields ] . the data points in the plots in figs . [ fig : finalpiplusyields ] and [ fig : finalpiminusyields ] have been normalized by the width of the momentum bins . the uncertainties in the table are in units of percent . we see that in most of the bins , the measurements are systematics limited and nearly all measurements have uncertainties estimated below 10% . fig . [ fig : finalpiratios ] shows the ratio , @xmath75 , of @xmath76 yields as a function of @xmath32 in slices of @xmath33 . table [ table : finalpionyields ] also lists the @xmath75-values measured in each bin along with the statistical and systematic uncertainties . correlated systematics between the positive and negative pion yields have been canceled out in the ratios . a measurement of @xmath68 and @xmath69 yields from 120 gev / c protons incident on a numi low - energy target has been performed across 124 and 119 bins of bins respectively using data collected in the mipp fixed - target experiment at fermilab . typical uncertainties on the measurements in each bin are a few percent . these data may be directly used to improve the calculation and the uncertainties on the calculation of the neutrino flux in the numi beam line . this work was supported by the us department of energy . we are grateful to the staff of fermilab , lawrence livermore national laboratory under contract de - ac52 - 07na27344 and argonne national laboratory for their contributions to this effort . + & & & * @xmath77 * & * @xmath77 * & & * @xmath78 * & * @xmath78 * & * @xmath79 * & * @xmath80 * & * @xmath80 * + * @xmath32 * & * @xmath33 * & * @xmath81 * & * stat+ * & * syst * & * @xmath82 * & * stat+ * & * syst * & * @xmath82 * & * stat+ * & * syst * + * ( gev / c ) * & * ( gev / c ) * & * pot * & * bkgd * & * ( % ) * & * pot * & * bkgd * & * ( % ) * & * @xmath73 * & * bkgd * & * ( % ) * + & & & * ( % ) * & & & * ( % ) * & & & * ( % ) * & + [ 0.30,0.50 ) & [ 0.00,0.10 ) & 3.32e-01 & 0.97 & 9.77 & 2.76e-01 & 0.86 & 9.09 & 0.84 & 1.29 & 5.19 + [ 0.30,0.50 ) & [ 0.10,0.20 ) & 3.25e-01 & 1.05 & 4.86 & 3.56e-01 & 0.75 & 4.93 & 1.08 & 1.29 & 4.64 + [ 0.30,0.50 ) & [ 0.20,0.30 ) & 1.93e-01 & 1.46 & 5.30 & 1.91e-01 & 1.09 & 5.24 & 1.00 & 1.82 & 4.59 + [ 0.30,0.50 ) & [ 0.30,0.40 ) & 1.06e-01 & 1.97 & 8.15 & 9.49e-02 & 1.68 & 7.39 & 0.95 & 2.59 & 4.61 + [ 0.50,0.62 ) & [ 0.00,0.10 ) & 9.20e-02 & 1.22 & 5.10 & 7.54e-02 & 1.16 & 4.98 & 0.83 & 1.69 & 4.59 + [ 0.50,0.62 ) & [ 0.10,0.20 ) & 1.37e-01 & 1.18 & 4.72 & 1.44e-01 & 0.85 & 4.70 & 1.05 & 1.46 & 4.59 + [ 0.50,0.62 ) & [ 0.20,0.30 ) & 9.03e-02 & 1.57 & 5.64 & 9.24e-02 & 1.14 & 5.92 & 0.99 & 1.94 & 4.59 + [ 0.50,0.62 ) & [ 0.30,0.40 ) & 4.92e-02 & 2.12 & 4.77 & 4.63e-02 & 1.76 & 5.24 & 0.89 & 2.76 & 4.58 + [ 0.50,0.62 ) & [ 0.40,0.50 ) & 1.42e-02 & 5.39 & 8.18 & 2.23e-02 & 2.81 & 8.33 & 1.46 & 6.08 & 4.58 + [ 0.62,0.75 ) & [ 0.00,0.10 ) & 6.02e-02 & 1.41 & 4.94 & 5.14e-02 & 1.31 & 4.85 & 0.87 & 1.93 & 4.59 + [ 0.62,0.75 ) & [ 0.10,0.20 ) & 1.11e-01 & 1.33 & 4.75 & 1.13e-01 & 0.94 & 4.74 & 1.01 & 1.63 & 4.58 + [ 0.62,0.75 ) & [ 0.20,0.30 ) & 8.26e-02 & 1.59 & 4.65 & 8.14e-02 & 1.12 & 4.63 & 0.98 & 1.94 & 4.58 + [ 0.62,0.75 ) & [ 0.30,0.40 ) & 4.32e-02 & 2.09 & 5.62 & 4.33e-02 & 1.52 & 5.43 & 0.98 & 2.58 & 4.58 + [ 0.62,0.75 ) & [ 0.40,0.50 ) & 2.19e-02 & 3.12 & 5.78 & 2.22e-02 & 2.47 & 5.47 & 1.00 & 3.98 & 4.58 + [ 0.75,0.88 ) & [ 0.00,0.10 ) & 4.27e-02 & 1.74 & 4.74 & 3.90e-02 & 1.55 & 4.68 & 0.92 & 2.33 & 4.58 + [ 0.75,0.88 ) & [ 0.10,0.20 ) & 9.19e-02 & 1.48 & 4.68 & 9.04e-02 & 1.05 & 4.65 & 0.98 & 1.82 & 4.58 + [ 0.75,0.88 ) & [ 0.20,0.30 ) & 7.46e-02 & 1.67 & 4.62 & 7.11e-02 & 1.18 & 4.61 & 0.95 & 2.05 & 4.58 + [ 0.75,0.88 ) & [ 0.30,0.40 ) & 4.21e-02 & 2.19 & 5.50 & 4.03e-02 & 1.54 & 5.68 & 0.96 & 2.68 & 4.58 + [ 0.75,0.88 ) & [ 0.40,0.50 ) & 2.16e-02 & 2.84 & 4.71 & 2.09e-02 & 2.26 & 4.77 & 0.95 & 3.63 & 4.58 + [ 0.88,1.00 ) & [ 0.00,0.10 ) & 3.42e-02 & 1.87 & 4.74 & 3.23e-02 & 1.62 & 4.69 & 1.01 & 2.48 & 4.58 + [ 0.88,1.00 ) & [ 0.10,0.20 ) & 7.50e-02 & 1.69 & 4.68 & 7.25e-02 & 1.22 & 4.62 & 0.99 & 2.09 & 4.58 + [ 0.88,1.00 ) & [ 0.20,0.30 ) & 6.68e-02 & 1.85 & 4.64 & 6.19e-02 & 1.33 & 4.60 & 0.94 & 2.28 & 4.58 + [ 0.88,1.00 ) & [ 0.30,0.40 ) & 3.82e-02 & 2.87 & 5.57 & 3.61e-02 & 1.58 & 4.68 & 0.95 & 3.27 & 4.58 + [ 0.88,1.00 ) & [ 0.40,0.50 ) & 1.99e-02 & 2.90 & 8.53 & 1.93e-02 & 2.15 & 5.04 & 0.93 & 3.61 & 4.59 + [ 1.00,1.20 ) & [ 0.00,0.10 ) & 3.91e-02 & 2.04 & 6.21 & 3.65e-02 & 1.71 & 5.33 & 0.96 & 2.66 & 4.59 + [ 1.00,1.20 ) & [ 0.10,0.20 ) & 9.98e-02 & 1.83 & 4.65 & 9.09e-02 & 1.31 & 4.64 & 0.93 & 2.25 & 4.59 + [ 1.00,1.20 ) & [ 0.20,0.30 ) & 9.27e-02 & 2.06 & 4.62 & 8.22e-02 & 1.41 & 4.60 & 0.89 & 2.50 & 4.59 + [ 1.00,1.20 ) & [ 0.30,0.40 ) & 5.20e-02 & 2.46 & 5.65 & 5.25e-02 & 1.67 & 4.74 & 1.00 & 2.97 & 4.59 + [ 1.00,1.20 ) & [ 0.40,0.50 ) & 2.85e-02 & 4.18 & 5.42 & 3.01e-02 & 2.05 & 4.63 & 1.04 & 4.65 & 4.58 + [ 1.20,1.50 ) & [ 0.00,0.10 ) & 4.23e-02 & 2.35 & 6.27 & 3.72e-02 & 1.91 & 8.47 & 0.89 & 3.02 & 4.59 + [ 1.20,1.50 ) & [ 0.10,0.20 ) & 9.12e-02 & 2.07 & 4.61 & 8.83e-02 & 1.50 & 4.62 & 0.98 & 2.56 & 4.58 + [ 1.20,1.50 ) & [ 0.20,0.30 ) & 1.01e-01 & 2.20 & 5.00 & 9.32e-02 & 1.60 & 4.90 & 0.93 & 2.72 & 4.59 + [ 1.20,1.50 ) & [ 0.30,0.40 ) & 7.65e-02 & 2.51 & 4.64 & 6.58e-02 & 1.80 & 4.60 & 0.85 & 3.08 & 4.59 + [ 1.20,1.50 ) & [ 0.40,0.50 ) & 4.76e-02 & 3.11 & 5.25 & 3.87e-02 & 2.19 & 5.40 & 0.80 & 3.80 & 4.59 + [ 1.50,2.00 ) & [ 0.00,0.10 ) & 4.01e-02 & 2.90 & 4.62 & 3.20e-02 & 2.58 & 4.59 & 0.82 & 3.88 & 4.58 + [ 1.50,2.00 ) & [ 0.10,0.20 ) & 8.80e-02 & 2.44 & 4.88 & 8.20e-02 & 1.79 & 4.93 & 0.94 & 3.03 & 4.59 + [ 1.50,2.00 ) & [ 0.20,0.30 ) & 1.26e-01 & 2.41 & 4.65 & 1.08e-01 & 1.73 & 4.65 & 0.89 & 2.97 & 4.59 + [ 1.50,2.00 ) & [ 0.30,0.40 ) & 9.94e-02 & 2.70 & 4.71 & 8.30e-02 & 1.95 & 4.66 & 0.85 & 3.33 & 4.59 + [ 1.50,2.00 ) & [ 0.40,0.50 ) & 6.99e-02 & 3.14 & 4.63 & 5.61e-02 & 2.31 & 4.63 & 0.80 & 3.90 & 4.59 + [ 4.00,6.00 ) & [ 0.00,0.10 ) & 1.54e-02 & 11.19 & 5.03 & 1.74e-02 & 7.94 & 5.83 & 1.06 & 13.73 & 4.58 + [ 4.00,6.00 ) & [ 0.10,0.20 ) & 5.12e-02 & 7.63 & 4.67 & 4.82e-02 & 4.23 & 4.91 & 0.85 & 8.72 & 4.58 + [ 4.00,6.00 ) & [ 0.20,0.30 ) & 6.47e-02 & 8.72 & 4.79 & 5.64e-02 & 4.48 & 4.86 & 0.88 & 9.81 & 4.59 + [ 4.00,6.00 ) & [ 0.30,0.40 ) & 6.79e-02 & 9.31 & 5.61 & 5.79e-02 & 5.64 & 4.73 & 0.71 & 10.89 & 4.60 + [ 6.00,8.00 ) & [ 0.00,0.10 ) & 7.94e-03 & 4.78 & 5.40 & 8.64e-03 & 3.46 & 6.40 & 1.01 & 5.90 & 4.58 + [ 6.00,8.00 ) & [ 0.10,0.20 ) & 2.69e-02 & 3.55 & 4.64 & 2.60e-02 & 2.57 & 4.95 & 0.90 & 4.38 & 4.58 + [ 6.00,8.00 ) & [ 0.20,0.30 ) & 3.93e-02 & 3.82 & 4.69 & 3.47e-02 & 2.49 & 4.62 & 0.87 & 4.56 & 4.58 + [ 6.00,8.00 ) & [ 0.30,0.40 ) & 4.01e-02 & 3.88 & 4.79 & 3.26e-02 & 3.14 & 4.71 & 0.76 & 4.99 & 4.59 + [ 6.00,8.00 ) & [ 0.40,0.50 ) & 3.11e-02 & 7.89 & 5.92 & 2.53e-02 & 6.11 & 5.24 & 0.75 & 9.98 & 4.59 + [ 8.00,10.00 ) & [ 0.00,0.10 ) & 3.99e-03 & 3.55 & 6.90 & 4.46e-03 & 3.22 & 7.02 & 1.12 & 4.79 & 4.58 + [ 8.00,10.00 ) & [ 0.10,0.20 ) & 1.39e-02 & 6.28 & 4.89 & 1.20e-02 & 8.47 & 4.84 & 0.88 & 10.54 & 4.58 + [ 8.00,10.00 ) & [ 0.20,0.30 ) & 2.32e-02 & 1.70 & 4.69 & 1.90e-02 & 1.60 & 4.73 & 0.90 & 2.34 & 4.58 + [ 8.00,10.00 ) & [ 0.30,0.40 ) & 2.46e-02 & 1.86 & 4.66 & 2.09e-02 & 1.91 & 4.68 & 0.81 & 2.67 & 4.58 + [ 8.00,10.00 ) & [ 0.40,0.50 ) & 2.14e-02 & 2.64 & 4.66 & 1.72e-02 & 2.87 & 4.68 & 0.78 & 3.90 & 4.58 + [ 8.00,10.00 ) & [ 0.50,2.00 ) & 5.80e-02 & 5.87 & 6.20 & 4.86e-02 & 6.33 & 5.19 & 0.75 & 8.63 & 4.60 + [ 10.00,12.00 ) & [ 0.00,0.10 ) & 2.35e-03 & 4.01 & 5.30 & 2.63e-03 & 4.12 & 8.46 & 0.95 & 5.75 & 4.58 + [ 10.00,12.00 ) & [ 0.10,0.20 ) & 8.25e-03 & 1.95 & 4.67 & 8.44e-03 & 2.01 & 4.97 & 0.95 & 2.80 & 4.58 + [ 10.00,12.00 ) & [ 0.20,0.30 ) & 1.50e-02 & 1.72 & 4.63 & 1.27e-02 & 1.78 & 4.62 & 0.86 & 2.47 & 4.58 + [ 10.00,12.00 ) & [ 0.30,0.40 ) & 1.66e-02 & 2.10 & 4.75 & 1.40e-02 & 1.91 & 4.69 & 0.78 & 2.84 & 4.58 + [ 10.00,12.00 ) & [ 0.40,0.50 ) & 1.45e-02 & 2.79 & 4.70 & 1.21e-02 & 2.62 & 4.80 & 0.83 & 3.83 & 4.58 + [ 10.00,12.00 ) & [ 0.50,2.00 ) & 3.78e-02 & 4.18 & 5.46 & 3.10e-02 & 4.26 & 5.17 & 0.80 & 5.97 & 4.59 + [ 12.00,14.00 ) & [ 0.00,0.10 ) & 1.62e-03 & 4.98 & 5.34 & 1.68e-03 & 5.10 & 5.28 & 1.03 & 7.12 & 4.58 + [ 12.00,14.00 ) & [ 0.10,0.20 ) & 5.84e-03 & 2.43 & 4.89 & 5.10e-03 & 2.61 & 5.08 & 0.85 & 3.56 & 4.58 + [ 12.00,14.00 ) & [ 0.20,0.30 ) & 1.01e-02 & 2.00 & 4.62 & 7.65e-03 & 2.11 & 4.72 & 0.73 & 2.91 & 4.58 + [ 12.00,14.00 ) & [ 0.30,0.40 ) & 1.12e-02 & 2.14 & 4.65 & 9.51e-03 & 2.12 & 4.61 & 0.83 & 3.02 & 4.58 + [ 12.00,14.00 ) & [ 0.40,0.50 ) & 1.01e-02 & 2.40 & 4.68 & 8.74e-03 & 2.38 & 4.74 & 0.89 & 3.38 & 4.58 + [ 12.00,14.00 ) & [ 0.50,2.00 ) & 2.79e-02 & 2.96 & 4.72 & 2.21e-02 & 2.96 & 4.70 & 0.73 & 4.19 & 4.58 + [ 14.00,17.00 ) & [ 0.00,0.10 ) & 1.78e-03 & 4.62 & 5.19 & 1.61e-03 & 5.05 & 5.45 & 0.87 & 6.85 & 4.58 + [ 14.00,17.00 ) & [ 0.10,0.20 ) & 5.51e-03 & 2.50 & 4.78 & 4.73e-03 & 2.68 & 6.56 & 0.73 & 3.66 & 4.58 + [ 14.00,17.00 ) & [ 0.20,0.30 ) & 9.61e-03 & 1.88 & 5.01 & 7.06e-03 & 2.09 & 4.68 & 0.78 & 2.81 & 4.58 + [ 14.00,17.00 ) & [ 0.30,0.40 ) & 1.19e-02 & 1.87 & 4.72 & 8.99e-03 & 2.01 & 4.69 & 0.81 & 2.74 & 4.58 + [ 14.00,17.00 ) & [ 0.40,0.50 ) & 1.08e-02 & 2.12 & 4.64 & 8.20e-03 & 2.27 & 4.72 & 0.79 & 3.10 & 4.58 + [ 14.00,17.00 ) & [ 0.50,2.00 ) & 3.02e-02 & 2.15 & 4.66 & 2.54e-02 & 2.01 & 4.61 & 0.81 & 2.95 & 4.58 + [ 17.00,20.00 ) & [ 0.00,0.10 ) & 1.01e-03 & 5.78 & 5.12 & 9.19e-04 & 6.85 & 6.20 & 0.81 & 8.96 & 4.58 + [ 17.00,20.00 ) & [ 0.10,0.20 ) & 3.39e-03 & 2.99 & 4.71 & 2.61e-03 & 3.66 & 5.45 & 0.66 & 4.73 & 4.58 + [ 17.00,20.00 ) & [ 0.20,0.30 ) & 5.73e-03 & 2.17 & 4.65 & 3.98e-03 & 2.68 & 5.10 & 0.64 & 3.45 & 4.58 + [ 17.00,20.00 ) & [ 0.30,0.40 ) & 7.56e-03 & 2.45 & 4.74 & 5.11e-03 & 2.73 & 4.72 & 0.73 & 3.67 & 4.58 + [ 17.00,20.00 ) & [ 0.40,0.50 ) & 7.07e-03 & 2.55 & 4.70 & 5.59e-03 & 2.69 & 4.71 & 0.77 & 3.71 & 4.58 + [ 17.00,20.00 ) & [ 0.50,2.00 ) & 2.20e-02 & 1.88 & 4.62 & 1.68e-02 & 2.01 & 4.69 & 0.81 & 2.75 & 4.58 + [ 20.00,24.00 ) & [ 0.00,0.10 ) & 9.77e-04 & 5.79 & 5.85 & 6.40e-04 & 7.74 & 10.03 & 0.53 & 9.66 & 4.58 + [ 20.00,24.00 ) & [ 0.10,0.20 ) & 2.84e-03 & 3.22 & 4.73 & 2.04e-03 & 4.12 & 4.66 & 0.70 & 5.23 & 4.58 + [ 20.00,24.00 ) & [ 0.20,0.30 ) & 4.82e-03 & 2.98 & 4.89 & 2.91e-03 & 3.87 & 4.67 & 0.64 & 4.89 & 4.58 + [ 20.00,24.00 ) & [ 0.30,0.40 ) & 6.28e-03 & 2.52 & 4.67 & 4.04e-03 & 3.10 & 5.05 & 0.60 & 4.00 & 4.58 + [ 20.00,24.00 ) & [ 0.40,0.50 ) & 6.53e-03 & 2.72 & 4.66 & 4.18e-03 & 2.95 & 4.64 & 0.62 & 4.01 & 4.58 + [ 20.00,24.00 ) & [ 0.50,2.00 ) & 2.01e-02 & 2.02 & 4.75 & 1.51e-02 & 1.89 & 4.65 & 0.70 & 2.77 & 4.58 + [ 24.00,28.00 ) & [ 0.00,0.10 ) & 6.89e-04 & 6.65 & 8.24 & 4.61e-04 & 9.09 & 4.68 & 0.90 & 11.27 & 4.58 + [ 24.00,28.00 ) & [ 0.10,0.20 ) & 1.70e-03 & 5.74 & 5.06 & 1.14e-03 & 7.25 & 5.09 & 0.66 & 9.25 & 4.58 + [ 24.00,28.00 ) & [ 0.20,0.30 ) & 2.75e-03 & 3.92 & 5.28 & 1.48e-03 & 6.47 & 5.29 & 0.54 & 7.56 & 4.58 + [ 24.00,28.00 ) & [ 0.30,0.40 ) & 3.57e-03 & 3.49 & 4.73 & 2.17e-03 & 3.92 & 4.78 & 0.68 & 5.25 & 4.58 + [ 24.00,28.00 ) & [ 0.40,0.50 ) & 3.96e-03 & 3.35 & 4.82 & 2.50e-03 & 3.49 & 4.67 & 0.64 & 4.83 & 4.58 + [ 24.00,28.00 ) & [ 0.50,2.00 ) & 1.45e-02 & 2.05 & 4.66 & 1.04e-02 & 2.01 & 4.67 & 0.78 & 2.87 & 4.58 + [ 28.00,34.00 ) & [ 0.00,0.10 ) & 5.65e-04 & 11.39 & 4.97 & 3.88e-04 & 13.69 & 5.79 & 0.55 & 17.81 & 4.58 + [ 28.00,34.00 ) & [ 0.10,0.20 ) & 1.81e-03 & 5.12 & 4.74 & 8.67e-04 & 8.86 & 5.79 & 0.42 & 10.24 & 4.58 + [ 28.00,34.00 ) & [ 0.20,0.30 ) & 2.50e-03 & 4.06 & 4.88 & 1.20e-03 & 5.64 & 4.74 & 0.50 & 6.95 & 4.58 + [ 28.00,34.00 ) & [ 0.30,0.40 ) & 3.24e-03 & 3.45 & 4.92 & 1.62e-03 & 5.02 & 4.65 & 0.55 & 6.09 & 4.58 + [ 28.00,34.00 ) & [ 0.40,0.50 ) & 3.36e-03 & 3.36 & 4.61 & 2.11e-03 & 3.81 & 5.13 & 0.57 & 5.08 & 4.58 + [ 28.00,34.00 ) & [ 0.50,2.00 ) & 1.34e-02 & 2.13 & 4.65 & 8.83e-03 & 2.00 & 4.61 & 0.65 & 2.93 & 4.58 + [ 34.00,40.00 ) & [ 0.00,0.10 ) & 4.64e-04 & 8.56 & 5.41 & 2.63e-04 & 11.23 & 4.60 & 0.64 & 14.12 & 4.58 + [ 34.00,40.00 ) & [ 0.10,0.20 ) & 1.18e-03 & 5.38 & 5.42 & 5.55e-04 & 7.33 & 4.84 & 0.51 & 9.09 & 4.58 + [ 34.00,40.00 ) & [ 0.20,0.30 ) & 1.62e-03 & 4.20 & 4.74 & 7.23e-04 & 6.14 & 5.10 & 0.39 & 7.44 & 4.58 + [ 34.00,40.00 ) & [ 0.30,0.40 ) & 1.83e-03 & 3.51 & 4.73 & 8.38e-04 & 5.63 & 4.74 & 0.47 & 6.64 & 4.58 + [ 34.00,40.00 ) & [ 0.40,0.50 ) & 2.05e-03 & 3.44 & 4.68 & 1.09e-03 & 4.98 & 4.78 & 0.54 & 6.05 & 4.58 + [ 34.00,40.00 ) & [ 0.50,2.00 ) & 8.81e-03 & 1.88 & 4.73 & 5.25e-03 & 2.36 & 4.69 & 0.59 & 3.02 & 4.58 + [ 40.00,48.00 ) & [ 0.00,0.10 ) & 3.98e-04 & 9.82 & 4.70 & 1.38e-04 & 16.11 & 5.20 & 0.32 & 18.87 & 4.58 + [ 40.00,48.00 ) & [ 0.10,0.20 ) & 7.99e-04 & 6.32 & 5.22 & 2.92e-04 & 10.10 & 4.90 & 0.44 & 11.92 & 4.58 + [ 40.00,48.00 ) & [ 0.20,0.30 ) & 1.15e-03 & 4.63 & 4.79 & 4.77e-04 & 7.49 & 4.61 & 0.40 & 8.80 & 4.58 + [ 40.00,48.00 ) & [ 0.30,0.40 ) & 1.33e-03 & 4.11 & 4.69 & 5.04e-04 & 7.10 & 5.62 & 0.44 & 8.21 & 4.58 + [ 40.00,48.00 ) & [ 0.40,0.50 ) & 1.54e-03 & 3.83 & 4.93 & 6.55e-04 & 6.21 & 5.07 & 0.42 & 7.29 & 4.58 + [ 40.00,48.00 ) & [ 0.50,2.00 ) & 7.02e-03 & 2.01 & 4.83 & 3.58e-03 & 2.76 & 4.63 & 0.50 & 3.41 & 4.58 + [ 48.00,56.00 ) & [ 0.00,0.10 ) & 1.87e-04 & 11.91 & 4.80 & & & & & & + [ 48.00,56.00 ) & [ 0.10,0.20 ) & 4.48e-04 & 10.33 & 4.99 & 1.55e-04 & 13.69 & 4.89 & 0.33 & 17.15 & 4.58 + [ 48.00,56.00 ) & [ 0.20,0.30 ) & 7.06e-04 & 5.87 & 4.96 & 2.15e-04 & 10.82 & 5.22 & 0.37 & 12.31 & 4.58 + [ 48.00,56.00 ) & [ 0.30,0.40 ) & 7.71e-04 & 5.31 & 4.92 & 2.19e-04 & 10.36 & 5.18 & 0.29 & 11.64 & 4.58 + [ 48.00,56.00 ) & [ 0.40,0.50 ) & 7.46e-04 & 5.21 & 4.69 & 3.05e-04 & 8.80 & 4.93 & 0.39 & 10.23 & 4.58 + [ 48.00,56.00 ) & [ 0.50,2.00 ) & 3.88e-03 & 2.46 & 4.67 & 1.80e-03 & 3.62 & 4.66 & 0.48 & 4.38 & 4.58 + [ 56.00,68.00 ) & [ 0.00,0.10 ) & 1.90e-04 & 11.86 & 4.92 & & & & & & + [ 56.00,68.00 ) & [ 0.10,0.20 ) & 4.24e-04 & 9.19 & 4.65 & 8.79e-05 & 16.67 & 5.70 & 0.19 & 19.03 & 4.58 + [ 56.00,68.00 ) & [ 0.20,0.30 ) & 5.80e-04 & 6.28 & 5.96 & 1.71e-04 & 11.59 & 4.69 & 0.36 & 13.18 & 4.58 + [ 56.00,68.00 ) & [ 0.30,0.40 ) & 6.27e-04 & 5.64 & 4.68 & 1.52e-04 & 11.85 & 5.76 & 0.30 & 13.12 & 4.58 + [ 56.00,68.00 ) & [ 0.40,0.50 ) & 4.84e-04 & 6.22 & 4.76 & 1.57e-04 & 11.15 & 5.32 & 0.37 & 12.77 & 4.58 + [ 56.00,68.00 ) & [ 0.50,2.00 ) & 2.81e-03 & 2.77 & 4.79 & 1.03e-03 & 4.51 & 5.04 & 0.36 & 5.29 & 4.58 + [ 68.00,80.00 ) & [ 0.20,0.30 ) & 2.36e-04 & 10.32 & 5.45 & & & & & & + [ 68.00,80.00 ) & [ 0.30,0.40 ) & 1.88e-04 & 10.39 & 4.89 & & & & & & + [ 68.00,80.00 ) & [ 0.40,0.50 ) & 1.88e-04 & 10.15 & 4.80 & & & & & & + [ 68.00,80.00 ) & [ 0.50,2.00 ) & 9.29e-04 & 4.69 & 4.75 & 3.33e-04 & 7.34 & 5.06 & 0.38 & 8.71 & 4.58 + p. adamson , et al . ( minos collaboration ) , `` neutrino and antineutrino inclusive charged - current cross section measurements with the minos near detector '' , _ phy . d _ * 81 * 072002 ( 2010 ) . + l. fields et al . ( minerva collaboration ) , `` measurement of muon antineutrino quasi - elastic scattering on a hydrocarbon target at [email protected] gev '' , _ phys . lett . _ * 111 * , 022501 ( 2013 ) . + g. a. fiorentini et al . ( minerva collaboration ) , `` measurement of muon neutrino quasi - elastic scattering on a hydrocarbon target at [email protected] gev '' , _ phys . lett . _ * 111 * , 022502 ( 2013 ) . + g. rai et al . , `` a tpc detector for the study of high multiplicity heavy ion collisions '' , _ ieee trans . sc . _ 37 56 - 64 ( 1990 ) . + j. uribe et al . , `` pion - pion correlations at low relative momentum produced in p - p collisions at 27.5 gev / c '' , _ phys . d _ * 49 * 4373 ( 1994 ) . + j. engelfried et al . , `` the selex phototube rich detector '' , _ nucl . meth . _ a * 431 * 53 - 69 ( 1999 ) . + nigmanov , t.s . ( mipp collaboration ) , `` electromagnetic and hadron calorimeters in the mipp experiment '' , _ nucl . meth . _ a * 598 * , , 394 - 399 ( 2009 ) . + d. c. christian et al . , `` high rate drift chambers '' , _ nucl . _ a * 345 * , 62 ( 1994 ) . + a. fasso , a. ferrari , j. ranft , and p.r . sala , `` fluka : a multi - particle transport code '' , cern-2005 - 10 ( 2005 ) , infn / tc 05/11 , slac - r-773 . + r. brun , f. bruyant , m. maire , a. c. mcpherson and p. zanarini , `` geant3 , '' cern - dd - ee-84 - 1 . + j. beringer et al . ( particle data group ) , _ phys . d _ _ 86 _ , 010001 ( 2012 ) . +

the sloan digital sky survey ii ( sdss - ii ) supernova survey discovered type ia supernovae ( sne ia ) in an almost unexplored intermediate redshift range of and provided densely sampled multi - color light curves for sn candidates . followup spectroscopy of this survey was carried out with the subaru telescope and spectra of 71 sn ia candidates were obtained . one spectrum was observed per candidate except for a peculiar variable . this paper presents the method for processing these spectra . the observed wavelength ranges of our spectra are 4000 to 9000 for year 2005 and 3600 to 9000 for year 2006 . most sn ia spectra have signal to noise ratios ( s / n ) between 4 and 10 per 2 averaged over the entire wavelength region . we developed a new code to extract the sn spectral component from spectra contaminated by the host galaxy . of 71 sn ia candidates , 59 are identified as normal sne ia and 3 are peculiar sne ia . the range of spectral phases varies from -7 days to + 30 days from maximum brightness . there are also 7 sne ii , 1 possible hypernova and 1 agn .

1101.1565

type ia supernovae ( sne ia ) have played a significant role in observational cosmology as a probe for the expansion history of the universe . this is because the majority of the observed sne ia have homogeneous light curves and spectra . recent nearby observations have enabled the construction of large datasets of sne ia obtained by a homogeneous procedure . many spectra have been stored in the supernova spectra ( suspect ) database . ] a supernova group in the center for astrophysics ( cfa ) presented ubvri photometry of 44 nearby sne ia @xcite and 185 sne ia @xcite and 32 optical spectra @xcite . the carnegie supernova project recently published @xmath1bvyjhk@xmath2 light curves of 35 nearby sne ia @xcite . the nearby supernova factory @xcite has been gathering spectrophotometry data of nearby sne ia . high redshift ( high - z ) sne ia have been observed by several groups : the high - z team @xcite , the supernova cosmology project @xcite , the supernova legacy survey @xcite and the equation of state : supernovae trace cosmic expansion survey @xcite . however , since wide field observations are required to obtain comparable numbers of sne ia , few have been observed in the intermediate redshift range ( @xmath3 ) . the sdss - ii supernova survey carried out a three year project @xcite to obtain multi - band light curves of sne ia at intermediate redshifts . the sdss 2.5 m telescope @xcite scanned 300 square deg with the sdss imaging camera @xcite and five filters ( @xmath4 , * ? ? ? * ) . the observed cadence was two days minimum for the same sky region . analysis was conducted in real - time to assign sn candidates for spectroscopic observations @xcite . follow - up spectroscopic observations were carried out at a large number of telescopes to determine at least the supernova type and redshift ; the 2.4 m hiltner , the 2.5 m not , the 3.5 m sdss arc , the 3.6 m ntt , the 4.2 m wht , the 8.2 m subaru , the 9.2 m het and the 10 m keck i telescopes . the spectroscopy of the first season ( year 2005 ) is presented in @xcite . @xcite present the spectra obtained at the ntt and the not during year 2006 and 2007 . @xcite provide high - quality sn light curves in @xmath4 for the first season . @xcite describes the keck followup spectra and presents a mismatch in the ultraviolet spectra of these sne ia from nearby counterparts . the final data release for all spectra obtained by these telescopes will be published in sako et al ( in preparation ) . in this paper , we present an observational summary of the subaru spectroscopic followup . [ observ ] describes our observations . [ reduc ] describes spectroscopic data reduction including the method to extract sn spectra from host galaxy contaminated spectra . [ datasets ] describes the dataset obtained . we summarize this paper in [ summary ] . we used the subaru telescope for follow - up spectroscopic observations of the sn candidates found by the sdss - ii supernova survey for the first two seasons ( years 2005 & 2006 ) . the faint object camera and spectrograph ( focas ; * ? ? ? * ) was used in the long - slit mode . the atmospheric dispersion corrector ( adc ) was used in our observations . the instrumental dispersions were 1.34 / pix . we used a 0.8 slit for sn candidates and a 2.0 slit for spectroscopic standard stars . the pixel scale was 0.104 and the data were read out with three pixel binning in the spatial direction . in order to obtain spectra in as wide a wavelength range as possible , we observed red and blue sides separately with different sets of grisms and filters . we obtained a red side spectrum first to identify the `` siii @xmath56355 '' feature which characterize sne ia and then a blue side spectrum unless the weather was bad or the observing time was limited . sn candidates were observed at the time when they were as close to zenith as possible and above an elevation of 30@xmath6 , so that they suffered the least from atmospheric absorption . for year 2005 , we adopted the following grating and filter combinations for the blue and the red sides ; ( i ) 300 line mm@xmath7 grating blazed at 5500 ( 300b ) and the l600 filter , which is sensitive in the wavelength window of 4000 to 6000 , and ( ii ) 300 line mm@xmath7 grating blazed at 7500 ( 300r ) and the sy47 filter , which is sensitive in the wavelength window of 5000 to 9000 . the configuration for year 2006 was the same as the first year except the filter for the blue side . we replaced the l600 filter with a l550 filter which is more efficient for shorter wavelengths , widening the spectral window to 3600 to 9000 . table [ setups ] summarizes our configurations . because of the subaru s large aperture and a superb seeing site , we observed the subset of sdss targets with large host galaxy contamination . we aligned the slit so as to go through each sn candidate and host center . the slit angles were initially determined from the images taken by the sdss 2.5 m telescope , but then adjusted by checking the @xmath8-band images taken by the subaru telescope at the time of each observation . this slit alignment enabled us to simultaneously determine the sn type as well as its redshift . the redshift was generally determined using spectral lines of the host galaxy ( see [ zz ] for details ) . the observations of the sn candidates were not typically made at the parallactic angle , and the observed spectra should be color corrected due to differential atmospheric refraction when the airmass was large . the atmospheric refraction was corrected by the adc . 89 % of our targets were observed at an airmass smaller than 1.5 . we usually observed two spectrophotometric standard stars each night for the flux calibration ; at the beginning and the end of observations . table [ stdstar ] summarizes the list of standard stars observed in years 2005 and 2006 . all the sn candidates were observed for @xmath9 seconds for both the blue and red sides . candidates spectra were observed only once . the only exception was a peculiar variable sn10450 . we observed sn10450 with an exposure of 900 seconds and 1,800 seconds ( [ zz ] , @xcite ) . we determined types and redshifts of each target via real - time analysis . this information was collected by our spectroscopic coordinator and reported to the international astronomical union ( iau ) . type identification of three targets in our list was delayed due to high contamination by the host galaxy and lower s / n . detailed off - line analysis identified these three targets as sne ia . the spectroscopic data were reduced using the image reduction and analysis facility ( iraf ) and our own programs written in c. the signals from objects were recorded in two - dimensional ( 2d ) ccd detectors ( called , frames " hereafter ) . we carried out the reduction to obtain calibrated 2d spectra along the spatial and wavelength axes with iraf as follows . * bias subtraction : * even when the exposure time was zero , the pixel adu counts were not zero but some positive value ; a bias voltage was applied when reading the ccd to avoid negative values due to readout noise . this bias value ( adu counts when the exposure time was zero ) had to be subtracted from every frame . we used an overscan region of 2d bias spectra to subtract the bias . * flat fielding : * we observed three dome flat fields for each grism and filter combination and created high s / n dome flats by taking the median of these three dome flat frames . we corrected for the non - uniformity by dividing the bias subtracted frame by the median dome flat field . * wavelength calibration : * we used night glow oh emission lines recorded in object spectra for the red side and the th - ar lamp emission lines taken separately from object frames for the blue side . the overall wavelength calibration was done using a legendre polynomial of third order for the red side and a legendre polynomial of fifth order for the blue side . the polynomial order was determined to minimize the rms of the wavelength residuals . * distortion correction : * our spectra are distorted in the spatial direction to some extent . we obtained a spatial distortion map for each frame by tracing the spatial location of spectrophotometric standard stars at each wavelength . we positioned the telescope so as to obtain spectra of objects and standard stars at the same position on the ccd . we then used the distortion map to correct for each object spectrum . * background subtraction : * background photons from night glow contaminated the frames . we subtracted this background contamination by masking the object region and fitting a low order polynomial function to the spatial profile along the wavelength axis . * flux calibration : * we observed spectrophotometric standard stars : bd+28@xmath64211 , feige110 , gd71 , bd+75@xmath6325 and feige34 . library spectra in iraf were used for bd+28@xmath64211 , feige110 , bd+75@xmath6325 and feige34 . a spectrum in the hst calibration database system ( calspec ) was used for gd71 to create the sensitivity function for the flux calibration , which is expressed as a fifth order cubic spline function . the best fit polynomial function was searched for each standard star spectrum . table [ stdspec ] is the list of spectrophotometric standard stars and the origin of the spectral energy distributions ( seds ) used in this reduction . the extinction file for the mauna kea telescope was used to remove the telluric extinction . * correction for dust extinction in our galaxy : * we corrected for dust extinction in our galaxy using @xmath10 and the color excess @xmath11 from @xcite . the extinction was assumed to follow the @xcite law . we developed c programs for the following processes : * telluric absorption removal : * any spectra observed from the ground suffers from telluric absorption lines due to water and oxygen molecules in the atmosphere , especially at wavelengths longer than 5000 . the deepest is the a - band line at @xmath127600 . for targets at @xmath13 , most of our sample , this absorption is superimposed around the siii @xmath55972 " or siii @xmath56355 " feature . since flux calibration is not perfect for the removal of sharp lines , we derived telluric - line correction factors which varies with wavelength using the following process : ( i ) composite spectra for each season ( year 2005 and 2006 ) were created from all the frames of a spectrophotometric standard star obtained during our campaign of the year . bd+28@xmath64211 was used for this purpose , since this was the best observed standard star during the relevant seasons ( table [ stdstar ] ) . ( ii ) a corresponding spectrum in the hst calspec database was chosen for comparison . library spectra of bd+28@xmath64211 are also available in the iraf package ; however , the size of the wavelength binning is smaller for the hst calspec spectrum ( 5 , black line in figure [ compbd ] ) , than the iraf spectrum ( larger than 50 , red pluses in figure [ compbd ] ) . moreover , the hst spectrum does not suffer from telluric absorption . note that up to a 10 % difference in flux between these two spectra ( bottom panel of figure [ compbd ] ) is observed . ( iii ) the extent of absorption at each wavelength was examined by comparison of these spectra . figure [ telrm ] shows telluric - line correction factors for year 2005 ( red ) and 2006 ( green ) . calibration uncertainties are up to 4% near 6000 between these correction factors , perhaps due to different telescope pointing directions and atmospheric conditions at the observation times of the standard star and object observations . this could produce a comparable uncertainty in the absolute flux calibration . . we performed the telluric - line correction by dividing an observed spectrum by correction factors from the same year . it should be noted that residuals of telluric - lines could remain to some extent after this process because of variable atmospheric conditions . * seeing measurement : * seeing is the full width at half maximum ( fwhm ) of the spatial profile recorded on the detector . we calculated the seeing using a 2d spectrum of a standard star , observed with a 2.0 " slit ( wide enough to pass through all the photons from a point - like object ) , along the spatial and wavelength axes . the spatial profile of the star was traced in the wavelength direction with a gaussian function to obtain the seeing size as a function of wavelength . we used bd+28@xmath64211 , since this was the best observed standard star . figure [ seeing ] is an example of the seeing size in arcsec measured for bd+28@xmath64211 taken on the night of oct . 27 , 2006 . the seeing is known to vary as an exponential with wavelength . we obtained the index of the seeing - wavelength relation for the subaru / focas instrument of @xmath14 : @xmath15 we used this relation for the sn extraction . the index for the focal reducer and low dispersion spectrograph ( fors ) attached to the very large telescope ( vlt ) was reported to be -0.29 @xcite . the subaru / focas result is comparable to their value . * sn extraction : * the observed sne suffered from host galaxy contamination . the best way to eliminate the galaxy light from the sn candidate may be to observe the same galaxy with the same telescope configuration after the candidate disappears ; however , this takes considerable time . as an alternative , we have developed algorithms to eliminate the galaxy contribution . two proposed methods had been discussed in the literature : one was to extract sn spectra from 1d galaxy - contaminated spectra with the aid of galaxy template spectra ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and the other was to extract sn spectra from 2d galaxy - contaminated spectra without the aid of galaxy template spectra @xcite . an advantage of the latter method is that no assumption of a host galaxy spectrum is necessary . for this method , one should determine the spatial profiles for the sn and host components for the extraction . we developed an algorithm of sn extraction based on the latter concept . the basic concept is similar to the technique used by the supernova legacy survey collaboration @xcite , but developed independently . a model spectrum @xmath16 is assumed to be the sum of the sn and host galaxy flux : @xmath17 where @xmath18 is the pixel index in the spatial direction and @xmath5 is the discretized wavelength . the function @xmath19 is the sn candidate flux density in the pixel ( @xmath20 ) and @xmath21 is the host galaxy flux density . the spatial profile of a point source @xmath19 is assumed to be a gaussian function : @xmath22 where @xmath23 is the bestfit gaussian width for the sn candidate . the spatial profile of the extended component @xmath21 is assumed to be the sum of two gaussian functions . the summing was introduced in order to approximate the bulge and disk components of the host galaxy . we should note that a convolution with the psf of more sophisticated functions such as the de vaucouleurs @xmath24 law or an exponential law are more suited to fit galaxy morphologies . although a sum of two gaussians is just an approximation , the @xmath25 of fitting tells us that this approximation is reasonable for the majority of our sample . for our hosts with complicated spatial profiles , we performed a @xmath25 fitting by adjusting two gaussian peak positions representing the hosts . a larger number of gaussians or a more generalized function @xcite would be required for hosts with more complex spatial profiles . in the fitting process , photon statistics was assumed to be the only source of the uncertainty @xmath26 . this uncertainty was assumed to be constant in the spatial direction @xmath27 and was estimated from the blank areas adjacent to the object . the width of the blank ( no sn nor hosts ) area was typically 25 arcsec on both sides , centered on the object . we minimized : @xmath28 where @xmath29 was the observed flux density at each pixel . the fitting was done with a two stage process . first , the model function @xmath16 was fitted to the averaged spatial profile created by averaging the 2d flux over the wavelength ranges ( 3600 to 5400 , 4000 to 6000 and 5000 to 9000 ) to determine the central positions and the widths of each component such as @xmath23 . the width @xmath23 should be identical to the seeing size . then , the spatial profile at each wavelength was fitted , with the spatial positions of the sn and its host fixed at the values obtained in the first step and the widths changing due to the seeing - wavelength relation of equation [ sw ] . after finding the parameters , a candidate spectrum @xmath30 , its uncertainty @xmath31 and the host galaxy spectrum @xmath32 were derived . if we created sn spectra by integrating flux over too wide a @xmath18 region , the s / n ratio became low . we restricted sn candidate flux within the fwhm ( @xmath33 ) . @xmath34 where @xmath35 , and the integral was calculated using the trapezoidal formula . the sn flux uncertainty @xmath31 was estimated from : @xmath36 therefore , @xmath37 host galaxy spectra were constructed by subtracting the sn component @xmath38 from @xmath29 . in figure [ sec ] , we show an example of the sn extraction procedure . we should note that the fitting would be affected by the presence of emission lines . our method fits spatial profiles at each wavelength with the sum of two or three gaussian functions . when a host galaxy spatial profile can be approximated by a gaussian , the total spatial profile would be the sum of two gaussians . for the galaxy approximated by two gaussians , the total would be that of three gaussians . spatial profiles at the lines could be too complicated for our model if there were strong lines from host galaxies . since sne ia do not show emission lines in their spectra , we can regard emission lines in the extracted sn spectrum , more than 80 % of our sample , as those originated in their host galaxies . we can regard absorption lines in these spectra as galaxy origin if its width is smaller than expanding velocities of sn ejecta ( @xmath39 km sec@xmath7 ) . a profile fit to a gaussian function might miss a few percent of the observed flux . an analysis of spatial profiles for standard stars showed a slight excess in flux at the outskirts of the spatial profile compared to the prediction of the gaussian function . we neglect this excess for our current analysis . * combining spectra : * in most cases , sn spectra were taken with two different combinations of grisms and filters : blue and red side spectra . we combined these two spectra by taking the average flux within the overlapping wavelength regions and scaling the red side to match the blue side flux and re - binning to 2 per pixel . the result is a contiguous science spectrum together with an uncertainty spectrum representing the statistical uncertainties in the flux in each binned pixel . * slitloss correction : * since seeing sizes were comparable to the slit width , some fraction of the object photons could not go through the slit . we corrected for this slitloss by estimating @xmath40 magnitudes at the spectroscopy date from light curves and spectra . magnitudes at the spectroscopy date were determined by interpolating the bestfit templates for the salt2 light curve fitter ( see the light curve fitting in [ salt2fit ] ) . magnitudes were synthesized from spectra using @xmath40 filter responses . we matched the interpolated magnitudes with the synthetic magnitudes to determine a second - order polynomial slitloss correction function ; @xmath41 where @xmath42 is a coefficient . this function was determined for each spectrum . [ zz ] describes the method for determining types and redshifts for all the subaru targets and the following sections are dedicated only to sne ia . one can determine the redshift of the supernova either from the supernova itself or from the host galaxy ; the latter was our primary choice . the representative emission lines are balmer lines ( h@xmath43 and h@xmath44 ) and forbidden lines of [ nii ] , [ oii ] , [ oiii ] and [ sii ] . redshifts were determined from the wavelengths determined by fitting gaussian functions to all the emission lines . the absorption lines we used were ca h&k , mgb ( sn12991 ) and nad ( sn12979 ) . we identified absorption lines by visual inspection . the final redshift was the average value of all measured lines with equal weight . lines and their rest - wavelengths are listed in table [ emit ] . if there were no spectral lines in the host galaxy spectra , we adapted the technique of spectra template fitting to the sn spectrum . this technique cross - correlates sn spectra with a series of sn library spectra ( spectral templates and high quality nearby sn spectra ) and search for the @xmath25 minimum to determine redshifts and types ; details are explained in @xcite . the redshifts for 64 objects were determined from spectral lines of their host galaxies and seven were from the fitting of sn spectra . figure [ sbr_z ] shows the redshift distribution for all our targets except 1 agn at @xmath45 . we assigned sn classification by visual inspection @xcite : sne ia are characterized by a deep absorption trough at @xmath46 produced by a blue - shift of `` siii @xmath56355 '' . sne ib show deep absorption of blueshifted hei at 4471,5876,6678 at similar phases . sne ic show several strong caii and oi 7774 absorption lines along with no hydrogen lines and a very weak siii @xmath56355 " line . sne ii all prominently exhibit hydrogen in their spectra . agn show very broad emission lines of hydrogen balmer lines and forbidden lines on a power - law continuum . for the seven spectra not classified visually due to low s / n , a spectral template fitting @xcite was used . our type classifications resulted in 62 confirmed sne ia , 3 of which have not been assigned iau names , 8 sne ii , and 1 agn . table [ sbrspecinfo ] describes sn candidates observed by the subaru / focas telescope . column 1 is the sdss identification number . we denote each sn by its sdss identification number in this paper , but show its iau name in column 2 . column 3 is the candidate type . the celestial coordinates of candidates are shown in columns 4 and 5 . column 6 is the color excess for our galaxy from the @xcite map . column 7 is the mean time of our observation . when both blue and red side spectra were taken , which is the case for most of our sample , we defined the observation date as the average time of both observations . column 8 is the redshift . the parenthesis in this column represents how the redshift was determined ; `` ge '' from galaxy emission lines , `` ga '' from galaxy absorption lines , `` sn '' from the sn spectrum fitting and `` snh '' from the h@xmath43 line of sne ii . column 9 is the mean airmass of our observations . the median value is 1.2 , with 23 % of the observations having an airmass larger than 1.3 . column 10 shows two ratios , @xmath47 , of host galaxy flux to reconstructed sn and host flux for the blue ( 4000 to 6000 or 3600 to 5400 ) and red side spectra ( 5000 to 9000 ) , respectively . ratios are determined from averaged spatial profiles over the wavelength region of the spectrum , which was used to determine the central position and width of each component in the sn extraction . the spatial region was integrated over the fwhm centered on the sn position . when either the blue or red side is missing , this is represented in the table by `` '' . when a sn occurs in a host core , sn and host profiles can be degenerate . this is especially the case when the brightness of the host is comparable to the sn . since these candidates can be a variable active galactic nucleus , a lower priority was put for these sn candidates for spectroscopic followup @xcite , and thus we have few such candidates . for 62 of 66 sn candidates with blue and red side spectra , the difference of host contamination between the two sides ( @xmath48 ) was less than 0.2 . the difference was larger than 0.4 for the remaining four candidates ( sn5751 , 16779 , 16938 , and 17081 ) . either side of the first three ( sn5751 , 16779 and 16938 ) shows a degenerate spatial profile and have large host contamination ( around 0.6 ) . the bestfit gaussian width @xmath23 for the spatial profile of these sne was larger than the width of their host . our sn extraction method did not work sufficiently . a different subtraction method based on the principle component analysis is adapted for these spectra ( sako et al . in prep ) . the gaussian width for sn17081 was normal ( @xmath49 ) , corresponding to the fwhm of @xmath50 . from its negative value of @xmath51 ( -0.40 ) , the lowest contamination ratio of these four candidates , we speculate that sn17081 occurred in a red galaxy . for each sn , we measured the average s / n over the entire spectral range in 2 bins . figure [ sn_sn1a ] shows the s / n distribution for the 62 sne ia . most spectra of the sample show s / n between 4 and 10 . the very high s / n spectra are those of sne ia at @xmath52 . figure [ spec_ex ] gives examples of sn ia spectra obtained by the subaru / focas telescope : two high quality sn ia spectra are shown , a spectrum of a nearby sn ia at @xmath53 one month after from the maximum date in the top left , and a near maximum spectrum of mid - z sn ia at @xmath54 in the top right . also a sn ia spectrum at @xmath55 of low quality one week after from the maximum date is given in the bottom left . the subaru telescope observed 71 sn candidates in total with the focas instrument and confirmed 59 normal sne ia and three probable sne ia . table [ obsf ] summarizes the observational results of the subaru / focas sdss sn campaign . the template fitting of subaru spectra has provided the bestfit sn type ( and redshifts in case of no lines from their hosts ) for ambiguous targets based on the reduced @xmath25 . the following are notes on those targets . ( i ) sn6471 . the bestfit type among our template spectra is a sn iip , matching the sn1999em and sn1992h spectra well . we could not detect emission or absorption lines of this host . the h@xmath43 emission of sn6471 shows that sn6471 is one of the farthest sne iip discovered by the sdss - ii survey , @xmath56 . ( ii ) sn10450 . the h@xmath44 and [ oii ] emission lines of this host show that this is one of the farthest sne discovered by the sdss - ii sn survey , @xmath57 . our library does not contain spectra that match all the spectral features of sn10450 . an emission line around 7500 in the observer frame may be broader than nebular emission . if this is the case , sn10450 is a peculiar sniin . sn10450 is around 1 mag brighter in the rest @xmath8 band than normal sne ia at this redshift . ( iii ) sn12844 . the bestfit type among our template spectra is sn2005gj , which shows an h@xmath43 feature evolving with time and called a type ia / iin sn @xcite . the spectrum of sn12844 shows strong emission lines of [ oii ] , [ oiii ] , h@xmath44 , h@xmath43 , [ nii ] and [ sii ] . ( iv ) sn12979 . the bestfit type among our template spectra is an intrinsically faint sn1991bg . ( v ) sn14475 . the bestfit spectrum is sn1998bw @xcite at @xmath58 days after the maximum date . other matches are sn1997ef @xcite and sn2002ap @xcite . although all the details of the absorption lines of the bestfit spectra are not completely identical , they are all hypernovae . thus it can be said that sn14475 is a possible hypernova . ( vi ) sn15170 . the bestfit spectrum is the intrinsically bright sn1991 t spectrum around the maximum date . we used the salt2 light curve fitting code developed by the snls collaboration @xcite to derive light curve parameters . this code employs a two dimensional surface in time and wavelength that describes the temporal evolution of a sn ia spectrum @xmath59 ; @xmath60 \times \exp(c_{salt } cl(\lambda)),\ ] ] where @xmath61 is the sn phase in restframe days , the elapsed time from maximum luminosity , @xmath62 is the average sed , and @xmath63 is a normalization . the first - order spectral deviation @xmath64 is included in a linear fashion with a coefficient @xmath65 . we used the spectral surfaces of @xmath66 ( @xmath67 ) that we also used for the cosmological analysis of @xcite . the parameter @xmath65 is the normalized deviation from the typical spectrum @xmath62 and can be approximated with @xmath68 as @xmath69@xmath65 @xcite . the function @xmath70 is the color law of a third order polynomial function monotonically increasing with wavelength and @xmath71 , the deviation from the mean sn ia ( b - v ) color , is its coefficient which is sensitive to both intrinsic color diversity and the host - dust reddening . we selected every photometric point of the @xmath40 band light curves unless a bad @xmath72 was assigned ; this happened when the photometric scaling factor was low , the rms magnitude of calibration stars was high , the variation in sky brightness was high , the frame fit quality for the sn photometry was poor , or no calibration stars were available in the frame ( corresponding to the @xmath73 as recommended in @xcite ) . we did not use @xmath74 and @xmath75 band data , since their s / n were not good enough for detailed analysis . the output parameters of this code were rest-@xmath76 band maximum magnitude @xmath77 , which is derived from @xmath63 , spectral deviation @xmath65 , color excess @xmath71 , and the date of maximum luminosity @xmath78 . in figure [ sbrdist ] , we present the distribution of sn phases at the spectroscopic epoch ( upper left ) , the redshift ( upper right ) , the spectral deviation @xmath65 ( lower left ) and the color excess @xmath71 ( lower right ) for the subaru sample . the sn phase is defined as the difference of dates between the spectral observation and maximum brightness divided by @xmath79 . we can see that the range of phases varies from -7 days to + 30 days and that the highest number is in the bin @xmath80 after maximum brightness ( top left panel ) . the redshift distribution reaches to the @xmath81 bin and it peaks at the @xmath82 bin ( top right ) . the lightcurve - width histogram resembles a right rectangle shape , with its oblique side decreasing with the value of @xmath65 . the number shortage of sne ia with large @xmath65 values is due to a bias that arose when the subaru targets were selected ( left bottom ) . most of the sample fall in the color range from -0.2 to 0.2 . the salt2 fitting gives large positive values of this parameter for ambiguous targets sn12844 , sn12979 and sn14475 , which are @xmath71 @xmath83 , @xmath84 , and @xmath85 , respectively . these might be cool or dusty sne . we carried out followup spectroscopy of 71 sn ia candidates discovered by the sdss - ii supernova survey . nearly 90 % of the observed spectra were identified as sne ia , showing the high efficiency of the photometric selection . sn cosmology and related work , such as the study of spectroscopic diversity , will benefit from our observations . the number of sne ia in the intermediate redshift range has increased substantially . although observed spectra are a combination of the sn flux and the flux of their host galaxies , we succeeded in extracting the sn component from the observed spectra using code developed by the author . acknowledgements we thank harold spinka for his careful reading of the manuscript for publication . thanks the coe program `` the quantum extreme systems and their symmetries '' for fiscal 2007 , the global coe program `` the physical sciences frontier '' for fiscal 2008 - 2009 , mext , japan and the jasso scholarship for fiscal 2007 - 2009 . l. is partially supported by the spanish ministry of science and innovation ( micinn ) through the consolider ingenio-2010 program , under project csd2007 - 00060 `` physics of the accelerating universe ( pau ) '' . funding for the sdss and sdss - ii was provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss web site is http://www.sdss.org/[the sdss web site is http://www.sdss.org/ ] . the sdss is managed by the astrophysical research consortium ( arc ) for the participating institutions . the participating institutions are the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , university of pittsburgh , princeton university , the united states naval observatory , and the university of washington . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . aldering , g. , et al . 2002 , spie , 4836 , 61 aldering , g. , et al . 2006 , , 650 astier , p. , et al . 2006 , , 447 , 31 baumont , s. , et al . 2008 , , 491 , 567 blondin , s. , walsh , j.r . , leibundgut , b. , & sainton , g. 2005 , , 431 , 757 bohlin , r.c . , dickinson , m.e . & calzetti , d. 2001 , , 122 , 2118 cardelli , j.a . , clayton , g.c . , & mathis , j.s . 1989 , , 345 , 245 contreras , et al . 2010 , , 139 , 519 ellis , r.s . , et al . 2008 , , 674 , 51 fillipenko , a.v , 1997 , , 1997 , 35 foley , r. j. , et al . 2003 , , 115 , 1220 foley , r.j . , et al . , 2008 , , 684 , 68 foley , r.j . , et al . 2010 , arxiv : 1010.2749 frieman , j.a . , et al . 2008 , , 135 , 338 fukugita , m. , ichikawa , t. , gunn , j. e. , doi , m. , shimasaku , k. , & schneider , d. p. 1996 , , 111 , 1748 galama , t.j . , et al . 1998 , , 395 , 670 gal - yam , a. , ofek , e. o. , & shemmer , o. 2002 , , 332 , l73 gunn , j. e. , et al . 1998 , , 116 , 3040 gunn , j. e. , et al . 2006 , , 131 , 2332 guy , j. , et al . 2007 , , 466 , 11 hicken , m. , et al . 2009 , , 700 , 331 holtzman , j.a . , et al . 2008 , , 136 , 2306 hflich , p. , wheeler , j. c. , & thielemann , f. k. 1998 , , 495 , 617 howell , d.a . , et al . 2005 , , 634 , 1190 iwamoto , k. , et al . 2000 , , 534 , 660 jha , s. , et al . 2006 , , 131 , 527 kashikawa , n. , et al . 2002 , , 54 , 819 kessler , r. , et al . 2009 , , 185 , 32 konishi , k. , et al . 2011b in prep . massey , p. , strobel , k. , barnes , j.v . & anderson , e. 1988 , , 328 , 315 matheson , t. , et al . 2008 , , 135 , 1598 mazzali , p. a. , et al . 2002 , , 572 , l61 miknaitis , g. , et al . 2007 , , 666 , 674 oke , j.b . 1990 , , 99 , 1621 stman , l. et al . , 2011 , , 526 , a28 patat , f. , et al . , 2001 , , 555 , 900 perlmutter , s. , 1999 , , 517 , 565 prieto , j.l . 2007 , arxiv : 0706.408 riess , a. g. , et al . 1998 , , 116 , 1009 sako , m. , et al . 2008 , , 135 , 348 schlegel , d.j . , finkbeiner , d.p . & davis , m 1998 , , 500 , 525 tokita , k. 2009 , ph - d thesis , univ . of tokyo zheng , c. , et al . 2008 , , 135 , 1766 -band image ; the sn ia ( sn16953 ) occurs on the right side of the host galaxy . ( b ) a part of the 2d spectrum centered on an emission line : before ( left ) and after ( right ) sn extraction . ( c ) spatial profile of the spectra : the observed profile ( red ) is fitted with gaussians for the sn ( black ) and galaxy ( green ) components . the sum is shown with the pink curve . ( d ) extracted sn spectrum ( black ) and uncertainty flux ( red ) . [ sec ] , title="fig:",width=264,height=188 ] -band image ; the sn ia ( sn16953 ) occurs on the right side of the host galaxy . ( b ) a part of the 2d spectrum centered on an emission line : before ( left ) and after ( right ) sn extraction . ( c ) spatial profile of the spectra : the observed profile ( red ) is fitted with gaussians for the sn ( black ) and galaxy ( green ) components . the sum is shown with the pink curve . ( d ) extracted sn spectrum ( black ) and uncertainty flux ( red ) . [ sec ] , title="fig:",width=264,height=188 ] + -band image ; the sn ia ( sn16953 ) occurs on the right side of the host galaxy . ( b ) a part of the 2d spectrum centered on an emission line : before ( left ) and after ( right ) sn extraction . ( c ) spatial profile of the spectra : the observed profile ( red ) is fitted with gaussians for the sn ( black ) and galaxy ( green ) components . the sum is shown with the pink curve . ( d ) extracted sn spectrum ( black ) and uncertainty flux ( red ) . [ sec ] , title="fig:",width=264,height=188 ] -band image ; the sn ia ( sn16953 ) occurs on the right side of the host galaxy . ( b ) a part of the 2d spectrum centered on an emission line : before ( left ) and after ( right ) sn extraction . ( c ) spatial profile of the spectra : the observed profile ( red ) is fitted with gaussians for the sn ( black ) and galaxy ( green ) components . the sum is shown with the pink curve . ( d ) extracted sn spectrum ( black ) and uncertainty flux ( red ) . [ sec ] , title="fig:",width=264,height=188 ] + in the top left , a near maximum spectrum of mid - z sn ia at @xmath86 in the top right , and one low quality sn ia spectrum ( one week after the maximum date ) at @xmath55 in the bottom left . the lower and upper horizontal axes are observed and rest wavelengths . the vertical axis is the flux density . spectra are corrected for slitloss if the @xmath40 magnitudes can be interpolated at the spectral date ( labeled as dseab " in the figures ; @xmath87 , @xmath8 and @xmath88 magnitudes are shown in blue , green and red colors ) or not ( labeled as deab " in the figure ) . [ spec_ex ] , title="fig:",width=264,height=188 ] in the top left , a near maximum spectrum of mid - z sn ia at @xmath86 in the top right , and one low quality sn ia spectrum ( one week after the maximum date ) at @xmath55 in the bottom left . the lower and upper horizontal axes are observed and rest wavelengths . the vertical axis is the flux density . spectra are corrected for slitloss if the @xmath40 magnitudes can be interpolated at the spectral date ( labeled as dseab " in the figures ; @xmath87 , @xmath8 and @xmath88 magnitudes are shown in blue , green and red colors ) or not ( labeled as deab " in the figure ) . [ spec_ex ] , title="fig:",width=264,height=188 ] + in the top left , a near maximum spectrum of mid - z sn ia at @xmath86 in the top right , and one low quality sn ia spectrum ( one week after the maximum date ) at @xmath55 in the bottom left . the lower and upper horizontal axes are observed and rest wavelengths . the vertical axis is the flux density . spectra are corrected for slitloss if the @xmath40 magnitudes can be interpolated at the spectral date ( labeled as dseab " in the figures ; @xmath87 , @xmath8 and @xmath88 magnitudes are shown in blue , green and red colors ) or not ( labeled as deab " in the figure ) . [ spec_ex ] , title="fig:",width=264,height=188 ] in the top left , a near maximum spectrum of mid - z sn ia at @xmath86 in the top right , and one low quality sn ia spectrum ( one week after the maximum date ) at @xmath55 in the bottom left . the lower and upper horizontal axes are observed and rest wavelengths . the vertical axis is the flux density . spectra are corrected for slitloss if the @xmath40 magnitudes can be interpolated at the spectral date ( labeled as dseab " in the figures ; @xmath87 , @xmath8 and @xmath88 magnitudes are shown in blue , green and red colors ) or not ( labeled as deab " in the figure ) . [ spec_ex ] , title="fig:",width=264,height=188 ] + lccc 2005 + bd+28@xmath64211(300b & y47 ) & sep26 & 35.0 & 10.0 + bd+28@xmath64211(300r & l600 ) & sep26 & 34.7 & 20.0 + feige110(300b & y47 ) & sep26 & 31.4 & 30.0 + feige110(300r & l600 ) & sep26 & 31.9 & 40.0 + bd+28@xmath64211(300b & y47 ) & sep27 & 39.0 & 20.0 + bd+28@xmath64211(300r & l600 ) & sep27 & 38.2 & 10.0 + gd71(300b & y47 ) & sep27 & 5.8 & 90.0 + gd71(300r & l600 ) & sep27 & 5.3 & 90.0 + bd+28@xmath64211(300b & y47 ) & oct25 & 17.8 & 15.0 + bd+28@xmath64211(300r & l600 ) & oct25 & 16.7 & 10.0 + bd+75@xmath6325(300b & y47 ) & oct25 & 55.4 & 5.0 + bd+75@xmath6325(300b & y47 ) & oct25 & 55.4 & 5.0 + bd+28@xmath64211(300b & y47 ) & oct26 & 18.1 & 15.0 + bd+28@xmath64211(300r & l600 ) & oct26 & 17.8 & 10.0 + bd+75@xmath6325(300b & y47 ) & oct26 & 55.3 & 4.0 + bd+75@xmath6325@xmath89(300r & l600 ) & oct26 & 55.3 & 4.0 + bd+28@xmath64211(300b & y47 ) & oct27 & 16.8 & 30.0 + bd+28@xmath64211(300r & l600 ) & oct27 & 16.1 & 20.0 + gd71(300b & y47 ) & oct27 & 25.0 & 60.0 + gd71(300r & l600 ) & oct27 & 25.6 & 60.0 + feige110(300b & y47 ) & nov25 & 26.4 & 60.0 + feige110(300r & l600 ) & nov25 & 25.9 & 40.0 + feige34(300b & y47 ) & nov25 & 27.3 & 15.0 + feige34(300r & l600 ) & nov25 & 27.6 & 25.0 + feige110(300b & y47 ) & nov26 & 26.6 & 60.0 + feige110(300r & l600 ) & nov26 & 26.4 & 40.0 + 2006 + bd+28@xmath64211(300b & l550 ) & sep15 & 36.0 & 3.0 + bd+28@xmath64211(300b & l550 ) & sep16 & 43.3 & 15.0 + bd+28@xmath64211(300r & y47 ) & sep16 & 41.3 & 15.0 + bd+28@xmath64211(300b & l600 ) & sep16 & 40.7 & 10.0 + bd+28@xmath64211(300b & l550 ) & sep17 & 45.8 & 30.0 + bd+28@xmath64211(300b & l600 ) & sep17 & 41.3 & 10.0 + bd+28@xmath64211(300r & y47 ) & sep17 & 42.0 & 15.0 + gd71(300b & l550 ) & sep17 & 16.8 & 20.0 + gd71(300b & l600 ) & sep17 & 14.9 & 10.0 + gd71(300r & y47 ) & sep17 & 15.3 & 15.0 + bd+28@xmath64211(300b & l550 ) & oct14 & 23.4 & 10.0 + bd+28@xmath64211(300r & y47 ) & oct14 & 23.7 & 15.0 + bd+28@xmath64211(300b & l550 ) & nov25 & 12.9 & 5.0 + bd+28@xmath64211(300r & y47 ) & nov25 & 13.1 & 15.0 + feige34(300b & l550 ) & nov25 & 29.5 & 40.0 + feige34(300r & y47 ) & nov25 & 29.2 & 60.0 + feige110(300b & l550 ) & nov26 & 35.2 & 15.0 + feige110(300r & y47 ) & nov26 & 35.5 & 30.0 + ll balmer lines & + h@xmath44 & 4862.68 + h@xmath43 & 6564.61 + forbidden lines & + @xmath90oii@xmath91 $ ] & 3727.092 , 3729.875 + @xmath90oiii@xmath91 $ ] & 4364.436 , 4932.60 , 4960.295 , 5008.240 + @xmath90oi@xmath91 $ ] & 6302.046 + @xmath90nii@xmath91 $ ] & 6549.86 , 6585.27 + @xmath90sii@xmath91 $ ] & 6718.29 , 6732.67 + ca h&k & 3934.777 , 3969.588 + mgb & 5176.7 + nad & 5895.6 lllrrrrlll 1119 & 2005fc & ia & -39.58644 & 0.89466 & 0.060 & 53640.26 & 0.2974 ( ge ) & 1.10 & 0.00,0.09 + 1166 & & ia & 9.35564 & 0.97328 & 0.019 & 53641.39 & 0.3814 ( ga ) & 1.14 & 0.12,0.00 + 1686 & & ia & 2.24835 & -0.21039 & 0.055 & 53640.42 & 0.1362 ( ge ) & 1.07 & 0.41,0.39 + 1688 & & ia & -38.64223 & 0.32454 & 0.050 & 53641.25 & 0.3591 ( ge ) & 1.18 & 0.24,0.30 + 2165 & 2005fr & ia & 17.09174 & -0.09627 & 0.033 & 53641.42 & 0.2841 ( sn ) & 1.11 & 0.00,0.00 + 2330 & 2005fp & ia & 6.80711 & 1.12060 & 0.024 & 53641.35 & 0.2129 ( ge ) & 1.27 & 0.34,0.29 + 2422 & 2005fi & ia & 1.99454 & 0.63820 & 0.039 & 53641.32 & 0.2628 ( ge ) & 1.42 & 0.16,0.20 + 2661 & & iip & -6.79249 & 0.09721 & 0.038 & 53641.28 & 0.1928 ( ge ) & 1.50 & 0.52,0.42 + 2635 & 2005fw & ia & 52.70433 & -1.23808 & 0.096 & 53670.60 & 0.1434 ( ge ) & 1.39 & 0.45,0.25 + 2789 & 2005fx & ia & -15.79855 & 0.40110 & 0.051 & 53640.38 & 0.2862 ( ga ) & 1.06 & 0.00,0.07 + 2992 & 2005gp & ia & 55.49706 & -0.78267 & 0.087 & 53669.63 & 0.1261 ( ge ) & 1.56 & 0.79,0.54 + 3080 & 2005ga & ia & 16.93234 & -1.03950 & 0.048 & 53670.36 & 0.1743 ( ge ) & 1.10 & 0.36,0.51 + 3451 & 2005gf & ia & -25.93073 & 0.70815 & 0.056 & 53640.31 & 0.2497 ( ge ) & 1.10 & 0.19,0.33 + 3452 & 2005gg & ia & -25.32850 & 0.63919 & 0.059 & 53640.34 & 0.2311 ( ge ) & 1.06 & 0.35,0.38 + 5391 & 2005hs & ia & 52.34197 & -1.09470 & 0.111 & 53669.56 & 0.3003 ( ge ) & 1.18 & 0.04,0.01 + 5533 & 2005hu & ia & -31.33006 & 0.41343 & 0.070 & 53669.25 & 0.2199 ( ge ) & 1.06 & 0.21,0.27 + 5717 & 2005ia & ia & 17.89596 & -0.00589 & 0.031 & 53670.28 & 0.2521 ( ge ) & 1.44 & 0.25,0.22 + 5737 & 2005ib & ia & 22.85704 & -0.60339 & 0.038 & 53670.32 & 0.3933 ( ge ) & 1.25 & 0.39,0.62 + 5751 & 2005hz & ia & 11.63416 & 0.83823 & 0.022 & 53701.33 & 0.1308 ( ge ) & 1.08 & 0.06,0.65 + 5844 & 2005ic & ia & -32.21379 & -0.84294 & 0.121 & 53669.29 & 0.3120 ( ge ) & 1.09 & 0.16,0.15 + 5944 & 2005hc & ia & 29.19980 & -0.21372 & 0.030 & 53701.49 & 0.0445 ( ge ) & 1.84 & 0.12,0.09 + 5957 & 2005ie & ia & 34.76058 & -0.27283 & 0.036 & 53669.60 & 0.2796 ( ge ) & 1.86 & 0.00,0.00 + 6057 & 2005if & ia & 52.55362 & -0.97458 & 0.119 & 53700.54 & 0.0669 ( ge ) & 1.56 & 0.26,0.29 + 6108 & 2005ih & ia & 1.80660 & 0.34898 & 0.054 & 53671.33 & 0.2599 ( ge ) & 1.07 & 0.32,0.29 + 6196 & 2005ig & ia & -22.36885 & -0.50266 & 0.063 & 53670.25 & 0.2811 ( ga ) & 1.10 & 0.39,0.41 + 6249 & 2005ii & ia & 3.26555 & -0.62012 & 0.041 & 53671.39 & 0.2947 ( ge ) & 1.10 & 0.34,0.40 + 6406 & 2005ij & ia & 46.08861 & -1.06296 & 0.081 & 53700.50 & 0.1244 ( ge ) & 1.41 & 0.61,0.63 + 6471 & & iip & -50.61388 & 0.49495 & 0.075 & 53671.23 & 0.202 ( snh ) & 1.08 & 0.00,0.00 + 6699 & 2005ik & ia & -37.18502 & -1.05699 & 0.053 & 53671.29 & 0.3105 ( ge ) & 1.12 & 0.04,0.18 + 9032 & 2005le & ia & -22.11544 & -0.49357 & 0.060 & 53700.23 & 0.2541 ( ge ) & 1.17 & 0.00,0.02 + 9207 & 2005lg & ia & 19.08363 & -0.80780 & 0.038 & 53700.47 & 0.3496 ( ge ) & 1.82 & 0.46,0.40 + 10449 & 2005ll & ia & -22.97138 & -1.12817 & 0.062 & 53701.23 & 0.2415 ( ge ) & 1.10 & 0.34,0.39 + 10450 & & iin - pec & -22.72438 & -1.12867 & 0.064 & 53700.23,53701.27 & 0.5399 ( ge ) & 1.10 & 0.06,0.02 + 10550 & 2005lf & ia & -10.32462 & -1.20487 & 0.036 & 53700.30 & 0.3000 ( ge ) & 1.19 & 0.08,0.20 + 12844 & 2006fe & 05gj ? & -46.96182 & -0.51113 & 0.101 & 53995.25 & 0.0702 ( ge ) & 1.14 & 0.42,0.36 + 12855 & 2006fk & ia & -29.74447 & 0.71624 & 0.052 & 53993.27 & 0.1722 ( ga ) & 1.22 & 0.00, + 12860 & 2006fc & ia & -36.30580 & 1.17591 & 0.069 & 53993.28 & 0.1210 ( ge ) & 1.19 & 0.13, + 12869 & 2006ge & ia & -32.60088 & 0.00108 & 0.140 & 53996.24 & 0.2841 ( ge ) & 1.15 & 0.07,0.12 + 12883 & 2006fr & ia & -47.37411 & 0.39867 & 0.102 & 53995.30 & 0.3060 ( ge ) & 1.06 & 0.53,0.32 + 12972 & 2006 ft & ia & 7.95859 & -0.38296 & 0.023 & 53995.43 & 0.2606 ( ge ) & 1.09 & 0.00,0.00 + 12977 & 2006gh & ia & 13.69561 & -0.25086 & 0.025 & 53996.50 & 0.2474 ( ge ) & 1.08 & 0.12,0.19 + 12979 & 2006gf & 91bg ? & 11.60148 & 0.00348 & 0.020 & 53996.46 & 0.1164 ( ga ) & 1.06 & 0.41,0.22 + 12991 & 2006gd & iip & 17.60995 & -1.06803 & 0.073 & 53996.53 & 0.1539 ( ge ) & 1.12 & 0.19,0.08 + 13000 & & iin & 20.74847 & -0.63447 & 0.037 & 53995.58 & 0.3513 ( ge ) & 1.25 & 0.34,0.07 + 13025 & 2006fx & ia & -18.43267 & 0.41590 & 0.083 & 53995.35 & 0.2240 ( ge ) & 1.09 & 0.29,0.16 + 13099 & 2006 gb & ia & -0.18127 & -1.25038 & 0.039 & 53995.39 & 0.2655 ( ge ) & 1.12 & 0.32,0.30 + 13152 & 2006gg & ia & 7.05211 & 0.11801 & 0.021 & 53996.40 & 0.2031 ( ge ) & 1.13 & 0.35,0.15 + 13174 & 2006ga & ia & 13.23469 & 0.44786 & 0.026 & 53995.48 & 0.2357 ( ge ) & 1.06 & 0.30,0.23 + 13327 & 2006jf & ia & -21.27261 & 0.00204 & 0.057 & 53996.32 & 0.2814 ( ge ) & 1.13 & 0.55,0.47 + 13346 & & agn & -20.86334 & 0.32819 & 0.066 & 53996.36 & 1.3716 ( agn ) & 1.06 & , + 13376 & 2006gq & ii ? & 26.30225 & 0.35644 & 0.037 & 53996.60 & 0.0696 ( ge ) & 1.29 & 0.15,0.26 + 13934 & 2006jg & ia & -17.88899 & -0.43527 & 0.102 & 54023.41 & 0.3310 ( ge ) & 1.33 & 0.12,0.41 + 14261 & 2006jk & ia & -31.75960 & 0.25379 & 0.094 & 54023.27 & 0.2853 ( ge ) & 1.06 & 0.00,0.00 + 14298 & 2006jj & ia & -45.10495 & 1.22326 & 0.076 & 54023.23 & 0.2692 ( sn ) & 1.06 & 0.00,0.00 + 14456 & 2006jm & ia & -16.44890 & 1.05066 & 0.090 & 54023.46 & 0.3283 ( ge ) & 1.77 & 0.28,0.35 + 14475 & & hyp ? & -23.87125 & 0.20341 & 0.084 & 54023.30 & 0.1437 ( ge ) & 1.06 & 0.27,0.21 + 15170 & 2006jx & 91 t ? & 58.05858 & 0.29207 & 0.245 & 54023.57 & 0.3999 ( sn ) & 1.11 & ,0.06 + 15217 & 2006jv & ia & 22.63421 & 0.21980 & 0.029 & 54023.49 & 0.371 ( sn ) & 1.18 & 0.00,0.00 + 15219 & 2006ka & ia & 34.61115 & 0.22669 & 0.037 & 54023.54 & 0.2467 ( ge ) & 1.23 & 0.13,0.14 + 16631 & 2006pv & ia & 54.62279 & -0.67557 & 0.128 & 54065.53 & 0.207 ( sn ) & 1.55 & ,0.37 + 16758 & 2006pw & ia & -18.38765 & 1.15081 & 0.085 & 54065.25 & 0.3261 ( ge ) & 1.12 & 0.02,0.01 + 16776 & 2006qd & ia & -12.97859 & -0.17608 & 0.044 & 54066.30 & 0.2667 ( ge ) & 1.31 & 0.32,0.37 + 16779 & 2006qa & ia & 45.29054 & -0.02624 & 0.105 & 54065.44 & 0.3983 ( ge ) & 1.16 & 0.59,0.04 + 16781 & 2006qb & ia & 52.06103 & -0.20096 & 0.111 & 54065.48 & 0.3257 ( ge ) & 1.29 & 0.01,0.01 + 16840 & 2006qf & ii ? & 57.98849 & -0.39658 & 0.160 & 54065.56 & 0.1662 ( ge ) & 1.85 & ,0.07 + 16847 & 2006px & ia & -6.67875 & 0.96110 & 0.033 & 54066.30 & 0.2767 ( ge ) & 1.44 & 0.24,0.28 + 16938 & 2006qe & ia & -11.91674 & -0.53442 & 0.047 & 54065.29 & 0.386 ( sn ) & 1.19 & 0.61,0.12 + 16953 & 2006pp & ia & 12.24364 & 0.47695 & 0.024 & 54065.33 & 0.3387 ( ge ) & 1.11 & 0.27,0.19 + 17048 & 2006qi & ia & 27.58151 & 0.88490 & 0.025 & 54065.36 & 0.189 ( sn ) & 1.10 & 0.19,0.38 + 17081 & 2006ql & ia & 27.53797 & 0.42048 & 0.038 & 54065.40 & 0.2749 ( ga ) & 1.21 & 0.02,0.42 + 17117 & 2006qm & ia & 40.60031 & -0.79652 & 0.032 & 54066.38 & 0.1404 ( ge ) & 1.08 & 0.00,0.00 + lll supernova & & + - normal ia & 59 ( @xmath92 ) & 2005/2006yr incl . non - iau named ia ( 3@xmath89 ) + - peculiar ia & 3 & sn12844@xmath93 , sn12979@xmath94 , sn15170@xmath95 + - type iin & 7 & sn10450 ( peculiar iin ) + probable hypernova & 1 & sn14475@xmath96 + agn & 1 & sn13346 +

we analyze a high - resolution spectrum of a microlensed g - dwarf in the galactic bulge , acquired when the star was magnified by a factor of 110 . we measure a spectroscopic temperature , derived from the wings of the balmer lines , that is the same as the photometric temperature , derived using the color determined by standard microlensing techniques . we measure [ fe / h]= , which places this star at the upper end of the bulge giant metallicity distribution . in particular , this star is more metal - rich than any bulge m giant with high - resolution abundances . we find that the abundance ratios of alpha and iron - peak elements are similar to those of bulge giants with the same metallicity . for the first time , we measure the abundances of k and zn for a star in the bulge . the [ k / mg ] ratio is similar to the value measured in the halo and the disk , suggesting that k production closely tracks production . the [ cu / fe ] and [ zn / fe ] ratios support the theory that those elements are produced in type ii sne , rather than type ia sne . we also measured the first c and n abundances in the bulge that have not been affected by first dredge - up . the [ c / fe ] and [ n / fe ] ratios are close to solar , in agreement with the hypothesis that giants experience only canonical mixing .

0801.2159

the galactic bulge underwent an intense burst of star formation early in the formation of the galaxy , leading to a very different stellar population and chemical evolution history than found in the milky way disk or halo ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in particular , massive stars may dominate the pollution at almost all metallicities , leading to unique abundance patterns ( e.g. , * ? ? ? * ) . similar events are thought to mark the formation of other galactic spheroids , making the bulge stellar population a template for interpreting extragalactic observations . as a result of its unique formation history in the galaxy , the bulge has been the subject of intensive study . the detection of rr lyrae stars @xcite first indicated that the bulge contained old stars . with deeper photometry , the main sequence turnoff ( msto ) of the bulge was detected . @xcite found a mean age of 11 - 14 gyr for stars in baade s window , with a negligible fraction of stars with ages @xmath2 5 gyr . photometry reaching more than 2 magnitudes below the msto with _ hubble space telescope _ confirmed the generally old nature of the bulge @xcite , although @xcite included a reminder that a young metal - rich population has similar msto colors and luminosities to an older , more metal - poor population . therefore , deriving ages reliably from photometry of the msto requires adequate knowledge of the bulge metallicity distribution function ( mdf ) , specifically for the msto stars . however , because of the faintness of those stars , the measurement of the bulge mdf has historically relied on giants . since the discovery of both m giants and rr lyr stars , it has been known that the bulge giants span a wide range in metallicity . low - dispersion spectra provided the first quantitative measure of the mdf @xcite . @xcite measured indices from low - dispersion spectra of 268 k giants ( both red clump stars and first ascent giants ) to derive a mean metallicity @xmath3 log@xmath4(n@xmath5/n@xmath6)@xmath7 log@xmath4(n@xmath5/n@xmath6)@xmath8 ) @xmath9}\rangle = -0.11 $ ] , with a dispersion of 0.46 dex . recalibration by @xcite based on high - resolution spectra of 15 stars in common with the @xcite sample reduced the mean metallicity to @xmath10 . @xcite measured @xmath9}\rangle = -0.21 $ ] from low - dispersion near - infrared spectra for 72 m giants in the inner bulge . the good agreement between @xcite result and the recalibrated @xcite result is somewhat surprising . at the bright end of the giant branch , only metal - rich first ascent giants become m giants . however , both k giants and m giants become red clump stars , and lower luminosity metal - rich giants are k stars as well . therefore , the inclusion of red clump stars and fainter giants in the @xcite sample make the biases in their sample more similar to those of the @xcite study . @xcite measured both the mdf of the bulge and the age of the stars using deep photometry of the bulge in the optical and near - infrared wavelengths . they constructed the m@xmath11 and ( v - k)@xmath12 color magnitude diagram for a low - reddening window at ( @xmath13)=(0.277 , @xmath146.167 ) . because the slope of the red giant branch ( rgb ) in these colors depends on the metallicity , rgb stars with different metallicities fall on different parts of the color - magnitude diagram . they could therefore use their photometry to derive the mdf of 503 giants by comparing the positions of bright ( m@xmath15 ) rgb stars with globular cluster fiducials of known metallicity . after correcting for small biases in their mdf caused by their magnitude cutoff , they find an mdf with a peak at [ m / h]@xmath16 , a sharp cutoff at [ m / h]@xmath17 and few stars with [ m / h]@xmath18 . adopting the metallicities derived from the giants for the msto dwarfs , they estimated that the bulge is coeval with the halo and argued that the lack of stars above the prominent msto of the bulge ruled out a significantly younger population . these measurements of the bulge giant mdf can be improved by metallicity measurements from high - dispersion measurements of many stars in several fields throughout the bulge . recently , @xcite and @xcite have obtained a total of @xmath19 k giant spectra at ( r@xmath2020,000 ) in 4 bulge windows . they confirm the metal - rich nature of the bulge . with such a metal - rich population having been reached so quickly after star formation began , the bulge is expected to have a distinct chemical evolution compared to other galactic populations , such as the halo or the disk because , for example , the contributions of longer - lived polluters , such as type ia supernovae ( sne ) or low - mass agb stars , should be small . indeed , @xcite measured high [ @xmath1/fe ] ratios in giants , in particular , high [ mg / fe ] for [ fe / h ] values up to solar , arguing for little type ia sn contribution of fe compared to the thick or thin disks . @xcite confirmed the overall enhancement in [ mg / fe ] and strengthened the conclusion of @xcite that the other @xmath1 abundance ratios do not track [ mg / fe ] exactly . [ o / fe ] , [ si / fe ] , [ ca / fe ] and [ ti / fe ] begin to decrease around [ fe / h]=0 , while [ mg / fe ] does so at supersolar [ fe / h ] . @xcite suggested that metallicity - dependent type ii sn yields could explain the different behaviors of the @xmath1 elements . @xcite explained the low [ o / mg ] ratios in metal - rich bulge giants through a different metallicity - dependent mechanism : wolf - rayet winds leading to less effective o production in metal - rich massive stars . by [ fe / h]@xmath200.2 , all the [ @xmath1/fe ] ratios have begun to decline , probably indicating the introduction of large amounts of fe from type ia sne ( e.g. , * ? ? ? several studies have looked at the abundances of the light elements na and al in the bulge , two elements whose production should depend on the metallicity of the massive stars that exploded as type ii sne . @xcite measured na in 7 k and m giants and found the predicted increase in [ na / fe ] and [ na / o ] in the most metal - rich stars . @xcite found supersolar [ al / fe ] at all metallicities and , for [ fe / h ] @xmath210 , enhanced [ na / fe ] compared to the ratios in disk stars . interestingly , at higher metallicities , the scatter in [ al / fe ] and [ na / fe ] increased and became larger than could be explained by observational errors . interpreting the na and al abundances as the result of type ii sn production may be problematic . the surface abundances of al and na have been shown to be increased by large amounts of internal mixing in metal - poor globular cluster stars ( e.g. * ? ? ? * ) , where the products of proton - capture reactions deep inside the star are mixed up to the surface , leading to enhancements in these two elements . however , @xcite argued that the c and n abundances in the giants they studied were consistent with only mild mixing and , therefore , that the high na and al had to be due to the overall chemical evolution of the bulge . @xcite also found evidence for mild mixing in giants , affecting c and n , but not o , na , or al . finally , there is little information on the neutron - capture elements in the bulge . the absorption lines for these elements are concentrated in the blue part of the optical spectrum , where the crowding from fe , cn , and other lines is severe . near - ir spectra have essentially no lines of these elements . while there are a few lines of ba in the red , these lines in metal - rich giants are so saturated that reliable measurements are very difficult . as a result , only the neutron - capture element eu has published results so far . @xcite found [ eu / fe]@xmath210 in bulge giants , which is likely because of the production of eu in the r - process . measuring additional neutron - capture element abundances would test this idea , because the r - process is better at making some elements ( e.g. , eu ) than others ( e.g. , ba ) . our knowledge of the metallicity and abundance ratios of bulge stars has generally been confined to the bright giants , which are usually the only ones accessible to high - resolution observations . but studying the dwarfs has several advantages . their abundances of the elements such as c and n are unaffected by dredge - up processes on the giant branch . we can measure elements , such as s and zn , that not measured in giants , because the hotter temperatures of the dwarfs decrease the strength of cn and increase the strength of certain atomic lines . in addition , it is critical to know the metallicity of stars at the msto to accurately measure the ages from their color and luminosity . finally , individual ages can be assigned to dwarf stars near the turnoff . the advent of large surveys to identify and follow - up microlensing events , such as the microlensing observations in astrophsycis ( moa ) collaborationiabond / alert / alert.html ] , the optical gravitational lens experiment ogle / ogle3/ews / ews.html ] ( ogle ) , the microlensing follow up networkmicrofun/ ] ( @xmath22fun ) and the probing lensing anomalies network ( planet ) , provides an opportunity to study otherwise unobservable bulge dwarfs . during high - magnification microlensing events , it is possible to obtain high - resolution , high signal - to - noise ratio spectra of faint stars with a huge savings in observing time : a factor @xmath23 where @xmath24 is the magnification and @xmath25 is the number of magnitudes below sky of the unmagnified star . in @xcite , we reported the detailed abundances for a highly - magnified bulge dwarf , , which from a 15 minute exposure at magnification @xmath26 , was shown to be one of the most metal - rich stars ever observed . it also provided the first measurements of s and cu in the bulge . here we present a high - resolution spectrum of the bulge g - dwarf moa-2006-blg-99s , taken at magnification @xmath27 . finally , we respond the challenge : `` ask not what microlensing can do for stellar spectroscopy ask what stellar spectroscopy can do for microlensing . '' there is one important way that the spectroscopic study of bulge dwarfs can benefit microlensing . whenever a source approaches or transits a `` caustic '' ( line of infinite magnification ) caused by the lens , one can measure @xmath28 , the ratio of the angular source radius to angular einstein radius @xmath29 , from the microlens lightcurve . then @xmath30 is inferred from the dereddened color and magnitude of the source to yield @xmath31 , which in turn provides important constraints on the lens properties . because spectroscopy is not normally available for these microlensed sources , the dereddened color and magnitude are estimated by comparing the source position on an instrumental color - magnitude diagram with that of the clump and then assuming that the bulge clump is similar to the local clump as measured by _ hipparcos _ @xcite . this procedure undoubtedly suffers some statistical errors and could suffer systematic errors as well . for example , the bulge clump may have a different color from the local one . high - resolution spectra of an ensemble of microlensed bulge sources will test this procedure for both statistical and systematic errors . for ogle-2006-blg-265s , the standard microlensing procedure yielded @xmath32 , whereas @xcite obtained @xmath33 from high - resolution spectroscopy . this difference hints at a possible discrepancy , but only by repeating this procedure on a number of dwarfs can this be confirmed . was alerted as a probable microlensing event toward the galactic bulge ( j2000 ra = 17:54:10.99 , dec = @xmath1435:13:38.0 ; @xmath34 , @xmath35 ) by moa on 22 july 2006 . on 23 july , moa issued a further alert that this would be a high - magnification event , with @xmath36 . intensive photometric observations were then carried by several collaborations , including @xmath22fun , primarily with the aim of searching for planets @xcite . results of that search will be presented elsewhere . the event actually peaked on 23 july ( hjd 2453940.349 ) at @xmath37 . one of us ( bsg ) happened to be at the clay 6.5-m telescope at the magellan observatory when he received the flurry of @xmath22fun emails describing this event . he then interrupted his normal program to obtain a 20-minute exposure of this event at the beginning of the night , just after peak , when the magnification was @xmath38 . unfortunately , the atmospheric transparency was poor , and the observation had to be interrupted after 1089 seconds when the cloud cover became too thick . conditions cleared several hours later , and he obtained two 20-minute exposures centered at ut 02:09:36 and ut 02:30:34 24 july , when the magnification was @xmath39 and @xmath40 . we base our results on these higher quality spectra . the observations of moa-2006-blg-99 were made using the magellan inamori kyocera echelle ( mike ) double spectrograph @xcite mounted on the clay telescope on las campanas . with seeing of 0.71.0 arcsec . we used the 1.0 arcsec slit , which produces r@xmath2025,000 on the blue side and r@xmath2019,000 on the red side . the data were reduced using the mike python data reduction pipeline ( d. kelson , 2007 , private communication ) , with the exception of the bluest orders containing the ch and cn lines , which were reduced using the iraf _ echelle _ package . the bias and overscan were subtracted . the wavelength calibration was derived from th - ar data . flatfields were taken through a diffusor slide to create `` milky flats '' that made the orders sufficiently wide to get good flatfields along the edges of the orders of the data frames . parts of orders with overlapping wavelength coverage were coadded together before analysis . these overlap regions are larger for the bluer parts of the spectrum . over the wavelength region where individual lines were measured ( 5300 - 8000 ) , the signal - to - noise ( s / n ) is @xmath20 30 . the cn and ch bandhead regions had lower s / n ( s / n@xmath2010 ) . we analyzed both the spectrum of and the spectrum of the sun @xcite using the same set of lines , line parameters , and model atmosphere grids . we used turbospectrum @xcite , a 1-dimensional lte code , to derive abundances . turbospectrum uses the recent treatment of damping from @xcite . we interpolated the model atmosphere grid of atlas9 models updated with new opacity distribution functions @xcite . the abundances of most elements were determined from analysis of the equivalent widths ( ews ) . the ews for both and the sun are presented in table 1 . we restricted the analysis to lines with ew@xmath41m . the ews were measured using spectre ( c. sneden , 2007 , private communication ) . we compared synthetic with observed spectra to determine abundances for lines that were blended , lines that had substantial hyperfine splitting and for c and n that were measured from ch and cn lines , respectively . we used the solar atlas of @xcite to check for blending with telluric features and eliminated the few lines that were affected . the linelists for ch and cn are from b. plez ( 2006 , private communication ) . the effect of hyperfine splitting ( hfs ) was included for sc , mn , cu and ba . the hfs constants were taken from the sources listed in @xcite . hfs information was not available for the na or al lines , so neither this study nor the literature studies we use for comparison can correct for those effects . therefore , the comparison between [ na / fe ] and [ al / fe ] values for and the literature values is robust , but the absolute values of these abundance ratios for all studies have a systematic error . table 1 lists the transition probabilities ( listed as log @xmath42-values ) and sources for all the atomic lines . we measured a t@xmath43=5800k using the wings of the h-@xmath1 and h-@xmath44 lines ( figure [ fig : halpha ] ) . our uncertainty in the temperature is based on the uncertainty in this fit , in particular the uncertainty in the level of the continuum . temperatures derived from the balmer lines for metal - rich dwarfs show good agreement with temperatures derived from the infrared flux method and from @xmath45 colors @xcite . for , we compared the temperature derived from the balmer lines with that derived from excitation equilibrium of the lines . the temperature derived from the lines was very uncertain because the abundances derived from individual lines were poorly determined because of the s / n of the spectrum . the excitation equilibrium favored a higher temperature ( + 200k ) with an uncertainty of 300k.given the uncertainties , this temperature is in agreement with the balmer line temperature . next , we measured the microturbulent velocity , @xmath46 , by ensuring that the abundances derived from the lines do not depend on their reduced ew . our best fit value was @xmath46=1.5 km / s . changing @xmath46 by @xmath470.3 km / s gave marginal fits to the data , and we adopt that as our uncertainty in @xmath46 . the gravity was measured by ionization balance for and and for and . as a sanity check , in figure [ fig : iso ] , we compare the atmosphere parameters for , , and the sun to the parameters from the yonsei - yale @xcite isochrones . finally , the metallicity of the model atmosphere , [ m / h ] , was set equal to the [ fe / h ] given by the lines . because the log @xmath48 of the model atmosphere affects the [ m / h ] and the [ m / h ] affects the abundances ( and therefore the log @xmath48 measurement ) , we iterated to find a solution for which and were equal and the [ m / h ] of the model was equal to [ /h ] . [ fig : halpha ] = 0.385 and for ages 2 , 3 , 4 , 5 and 7 gyr . the dashed line shows a 5 gyr isochrone for a solar metallicity.,width=604 ] using standard microlensing techniques ( e.g. , @xcite ) , @xmath22fun determined that the dereddened color and magnitude of the source were @xmath49 , @xmath50 . the error is due to possible differential reddening between the microlensed source and the red clump , which is assumed to have the same @xmath51 as the local hipparcos clump . if ( as seems likely ) the source lies at approximately the galactocentric distance , then its absolute magnitude is @xmath52 and its radius is 1.2@xmath53 . that is , it is a solar - type star . we combined the @xmath22fun color and the @xcite color - temperature relation ( an update of the @xcite relation ) to derive an estimate of the temperature of 5806@xmath47 200 k. this photometric temperature agrees with the temperature from the balmer lines . for , the spectroscopic and photometric temperatures were different , suggesting possible systematic errors in estimating the dereddened color and magnitude of the source star . as outlined in the introduction , quantifying any systematic errors is important for obtaining the most accurate information about microlensing events , and additional data will be crucial for quantifying any systematic error . our uncertainties are 200k for t@xmath43 and 0.3 km / s for @xmath46 . the distribution of ews and excitation potential for the lines were sufficiently uncorrelated that @xmath46 did not depend on t@xmath43 . our measurements of t@xmath43 and @xmath46 therefore did not depend on accurate assessments of the other model atmosphere parameters , namely log @xmath48 and [ m / h ] . our measurement of log @xmath48 and [ m / h ] , on the other hand , depended strongly on the other model atmosphere parameters . because we set the gravity by ionization equilibrium , log @xmath48 depends on t@xmath43 , @xmath46 , and [ m / h ] , as well as the abundance measured from the and lines . therefore , the uncertainty in log @xmath48 depends on the uncertainties in those quantities . the uncertainty in [ m / h ] , in turn , depends on the uncertainty in t@xmath43 , log @xmath48 , and @xmath46 , as well as the scatter in the abundance given by different lines for a particular model atmosphere . the standard error of the mean for the gravity derived from the 35 lines for a single model atmosphere was 0.04 dex , and for the 4 lines was 0.14 . we adopt 0.04 dex as the uncertainty from ew and log @xmath42 errors for for inclusion in the [ m / h ] uncertainty . adding the and uncertainties in quadrature give us 0.15 dex as our uncertainty in the difference between the and the log@xmath54(a ) @xmath3 log@xmath4(n@xmath5/n@xmath55 ) + 12.0 ] values arising from the ew and oscillator strength uncertainties . the total uncertainty in log @xmath48 is 0.66 dex and in [ m / h ] is 0.21 dex . we ran the and ews through a series of models : @xmath47 200k , @xmath47 0.3 km / s , @xmath470.3 dex for log @xmath48 , and 0.13 dex in [ m / h ] ( smaller because of the limits of the kurucz grid ) , and calculated the difference in fe abundance with these different model atmospheres . finally , we calculate uncertainties using modified equations ( a5 ) and ( a20 ) from @xcite . we considered the covariance between t@xmath43 : log @xmath48 , t@xmath43 : [ m / h ] , log @xmath48 : [ m / h ] , @xmath46 : log @xmath48 , and @xmath46 : [ m / h ] . the covariances were calculated by a monte carlo technique . for example , to calculate the covariance between t@xmath43 and log @xmath48 , we first found @xmath56 and noted the remaining scatter that was caused by uncertainty in @xmath46 , etc . next , we randomly picked 1000 t@xmath43 from a gaussian distribution with a @xmath57 of 200k . we used the derivative to calculate log @xmath48 and then added an extra random @xmath58log @xmath48drawn from a gaussian distribution with a @xmath57 equal to the uncertainty from non - t@xmath43 causes . we calculated the covariance using these 1000 t@xmath43-log @xmath48 pairs . a similar calculation was done for the other covariances . in table 2 , we summarize the abundances measured for 17 elements in . we include both log@xmath54 and its error , as well as [ x / fe ] and its error . to give an idea of the uncertainty due to scatter from the lines , rather than from atmospheric parameters , we give @xmath57 , the rms of abundances derived from individual lines as well as the number of lines . we also give our measurements of the solar abundances , which we will use to calculate ratios . for reference , we include the @xcite solar abundances in the final column . we measure [ fe / h]@xmath59 for . in @xcite , we measured [ fe / h]@xmath60 for the dwarf . the stars that are microlensed are unbiased in metallicity . the criterion for spectroscopic follow - up is that the unmagnified source be faint enough to be a bulge dwarf , regardless of color . therefore , especially considering the large and variable reddening toward the bulge , we are not biased in our high - resolution follow - up toward high metallicity sources . the high metallicities of these two dwarfs is surprising given that work on giants has indicated an average metallicity near solar . in figure [ fig : mdf1 ] , we compare the metallicity distribution function ( mdf ) of the two dwarfs with several mdfs based on studies of giant stars . the mdf of @xcite , which is based on high - resolution analysis of m giants and should be biased _ toward _ the highest metallicity objects , lacks any giants as metal - rich as these dwarfs . this is particularly notable , since the work of @xcite on k giants with low - dispersion spectra shows some extremely metal - rich stars . a complicating factor is the possible presence of a metallicity gradient in the bulge . in figure [ fig : mdf2 ] , we compare the metallicities of the dwarfs with mdfs derived by @xcite from high - resolution spectra for giants in three bulge fields : 4 , 6 , and 12away from the galactic center . the inner field is more metal - rich than the outer . however , these two dwarfs are located 6.5 ( ) and 4.9 ( ) away from the galactic center , and therefore gradients can not explain their anomalously high metallicities . these results hint that the mdfs of the bulge giants and dwarfs may be different . whether this is true can be established by more observations of bulge dwarfs and by resolution of the discrepancies among the mdfs derived for giants , particularily between the low - dispersion and high - dispersion studies . because the microlensed dwarfs are found at a range of distances from the galactic center , comparison of the giant and dwarf mdfs also depends on measuring the metallicity gradient ( and the size of deviations from that gradient ) in the bulge . ideally , the mdf for giants in the same field as the microlensed dwarf would be measured . with our measurements of t@xmath43 , log @xmath48 , and [ fe / h ] from spectroscopy , we can compare the position of on the hertzsprung - russell diagram with theoretical isochrones ( fig . [ fig : iso ] ) . we use the yonsei - yale isochrones @xcite with [ fe / h]=0.385 , [ @xmath1/fe]=0 for comparison . the best fit age for this star is @xmath205 gyr . instead of log g , we can also use the i - band magnitude to plot the star on the h - r diagram . we adopt a distance of 8.5 kpc , placing this dwarf on the far side of the bulge , its most likely position because the optical depth for lensing is larger there . figure [ fig : imag ] shows that a similar age is obtained . indeed , shifting the position of in the vertical direction , by changing its distance or luminosity has little effect on the young age that we derive for this star , because , at this metallicity , there should be no stars this hot with ages @xmath61 6 gyr . however , the uncertainty in the temperature combined with the effect that changing the temperature has on the derived metallicity produce large uncertainties in the age . if we instead adopt the t@xmath43 on the lower edge of our range ( 5600k ) , the metallicity calculated from the lines drops to [ fe / h]=0.16 dex . using isochrones of this metallicity , the new t@xmath43 gives an age of @xmath209 gyrs . improvements in the accuracy of temperatures are needed to get better age constraints for bulge dwarfs , but in principal , ages can be measured for individual stars near the main - sequence turnoff . -t@xmath43 ) . m@xmath62 was calculated using the i@xmath12 magnitude and assuming a distance of 8.5 kpc . the t@xmath43 is the temperature from the balmer lines . we also show isochrones from @xcite the solid lines show isochrones for [ fe / h]=0.385 and for ages 2 , 3 , 4 , 5 and 7 gyr . the dashed line shows a 5 gyr isochrone for a solar metallicity.,width=604 ] as stars move up the giant branch , they pass through first dredge - up , which brings up material that has been processed in the cn cycle . the c and n abundances measured in giants no longer represent the original c and n endowments of the stars , although c+n will remain constant as long as only material processed in the cn cycle , and not the on cycle , is mixed to the surface . @xcite and @xcite measured c and n in giants in the bulge . they found that the giants lie to the n - rich side of the line defined by the c / n ratio of the sun ( figure [ fig : mix ] ) . @xcite concluded that a small amount of mixing had occurred in the giants . this conclusion is only valid if the original abundances in the giants lie close to the line . otherwise , if the bulge dwarfs have non - solar c / n ratios , the c / n ratios measured in bulge giants could imply either no mixing ( and a n - rich original composition ) or substantial mixing ( and a c - rich original composition ) . the abundances of c and n for are also plotted on figure [ fig : cn ] and shows that the assumption of @xcite is justified and that the expected amount of mixing has occurred in the giants . li has been created since the big bang by stellar nucleosynthesis and by cosmic ray spallation . a li abundance for a star in the bulge , measuring how fast li was made in the early galaxy , would be very interesting . however , most stars no longer have the same amount of li on their surfaces as was present in their natal gas clouds . li is easily burned during pre - main sequence and main - sequence phases of stars and is either destroyed throughout the convective envelope during the rgb phase or ( for a brief time ) created in the star itself and dredged up . we have no detection of li in this star , only a 3-@xmath57 upper limit of log@xmath54(li)=1.84 dex based on a @xmath63 fit to the data ( see @xcite for more details ) . in figure [ fig : li ] we show this upper limit compared with li measurements in open cluster stars having a range of ages as well as field stars from @xcite . we also include the upper limit from . lower values can be expected in field stars because of astration on the main - sequence and because they are often older than the clusters featured in figure [ fig : li ] and were formed out of gas that had not been polluted by as much li . figure [ fig : li ] shows that the li upper limits in the bulge dwarfs are consistent with the upper limits in field dwarfs . a dwarf with t@xmath43@xmath64k is probably needed to measure the amount of li produced by spallation in the bulge . vs. log@xmath54(li ) for the hyades and m67 from @xcite and for field stars from @xcite compared with the upper limit for and . the bulge dwarf li limits are consistent with disk stars , which is not surprising given the age and temperature of these stars.,width=604 ] the bulge has a different star formation history than the halo / disk . the ratios of type ii / type ia pollution or type ii / agb star pollution at a given [ fe / h ] are therefore different as well , and the abundance ratios reflect this . we compare the abundances for both and with bulge giants and field stars from the thick / thin disk and halo from literature sources . in table 3 , we summarize the literature sources we use for each element . for many elements , observing red giant stars in the bulge is an effective method of measuring their abundances . however , as shown in 5.2.1 , internal mixing on the rgb alters the abundances of c and n. the abundances of c and n that we measure in therefore represent the first observations of the primordial c and n produced by the chemical evolution of the bulge . has [ c / fe]=0.04@xmath65 and [ n / fe][email protected]@xmath66 ( fig . [ fig : cn ] ) . the solar values of [ c / fe ] and n / fe ] show that c and n production kept pace with the fe production in the bulge . there are many sources of c in the universe ( * ? ? ? type ii sne and agb stars certainly contribute substantial amounts of c and n ; the roles of novae and wolf - rayet stars are less clear . these contributions , whatever they are , track the production of fe in the chemical evolution of the bulge . finally , we attempted to measure the @xmath67c/@xmath68c ratio , which is sensitive to the source of c , being low for low - mass agb stars and high for type ii sne . we could only set an uninteresting limit ( @xmath67c/@xmath68c > 1 ) on this very interesting number . and [ n / fe ] for ( filled black square ) compared to bulge giants ( filled blue circles ) and disk dwarfs ( open red triangles ) . the low [ c / fe ] values for the bulge giants are the result of internal mixing . the [ c / fe ] and [ n / fe ] values in , on the other hand , are the result of the pollution of the gas of the bulge by previous generations of stars . the solar ratios for these elements impose constraints on the inefficiency of c and n production in the bulge.,width=604 ] the abundances of na and al are elevated in bulge giants @xcite compared to disk stars . metal - rich type ii sne are predicted to produce more of the odd - z elements such as na and al than metal - poor type ii sne and could potentially be the explanation of the difference between the bulge and the disk . in figure [ fig : naal ] , we show the [ na / fe ] and [ al / fe ] value for and . these two unmixed stars are on the lower end of the scatter seen in the giants . this could be a hint that the larger [ na / fe ] and [ al / fe ] values seen in giants are due to internal mixing , but because the dwarf values fall within the scatter outlined in the scatter indicates that more measurements in dwarf stars are needed before any differences in the distribution of [ na / fe ] and [ al / fe ] in dwarfs and giants can be seen . and [ al / fe ] for and ( filled black squares ) compared to bulge giants ( filled blue circles ) and field stars ( open red triangles ) . the dwarfs fall within the distribution of [ na / fe ] and [ al / fe ] values seen in the giants , but at the lower edge of that distribution . , width=604 ] figure [ fig : alpha ] shows the [ o / fe ] , [ mg / fe ] , [ si / fe ] , and [ ca / fe ] for compared with halo stars , thin and thick disk stars , and bulge giants . the metallicity of is in the range of metallicities measured for bulge giants , allowing direct comparison of [ @xmath1/fe ] ratios between dwarfs and giants . the agreement is good , as both the giants and the dwarf have [ @xmath1/fe ] below the high [ @xmath1/fe ] values of the more metal - poor ( [ fe / h]@xmath690 ) stars . this decline in all [ @xmath1/fe ] ratios suggests that fe from type ia sne is being added and that the more metal - rich stars formed sufficiently later to have this ejectum in their gas . , [ mg / fe ] , [ si / fe ] and [ ca / fe ] for and ( filled black squares ) compared to bulge giants ( filled blue circles ) and field stars ( open red triangles ) . the [ @xmath1/fe ] ratios in agree well with the values measured in giants of similar metallicity . , width=453 ] k is an odd - z element and is predicted in nucleosynthesis models to be underproduced relative to the @xmath1 elements in metal - poor sne . there are few measurements in the literature , and those that exist show the opposite trend of increasing [ k / fe ] with decreasing [ fe / h ] @xcite . however , the only k line available for study in most stars , the resonance line at 7698 , is affected by non - lte effects , and these corrections have not yet been applied to large samples . @xcite derived nlte corrections for each star in their sample . the [ k / fe ] values were still supersolar at low metallicities , with the thin disk stars showing a drop in [ k / fe ] for[fe / h]@xmath70 . the [ k / fe ] ratios in the thick disk stars remain high . however , the [ k / mg ] ratios showed much smaller variations among the different galactic populations . they argued that the constant [ k / mg ] ratio ( [ k / mg]=@xmath71 ) in the stars indicated that the nucleosynthesis of k is closely coupled to that of the @xmath1-elements , which is somewhat surprising given the theoretical predictions . we measured the 7698 line in and the sun . we also measured the k abundance in using turbospectrum and the model atmosphere described in @xcite . the [ k / fe ] we measure for is @xmath72 figure [ fig : k ] compares the results for the bulge to the @xcite results . we applied no nlte correction , but assumed that the lte abundances in the sun and would be affected by the same amount . the [ k / mg ] values are 0.07@xmath73 for and 0.12@xmath74 for ; the [ k / mg ] abundance is still within a narrow range , even in this very different chemical evolution history . for and ( filled black squares ) compared to field stars ( open red triangles ) . the [ k / fe ] values in the dwarfs fall in the range seen in the disk stars.,width=604 ] the abundances for ti , sc , mn and ni are shown in figure [ fig : iron ] . @xcite found that [ ti / fe ] behaves like [ o / fe ] in the bulge , with supersolar [ ti / fe ] ratios for many stars , followed by [ ti / fe ] decreasing to solar for [ fe / h @xmath75 . the abundance of ti is also enhanced in halo stars , leading it to be classified as an `` @xmath1-element '' for observational purposes . in , [ ti / fe ] is close to solar , in line with the @xmath1-elements discussed above . the ratios of [ sc / fe ] and [ ni / fe ] are observed to be close to solar for a wide range of populations : halo , thick and thin disk and bulge . the data for show that abundances in this bulge dwarf agree with this picture . for [ fe / h ] @xmath76 in the galactic halo / disk , there is a plateau at [ mn / fe]@xmath77 @xcite . because type ia sne have not polluted the most metal - poor stars in the galaxy , we can derive the ratio of mn / fe produced in ( metal - poor ) type ii sne from this plateau . [ mn / o ] starts to rise before [ o / fe ] starts to drop in the disk . because the drop in [ o / fe ] signals the onset of substantial type ia sn contribution , the rise in mn relative to o can not be due to type ia sne , but rather to increased production of mn by more metal - rich type ii sne @xcite . @xcite also measured mn in 13 stars in the sagittarius dwarf galaxy . these stars have substantial type ia sne contributions to their gas , but [ mn / fe ] values about 0.2 dex below the trend seen in the galactic disk , providing additional evidence that metal - rich type ii sne are responsible for mn production . the near solar [ mn / fe ] values for and ogle-99 are also in support of this idea , because they were seen in an environment that had many type ii sne occur , unlike the sagittarius stars . , [ ti / fe ] , [ mn / fe ] , and [ ni / fe ] for and ( filled black squares ) compared to bulge giants ( filled blue circles ) and field stars ( open red triangles).,width=453 ] the [ ti / fe ] value in agrees well with [ ti / fe ] ratios measured in bulge giants of similar metallicity . within the error bars , the other [ iron - peak / fe ] ratios follow the trends seen the disk stars . [ fig : iron ] and [ zn / fe ] for and ( filled black squares ) compared to field stars ( open red triangles ) . the @xcite models predict [ zn / fe]@xmath78 for high - metallicity stars in the bulge . therefore , the low [ zn / fe ] disagrees with the theory that zn is produced in large amounts of type ia sne . the solar and supersolar [ cu / fe ] values in the bulge are consistent with either type ia sn production or metal - rich type ii sn production.,width=604 ] the percent of cu and zn production to be ascribed to different nucleosynthesis sites ( e.g. , type ii sne , agb stars , type ia sne ) is uncertain . the observations show that [ cu / fe ] @xmath79 at the lowest metallicities and then rises to solar by [ fe / h]@xmath80 ( e.g. * ? ? ? [ zn / fe ] , on the other hand , has supersolar values at the lowest metallicities and then decreases to closer to solar ( e.g. * ? ? ? @xcite used new weak s - process calculations and available sn models to argue that approximately two - thirds of the zn and cu production in the universe is due to type ia sne . the rest of the zn is from a primary process in massive stars , while the cu comes from a secondary ( = metallicity - dependent ) process in massive stars . using this model and a chemical evolution model for the bulge , @xcite predicted that both [ cu / fe ] and [ zn / fe ] would be @xmath20 0.2 at [ fe / h]=0.3 . the sn models available to @xcite did not include important effects , such as detailed calculation of neutron - capture elements beyond fe . @xcite used updated results and considered zn and cu production by neutron - capture in the o - rich parts of type ii sne ( `` weak sr - process '' ) . in their analysis , cu is mostly produced in this weak sr - process , a secondary process . a small amount of primary cu is made as radioactive zn in the inner regions of type ii sne . zn production is also due to massive star nucleosynthesis , but here there is a large primary production in the @xmath1-rich freeze out in type ii sne , which is supplemented at higher metallicities by a secondary contribution from the weak - sr process . @xcite find no need for contributions to cu and zn from type ia sne or agb stars . the measurements of [ cu / fe ] and [ zn / fe ] in support the @xcite model and argue against the production of large amounts of zn in type ia sne . because the abundance of cu is dominated by a secondary process , the solar [ cu / fe ] at [ fe / h]=0.36 and the supersolar value at [ fe / h]=0.56 are the result of copious cu production in metal - rich type ii sne . however , the primary production of zn and the smaller secondary contribution is not sufficient to keep up with the fe from both type ii sne and type ia sne . the observations also show again that the chemical evolution of zn is separate from fe . for and ( filled black squares ) compared to field stars ( open red triangles ) . [ ba / fe ] shows the largest deviation from the trends seen in the disk stars of any element studied in this paper . this low ba value is consistent with the idea that the r - process is the dominate producer of heavy elements in the bulge . it also puts constraints on s - process contributions to ba ( and the accompanying c and n contributions ) from agb stars . , width=604 ] the [ ba / fe ] for falls considerably below the solar value ( [ ba / fe]=@xmath81 ( fig . [ fig : ba ] ) . it falls in the range not seen in other parts of galaxy , except for the metal - poor halo . we note that we could only measure 1 line of ba in , and the solar value we measure is the most discrepant from the solar value in @xcite . however , even if the grevesse & sauval value is used , the [ ba / fe ] for is still subsolar ( [ ba / fe][email protected] ) . if we use the solar value of ba ( log(@xmath54(ba)=2.39 ) reported in @xcite , has a [ ba / fe]@xmath82 . the paucity of ba in the halo stars is explained by the fact that the r - process is the only available channel for producing the heavy elements in the early univere , and the r - process is not an efficient producer of ba . it is tempting to ascribe the low ba in to the same cause , especially in light of the high [ eu / fe ] measurements in bulge giants by @xcite . stellar populations dominated by type ii sne and r - process production should have high [ eu / fe ] and [ eu / ba ] before type ia sne and agb stars eventually add fe and ba to the ism . whether the ba deficiency in can be explained by a lack of contributions from agb stars depends on the relative timescales of type ia sn pollution and agb pollution and whether the low [ @xmath1/fe ] values in are the result of type ia pollution . if we assume that type ia sne have contributed significantly to the abundances of , which is reasonable , then agb pollution must trail type ia sn production . the evidence on this point is mixed . @xcite saw , in addition to a wide range at any given metallicity , a rise in the [ la / fe ] ratios at [ fe / h ] @xmath83 . because la , like ba , is mostly due to the s - process , this would indicates the s - process from agb stars is added before fe from type ia sne causes the [ @xmath1/fe ] ratios to turn over . on the other hand , @xcite argued based on the isotope ratios of mg that 3 - 6 m@xmath84 stars did not start contributing to the halo until [ fe / h]@xmath85 , at the same time or later than the type ia sne . the [ ba / fe ] measured in is closer to the solar value , and suggests that by [ fe / h]=0.5 the bulge had reached the same point in chemical evolution as the solar neighborhood , with substantial amounts of ba supplied by the s - process in agb stars balancing the iron supplied by type ia sne . ba abundances for bulge stars with a wide range of metallicity would help clarify the origin of the ba abundance in the bulge . the results for demonstrate the unique information that can be obtained from high - resolution spectra of microlensed dwarfs in the bulge . because microlensing is not biased in metallicity , we can measure the mdf of the bulge main - sequence stars when sufficient number of highly magnified bulge dwarfs have been observed with high - dispersion spectrographs . however , the two dwarfs we have studied so far both have [ fe / h]@xmath86 , which makes them more metal - rich than any of the m giants observed at high - resolution . if the temperatures and metallicities measured for the two dwarfs are accurate , then these stars are younger than the bulk of the bulge population . for many elements , the abundance ratios we measure in are not surprising . the iron - peak and @xmath1-elements ( o , mg , si , ca , and ti ) fall on the trend observed in the bulge giants . the solar values for the [ @xmath1/fe ] ratios suggest that type ia sn ejecta are responsible for some of the fe , which would be expected if these metal - rich stars were part of a younger population in the bulge . the [ c / fe ] and [ n / fe ] ratios are also @xmath20 0 , suggesting that agb stars have also begun to contribute these light elements to offset the contribution of fe from type ia sne . it is therefore unexpected to find [ ba / fe]@xmath20-@xmath87 . such low [ ba / fe ] values , when found in the halo , imply that only type ii sne and the r - process have polluted the gas , and not the s - process from agb stars . the c and n abundances we measure in are unaffected by any mixing or dilution and therefore are the result of the chemical evolution of the bulge . comparing the total amount of c+n to that measured in bulge giants , we conclude that only mild mixing , as expected in theoretical models , has occurred in the bulge . this agrees with the assertions by @xcite , @xcite , and @xcite that the o , na and al abundances in bulge giants should not have been affected by mixing . the [ na / fe ] and [ al / fe ] ratios for fall on the lower end of the values seen in bulge giants of this metallicity . the [ na / fe ] and [ al / fe ] ratios in giants are marked above all else by large scatter and more abundance ratios in dwarfs are needed before a comparison of scatter can be made . we can use our measurements of the abundances of k , cu and zn , rare for bulge stars , to exploit the different star formation history of the bulge to study the nucleosynthesis sites of these elements . the supersolar [ cu / fe ] and subsolar [ zn / fe ] values agree with the model of @xcite and the production of these two elements predominantly in type ii sne . the chemical evolution of k is more difficult to understand , because the [ k / mg ] value in the bulge is similar to that in the metal - poor halo , although the production of k should be enhanced in metal - rich sne . finally , we note that the standard microlensing technique of estimating the intrinsic color and magnitude of the source star by using the offset from the position of the red clump can be tested by deriving a temperature and gravity from the spectrum . for , the photometric and spectroscopic temperature agree very well , although the agreement was considerably less good for . in summary , the abundances in provide unique information about the formation and evolution of the bulge . larger samples of dwarfs observed in this way would allow the derivation of the dwarf mdf and the trends in abundance ratios over a wide range of metallicity . by taking advantage of microlensing events in the bulge , we can achieve this goal with a modest amount of telescope time . our thanks to ian thompson for help with the echelle observations at las campanas . thomas masseron and bertrand plez provided invaluable support in installing and running turbospectrum . we would also like to thank daniel kelson for codes and support for reducing the mike data . jon fulbright , manuela zoccali , and solange ramiez kindly provided metallicity data for the bulge giant metallicity distribution functions . we acknowledge support from : nsf ast-042758 and nasa nng04gl51 g ( ag ) . work by b.s.g . was partially supported by a menzel fellowship from the harvard college observatory . lrrrrrr [ tab : ew ] & 7771.941 & 9.15 & 0.370 & 101.5 & 70.2 & 1 + & 7774.161 & 9.15 & 0.220 & 86.7 & 60.1 & 1 + & 7775.390 & 9.15 & 0.000 & 71.1 & 47.4 & 1 + & 5682.633 & 2.10 & @xmath140.710 & 148.1 & 96.9 & 2 + & 5688.194 & 2.10 & @xmath140.400 & 150.6 & 118.7 & 2 + & 6154.226 & 2.10 & @xmath141.570 & 60.9 & 35.9 & 2 + & 6160.747 & 2.10 & @xmath141.270 & 82.3 & 55.8 & 2 + & 5711.088 & 4.33 & @xmath141.870 & 140.6 & 112.7 & 2 + & 7387.689 & 5.75 & @xmath141.270 & 141.2 & 70.6 & 2 + & 7836.134 & 4.02 & @xmath140.650 & 93.1 & 54.2 & 2 + & 5665.555 & 4.93 & @xmath141.940 & 89.2 & 40.2 & 2 + & 5690.425 & 4.93 & @xmath141.770 & 57.4 & 49.7 & 2 + & 5772.146 & 5.08 & @xmath141.650 & 69.9 & 51.9 & 2 + & 6125.021 & 5.61 & @xmath141.520 & 57.7 & 30.4 & 2 + & 6142.483 & 5.62 & @xmath141.500 & 61.8 & 32.3 & 2 + & 6155.134 & 5.62 & @xmath140.720 & 124.5 & 84.5 & 2 + & 6237.319 & 5.61 & @xmath141.050 & 90.9 & 63.9 & 2 + & 6244.466 & 5.62 & @xmath141.320 & 77.6 & 47.3 & 2 + & 7415.948 & 5.61 & @xmath140.850 & 115.0 & 85.5 & 2 + & 7918.384 & 5.95 & @xmath140.510 & 157.5 & 84.5 & 2 + & 7932.348 & 5.96 & @xmath140.370 & 131.8 & 92.3 & 2 + & 7944.001 & 5.98 & @xmath140.210 & 140.8 & 107.0 & 2 + & 7698.974 & 0.00 & @xmath140.170 & syn & syn & 1 + & 5581.965 & 2.52 & @xmath140.530 & 111.8 & 94.7 & 3 + & 5590.114 & 2.52 & @xmath140.571 & 148.4 & 91.0 & 3 + & 6161.297 & 2.52 & @xmath141.266 & 77.3 & 59.2 & 3 + & 6166.439 & 2.52 & @xmath141.142 & 83.2 & 70.2 & 3 + & 6169.042 & 2.52 & @xmath140.797 & 110.2 & 88.7 & 3 + & 6169.563 & 2.52 & @xmath140.478 & 147.8 & 105.2&3 + & 6455.598 & 2.52 & @xmath141.290 & 59.6 & 56.9 & 3 + & 6471.662 & 2.52 & @xmath140.686 & 116.1 & 90.2 & 3 + & 6499.650 & 2.52 & @xmath140.818 & 97.5 & 84.1 & 3 + & 5526.820 & 1.77 & 0.02 & syn & syn & 5 + & 5657.880 & 1.51 & @xmath140.600 & syn & syn & 5 + & 6245.620 & 1.51 & @xmath141.070 & syn & syn & 5 + & 5866.451 & 1.07 & @xmath140.780 & 60.6 & 48.1 & 2 + & 6126.216 & 1.07 & @xmath141.370 & 59.7 & 23.4 & 2 + & 6258.102 & 1.44 & @xmath140.300 & 72.9 & 49.8 & 2 + & 6261.098 & 1.43 & @xmath140.420 & 61.1 & 50.1 & 2 + & 5381.015 & 1.57 & @xmath141.920 & 80.4 & 59.3 & 4 + & 5470.270 & 2.14 & @xmath141.460 & syn & syn & 5 + & 6013.520 & 3.07 & @xmath140.250 & syn & syn & 5 + & 6021.710 & 3.07 & 0.03 & syn & syn & 5 + & 5307.361 & 1.61 & @xmath142.987 & 105.7 & 84.5 & 1 + & 5321.108 & 4.43 & @xmath140.951 & 65.1 & 40.8 & 1 + & 5322.041 & 2.28 & @xmath142.803 & 105.4 & 59.8 & 1 + & 5329.989 & 4.08 & @xmath141.189 & 128.7 & 55.7 & 1 + & 5379.574 & 3.70 & @xmath141.510 & 74.2 & 60.5&1 + & 5501.465 & 0.96 & @xmath143.047 & 127.1 & 115.4&1 + & 5506.779 & 0.99 & @xmath142.797 & 149.4 & 115.1&1 + & 5701.545 & 2.56 & @xmath142.120 & 114.7 & 81.9 & 1 + & 5705.465 & 4.30 & @xmath141.500 & 54.3 & 38.7 & 1 + & 6027.051 & 4.08 & @xmath141.090 & 78.4 & 65.1 & 1 + & 6082.711 & 2.22 & @xmath143.573 & 63.5 & 34.0 & 1 + & 6165.360 & 4.14 & @xmath141.470 & 72.3 & 43.7 & 1 + & 6173.336 & 2.22 & @xmath142.880 & 119.9 & 68.7 & 1 + & 6219.281 & 2.20 & @xmath142.420 & 114.4 & 86.6 & 1 + & 6252.555 & 2.40 & @xmath141.690 & 148.8 & 115.3&1 + & 6265.134 & 2.18 & @xmath142.550 & 107.8 & 87.2 & 1 + & 6297.793 & 2.22 & @xmath142.740 & 94.6 & 72.8 & 1 + & 6301.501 & 3.65 & @xmath140.718 & 141.5 & 121.6&1 + & 6311.500 & 2.83 & @xmath143.141 & 49.0 & 27.1 & 1 + & 6322.685 & 2.59 & @xmath142.426 & 100.1 & 79.0 & 1 + & 6335.331 & 2.20 & @xmath142.180 & 127.5 & 95.8 & 1 + & 6344.149 & 2.43 & @xmath142.920 & 91.7 & 58.4 & 1 + & 6380.743 & 4.19 & @xmath141.380 & 82.6 & 51.6 & 1 + & 6408.018 & 3.69 & @xmath141.018 & 131.7 & 94.7 & 1 + & 6411.649 & 3.64 & @xmath140.720 & 139.5 & 128.7&1 + & 6498.939 & 0.96 & @xmath144.700 & 49.8 & 45.0 & 1 + & 6592.914 & 2.73 & @xmath141.470 & 129.6 & 107.3&1 + & 6593.870 & 2.43 & @xmath142.420 & 117.3 & 81.7 & 1 + & 6804.001 & 4.65 & @xmath141.496 & 67.2 & 21.1 & 1 + & 6810.263 & 4.61 & @xmath140.986 & 71.4 & 48.9 & 1 + & 6855.162 & 4.56 & @xmath140.740 & 102.2 & 69.0 & 1 + & 6858.150 & 4.61 & @xmath140.930 & 79.5 & 50.8 & 1 + & 7401.685 & 4.19 & @xmath141.350 & 58.7 & 39.7 & 1 + & 7710.364 & 4.22 & @xmath141.110 & 103.9 & 63.5 & 1 + & 7941.089 & 3.27 & @xmath142.580 & 59.8 & 42.0 & 1 + & 6247.557 & 3.89 & @xmath142.430 & 86.3 & 51.8 & 2 + & 6416.919 & 3.89 & @xmath142.880 & 65.0 & 41.0 & 2 + & 6432.680 & 2.89 & @xmath143.690 & 61.3 & 40.0 & 2 + & 6456.383 & 3.90 & @xmath142.190 & 71.7 & 61.2 & 2 + & 5754.655 & 1.93 & @xmath141.850 & 133.5 & 75.4 & 2 + & 5805.213 & 4.17 & @xmath140.620 & 60.5 & 42.1 & 2 + & 6108.107 & 1.68 & @xmath142.430 & 68.5 & 63.3 & 2 + & 6111.066 & 4.09 & @xmath140.820 & 70.0 & 32.7 & 2 + & 6128.963 & 1.68 & @xmath143.360 & 46.8 & 24.8 & 2 + & 6175.360 & 4.09 & @xmath140.500 & 63.5 & 46.9 & 2 + & 6176.807 & 4.09 & @xmath140.260 & 95.1 & 63.4 & 2 + & 6314.653 & 1.94 & @xmath142.000 & 92.1 & 72.4 & 2 + & 6378.247 & 4.15 & @xmath140.810 & 59.5 & 29.9 & 2 + & 6482.796 & 1.93 & @xmath142.760 & 54.8 & 40.7 & 2 + & 6598.593 & 4.24 & @xmath140.910 & 42.0 & 24.8 & 2 + & 6643.629 & 1.68 & @xmath141.910 & 122.2 & 95.0 & 2 + & 6767.768 & 1.83 & @xmath142.100 & 106.2 & 80.5 & 2 + & 6772.313 & 3.66 & @xmath140.940 & 83.8 & 49.8 & 2 + & 7110.892 & 1.94 & @xmath142.880 & 57.8 & 36.0 & 2 + & 7715.583 & 3.70 & @xmath141.010 & 105.7 & 49.7 & 2 + & 7748.891 & 3.71 & @xmath140.330 & 164.4 & 84.0&2 + & 7788.936 & 1.95 & @xmath141.750 & 123.4 & 92.2 & 2 + & 7797.586 & 3.90 & @xmath140.320 & 103.6 & 79.2 & 2 + & 5782.050 & 1.64 & @xmath142.920 & syn & syn & 5 + & 4722.153 & 4.03 & @xmath140.338 & syn & syn & 1 + & 4810.528 & 4.08 & @xmath140.137 & syn & syn & 1 + & 6496.410 & 0.60 & @xmath140.380 & syn & syn & 5 + lrrrrrrrr [ tab : abund ] c ( ch ) & 8.8 & 0.26 & 0.04 & 0.22 & 0.20 & & 8.4 & 8.52 + n ( cn ) & 8.2 & 0.56 & @xmath140.06 & 0.43 & 0.20 & & 7.9 & 7.92 + & 9.09 & 0.11 & @xmath140.16 & 0.22 & 0.33 & 3 & 8.89 & 8.83 + & 6.69 & 0.13 & 0.09 & 0.16 & 0.12 & 4 & 6.24 & 6.33 + & 8.18 & 0.21 & 0.16 & 0.25 & 0.17 & 2 & 7.66 & 7.58 + & 6.98 & 0.16 & 0.11 & 0.22 & 0.10 & 1 & 6.51 & 6.47 + & 7.98 & 0.10 & 0.06 & 0.13 & 0.10 & 12 & 7.56 & 7.55 + & 5.60 & 0.28 & 0.14 & 0.29 & 0.20 & 1 & 5.10 & 5.12 + & 6.58 & 0.29 & @xmath140.09 & 0.16 & 0.23 & 9 & 6.31 & 6.36 + & 3.25 & 0.31 & @xmath140.30 & 0.23 & 0.13 & 3 & 3.19 & 3.17 + & 5.25 & 0.23 & @xmath140.05 & 0.17 & 0.27 & 4 & 4.94 & 5.02 + & 5.18 & 0.36 & @xmath140.22 & 0.28 & 0.20 & 1 & 5.04 & 5.02 + & 5.62 & 0.13 & @xmath140.09 & 0.12 & 0.08 & 3 & 5.34 & 5.39 + & 7.81 & 0.18 & 0.36 & & 0.25 & 35 & 7.45 & 7.50 + & 7.81 & 0.28 & @xmath140.18 & 0.19 & 0.25 & 4 & 7.63 & 7.50 + & 6.61 & 0.10 & 0.02 & 0.09 & 0.26 & 19 & 6.23 & 6.25 + & 4.40 & 0.23 & @xmath140.01 & 0.20 & 0.10 & 1 & 4.05 & 4.21 + & 4.63 & 0.23 & @xmath140.28 & 0.21 & 0.10 & 2 & 4.55 & 4.60 + & 2.25 & 0.43 & @xmath140.61 & 0.32 & 0.20 & 1 & 2.50 & 2.13 + lcccccccccccccccc [ tab : lit ] + @xcite & & & x & x & x & x & x & & x & & & & & & & + @xcite & & & & & & & & & & & & & & & & + @xcite & x & & x & & x & & x & & x & & x & & & & & + @xcite & & & & & & & & & & & & & & & & + @xcite & x & x & x & x & & & & & & & x & & & & & + + @xcite & x & & x & x & x & x & x & & x & x & x & x & x & x & x & x + @xcite & x & x & x & x & x & x & x & & x & x & x & x & x & x & x & x + @xcite & & & & & & & & & & & & x & & & & + @xcite & x & & & & & & & & & & & & & & & + @xcite & & & & & x & & x & & & & x & & x & & x & x + @xcite & & & x & & & & & & & & & & & & & + @xcite & & & & x & x & & & & & & & & & & & + @xcite & & & & x & x & x & x & & x & x & x & x & x & & & x + @xcite & x & x & & & & & & & & & & & & & & + @xcite & & & & & & & & x & & & & & & & & +

a principal scientific goal of the gemini planet imager ( gpi ) is obtaining milliarcsecond astrometry to constrain exoplanet orbits . however , astrometry of directly imaged exoplanets is subject to biases , systematic errors , and speckle noise . here we describe an analytical procedure to forward model the signal of an exoplanet that accounts for both the observing strategy ( angular and spectral differential imaging ) and the data reduction method ( karhunen - love image projection algorithm ) . we use this forward model to measure the position of an exoplanet in a bayesian framework employing gaussian processes and markov chain monte carlo ( mcmc ) to account for correlated noise . in the case of gpi data on pic b , this technique , which we call bayesian klip - fm astrometry ( bka ) , outperforms previous techniques and yields 1-errors at or below the one milliarcsecond level . we validate bka by fitting a keplerian orbit to twelve gpi observations along with previous astrometry from other instruments . the statistical properties of the residuals confirm that bka is accurate and correctly estimates astrometric errors . our constraints on the orbit of pic b firmly rule out the possibility of a transit of the planet at 10- significance . however , we confirm that the hill sphere of pic b will transit , giving us a rare chance to probe the circumplanetary environment of a young , evolving exoplanet . we provide an ephemeris for photometric monitoring of the hill sphere transit event , which will begin at the start of april in 2017 and finish at the end of january in 2018 .

1607.05272

astrometry is an essential tool for characterizing directly imaged exoplanets and their physical relationship to other elements of the planetary system in which they reside . the gemini planet imager ( gpi ; * ? ? ? * ) was designed with a goal of achieving @xmath21.8 mas astrometric accuracy @xcite , necessary for characterizing the eccentricity distribution of exoplanet orbits from the gpi exoplanet survey @xcite . to do so , the astrometric calibration of gpi has continually been benchmarked to well - calibrated astrometric fields @xcite . this had led to some of the most precise astrometry on directly - imaged exoplanet systems to date @xcite , allowing us to constrain or fit the first ever orbit of some of these directly - imaged exoplanets . however , limited by either the signal to noise ratio ( snr ) of these exoplanets or by biases in the various data analysis algorithms , so far no astrometric study with gpi has achieved the design goal of 1.8 mas precision . the importance of understanding planetary orbits is highlighted by the @xmath0 pictoris system , a young ( @xmath3 myr ; * ? ? ? * ; * ? ? ? * ) and nearby system that has been extensively studied . @xmath0 pic harbors a near edge - on debris disk that was first imaged by @xcite and which was subsequently observed to have a warp thought to be induced by a planet whose orbit is inclined relative to the debris disk @xcite . additional indirect signatures of a planet were derived from variable spectral features modeled as infalling comets @xcite and a peculiar light curve anomaly detected in 1981 @xcite . @xcite then discovered @xmath0 pic b , a planet at an appropriate mass ( @xmath4 @xmath5 ) and semi - major axis ( @xmath6 au ) to be responsible for the previously observed indirect signatures of planets . a key focus of subsequent observations was determining the alignment of @xmath0 pic b relative to the main outer disk and the warp to determine if @xmath0 pic b is causing the warp . by observing the disk and planet simultaneously , @xcite concluded that the planet is misaligned from the main outer disk and consistent with being responsible for the warp . additionally , @xcite ruled out the possibility of having another giant planet in the system massive enough to cause the warp instead . thus , @xmath0 pic b is responsible for the warp in the debris disk . this was confirmed in astrometric monitoring campaigns @xcite which used homogeneous datasets to limit systematics and constrain the orbit of @xmath0 pic b. refining the orbital elements of @xmath0 pic b is not only crucial for investigating the dynamical link between the planet and the disk warp , but also because @xmath0 pic b may transit its host star once every @xmath720 years . to date , there are no other known systems where the physical properties of an exoplanet can be characterized by using both the direct imaging and transit techniques . currently , the tightest constraint on the probability of transit is @xmath70.06% , obtained with a dedicated astrometric monitoring campaign with gpi @xcite . however , @xcite point out that gpi measurements from @xcite have a higher position angle than astrometry from previous measurements , which could arise from a possible systematic calibration offset between gpi and previous instruments instead of actually due to @xmath0 pic b s orbit being slightly inclined away from edge on . we note there currently is no evidence of a position angle offset with the gpi astrometric calibration , and the gpi astrometry of hd 95086 b are consistent with previous astrometry from other instruments @xcite . however , it is important to more accurately compute the orbit of @xmath0 pic b because in late 2017 to early 2018 @xcite , it will be at its closest projected separation from the star . the transit of the planet and/or any circumplanetary material orbiting around it could therefore be detectable . one of the obstacles in characterizing directly imaged exoplanets is that even with the newest instrumentation , the glare of the host star covers the signal of the planet . in order to subtract the point spread function ( psf ) of the star and maximize the snr of the planet , observing techniques such as angular differential imaging ( adi ; * ? ? ? * ) and spectral differential imaging ( sdi ; * ? ? ? * ) and data reduction algorithms like karhunen - love image projection ( klip ; * ? ? ? * ; * ? ? ? * ) are used in combination to disentangle the psf of the star from potential astrophysical sources . however , these techniques distort the planet signal and create data reduction artifacts , which usually are nuisance parameters that need to be calibrated out to obtain unbiased astrometry . forward modelling effects of observing techniques and data reduction algorithms on the psf of the planet was first done in the context of adi and loci , showing significant improvements in astrometry and photometry for simulated planets @xcite . in similar contexts with adi and loci , @xcite and @xcite used forward modelling to correct for flux loss of exoplanets and the flux and morphology of disks respectively . for classical adi ( cadi ) , @xcite showed that forward modelling techniques can improve the sensitivity of cadi and mitigate biases . the use of markov - chain monte carlo ( mcmc ) in conjunction with forward modelling was presented in @xcite for reference differential imaging . recently , @xcite introduced a method , called klip - fm , to analytically forward model the degradation of a faint astrophysical signal that occurs when using least squares - based psf subtraction algorithms such as klip that is also generally applicable to any observing strategy . additionally , the computation of the forward model with klip - fm is much quicker than negative fake planet injection methods @xcite , as the stellar psf subtraction algorithm , klip , needs only to be run once . in this paper , we demonstrate the advantages of klip - fm for precise astrometry and constraining planetary orbits by applying it to gpi observations of @xmath0 pic b reduced using klip and adi+sdi . in section [ sec : obs ] , we describe our new astrometry technique , bayesian klip - fm astrometry ( bka ) , in which we forward model the psf of the planet with klip - fm and then use the forward model in a bayesian framework to measure the position of the planet while also modelling the correlated nature of the noise . in section [ sec : valid ] , we validate our technique by fitting an orbit to the data and showing that our astrometry and uncertainties are consistent with keplerian motion with no obvious systematics . finally , in section [ sec : orbit ] , we apply our new astrometry to constrain the orbit of @xmath0 pic b and place the tightest constraints yet on the transit of the planet and its hill sphere . to obtain a large temporal baseline of gpi astrometric points , we compiled gpi data of @xmath0 pic from commissioning ( gemini program gs - eng - gpi - com ) , an astrometric monitoring campaign of @xmath0 pic b ( gemini programs gs-2015a - q-21 and gs-2015b - q-9 ; pi : graham ) , the gpi exoplanet survey ( gs-2014b - q-500 ; pi : macintosh ) , and a gemini large and long program to characterize debris disks ( gs-2015b - lp-6 : pi : chen ) . all the data used in the following analysis are listed in table [ table : obs ] . most of these data were published and analyzed in a previous study characterizing the orbit of @xmath0 pic b by @xcite . we have reprocessed those data with the new astrometry methods presented in this paper and combined them with five additional new epochs . there were three datasets in @xcite that we did not use . we did not use the polarimetry dataset as there was no instrumental psf obtained with the data to forward model . the dataset on 2014 march 23 was taken during tests of the adaptive optics ( ao ) system , causing the instrumental psf to be highly varying and making it unsuitable for forward modelling . the 2015 - 01 - 24 dataset contained artifacts in the construction of the forward model that we could not remove . we chose not to include the measured astrometry from @xcite for the datasets we omitted in order to maintain homogeneity in astrometric datasets and reduce potential systematic errors . we also note that in the last dataset taken on 2016 - 01 - 21 , there was saturation on the edge of the occulting mask due to bad seeing . this affected the forward modelling of @xmath0 pic b , which was also near the occulting mask so a large portion of the frames taken could not be used in this analysis . clcccc|cccc 2013 nov 16 & gs - eng - gpi - com & _ k1 _ & 1789 & 26 & 1.09 & 7 & 4 & no & 13 + 2013 nov 16 & gs - eng - gpi - com & _ k2 _ & 1253 & 18 & 0.93 & 7 & 4 & no & 13 + 2013 nov 18 & gs - eng - gpi - com & _ h _ & 2446 & 32 & 0.68 & 7 & 4 & no & 13 + 2013 dec 10 & gs - eng - gpi - com & _ h _ & 1312 & 39 & 0.77 & 7 & 4 & no & 13 + 2013 dec 10 & gs - eng - gpi - com & _ j _ & 1597 & 19 & 0.70 & 7 & 4 & no & 13 + 2013 dec 11 & gs - eng - gpi - com & _ h _ & 410 & 65 & 0.46 & 7 & 4 & no & 11 + 2014 nov 8 & gs - eng - gpi - com & _ h _ & 2147 & 25 & 0.77 & 7 & 4 & no & 11 + 2015 apr 2 & gs-2015a - q-21 & _ h _ & 1312 & 10 & 0.51 & 2 & 2 & no & 11 + 2015 nov 6 & gs-2014b - q-500 & _ h _ & 2207 & 28 & - & 7 & 4 & no & 11 + 2015 dec 5 & gs-2015b - lp-6 & _ j _ & 4948 & 66 & 0.92 & 7 & 4 & yes & 11 + 2015 dec 22 & gs-2015b - q-9 & _ h _ & 2088 & 19 & 0.76 & 7 & 4 & no & 11 + 2016 jan 21 & gs-2015b - q-9 & _ h _ & 954 & 17 & 1.18 & 10 & 2 & yes & 7 + the raw integral field spectrograph ( ifs ) data were reduced to form 3-d spectral data cubes using the gpi data reduction pipeline ( drp ) version 1.2.1 or 1.3 @xcite . there were no significant changes between the two versions of the gpi drp that impacted the astrometry of @xmath0 pic b. we used the same data reduction process as in @xcite and will summarize them here . first , dark subtraction and bad pixel correction were applied to each 2-d image . for earlier datasets in which cryocooler vibration induced correlated noise on the detector , the frames were `` destriped '' to remove this noise @xcite . then , we corrected for instrument flexure using an argon arc lamp taken before each sequence to align each individual spectrum for extraction @xcite . then each 2-d frame was turned into a spectral data cube , corrected for any remaining bad pixels , and corrected for distortion @xcite . for the _ k_-band data , thermal background frames were taken along the sequence . we constructed thermal background cubes in the same fashion and subtracted them from the _ k_-band data . to spatially register and photometrically calibrate our data , we used the gpi drp to measure the flux and location of the `` satellite '' spots , which are centered on the occulted star and imprinted with its attenuated spectrum @xcite . the satellite spot fluxes were used to derive a flux calibration in each spectral channel , which we used when constructing the forward models in section [ sec : fm ] . the location of the occulted star at each wavelength in each datacube was found using a least squares fit to all of the satellite spots positions and the magnitude of the atmospheric differential refraction . the occulted star center is used to align all the images together before psf subtraction , and is crucial for determining the astrometry of @xmath0 pic b relative to its host star . the precision on the star center is 0.05 pixels ( 0.7 mas ) for satellite spots with signal - to - noise ratios ( snr ) @xmath8 20 @xcite , which is certainly the case for all of our data on bright stars like @xmath0 pic . to subtract the stellar psf from each individual datacube , we used ` pyklip ` @xcite , a python implementation of the klip algorithm . we used both adi and sdi to decorrelate the stellar psf from the psf of @xmath0 pic b. as we were only concerned with @xmath0 pic b , we applied our psf subtraction only on the annulus that included the planet , rather than the full image . we adjusted three main parameters for the psf subtraction , depending on the dataset . the first was the number of modes used from the karhunen - love ( kl ) transform to model the stellar psf . the second was an exclusion criteria for picking reference psfs . the exclusion criteria is similar to the quantity @xmath9 in @xcite and is defined by the number of pixels that @xmath0 pic b would move azimuthally and radially in an observing sequence due to adi and sdi . thirdly , we toggled an @xmath711 pixel wide spatial high - pass filter that was applied to some datasets before psf subtraction . the high - pass filter was implemented using a gaussian filter in fourier space and the @xmath711 pixel size was chosen to remove low frequency background without significantly distorting any point sources in the image . in general , increasing the number of kl modes , decreasing the exclusion criteria , and applying a high - pass filter improves the subtraction of the stellar psf . however , taken to the extreme , all three options attenuate signal from the planet . additionally , for a planet as bright as @xmath0 pic b , the forward modelling described in section [ sec : fm ] may not be valid when the psf subtraction becomes too aggressive @xcite . thus , to optimize the signal of the planet while maintaining the validity of the forward modelling , we varied these parameters for each dataset . as @xmath0 pic b is bright , in most of the datasets we used an exclusion criteria of 4 pixels , which is slightly greater than 1 @xmath10 , where @xmath11 is the wavelength and @xmath12 is the diameter of the telescope . when there was little field rotation for adi , we decreased the exclusion criteria to not overly restrict our reference psfs , but also decreased the number of kl modes used to avoid being too aggressive when the speckle noise was too bright due to observing at a shorter wavelength or when the planet moved closer to the star , we applied a high pass filter to remove some of the diffracted starlight . we list the chosen parameters for each dataset in table [ table : obs ] . after psf subtraction , all images were rotated so that north is up and east is left . then , all the frames were mean combined in both the spectral and temporal directions , resulting in a single psf - subtracted frame for each dataset . after stellar psf subtraction , the psf of the planet is distorted by over - subtraction , caused by the presence of the planet in the data we are subtracting , and self - subtraction , caused by the presence of the planet in the reference images . over and self - subtraction perturbs the astrometric and photometric properties of the planet s psf and prevents a straightforward measurement of its position . typically , the biases and uncertainties in the astrometry caused by stellar psf subtraction are estimated by injecting fake planets into the data at other position angles and comparing the retrieved position to the injected position . however , the over - subtraction and self - subtraction that distorts the planet psf is deterministically caused by the existence of a planet in the data and its apparent motion in the reference images induced by adi and sdi . in turn , these features , if modeled , can inform us about the location of the planet and improve our astrometric precision and accuracy . recently , @xcite introduced klip - fm , an analytic framework to compute the effect of a planet on the kl modes and use these perturbations to reconstruct the over - subtraction and self - subtraction features . using this technique , we are able to generate the psf of the planet after psf subtraction ( see appendix [ appendix : fmpsf ] for a detailed description of the procedure ) . briefly , we use the psf of the planet before psf subtraction , the apparent motion of the planet due to adi and sdi , and a model of the planet spectrum to compute the distorted psf of the planet after psf subtraction . the result is a 2-d broadband planet psf , which we call @xmath13 , centered at ( @xmath14 , @xmath15 ) , our initial estimate for the location of the planet . for our gpi data , the forward models were generated using the implementation of klip - fm in ` pyklip ` . for each dataset , we used the same parameters as our psf subtraction to construct the forward model . to construct a model of the instrumental psf at each wavelength , we used the average psf of the satellite spots across all images for that wavelength . we used a basic center of light centroiding routine to measuring an approximate @xmath16 and @xmath17 positions of the planet ( @xmath14 and @xmath15 ) in the psf subtracted image . this initial estimate for the position is good to within a pixel for our gpi data . for the input planet spectrum of the forward model , we used the normalized best - fit model spectrum from @xcite , which has an effective temperature of 1650 k and a @xmath18 ( cgs units ) . we scaled the spectrum by eye to approximately match the contrast of @xmath0 pic b in our data and used the satellite spot fluxes to convert from contrast to digital numbers . we found that varying the spectrum and photometry had negligible impacts on our measured astrometry . even a spectrum differing by @xmath725% in shape changed the astrometry by @xmath19 mas , significantly smaller than the uncertainties we find in the following analysis . with this input information , we used ` pyklip ` to generate one forward modeled psf , @xmath13 , for each dataset . in figure [ fig : data - fm - compare ] , we show an example forward modeled psf ( not optmized to fit the data ) and comparison to data for our 2014 november 18 dataset on @xmath0 pic b. qualitatively , the forward model matches all the features seen in data , including the prominent negative self - subtraction lobes on either side of the planet . as ` pyklip ` parallelizes the computation , the generation of the forward models is quick . on a 32-core machine with amd opteron 6378 processors clocked at 2.3 ghz , forward models for all 37 channels of a representative 37-cube gpi dataset were generated in 15 minutes : 4 minutes of overheads for preprocessing and generating the instrumental psfs and 11 minutes to execute klip - fm and create the forward model . we note that we chose to run the forward model on a large annulus to examine noise properties in the data and that the computation time for klip - fm decreases @xmath720 - 30% if a small sector around the planet was used instead . to use the forward modeled psf , @xmath13 , to perform astrometry , we developed a bayesian framework to fit @xmath13 to the data , account for correlated noise , and estimate our fitting uncertainties . first , ignoring the correlated noise , we can use three parameters to fit the forward model to the data : the location on the planet in @xmath16 ( @xmath20 ) , the location of the planet in @xmath17 ( @xmath21 ) , and a scale factor ( @xmath22 ) to match the flux of @xmath13 to the data . we can then write the posterior probability for @xmath20 , @xmath21 , and @xmath22 given the data @xmath12 using bayes theorem as @xmath23 ignoring normalization constants . the first term on the right - hand side is the likelihood and the second is the prior . to construct the likelihood , we must first use our input parameters and @xmath13 to generate a model to compare to the data . we scale @xmath13 by @xmath22 and recenter it from its guessed location ( @xmath14 , @xmath15 ) to ( @xmath20 , @xmath21 ) . the residual , @xmath24 , between the model and the data , @xmath12 , in fitting region @xmath25 is calculated by @xmath26 where fitting region @xmath25 is a fixed rectangular box centered at the approximate location of the planet in the data . we pick the size of the fitting region to be a few @xmath10 to encompass the psf and the self - subtraction wings . we varied the fitting box size for each of our datasets in order to keep the fit focused on the area where signal from the planet can be seen . we list the size of the fitting box for each dataset in table [ table : obs ] . note that we do not fit any background term to the image , as one of the first steps of klip is to subtract off the mean of the image , removing any spatially constant background . for data with uncorrelated errors , the log of the likelihood function of the data for a particular model is @xmath27 , \end{aligned}\ ] ] where @xmath28 is the uncertainty in pixel i and @xmath29 is the number of pixels in the fitting region . however , the assumption of uncorrelated errors does not hold for images limited by speckle noise . except in the cases of very aggressive psf subtraction or at much greater separations from the star , the residual noise after psf subtraction in gpi data of a bright star like @xmath0 pic is still dominated by correlated speckle noise , which has a correlation scale that depends on the aggressiveness of the psf subtraction . due to the bright nature of @xmath0 pic b , we could not use a very aggressive reduction to ensure that the planet remained a perturbation on our kl modes , thereby preserving the validity of the analytical forward modelling technique . the conservative psf subtraction combined with the close separation of @xmath0 pic b from its bright host star required us to capture the correlated nature of the noise . we thus write our likelihood function instead as @xmath30 where @xmath31 is the covariance matrix of size @xmath32 and @xmath24 is a @xmath33 matrix . we applied a gaussian process framework to characterize the covariance in the noise ( see @xcite for an in - depth explanation of the application of gaussian processes to astronomical data ) . we only aim to model the correlations within each individual speckle , which spans @xmath10 in spatial extent . while in a single unprocessed frame speckle noise also has addition correlations on much larger scales , there are no significant correlations between speckles at larger spatial scales in our small fitting region due to stellar psf subtraction with klip and averaging uncorrelated speckles together when collapsing the frames in our adi+sdi sequence . thus , within our fitting region , the dominant correlation in our noise is from pixels within a single speckle , and thus is the one correlation we modelled . following the procedure in @xcite for fitting 1-d correlations in stellar spectra , we used the matrn covariance function parametrized with @xmath34 to model the correlated speckle noise . we chose the matrn function with @xmath34 as it better fits the correlations at larger separations compared to a simple squared exponential relation . however , we note that switching between covariance functions that have similar shapes does not significantly alter the error bars . we also chose to assume symmetric noise as we did not find any difference in the correlation scale of our noise in the radial and azimuthal directions . this is likely be due to the fact that we used both adi and sdi to model the stellar psf and thus are better at subtracting speckle noise , which before psf subtraction is more correlated radially than azimuthally due to the finite spectral bandwidth of the data . in instruments that are not able to utilize sdi , it might be necessary to separate the noise into radial and azimuthal components , each with its own characteristic correlation length . thus for our purposes , we chose the symmetric matrn covariance function with @xmath34 , which defines the covariance between two pixels @xmath35 and @xmath36 separated by a distance of @xmath37 as @xmath38 where @xmath28 and @xmath39 are the uncertainties for each pixel and @xmath40 is a characteristic correlation length scale that increases when noise is correlated at larger spatial scales . we calculated the uncertainty for each pixel by computing the standard deviation of pixels in an annulus with a width of @xmath41 ( 6 pixels ) , centered on the star , and with a mean radius equal to the distance between that pixel and the star . any pixels containing signal from the planet were masked and not used in estimating the noise in the annulus . the correlation length @xmath40 is not fixed , but rather kept as a hyperparameter parameter in our bayesian framework that we will marginalize over in the end . thus the final posterior we are trying to calculate is really @xmath42 where @xmath22 and @xmath40 are both hyperparameters as we are only interested in the astrometry . compared to our likelihood function , our prior is relatively straightforward . we allow for uniform priors in @xmath16 and @xmath17 within 2 pixels from our initial guess location . similarly , we allow for an uniform prior between 0 and 5 for @xmath22 to determine how much to scale @xmath13 , which was already scaled to an approximate contrast of @xmath0 pic b. the correlation length @xmath40 has a uniform prior between 0 and 10 pixels , which provides an ample range to explore the correlations within individual speckles of size @xmath43 pixels . we used the @xcite affine invariant mcmc sampler implemented in the ` emcee ` python package @xcite to sample the posterior distribution and custom ` cython ` code to quickly generate the covariance matrix as we vary @xmath40 . the mcmc sampler was run for 800 steps using 100 walkers , with a `` burn - in '' of 200 steps beforehand . in figure [ fig : astrom - tri ] , we show the posterior distributions from the mcmc fit for the 2014 november 18 _ h_-band dataset as a representative posterior distribution . the value of @xmath40 is close to @xmath44 pixels , indicating we are accurately fitting the correlated speckle noise . we use the resulting posterior distributions to calculate the most - likely values and uncertainties for the location of @xmath0 pic b in our image at each epoch . to convert our results to more useful physical values , we need to convert our measured location of @xmath0 pic b to its positional offset from its host star in right ascension ( ra ) and declination ( dec ) . as our images are already rotated so that @xmath45 is positive ra and @xmath46 is positive dec , it is straightforward to convert from image fates to sky coordinates . we use the satellite spots to measure the location of the star behind the coronagraph , which has a precision of 0.7 mas @xcite . this allows us to derive the separation of the planet from the star in pixels . to convert from pixel separations to physical separations in ra and dec , we use the most - recent astrometric calibration numbers from @xcite : a plate scale of @xmath47 mas lenslet@xmath48 and a residual north angle offset of @xmath49 from the north angle value used in the gpi drp ( versions 1.2.1 to 1.3 , the current version ) . these astrometric calibration numbers show no significant change over time , so we apply them to all our epochs of data . then , we assume all of these error terms are uncorrelated and add them in quadrature with our measurement errors from our mcmc fit to determine our full astrometric precision . the combination of the forward modeled psf and the bayesian framework makes up the bayesian klip - fm astrometry technique we introduce in this work . we apply bka for all twelve gpi datasets and report the measured astrometry and error budgets in table [ table : astrometry ] . the best fit forward models and residuals to the fit are shown in appendix [ appendix : res ] . on most of our datasets , we are not limited just by the uncertainty in the location of the planet , which was as low as 0.3 mas . the uncertainties in the location of the star and north angle also make significant contributions to the error budget . typically , we achieved @xmath71 mas precision on the relative astrometry between @xmath0 pic b and its host star . this is a factor of @xmath724 improvement over previous techniques @xcite using the same data , indicating this technique can be useful in reanalysis of archival data to obtain better astrometry in cases where the limiting factor is the uncertainty on the planet position . in two of the later datasets where the planet is observed closer in , we were limited by the snr of the planet and unable to achieve 1 mas precision . in the 2015 december 5 dataset , the noise was higher due to the planet being fainter relative to the star in _ j_-band . in the 2016 january 21 dataset , a combination of poor seeing and a small amount of usable data limited our astrometric precision . overall though , this gpi @xmath0 pic b data is an excellent demonstration for bayesian klip - fm astrometry as the planet is bright enough that the extended psf features , such as the negative self - subtraction lobes , are clearly seen and provide significant information to constrain the position of the planet . for fainter planets , the extended features are harder to distinguish from the noise . as one of the main advantages of bka over techniques that do not forward model the psf is being able to forward model the extended self - subtraction lobes , the astrometric improvement would not be as large for lower signal - to - noise ratio planets . there still should be some improvement though due to accurately modelling the over - subtraction on the core of the psf and small contributions from the extended features even if they are hard to distinguish from noise . regardless , in addition to the improved precision , bka should also more accurately estimate the uncertainties as it fits for the correlation scale of the noise at the location of the planet . lcccc|cccc 2013 nov 16 @xmath50 & 0.6/0.7 & 0.7/0.7 & 0.3 & 0.13 & -228.5 @xmath51 1.3 & -366.2 @xmath51 1.1 & 431.6 @xmath51 1.0 & 212.0 @xmath51 0.2 + 2013 nov 16 @xmath52 & 0.5/0.4 & 0.7/0.7 & 0.3 & 0.13 & -229.2 @xmath51 1.2 & -364.5 @xmath51 1.0 & 430.6 @xmath51 0.9 & 212.2 @xmath51 0.2 + 2013 nov 18 @xmath53 & 0.3/0.3 & 0.7/0.7 & 0.3 & 0.13 & -229.1 @xmath51 1.1 & -364.7 @xmath51 1.0 & 430.6 @xmath51 0.8 & 212.1 @xmath51 0.2 + 2013 dec 10 @xmath53 & 0.4/0.4 & 0.7/0.7 & 0.3 & 0.13 & -227.9 @xmath51 1.2 & -359.9 @xmath51 1.0 & 426.0 @xmath51 0.9 & 212.3 @xmath51 0.2 + 2013 dec 10 @xmath54 & 0.6/0.7 & 0.7/0.7 & 0.3 & 0.13 & -227.2 @xmath51 1.3 & -361.1 @xmath51 1.2 & 426.6 @xmath51 1.1 & 212.2 @xmath51 0.2 + 2013 dec 11 @xmath53 & 0.5/0.4 & 0.7/0.7 & 0.3 & 0.13 & -227.8 @xmath51 1.2 & -359.2 @xmath51 1.0 & 425.4 @xmath51 0.9 & 212.4 @xmath51 0.2 + 2014 nov 8 @xmath53 & 0.5/0.5 & 0.7/0.7 & 0.2 & 0.13 & -194.0 @xmath51 1.1 & -299.1 @xmath51 1.0 & 356.5 @xmath51 0.9 & 213.0 @xmath51 0.2 + 2015 apr 2 @xmath53 & 0.4/0.5 & 0.7/0.7 & 0.2 & 0.13 & -172.1 @xmath51 1.0 & -266.5 @xmath51 1.0 & 317.2 @xmath51 0.9 & 212.9 @xmath51 0.2 + 2015 nov 6 @xmath53 & 0.7/0.7 & 0.7/0.7 & 0.2 & 0.13 & -137.8 @xmath51 1.1 & -207.2 @xmath51 1.0 & 248.8 @xmath51 1.0 & 213.6 @xmath51 0.3 + 2015 dec 5 @xmath54 & 1.2/1.3 & 0.7/0.7 & 0.2 & 0.13 & -133.9 @xmath51 1.5 & -200.5 @xmath51 1.5 & 241.1 @xmath51 1.4 & 213.7 @xmath51 0.4 + 2015 dec 22 @xmath53 & 0.5/0.5 & 0.7/0.7 & 0.2 & 0.13 & -130.0 @xmath51 1.0 & -194.7 @xmath51 0.9 & 234.1 @xmath51 0.9 & 213.7 @xmath51 0.2 + 2016 jan 21 @xmath53 & 1.8/1.6 & 0.7/0.7 & 0.2 & 0.13 & -126.8 @xmath51 2.0 & -185.8 @xmath51 1.8 & 225.0 @xmath51 1.8 & 214.3 @xmath51 0.5 + ccccccc semi - major axis ( @xmath55 ) & au & @xmath56 & uniform in @xmath57 & 9.02 & 9.66 & 10.78 + epoch of periastron ( @xmath58 ) & - & @xmath59 & uniform in @xmath58 & 0.32 & 0.73 & 0.87 + argument of periastron ( @xmath60 ) & & @xmath61 & uniform in @xmath60 & 192.8 & 205.8 & 258.4 + position angle of the ascending node ( @xmath62 ) & & @xmath63 & uniform in @xmath62 & 31.67 & 31.76 & 31.84 + inclination ( @xmath35 ) & & @xmath64 & uniform in @xmath65 & 88.70 & 88.81 & 88.93 + eccentricity ( @xmath66 ) & - & @xmath67 & uniform in @xmath66 & 0.027 & 0.080 & 0.171 + total mass ( @xmath68 ) & @xmath69 & @xmath70 & uniform in @xmath69 & 1.76 & 1.80 & 1.83 + + period ( @xmath71 ) & years & - & - & 20.21 & 22.47 & 26.24 + hill sphere ingress & mjd & - & - & 57,840 & 57,846 & 57,854 + @xmath72 hill sphere ingress & mjd & - & - & 57,916 & 57,924 & 57,934 + closest approach date & mjd & - & - & 57,986 & 57,996 & 58,008 + @xmath72 hill sphere egress & mjd & - & - & 58,056 & 58,069 & 58,082 + hill sphere egress & mjd & - & - & 58,132 & 58,147 & 58,162 + to explore the validity of our new astrometry technique , we fit a keplerian orbit to our twelve epochs of astrometry . since each epoch of astrometry is fit independently , and since the keplerian orbit is agnostic towards the exact data reduction methods , having all twelve epochs of astrometry fit the keperlain orbit would only be possible if all of the astrometry is accurate and precise . if there are errors in estimating the magnitude of the uncertainties or any remaining biases in our measurements , we expect them to become evident in the residuals of the keplerian fit either as systematic trends or fit outliers . we followed the same analysis as in @xcite to obtain the orbital elements of @xmath0 pic b using a mcmc fit with the parallel - tempered sampler implemented in ` emcee ` . we combined the twelve gpi astrometric points presented here along with the datasets presented in @xcite and @xcite . unlike @xcite , we did not explicitly include the radial velocity measurement of @xmath0 pic b from @xcite in order to limit potential systematics between instruments , but we do use it to constrain the prior on the position angle of the ascending node and thus the direction of the orbit ( i.e. , we know that that @xmath0 pic b has been moving towards us since @xmath72007 ) . our model of the orbit fits the same seven parameters as @xcite . for convenience , the parameters are listed in table [ tab : orb_params ] . the mcmc sampler was run for 30,000 steps using 1024 walkers at each of the 20 temperatures after 30,000 steps of `` burn - in '' to allow the walkers to converge . we thinned the chains to remove any remaining correlations , keeping every 75th step to result in effectively 400 samples per walker for a total of 409,600 samples to construct our posterior . the posteriors on the seven parameters in our model are shown in figure [ fig : orbit_triangle ] and the 16% , 50% , and 84% percentiles of the marginalized posterior distribution for each parameter is listed in table [ tab : orb_params ] . for comparison , we plot 500 randomly - chosen possible orbits along with the measured astrometry in figure [ fig : orbit - vs - time ] . the fit residuals of our measured gpi points are shown in figure [ fig : residuals ] . the residuals in ra and dec offset ( which are the parameters we use in our mcmc ) are consistent with zero and do not show any systematics . the residuals in radial separation and position angle ( pa ) , neither of which are used in our fit , are slightly further away from zero , but do not indicate any obvious errors in either our astrometry or our error estimates . to quantitatively assess the validity of our measurements , we use the reduced chi - squared ( @xmath73 ) statistic , which measures how consistent our astrometry is with a keplerian orbit . ideally , @xmath73 would be unity if all measurements and uncertainties were accurate . however , biases and improperly estimated errors would cause it to deviate from unity . due to systematic astrometric calibration offsets between instruments that have not been characterized , we expect @xmath73 to be slightly above unity . for example , @xcite reported a @xmath73 of @xmath74 when combining gpi astrometry , measured using standard techniques in the field , with previous astrometric measurements even with @xmath724 times smaller error bars on the gpi points , the @xmath73 of our accepted orbits is @xmath75 , unchanged from the @xmath74 reported in @xcite . with bka contributing twelve out of the thirty astrometry measurements , if the bka technique contained biases larger than one milliarcsecond , they would have caused a significant increase in @xmath73 . likewise , if we had been too optimistic with our error estimates , @xmath73 should have also increased as the reported uncertainties would not have matched the scatter in the measurements . the lack of increase in @xmath73 indicates that the more precise astrometry from bka is not biased and has accurate uncertainties . in addition to the keplerian orbit fit , we can examine the accuracy of our estimated one milliarcsecond uncertainties by looking at the astrometry measured on the same or consecutive days . as we do not expect the planet s position to change significantly in the span of a single day , the three measurements in november of 2013 and the three measurements in december of 2013 ought to be consistent with each other . indeed , our measurements in table [ table : astrometry ] indicate that in both sets of astrometry , the measurements agree at the milliarcsecond level , confirming that our estimated uncertainties are accurate . thus , the well - behaved residuals of our gpi measurements and the lack of change in @xmath73 from @xcite even with significantly smaller error bars lead us to conclude the measured astrometry using bka are accurate and free from biases . the fact the residuals are consistent with zero , the lack of change in @xmath73 , and the consistency of repeated measurements taken around the same time all indicate that the one milliarcsecond uncertainties estimated from bka are also accurate . together , these multiple assessments of the validity of bka all indicate that this technique is both accurate and precise . having demonstrated the accuracy and precision of this new technique , we now analyze the new constraints on the orbit of @xmath0 pic b. not surprisingly , the estimates for @xmath55 and @xmath66 have not changed significantly from @xcite since the gpi points reanalyzed in this paper do not have a sufficiently long time baseline to constrain these parameters . as seen in figure [ fig : orbit - vs - time ] , all but one astrometric measurement is on one half of the orbit curve . the other half of the orbit is not as well constrained , leaving a degeneracy in @xmath55 and @xmath66 . this degeneracy can be broken with more measurements obtained when the planet appears on the other side of the star . better constraints on @xmath55 and @xmath66 will provide better insight on how @xmath0 pic b interacts with the debris disk and potential unseen planets in the system . the new total mass of the system , @xmath76 @xmath69 , is significantly higher than the @xmath77 @xmath69 from @xcite . this new total system mass , which effectively measures the mass of the star at this precision , is in better agreement with the stellar mass of @xmath78 @xmath69 derived from stellar photometry @xcite . our measurement of the position angle of the ascending node , @xmath62 , slightly improves upon the value obtained by @xcite , but the overall value remains consistent . thus , @xmath0 pic b is still consistent with being the planet responsible for the known warp in the debris disk ( e.g. * ? ? ? compared to @xcite , the argument of the periastron , @xmath60 in our fit has increased from @xmath79 to @xmath80 . this new value is consistent with @xmath81 that is required for the falling evaporating bodies ( febs ) scenario proposed by @xcite to explain redshifted absorption features in @xmath0 pic s spectrum . note that under previous definitions of the orbital parameters , this has been expressed as @xmath82 from the line of sight . this scenario also requires a slightly eccentric orbit , which is consistent with our derived orbital parameters . for a more in - depth discussion of @xmath0 pic b s relationship to the debris disk and the feb scenario , we direct the reader to @xcite . the biggest improvement in our understanding of the orbit of @xmath0 pic b is the improved constraint on the inclination of the orbit . we find the inclination to be @xmath83 which allows us to place the tightest constraints on the probability that @xmath0 pic b will transit its host star . assuming an angular diameter of the star of 0.736 mas @xcite and considering the range of @xmath55 from our orbit fit , we find that in most cases , we need @xmath84 in order for the planet to transit . with our current constraints on the inclination , we have ruled out the possibility that @xmath0 pic b will transit at 10-@xmath1 significance . this tight constraint on the inclination and transit probability is due to the slightly longer time baseline and the improved precision in the measured position angle of the gpi astrometry compared to @xcite . for an edge - on orbit that transits the star , we should see no significant change in pa over time . however , figure [ fig : orbit - vs - time ] shows that the gpi points alone reveal a significant increase in pa over time . thus , regardless of systematic astrometric calibration errors between instruments , we conclude that @xmath0 pic b will not transit its star . unlike the planet , @xmath0 pic b s hill sphere , the region around the planet that could contain gravitationally bound circumplanetary material , will transit the star . we define the radius of the hill sphere as @xmath85{\left(\frac{m}{3m}\right)}$ ] using the approximate form proposed in @xcite where @xmath86 is the mass of the planet and @xmath87 is the mass of the star . assuming @xmath0 pic b follows ahot - start " evolutionary track with a mass of @xmath88 @xmath5 @xcite and using the range of semi - major axes , eccentricities , and stellar masses from our mcmc orbit fit , @xmath89 au ( @xmath90 mas ) . given that our prediction for the closest approach of @xmath0 pic b will be @xmath91 mas ( @xmath92 projected au ) from the star , the hill sphere of @xmath0 pic b will transit the star , as shown in figure [ fig : hillsphere_transit ] . from our mcmc orbit fit , we can compute when the transit of the hill sphere will occur . we pick five notable events to focus on : two are the ingress and egress of the hill sphere which are the extrema in time between which any circumplanetary material could transit ; another is the date of closest approach , which gives the opportunity to probe material closest in to the planet ; and the last two are the transit of the sphere that is @xmath72 @xmath93 in radial extent . almost all stable prograde circumplanetary orbits reside within @xmath72 @xmath93 @xcite , so it is more likely to find material within half a hill sphere . in table [ tab : orb_params ] , we list our constraints on the date of these events , and in figure [ fig : orbit_triangle ] , we plot the posterior distribution for these events . the duration of the hill - sphere transit will be long : ingress is @xmath72017 april 3 and egress is @xmath72018 january 29 . the 1/2 hill sphere begins transit @xmath72017 june 20 and ends transit on @xmath72017 november 12 , with closest approach on @xmath72017 august 31 . given the ra of @xmath0 pic ( @xmath94 ) , the star will not be visible from most ground - based observatories during almost the entire time between ingress of the hill sphere and closest approach . ground - based telescopes in antarctica , airborne observatories capable of travelling to antarctica , and space - based observatories would provide the only opportunities to observe @xmath0 pic during this time period . the second half of the hill - sphere transit will be visible from most southern - hemisphere ground - based observatories , making @xmath0 pic a well - suited candidate for photometric monitoring in late 2017 . as it is a rare opportunity to probe circumplanetary material , it is not certain what will be seen when the hill sphere of @xmath0 pic b transits the star . one possibility is that satellites could reside in the hill sphere . to approximate the photometric transit depths , the 0.736 mas angular diameter of the star corresponds to 1.53 @xmath95 or 1.065@xmath9610@xmath97 km . moons as large as ganymede ( @xmath98 = 2630 km ) or io ( @xmath98 = 1820 km ) would give transit depths of 2.473@xmath9610@xmath99 ( 2.7 mmag ) and 1.711@xmath9610@xmath99 ( 1.9 mmag ) , respectively . detecting these photometric signatures will require a high cadence , as any single satellite orbiting @xmath0 pic b will have a transit duration of @xmath72 days . additionally , @xmath0 pic is a variable star with pulsation timescales of @xmath70.5 hours and variability amplitudes in _ b_-band of @xmath100 mmags @xcite , so careful modelling of stellar activity is necessary to be sensitive to these transit depths . another possibility is that , as @xmath0 pic b is still young and evolving , it may harbor a circumplanetary disk or ring system comprised of leftover material from planet formation . such a hypothesis is not unprecedented as @xcite found evidence for a large circumplanetary disk around an unseen planet in the 1swasp j140747.93 - 394542.6 ( hereafter j1407 ) system , which has a similar age ( @xmath716 myr ; * ? ? ? * ) and thus likely at a similar stage in its evolution . @xcite interpreted the series of complex and deep eclipses in the j1407 lightcurve as from a circumplanetary disk 0.6 au in radial extent in the process of forming rings due to newly formed satellites . it is plausible that @xmath0 pic b can harbor a similar disk as a 0.6 au disk would be @xmath72 @xmath93 in extent and consistent with where we would expect stable orbits to reside around @xmath0 pic b. additionally , since @xmath0 pic b is young , it is plausible that there is a large amount of circumplanetary material which has yet to be cleared out dynamically . such a large disk would transit the star and be suitable for detection through photometric monitoring of the host star if the disk is inclined by @xmath101 with respect to the orbital plane of @xmath0 pic b. to maintain such an inclination , the disk needs to be massive enough to prevent stellar tidal forces from aligning the disk to the orbital plane @xcite . so far , there have not been many observational constraints of circumplanetary material around @xmath0 pic b. @xcite reported a photometric event in 1981 and hypothesized it could be due to the transit of a planet that cleared out a hole in the debris disk around its hill sphere . however , @xcite show that the planet is not embedded in the debris disk and we definitely show the planet itself will not transit . still , our orbit models give a @xmath102% and @xmath103% chance the photometric event coincided with the transit of the hill sphere and 1/2 hill sphere respectively , during which time material around the planet could have passed in front of the star . additionally , a circumplanetary disk or ring may also be detectable through the planet s spectral energy distribution ( sed ) . the planet s near - infrared spectrum would experience extinction if the dust resides between us and the planet , with the magnitude of extinction depending on the amount of dust . in the near infrared , the extinction would be greater at shorter wavelengths due to the increased scattering and absorption by dust and would produce a spectral slope in the planet s near - infrared sed . dust around the planet would also scatter starlight , causing the planet s sed to appear brighter in the optical , as has been postulated for fomalhaut b @xcite . furthermore , the dust will produce millimeter emission . however , detecting circumplanetary material in the planet s sed will require being able to distinguish it from the circumstellar disk with precise spectral data . in the first part of this work , we have presented a new technique for more precise and accurate astrometry of directly imaged exoplanets using a new analytical forward modelling approach in a robust statistical framework . * using the klip - fm framework presented in @xcite , we are able to analytically forward model the psf of the planet through the data reduction process , giving us better information on the location of the planet . we apply klip - fm to gpi data on @xmath0 pic and forward model the psf of @xmath0 pic b using the open - source ` pyklip ` package . * for a close - in planet orbiting a bright star like in the case of @xmath0 pic b , we are limited by correlated speckle noise in our data . we developed a bayesian framework utilizing gaussian processes and mcmc to account for the correlated noise and to find the position of the planet simultaneously . * with this technique , we have achieved the most precise astrometry on @xmath0 pic b to date . on most of our gpi datasets , we achieve @xmath71 mas precision on the relative separation between @xmath0 pic b and its host star , a @xmath72 - 4 fold improvement over previous techniques using the same data @xcite . * in datasets where the astrometry is limited by noise and not by astrometric calibration uncertainty , bayesian klip - fm astrometry approach should improve astrometric precision . in the second part of this work , we apply our bayesian klip - fm astrometry technique to the orbit of @xmath0 pic b. * to validate this new astrometric technique , we used it to measure the position of @xmath0 pic b in twelve epochs of gpi data . we combined these twelve astrometric points with two previous astrometric monitoring campaigns and fit a keplerian orbit using mcmc methods . we find the residuals to the fit are consistent with zero and show no apparent systematic trends , indicating that our fit is accurate and the uncertainties we estimate are reliable . * due to the improved pa measurements from our technique , we have the tightest constraints on the inclination of the orbit and can exclude a possible transit of @xmath0 pic b at 10-@xmath1 significance . * while the planet will not transit , we are confident the hill sphere around @xmath0 pic b will transit . the hill sphere will begin transit at the start of april in 2017 and finish transiting at the end of january in 2018 with closest approach in the end of august in 2017 . the transit of @xmath0 pic b s hill sphere should be our best chance in the near future to investigate young circumplanetary material . in the future , this mcmc forward modelling technique can be applied to photometry and spectral extraction alongside of astrometry of directly imaged exoplanets , allowing for improved characterization of their atmospheres . for @xmath0 pic b , continued monitoring of its orbit will yield more insight into the dynamics of the star system , although the planet will soon be too close to its star to be seen with current direct imaging instrumentation . however , once the planet appears on the other side , continued astrometric monitoring should be able to constrain the semi - major axis and eccentricity of the orbit much better , which will improve our understanding of how @xmath0 pic b perturbs the disk and if there are other planets perturbing @xmath0 pic b. the gpi project has been supported by gemini observatory , which is operated by aura , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the nsf ( usa ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , mcti ( brazil ) and mincyt ( argentina ) . we thank the national science foundation ( nsf ast-1518332 ) and nasa ( nnx15ac89 g and nexss program nnx15ad95 g ) for contributing to support of this research . portions of this work were performed under the auspices of the u.s . department of energy by lawrence livermore national laboratory under contract de - ac52 - 07na27344 . this research has made use of the simbad database , operated at cds , strasbourg , france . the posterior distribution plots were made with ` corner ` @xcite . using the notation of @xcite , we can analytically forward model the psf of the planet through the process of stellar psf subtraction . let us denote the target image as @xmath104 , which is taken at wavelength @xmath105 and time @xmath106 and contains the pixels @xmath107 from which we want to subtract the stellar psf . to forward model the psf of the planet , we need a model of the psf of the planet prior to stellar psf subtraction . for our gpi data , we use the satellite spots to generate a realization of an unocculted and unprocessed point source as a function of wavelength . using this model , we can generate @xmath108 , a frame that consists solely of the unprocessed psf of the planet in the target frame . the initial estimate for the planet s position in the frame is determined using an input separation and pa of the planet and the location of the star . similarly , we can also generate an array of frames , each containing the unprocessed psf of the planet for each image in the reference library , accounting for the fact that the planet position changes due to adi field rotation and due to sdi rescaling to align speckles . in these frames , the psf shape only depends on the wavelength of the frame and the position of the psf depends on the apparent movement of the planet in the data due to adi and sdi . this array of unprocessed psfs of the planet will be used to calculate the perturbation of the kl modes due to the existence of the planet . assuming a spectrum , @xmath109 , of the planet in the case of ifs data ( for imaging data , @xmath109 is just a scalar ) , we can compute @xmath110 , the perturbation on the @xmath111th kl mode , @xmath112 , due to the existence of a planet , using equation e18 or e20 of @xcite . we will not write out explicitly the exact formula to compute @xmath113 here since we have not made any modifications to it . we note that we use equation e18 from @xcite to compute @xmath113 as it is computationally faster than equation e20 for a fixed input spectrum . we calculate the perturbed @xmath113 for the first @xmath114 kl modes , where @xmath114 is the number of kl modes we choose to use in our psf subtraction . from here , the forward - modeled psf of the planet for this frame is computed as @xmath115 \\ - \left [ \sum_{k=1}^{k_{klip } } \left ( < t_{\lambda_{p},t_p}(\mathbf{x } ) , z_k ( \mathbf{x } ) > \delta z_k ( \mathbf{x } ) \right . \\ + \left . \left . < t_{\lambda_{p},t_p}(\mathbf{x } ) , \delta z_k ( \mathbf{x } ) > z_k ( \mathbf{x } ) \right ) \vphantom{\sum_{k=1}^{k_{klip } } } \right],\end{gathered}\ ] ] where @xmath116 is the inner product . equation [ fmmath ] is very similar to eq . f7 in @xcite , but is focused on generating the forward model with a fixed input spectrum and not concerned with keeping the planet s spectrum as a free parameter for spectral extraction . as mentioned in @xcite , the term in the first square bracket is the over - subtraction term that is due to the projection of the kl modes on data with a planet in it , and the terms in the second square bracket are the self - subtraction terms due to the presence of a planet in the reference images influencing the kl modes . after generating forward models for each frame , we take the mean of all the forward models in time to give us a single forward modeled psf cube , @xmath117 that still has spectral information in the third dimension . in principle , this psf cube can be used to retrieve spectral information from the data , but it is outside of the scope of this paper . for our astrometry purposes , we also take the mean in wavelength , generating a single 2-d forward modeled psf , which we call @xmath13 , for each dataset . in figure [ fig : astrometry_residuals ] , we plot the residuals to the best fit model for each epoch using bka . as @xmath0 pic b moves closer to its host star , the magnitude of speckle noise increases relative to the signal of @xmath0 pic b. this is especially true in the last epoch when @xmath0 pic b was closest to its star . however , the fit was still accurate as the bright positive core of the psf of the planet was successfully modeled and not seen in the residuals , which are consistent with noise . apai , d. , schneider , g. , grady , c. a. , et al . 2015 , , 800 , 136 beust , h. , & morbidelli , a. 2000 , , 143 , 170 binks , a. s. , & jeffries , r. d. 2016 , , 455 , 3345 bottom , m. , kuhn , j. , mennesson , b. , et al . 2015 , , 809 , 11 brandt , t. d. , mcelwain , m. w. , turner , e. l. , et al . 2013 , , 764 , 183 burrows , c. j. , krist , j. e. , stapelfeldt , k. r. , & wfpc2 investigation definition team 1995 , bulletin of the american astronomical society , 27 , 32.05 cantalloube , f. , mouillet , d. , mugnier , l. m. , et al . 2015 , , 582 , a89 chauvin , g. , lagrange , a .- m . , beust , h. , et al . 2012 , , 542 , a41 chilcote , j. , barman , t. , fitzgerald , m. p. , et al . 2015 , , 798 , l3 crifo , f. , vidal - madjar , a. , lallement , r. , ferlet , r. , & gerbaldi , m. 1997 , , 320 , l29 czekala , i. , andrews , s. m. , mandel , k. s. , et al . 2015 , , 812 , 128 dawson , r. i. , murray - clay , r. a. , & fabrycky , d. c. 2011 , , 743 , l17 de rosa , r. j. , nielsen , e. l. , blunt , s. c. , et al . 2015 , , 814 , l3 dent , w. r. f. , wyatt , m. c. , roberge , a. , et al . 2014 , science , 343 , 1490 defrre , d. , lebreton , j. , le bouquin , j .- b . , et al . 2012 , , 546 , l9 esposito , t. m. , fitzgerald , m. p. , graham , j. r. , & kalas , p. 2014 , , 780 , 25 foreman - mackey , d. , hogg , d. w. , lang , d. , goodman , j. 2013 , 125 , 306 foreman - mackey d. , vousden w. , price - whelan a. , et al . 2016 , corner.py , v1.0.2 , zenodo , doi:10.5281/zenodo.45906 galicher , r. , & marois , c. 2011 , in second international conference on adaptive optics for extremely large telescopes ( victoria , bc : victoria conference centre ) , p25 , http://ao4elt2.lesia.obspm.fr graham , j. r. 2009 , gemini planet imager operational concept definition document v4.1 , ( hilo , hi : gemini observatory ) goodman , j. & weare , j. , 2010 , comm . math . comp . , 5 , 65 hamilton , d. p. , & burns , j. a. 1992 , , 96 , 43 heap , s. r. , lindler , d. j. , lanz , t. m. , et al . 2000 , , 539 , 435 ingraham , p. , perrin , m. d. , sadakuni , n. , et al . 2014 , , 9147 , 91477o kalas , p. , graham , j. r. , chiang , e. , et al . 2008 , science , 322 , 1345 kenworthy , m. a. , & mamajek , e. e. 2015 , , 800 , 126 koen , c. , balona , l. a. , khadaroo , k. , et al . 2003 , , 344 , 1250 konopacky , q. m. , thomas , s. j. , macintosh , b. a. , et al . 2014 , , 9147 , 914784 lafrenire d. , marois c. , doyon , r. , nadeau d. , & artigau . 2007 , , 660 , 770 lagrange , a .- m . , gratadour , d. , chauvin , g. , et al . 2009 , , 493 , l21 lagrange , a .- m . , bonnefoy , m. , chauvin , g. , et al . 2010 , science , 329 , 57 lagrange , a .- m . , boccaletti , a. , milli , j. , et al . 2012 , , 542 , a40 lecavelier des etangs , a. , vidal - madjar , a. , burki , g. , et al . 1997 , , 328 , 311 lecavelier des etangs , a. , & vidal - madjar , a. 2016 , , 588 , a60 macintosh , b. a. , graham , j. r. , ingraham , p. , et al . 2014 , pnas , 111 , 35 males , j. r. , close , l. m. , morzinski , k. , m. , et al . 2014 , , 786 , 32 mamajek , e. e. , bell , c. p. m. 2014 , , 445 , 2169 mamajek , e. e. , quillen , a. c. , pecaut , m. j. , et al . 2012 , , 143 , 72 marois , c. , doyon , r. , racine , r. , nadeau , d. 2000 , , 112 , 91 marois , c. , lafreniere , d. , doyon , r. , macintosh , b. , & nadeau , d. 2006a , , 641 , 556 marois , c. , lafrenire , d. , macintosh , b. , & doyon , r. 2006b , , 647 , 612 marois , c. , macintosh , b. , & vran , j .- 2010 , , 7736 , 77361j mawet , d. , milli , j. , wahhaj , z. , et al . 2014 , , 792 , 97 millar - blanchaer , m. a. , graham , j. r. , pueyo , l. , et al . 2015 , , 811 , 18 morzinski , k. m. , males , j. r. , skemer , a. j. , et al . 2015 , , 815 , 108 mouillet , d. , larwood , j. d. , papaloizou , j. c. b. , & lagrange , a. m. 1997 , , 292 , 896 nielsen , e. l. , liu , m. c. , wahhaj , z. , et al . 2014 , , 794 , 158 perrin , m. d. , maire , j. , ingraham , p. , et al . 2014 , , 9147 , 91473j pueyo , l. , soummer , r. , hoffman , j. , et al . 2015 , , 803 , 31 pueyo , l. 2016 , , 824 , 117 rameau , j. , nielsen , e. l. , de rosa , r. j. , et al . 2016 , , 822 , l29 shen , y. , & tremaine , s. 2008 , , 136 , 2453 sivaramakrishnan , a. , & oppenheimer , b. r. 2006 , , 647 , 620 smith , b. a. , & terrile , r. j. 1984 , science , 226 , 1421 snellen , i. a. g. , brandl , b. r. , de kok , r. j. , et al . 2014 , , 509 , 63 soummer , r. , pueyo , l. , & larkin , j. 2012 , , 755 , l28 thbault , p. , & beust , h. 2001 , , 376 , 621 van leeuwen , f. 2007 , , 474 , 653 wang , j. j. , rajan , a. , graham , j. r. , et al . 2014 , , 9147 , 914755 wang , j. j. , ruffio , j .- b . , de rosa , r. j. , et al . 2015 , astrophysics source code library , ascl:1506.001 wolff , s. g. , perrin , m. d. , maire , j. , et al . 2014 , , 9147 , 91477h zanazzi , j. j. , & lai , d. 2016 , arxiv:1605.02365

\(1 ) for any integers and , there is a closed 3-manifold which admits a distance heegaard splitting of genus except that the pair of is . furthermore , can be chosen to be hyperbolic except that the pair of is . ( 2 ) for any integers and , there are infinitely many non - homeomorphic closed 3-manifolds admitting distance heegaard splittings of genus . [ section ] [ thm]lemma [ thm]corollary [ thm]proposition [ thm]example [ section ] [ section ] : attaching handlebody , heegaard distance , subsurface projection . ams classification : 57m27

1302.5188

let @xmath9 be a compact surface with @xmath10 not a 4-punctured sphere . harvey @xcite defined the curve complex @xmath11 as follows : the vertices of @xmath11 are the isotopy classes of essential simple closed curves on @xmath9 , and @xmath12 distinct vertices @xmath13 determine a k - simplex of @xmath11 if and only if they are represented by pairwise disjoint simple closed curves . for two vertices @xmath14 and @xmath15 of @xmath11 , the distance of @xmath14 and @xmath15 , denoted by @xmath16 , is defined to be the minimal number of 1-simplexes in a simplicial path joining x to y. in other words , @xmath16 is the smallest integer @xmath17 such that there is a sequence of vertices @xmath18 such that @xmath19 and @xmath20 are represented by two disjoint essential simple closed curves on @xmath9 for each @xmath21 . for two sets of vertices in @xmath11 , say @xmath22 and @xmath23 , @xmath24 is defined to be @xmath25 as follows : the vertices of @xmath26 are the isotopy classes of essential simple closed curves on @xmath9 , and @xmath12 distinct vertices @xmath13 determine a k - simplex of @xmath11 if and only if @xmath20 and @xmath27 are represented by two simple closed curves @xmath28 and @xmath29 on @xmath9 such that @xmath28 intersects @xmath29 in just one point for each @xmath30 . let @xmath31 be a compact orientable 3-manifold . if there is a closed surface @xmath9 which cuts @xmath31 into two compression bodies @xmath32 and @xmath33 such that @xmath34 , then we say @xmath31 has a heegaard splitting , denoted by @xmath35 , where @xmath36 ( resp.@xmath37 ) means the positive boundary of @xmath32 ( resp.@xmath33 ) . we denote by @xmath38(resp.@xmath39 ) the set of vertices in @xmath11 such that each element of @xmath40 ( resp . @xmath39)is represented by the boundary of an essential disk in @xmath32(resp . @xmath33 ) . the distance of the heegaard splitting @xmath41 , denoted by @xmath42 , is defined to be @xmath43 . see @xcite . it is well known that a 3-manifold admitting a high distance heegaard splitting has good topological and geometric properties . for example , hartshorn@xcite and scharlemann@xcite showed that a 3-manifold admitting a high distance heegaard splitting contains no essential surface with small euler characteristic number ; scharlemann and tomova@xcite showed that a high distance heegaard splitting is a unique minimal heegaard splitting up to isotopy . by geometrilization theorem and hempel s work in @xcite , a 3-manifold @xmath31 admitting a distance at least 3 heegaard splitting is hyperbolic . from this view , studying heegaard distance is an active topic on heegaard splitting . the following is a brief survey on the existences of high distance heegaard splittings : hempel@xcite showed that for any integers @xmath1 , and @xmath44 , there is a 3-manifold which admitting a distance at least @xmath3 heegaard splitting of genus @xmath4 . similar results are obtained in different ways by @xcite and @xcite . minsky , moriah and schleimer @xcite proved the same result for knot complements , and li@xciteconstructed the non - haken manifolds admitting high distance heegaard splittings . in general , generic heegaard splittings have heegaard distances at least @xmath3 for any @xmath44 , see @xcite,@xcite,@xcite . by studying dehn filling , ma , qiu and zou@xcite proved that distances of genus 2 heegaard splittings cover all non - negative integers except 1 . recently , ido , jang and kobayashi@xcite proved that , for any @xmath45 and @xmath46 , there is a compact 3-manifold with two boundary components which admits a distance @xmath3 heegaard splitting of genus @xmath4 . and johnson@xcite proved that there are always existing closed 3-manifold admitting a distance @xmath47 , genus @xmath4 heegaard splitting . the main result of this paper is the following : * theorem 1 . * for any integers @xmath0 and @xmath1 , there is a closed 3-manifold @xmath2 which admits a distance @xmath3 heegaard splitting of genus @xmath4 except that the pair of @xmath5 is @xmath6 . furthermore , @xmath2 can be chosen to be hyperbolic except that the pair of @xmath5 is @xmath7 . * remark on theorem 1 . * ( 1 ) it is well known that there is not a distance 1 heegaard splitting of genus 2 . \(2 ) by the above argument , a 3-manifold admitting a distance at least 3 heegaard splitting is hyperbolic . hempel @xcite showed that any heegaard splitting of a seifert 3-manifold has distance at most 2 . now a natural question is : for any integer @xmath1 , is there a closed hyperbolic 3-manifold admitting a distance 2 heegaard splitting of genus g ? suppose first that @xmath48 . eudave - munoz@xcite proved that there is a hyperbolic @xmath49-knot in 3-sphere , say @xmath50 . in this case , the complement of @xmath50 , say @xmath51 , admits a distance 2 heegaard splitting of genus @xmath52 . by the main results in [ 1 ] , [ 14 ] and [ 31 ] , there is a slope @xmath53 on @xmath54 such that the manifold obtained by doing a surgery on @xmath55 along @xmath53 , say @xmath56 , is still hyperbolic . hence @xmath56 admits a distance 2 heegaard splitting of genus @xmath52 . maybe the answer to this question has been well known when @xmath57 . however we did not find published papers related to it . \(3 ) if @xmath31 admits a distance 1 heegaard splitting of genus 3 , then @xmath31 contains an essential torus . hence @xmath31 is not hyperbolic . \(4 ) the proof of theorem 1 implies the following fact : let @xmath3 be a positive integer , @xmath58 be a collection of closed orientable surfaces , and @xmath59 and @xmath60 be two subsets of @xmath61 . then , for any integers @xmath62 and @xmath63 , there is a compact 3-manifold @xmath31 admitting a distance @xmath64 heegaard splitting of genus @xmath4 , say @xmath35 , such that @xmath65 for @xmath66 , and @xmath67 for @xmath68 . we omit the proof . under the arguments in theorem 1 , we have the following result : * theorem 2 . * for any integers @xmath1 and @xmath8 , there are infinitely many non - homeomorphic closed 3-manifolds admitting distance @xmath3 heegaard splittings of genus @xmath4 . we organize this paper as follows : section 2 is devoted to introduce some results on curve complex . then we will prove theorem 1 when @xmath69 in section 3 , theorem 1 when @xmath70 in section 5 , and theorem 2 in section 4 . let @xmath9 be a compact surface of genus at least 1 , and @xmath71 be the curve complex of @xmath9 . we call a simple closed curve @xmath72 in @xmath9 is essential if @xmath72 bounds no disk in @xmath9 and is not parallel to @xmath73 . hence each vertex of @xmath74 is represented by the isotopy class of an essential simple closed curve in @xmath9 . for simplicity , we do not distinguish the essential simple closed curve @xmath72 and its isotopy class @xmath72 without any further notation . the following lemma is well known , see @xcite , @xcite , @xcite . * lemma 2.1 . * @xmath71 is connected , and the diameter of @xmath71 is infinite . we call a collection @xmath75 is a geodesic in @xmath74 if each @xmath76 and @xmath77 , for any @xmath78 . and the length of @xmath79 is denoted by @xmath80 is defined to be @xmath3 . by the connection of @xmath81 , there is always a shortest path in @xmath82 connecting any two vertices of @xmath74 . thus for any two distance @xmath3 vertices @xmath83 , we call a geodesic @xmath79 connecting @xmath84 if @xmath85 . now for any two sub - simplicial complex @xmath86 , we call a geodesic @xmath79 realizing the distance of @xmath22 and @xmath23 if @xmath79 connecting an element @xmath87 and an element @xmath88 such that @xmath89 . let @xmath90 be a compact surface of genus at least 1 with non - empty boundary . similar to the definition of the curve complex @xmath91 , we can define the arc and curve complex @xmath92 as follows : each vertex of @xmath92 is the isotopy class of an essential simple closed curve or an essential properly embedded arc in @xmath90 , and a set of vertices form a simplex of @xmath92 if these vertices are represented by pairwise disjoint arcs or curves in @xmath90 . for any two disjoint vertices , we place an edge between them . all the vertices and edges form 1-skeleton of @xmath92 , denoted by @xmath93 . and for each edge , we assign it length 1 . thus for any two vertices @xmath94 and @xmath95 in @xmath93 , the distance @xmath96 is defined to be the minimal length of paths in @xmath93 connecting @xmath94 and @xmath95 . similarly , we can define the geodesic in @xmath92 . when @xmath90 is a subsurface of @xmath9 , we call @xmath90 is essential in @xmath9 if the induced map of the inclusion from @xmath97 to @xmath98 is injective . furthermore , we call @xmath90 is a proper essential subsurface of @xmath9 if @xmath90 is essential in @xmath9 and at least one boundary component of @xmath90 is essential in @xmath9 . for more details , see @xcite . so if @xmath90 is an essential subsurface of @xmath9 , there is some connection between the @xmath92 and @xmath74 . for any @xmath99 , there is a representative essential simple closed curve @xmath100 such that the intersection number @xmath101 is minimal . hence each component of @xmath102 is essential in @xmath90 or @xmath103 . now for @xmath104 , let @xmath105 be isotopy classes of the essential components of @xmath102 . it is well defined * * @xmath106**since for any two isotopy class @xmath107 and @xmath108 of @xmath94 which both intersect @xmath109 minimally , either \(1 ) @xmath110 . then they bounds an annulus @xmath111 in @xmath9 . hence either ( 1.1 ) @xmath112 . if @xmath107 is essential in @xmath90 , then @xmath113 . hence @xmath114 . and @xmath115 is isotopic to @xmath116 in @xmath90 . if both @xmath115 and @xmath117 are inessential , then the essential components of @xmath115 and @xmath117 are @xmath118 . or , ( 1.2 ) @xmath119 . by minimality of intersection number , @xmath120 consists of squares . it is not hard to see that each component @xmath115 ( resp . @xmath117 ) is essential . and each component @xmath115 is isotopic to one component of @xmath117 . the reverse is true . or , \(2 ) @xmath121 . since the intersection number @xmath122 , by bigon criterion ( proposition 1.7@xcite ) , there is always an innermost bigon @xmath123 bounded by @xmath124 in @xmath9 . since there is no proper bigon in @xmath123 bounded by @xmath125 and @xmath109 ( the minimality of @xmath126 and @xmath127 ) , we can isotopy @xmath107 and @xmath108 such that @xmath128 ( resp . @xmath129 ) bounds an annulus in @xmath9 and @xmath130 ( resp . @xmath131 ) is minimal . and the bigon @xmath123 is vanishing . following ( 1 ) , we get that any essential component of @xmath115(resp . @xmath117 ) is isotopic to an essential component of @xmath132 ( resp.@xmath133 ) . and the reverse is true . we can do it again and again until there is no such bigon . then it returns to ( 1).*@xmath134 * for any @xmath135 , @xmath136 if and only if @xmath137 is the essential boundary component of a closed regular neighborhood of @xmath138 . specially , let @xmath139 . now let @xmath140 . then the map @xmath141 links the @xmath92 and @xmath74 , which is the defined subsurface projection map in @xcite . we say @xmath99 cuts @xmath90 if @xmath142 . if @xmath94 , @xmath143 both cut @xmath90 , we write @xmath144 . and if @xmath145 , then @xmath146 and @xmath147 , observed by h.masur and y.n.minsky at first . what if the two vertices @xmath94 and @xmath95 has distance @xmath148 in @xmath74 ? the following is immediately followed from the above observation . * lemma 2.2 . * let @xmath90 and @xmath9 be as above , @xmath149 be a geodesic of @xmath74 such that @xmath150 cuts @xmath90 for each @xmath151 . then @xmath152 . however , when @xmath148 is quite large , the lemma 2.2 can not provide more information . in general , masur - minsky @xcite proved the following result called bounded geodesic image theorem . * lemma 2.3 . * let @xmath90 be an essential sub - surface of @xmath9 , and @xmath153 be a geodesic segment in @xmath74 , such that @xmath154 for every vertex @xmath155 of @xmath153 . then there is a constant @xmath156 depending only on @xmath9 so that @xmath157 . when @xmath9 is closed with @xmath158 , there is always a compact 3-manifold @xmath31 with @xmath9 as its compressible boundary . let @xmath159 , called disk set for @xmath9 , be the subset of vertices of @xmath11 , where each element bounds a disk in @xmath31 . now an essential simple closed curve on @xmath9 , say @xmath72 , is said to be disk - busting if @xmath160 is incompressible in @xmath31 . since any two essential disks intersect in a typical way , it provides more information to study the subsurface projection of disk complex . the following disk image theorem is proved by t.li @xcite , h.masur and s.schleimer @xcite independently . * lemma 2.4 . * let @xmath31 be a compact orientable and irreducible 3-manifold . s is a boundary component of @xmath31 . suppose @xmath161 is incompressible . let @xmath162 be the disk complex of @xmath9 , @xmath163 be an essential subsurface . assume each component of @xmath109 is disk - busting . then either \(1 ) m is an i - bundle over some compact surface , @xmath90 is a horizontal boundary of the i - bundle and the vertical boundary of this i - bundle is a single annulus . or , \(2 ) the image of this complex , @xmath164 , lies in a ball of radius 3 in @xmath92 . in particular , @xmath164 has diameter 6 in @xmath92 . moreover , @xmath165 has diameter at most 12 in @xmath91 . * note . * for any i - bundle @xmath166 over a bounded compact surface @xmath167 , @xmath168 , where the vertical boundary @xmath169 is the i - bundle related to @xmath170 , and the horizontal boundary @xmath171 is the portion of @xmath172 transverse to the i - fibers . on the other side , j.hempel @xcite defined a full simplex @xmath22 on @xmath9 to be a dimension @xmath173 simplex in @xmath74 . hence , after attaching 2-handles and 3-handles along the vertices of @xmath22 in the same side of @xmath9 from the same side , we can get a handlebody , denoted by @xmath174 . * lemma 2.5 @xcite . * let @xmath9 be a closed , orientable surface of genus at least 2 . for any positive number @xmath175 , any full simplex @xmath22 of @xmath74 , there is another full simplex @xmath23 of @xmath74 such that @xmath176 . through subsurface projection , the bounded geodesic image theorem links the geodesic in curve complex and a proper subsurface . the example 1.5 @xcite shows that there is a geometry rigidity in curve complex . with the property of infinity of diameter of curve complex , we can construct any long geodesic in curve complex . furthermore , we also require that the constructed geodesic satisfying some condition , such as the first and last vertices are represented by separating essential simple closed curves . we organize our results as the following lemma which is a more stronger version of lemma 4.1 in @xcite . * lemma 2.6 . * let @xmath177 be be five integers such that @xmath178 , and @xmath179 . let @xmath180 be a closed surface of genus @xmath4 . then there are two essential separating curves @xmath94 and @xmath95 in @xmath180 such that @xmath181 , one component of @xmath182 has genus @xmath183 while one component of @xmath184 has genus @xmath185 . furthermore , there is a geodesic @xmath186 in @xmath187 such that \(1 ) @xmath188 is non - separating in @xmath180 for @xmath189 , and \(2 ) @xmath190 , where @xmath191 is the surface @xmath192 for @xmath193 . * let @xmath94 be an essential separating curve in @xmath9 such that one component of @xmath182 , say @xmath194 , has genus @xmath183 . suppose first that @xmath70 . let @xmath195 be a non - separating curve in @xmath180 which is disjoint from @xmath94 . let@xmath196 be the surface @xmath197 , where @xmath198 is a open regular neighborhood of @xmath195 in @xmath180 . then @xmath196 is a genus @xmath199 surface with two boundary components . furthermore , @xmath94 is an essential separating simple closed curve in @xmath196 . by lemma 2.1 , @xmath200 is connected and its diameter is infinite . hence there is an essential simple closed curve @xmath72 in @xmath196 with @xmath201 . note that @xmath202 . if @xmath72 is separating in @xmath196 , there is a non - separating essential simple closed curve @xmath203 in @xmath196 such that @xmath204 . hence @xmath205 , and @xmath206 . it means there is a non - separating slope @xmath72 in @xmath196 such that @xmath207 . let @xmath208 be a non - separating simple closed curve in @xmath196 such that @xmath208 intersects @xmath72 in one point , and @xmath209 be the boundary of the regular neighborhood of @xmath210 . then @xmath209 bounds a once - punctured torus @xmath211 containing @xmath208 and @xmath72 . since @xmath212 , there is an essential separating simple closed curve @xmath95 such that @xmath95 bounds a once - punctured surface of genus @xmath185 containing @xmath211 as a sub - surface . see figure 1 . figure 1 it is easy to see that @xmath95 is also separating in @xmath180 . now we prove that @xmath213 , and @xmath214 . since @xmath215 , @xmath216 and @xmath217 , @xmath218 , and @xmath219 . since @xmath220 , by the assumption on @xmath221 , @xmath222 . hence @xmath223 . for if @xmath224 , then , by lemma 2.3 , @xmath225 , a contradiction . similarly , @xmath226 . and @xmath227 and @xmath228 are two geodesics of @xmath229 . furthermore , @xmath79 satisfies the conclusion of lemma 2.6 . now we prove lemma 2.6 by induction on @xmath3 . * assumption . * let @xmath230 . suppose that there are two essential separating simple closed curves @xmath94 and @xmath95 , and a non - separating simple closed curve @xmath72 in @xmath180 such that @xmath231 , @xmath232 , and one component of @xmath182 has genus @xmath183 while one component of @xmath184 has genus @xmath185 . furthermore , there is a geodesic @xmath233 satisfying lemma 2.6(1 ) and ( 2 ) , and a geodesic @xmath234 is also a geodesic connecting @xmath94 to @xmath72 satisfying @xmath235 , for any @xmath236 and @xmath237 , let @xmath238 be the surface @xmath239 , where @xmath240 is a open regular neighborhood of @xmath72 in @xmath180 . since @xmath72 is non - separating in @xmath180 , @xmath238 is a genus @xmath199 surface with two boundary components . since @xmath234 is also a geodesic connecting @xmath94 to @xmath72 , @xmath241 is an essential non - separating simple closed curve in @xmath238 . by the above argument , there are a non - separating curve @xmath242 and a separating curve @xmath209 in @xmath238 such that @xmath209 bounds a once - punctured torus @xmath243 containing @xmath242 , @xmath244 , and @xmath245 . and there is aslo an essential separating simple closed curve @xmath153 which bounds a genus @xmath185 sub - surface of @xmath238 containing @xmath243 as a sub - surface . not hard to see @xmath153 is also separating in @xmath180 . since @xmath242 is disjoint from @xmath153 , @xmath246 now we prove that @xmath247 , @xmath248 . suppose , otherwise , that @xmath249 . then there exists a geodesic line @xmath250 . note that each of @xmath94 and @xmath242 is not isotopic to @xmath72 . with the length is less then or equal to @xmath50 . since @xmath232 , @xmath251 is not isotopic to @xmath72 for @xmath252 . that means @xmath251 cuts @xmath238 for each @xmath253 . by lemma 2.3 , @xmath254 . on the other side , since @xmath232 , @xmath255 is not isotopic to @xmath72 for @xmath256 . by using lemma 2.3 again , @xmath257 . then @xmath258 . it contradicts the choice of @xmath242 . now @xmath259 and @xmath260 are two geodesics satisfying the assumption . hence lemma 2.6 holds . end . in this section , we will prove the following proposition : * proposition 3.1 . * for any positive integers @xmath69 and @xmath1 , there is a closed 3-manifold which admits a distance @xmath3 heegaard splitting of genus @xmath4 except that the pair of @xmath5 is @xmath6 . furthermore , @xmath2 can be chosen to be hyperbolic except that the pair of @xmath5 is @xmath7 . * proof . * we first suppose that @xmath261 . let @xmath9 be a closed surface of genus @xmath4 . by lemma 2.6 , there are two separating essential simple closed curves @xmath94 and @xmath95 such that @xmath262 for @xmath261 . let @xmath32 be the compression body obtained by attaching a 2-handle to @xmath263 $ ] along @xmath264 , and @xmath33 be the compression body obtained by attaching a 2-handle to @xmath265 $ ] along @xmath266 . then @xmath41 is a heegaard splitting where @xmath9 is the surface @xmath267 , see figure 2 . since @xmath32 contains only one essential disk @xmath123 with @xmath268 up to isotopy , and @xmath33 contains only one disk @xmath269 with @xmath270 up to isotopy , @xmath271 . figure 2 let @xmath272 and @xmath273 be the components of @xmath274 , and @xmath194 and @xmath275 be the two components of @xmath276 . similarly , let @xmath277 and @xmath278 be the components of @xmath279 , and @xmath280 and @xmath281 be the two components of @xmath282 . now @xmath123 cuts @xmath32 into two manifolds @xmath283 and @xmath284 , and @xmath269 cuts @xmath33 into two manifolds @xmath285 and @xmath286 . see figure 2 . by lemma 2.6 , we may assume that @xmath280 is a once - punctured torus . we first consider the compression body @xmath32 . we may assume that @xmath287 , @xmath288 for @xmath289 . let @xmath290 be the natural homeomorphism such that @xmath291 for @xmath292 and @xmath293 . no doubt that @xmath294 is well defined . then , for any two essential simple closed curves @xmath295 , @xmath296 for @xmath292 . see figure 3 . hence @xmath294 induces an isomorphism from @xmath297 to @xmath298 , for any @xmath299 . denote the isomorphism by @xmath294 too . note that the shadow disk in figure 3 is @xmath123 . figure 3 let @xmath300 be the inclusion map for @xmath292 . note that @xmath301 contains only one component . if @xmath72 is an essential simple closed curve in @xmath302 , @xmath303 is also essential in @xmath304 . now , for any two essential simple closed curves @xmath305 @xmath306 for @xmath292 . hence @xmath307 induces a distance non - increasing map from @xmath308 to @xmath297 , for any @xmath299 . denote the inclusion map by @xmath307 too . then we can define a projection map : @xmath309 . since @xmath310 , @xmath311 . by the argument in section 2 , @xmath312 . hence @xmath313 . we start to attach a handlebody to @xmath32 along @xmath272 . then either \(1 ) @xmath272 is a torus . by lemma 2.1 , there is an essential simple closed curve @xmath53 in @xmath272 such that @xmath314 . let @xmath315 be a solid torus such that @xmath316 , and @xmath53 bounds a disk in @xmath315 . in this case , @xmath315 contains only one essential disk up to isotopy . let @xmath317 be the manifold @xmath318 . or , \(2 ) @xmath319 . by lemma 2.5 , there is a full simplex @xmath22 on of @xmath320 such that @xmath321 , where @xmath174 is the handlebody obtained by attaching 2-handles to @xmath272 along @xmath22 then 3-handles to cap off the possible 2-spheres . in this case , we denote by @xmath317 the manifold @xmath322 . in whole words , @xmath317 is a compression body with only one minus boundary component @xmath273 . see figure 4 . hence @xmath323 is a heegaard splitting . figure 4 * claim 3.2 . * the heegaard distance of @xmath323 , say @xmath324 , is @xmath3 . * proof . * suppose , otherwise , that @xmath325 . since @xmath33 contains only one essential disk @xmath269 up to isotopy such that @xmath270 , there is an essential disk @xmath326 in @xmath317 such that @xmath327 , i.e , there is a geodesic @xmath328 , where @xmath329 . * fact 3.3 . * @xmath330 , for any @xmath331 . suppose that @xmath332 for some @xmath331 . @xmath333 since @xmath334 , and @xmath335 and if @xmath336 , then @xmath337 , and @xmath326 is inessential in @xmath317 . @xmath338 since @xmath339 . hence there is a geodesic @xmath340 . it means that @xmath341 , a contradiction . by lemma 2.3 , furthermore , @xmath343 . depending on the way of intersection between @xmath326 and @xmath123 , either \(1 ) @xmath344 . since @xmath326 is not isotopic to @xmath123 , @xmath345 bounds an essential disk in @xmath174 or @xmath315 depending on @xmath346 , where @xmath174 and @xmath315 are constructed as above . it contradicts the choice of @xmath22 or @xmath53 . or , \(2 ) @xmath347 . let @xmath348 be an outermost arc of @xmath349 on @xmath326 . it means that @xmath348 , together with a sub - arc @xmath350 , bounds a disk @xmath351 such that @xmath352 . since @xmath123 cuts @xmath317 into a handlebody @xmath353 which contains @xmath272 and a @xmath59-bundle @xmath284 , @xmath354 . hence @xmath345 bounds an essential disk in @xmath174 or @xmath315 . by the argument in ( 1 ) , it is impossible . end.(claim 1 ) figure 5 now @xmath317 is a compression body which has only one minus boundary component @xmath273 . since @xmath310 , @xmath355 . by lemma 2.6 , there is always a full simplex @xmath23 on @xmath273 such that @xmath356 , where @xmath357 is the handlebody obtained by attaching 2-handles to @xmath273 along @xmath23 then 3-handles to cap off the possible 2-spheres , and @xmath358 is defined as before . let @xmath359 be the manifold obtained by attaching @xmath357 to @xmath317 along @xmath273 . see figure 5 . then @xmath359 is a handlebody . hence @xmath360 is also a heegaard splitting . * claim 3.4 . * the heegaard distance of @xmath360 , said @xmath361 , is @xmath3 . * proof . * suppose , otherwise , that @xmath362 . since @xmath33 contains only one essential disk @xmath269 up to isotopy such that @xmath270 , there is an essential disk @xmath363 in @xmath359 such that @xmath364 , i.e , there is a geodesic @xmath365 , where @xmath329 . by the proof of claim 1 , @xmath366 for @xmath256 . note that @xmath268 . depending on the way of intersection between @xmath363 and @xmath123 , either \(1 ) @xmath367 . since @xmath368 , @xmath363 is not isotopic to @xmath123 . by the proof of claim 1 , @xmath369 does not lie in @xmath194 . hence @xmath370 . it implies that @xmath371 bounds an essential disk in @xmath357 . by lemma 2.3 , @xmath372 . hence @xmath373 , and @xmath374 . it contradicts the choice of @xmath23 . or , \(2 ) @xmath375 . let @xmath376 be an outermost arc of @xmath377 on @xmath363 . this means that @xmath376 , together with a sub - arc @xmath378 , bounds a disk @xmath379 such that @xmath380 . by the proof of claim 4.1 ( 2 ) , @xmath381 . thus @xmath371 bounds an essential disk in @xmath357 . by the same argument in claim 1 , it is impossible . end . ( claim 2 ) until now , we get a distance @xmath3 genus @xmath4 heegaard splitting @xmath360 . in this case , @xmath359 is a handlebody , and @xmath33 contains only one essential disk @xmath269 such that @xmath270 . furthermore , we can cut @xmath9 along @xmath95 into two components @xmath280 and @xmath281 , and cut @xmath33 along @xmath269 into two manifolds @xmath285 and @xmath286 such that @xmath287 , and @xmath382 for @xmath383 . now the shadow disk in figure 3 is @xmath269 . let @xmath384 be the natural homeomorphism such that @xmath291 for @xmath383 . then , for any two essential simple closed curves @xmath385 , @xmath386 for @xmath383 , see figure 3 . hence @xmath294 induces an isomorphism from @xmath297 to @xmath298 , for any @xmath299 . denote the isomorphism by @xmath294 too . let @xmath387 be the inclusion map for @xmath383 . note that @xmath301 contains only one component . if @xmath72 is an essential simple closed curve in @xmath302 , @xmath303 is also essential in @xmath388 . now , for any two essential simple closed curves @xmath305 @xmath389 for @xmath383 . hence @xmath307 induces a distance non - increasing map from @xmath308 to @xmath297 , for any @xmath299 . denote the inclusion map by @xmath307 too . then we can define a projection map : @xmath309 . since @xmath69 , there are two cases : * case 1 . * @xmath261 . since @xmath360 is a distance @xmath3 heegaard splitting of genus @xmath4 , and @xmath33 contains only an essential disk @xmath269 up to isotopy , @xmath280 and @xmath281 are incompressible in @xmath359 . hence @xmath390 is disk - busting in @xmath359 . since @xmath57 , and @xmath391 , @xmath359 is not an i - bundle over some compact surface with @xmath302 a horizontal boundary of the i - bundle , and the vertical boundary of this i - bundle a single annulus for @xmath383 . by lemma 2.4 , @xmath392 for @xmath383 . hence @xmath393 . since @xmath277 is a torus , by lemma 2.1 , there is an essential simple closed curve @xmath394 in @xmath277 such that @xmath395 . let @xmath396 the be the manifold obtained attaching a solid @xmath397 to @xmath33 along @xmath277 so that @xmath394 bounds a disk in @xmath397 . then @xmath396 is a compression body . since @xmath57 , @xmath398 . by lemma 2.5 , there is a full simplex @xmath399 of @xmath400 such that @xmath401 , where @xmath402 is the handlebody obtained by attaching 2-handles to @xmath278 along @xmath399 then 3-handles to cap off the possible 2-spheres . in this case , let @xmath403 be the handlebody @xmath404 . now @xmath405 is a heegaard splitting of a closed 3-manifold . figure 6 * claim 3.5 . * the distance of @xmath405 , said @xmath406 , is @xmath3 . * proof.*let @xmath269 be the essential disk in @xmath407 bounded by @xmath95 . suppose , otherwise , that @xmath408 . then there is a geodesic @xmath409 , where @xmath329 , @xmath326 is a disk in @xmath410 , and @xmath411 is a disk in @xmath412 . @xmath413 , for any @xmath414 for if not , the distance of @xmath415 would be at most @xmath416 . similarly , @xmath411 is not isotopic to @xmath269 . then either \(1 ) @xmath417 . then @xmath418 lies in one of @xmath280 and @xmath281 , say @xmath280 . hence @xmath419 . by lemma 2.3 , @xmath420 . since @xmath421 , + @xmath422 . hence @xmath423 , a contradiction . or , \(2 ) @xmath424 , let @xmath72 be an outermost arc of @xmath377 on @xmath363 . this means that @xmath72 , together with a sub - arc @xmath425 , bounds a disk @xmath426 such that @xmath427 . we may assume that @xmath428 . by lemma 2.3 , @xmath429 . hence @xmath430 . note that @xmath431 . by the same argument in ( 1 ) , it is impossible . now we suppose that @xmath432 . let @xmath433 and @xmath434 be two 3-manifolds with homeomorphic connected boundary . let @xmath435 be the manifold obtained by gluing @xmath433 and @xmath434 along a homeomorphism from @xmath436 to @xmath437 . let @xmath438 be a minimal heegaard splitting for @xmath292 . in this case , @xmath435 has a natural heegaard called the amalgamation of @xmath439 and @xmath440 . the following facts are well known : \(1 ) if the gluing map @xmath441 is enough complicated , then the amalgamation of @xmath439 and @xmath440 is unstabilized , see @xcite , @xcite , @xcite , @xcite . \(2 ) if both @xmath439 and @xmath440 have high distance , then the amalgamation of @xmath439 and @xmath440 is unstabilized , see @xcite , @xcite . now let @xmath438 be a heegaard splitting of genus two such that @xmath442 is a torus , and @xmath443 for @xmath292 , then , by the main result in @xcite , the amalgamation of @xmath439 and @xmath440 , say @xmath41 , is unstabilized . furthermore , @xmath444 . suppose that @xmath445 . by the above argument , there are a heegaard splitting @xmath446 of genus @xmath199 such that @xmath447 , and @xmath448 , and a heegaard splitting @xmath440 of genus 3 such that @xmath449 , and @xmath450 . hence both @xmath433 and @xmath434 are hyperbolic . by the main result in @xcite , the amalgamation of @xmath439 and @xmath440 , say @xmath35 , is unstabilized . furthermore , @xmath451 . by thurston s theorem , @xmath31 is hyperbolic . end(proposition 3.1 ) * remark . * the strongly irreducible heegaard splitting @xmath41 where both @xmath32 and @xmath33 contain only one essential separating disk up to isotopy independently is always a minimal heegaard splitting of @xmath452 . @xcite defined a sub - complex @xmath453 , for @xmath454 and proved that for any handlebody @xmath353 attached to @xmath31 along @xmath272 , if @xmath455 is larger than a constant @xmath456 which depends on @xmath457 and @xmath353 , then the new generated heegaard splitting @xmath323 is still the minimal heegaard splitting of @xmath458 . similar to the other boundaries of @xmath31 . now in our construction of distance @xmath44 strongly irreducible heegaard splitting ( for n=2 , see section 5 ) , we can choose a full simplex @xmath22 in @xmath272 such that @xmath459 is large enough and @xmath460 is larger than @xmath456 . then the new heegaard splitting @xmath323 is still the minimal heegaard splitting of @xmath458 and has the same distance as the older one . we will prove theorem 2 in this section . * theorem 2 . * for any integers @xmath1 and @xmath8 , there are infinitely many non - homeomorphic closed 3-manifolds which admit distance @xmath3 heegaard splittings of genus @xmath4 . * proof . * let @xmath180 be a closed surface of genus @xmath4 . by lemma 2.6 , for each @xmath63 , there is a geodesic @xmath461 in @xmath187 such that \(1 ) @xmath188 is non - separating in @xmath180 for @xmath189 , @xmath94 and @xmath462 are two essential separating simple closed curves on @xmath180 for @xmath63 , \(2 ) @xmath190 , where @xmath191 is the surface @xmath192 for @xmath193 , and \(3 ) one component of @xmath463 has genus one . without loss of generality , we assume that @xmath464 . let @xmath465 be the manifold obtained by attaching two 2-handles to @xmath466 $ ] along @xmath467 and @xmath468 . we denote also by @xmath180 the surface @xmath469 . now @xmath465 has a heegaard splitting as @xmath470 , where @xmath471 is the compression body obtained by attaching a 2-handle to @xmath265 $ ] along @xmath467 , and @xmath472 is the manifold obtained by attaching a 2-handle to @xmath263 $ ] along @xmath468 . then @xmath473 contains two components @xmath272 and @xmath273 , and @xmath474 contains two components @xmath475 and @xmath476 . see figure 7 . furthermore , one component of @xmath475 and @xmath476 has genus one . figure 7 by the proof of theorem 1(1 ) , there is a closed 3-manifold @xmath477 which admits a distance @xmath3 heegaard splitting @xmath478 , where @xmath479 is obtained by attaching handlebodies @xmath480 and @xmath481 to @xmath471 along @xmath272 and @xmath273 , and @xmath482 is obtained by attaching handlebodies @xmath483 and @xmath484 to @xmath472 along @xmath475 and @xmath476 such that \(1 ) @xmath485 for @xmath292 , and \(2 ) @xmath486 for @xmath383 . replace @xmath477 , @xmath479 and @xmath482 by @xmath465 , @xmath471 and @xmath472 . now @xmath487 is also a geodesic of @xmath229 realizing the distance of @xmath488 . * claim 4.2 . * let @xmath489 be a geodesic of @xmath229 realizing the distance of @xmath470 . then @xmath490 for any @xmath491 . * * let @xmath194 and @xmath275 be the two components of @xmath182 . we assume that @xmath492 bounds a disk @xmath493 in @xmath471 , and @xmath494 bounds a disk @xmath495 in @xmath472 . we first prove that @xmath94(resp . @xmath462 ) is disjoint from @xmath496 ( resp . @xmath497 ) . suppose , otherwise , that @xmath498 . hence @xmath492 is not isotopic to @xmath499 . then either \(1 ) @xmath500 . let @xmath348 be an outermost arc of @xmath501 on @xmath493 . it means that @xmath348 , together a sub - arc of @xmath502 , bounds a disk @xmath351 such that @xmath352 . without assumption , we may assume that @xmath503 . by the argument in section 3 , @xmath504 bounds an essential disk in @xmath480 . but with @xmath505 , it implies that @xmath506 . hence @xmath507 . or , \(2 ) @xmath508 . by @xmath509 , @xmath493 is not isotopic to @xmath123 . then @xmath510 is essential in @xmath194 or @xmath275 . we assume that @xmath511 . the other case is similar . hence by ( 1 ) , @xmath507 . however , by heegaard distance is at least 4 and @xmath512 bounds an essential disk in @xmath513 , it means that @xmath94 is disk - busting for @xmath514 and @xmath514 can not be the i - bundle of compact surface with @xmath194 or @xmath275 as one of its horizontal boundary . then by lemma 2.4 , @xmath515 and @xmath516 . hence @xmath517 and @xmath518 . together with ( 1 ) and ( 2 ) , by triangle inequality , @xmath519 . it contradicts the choice of @xmath520 in @xmath272 . the other case is similar . let @xmath521 be a new geodesic realizing the distance of @xmath522 . now we prove that @xmath496 is isotopic to @xmath523 . suppose , otherwise , that @xmath496 is not isotopic to @xmath523 . note that @xmath524 is not isotopic to @xmath523 . otherwise , the distance of @xmath525 would be at most @xmath526 . let @xmath527 be the surface @xmath528 , where @xmath529 is a open regular neighborhood of @xmath523 on @xmath180 . by lemma 2.3 , @xmath530 . now let s consider the shorter geodesic @xmath531 which is a sub - geodesic of @xmath461 . due to the definition of geodesic in curve complex , @xmath532 is not isotopic to @xmath523 for any @xmath533 . by lemma 2.3 again , @xmath534 . hence @xmath535 . this contradicts our assumption on @xmath536 . hence @xmath496 is isotopic to @xmath523 . by induction on @xmath537 , the claim holds . end ( claim 4.2 ) replace @xmath488 by @xmath538 . the following claim reveals the connection between geodesics in curve complex and closed 3-manifolds . * claim 4.3 . * for any @xmath539 , either \(1 ) @xmath540 and @xmath541 are two different 3-manifolds up to homeomorphism . or , \(2 ) @xmath542 is homeomorphic to @xmath543 , but @xmath544 and @xmath545 are two different heegaard splittings of @xmath542 up to homeomorphic equivalence . * suppose that @xmath542 is homeomorphic to @xmath543 for some @xmath539 . if ( 2 ) fails , then @xmath544 and @xmath545 are homeomorphic . it means that there is a homeomorphism @xmath441 from @xmath542 to @xmath543 such that @xmath546 . we assume that @xmath547 and @xmath548 . the other case is similar . it is well known that @xmath441 induces an isomorphism from @xmath549 to @xmath550 , still denoted by @xmath441 . then for the geodesic @xmath551 which realizes the distance of @xmath552 , @xmath553 is also a geodesic in @xmath550 realizing the distance of @xmath545 . by claim 4.2 , @xmath554 is isotopic to @xmath555 for @xmath556 . as @xmath8 , we choose @xmath557 . since @xmath558 is isotopic to @xmath559 , we can perform an isotopy on @xmath560 such that the composition of @xmath441 with the isotopy gives an homeomorphism @xmath561 from @xmath562 to @xmath563 and @xmath564 . even more , @xmath565 and @xmath566 . it s still true that @xmath561 induces an automorphism from @xmath549 to @xmath550 , denoted by @xmath561 too . thus @xmath567 is also a geodesic realizing the distance of @xmath545 . by claim 4.2 again , for any @xmath568 , @xmath569 is still isotopic to @xmath555 . hence @xmath570 ( resp . @xmath571 ) is isotopic to @xmath572 ( resp . @xmath573 ) . let @xmath574 be the surface @xmath575 , where @xmath576 is an open regular neighborhood of @xmath557 on @xmath577 , and @xmath578 be the surface of @xmath579 . then @xmath580 and @xmath581 is a homeomorphism . hence @xmath561 also induces an isomorphism from @xmath582 to @xmath583 , still denoted by @xmath561 . now we can also assume @xmath584 and @xmath585 . thus @xmath586 and @xmath587 . then @xmath588 . on the other side , @xmath570 ( resp.@xmath589 ) must be isotopic to @xmath572 ( resp.@xmath590 ) in @xmath578 for @xmath591 if not , then after removing possible bigon capped by them , they bounds no annulus in @xmath578 , thus they bounds no annulus and bigon in @xmath560 . by bigon criterion ( proposition 1.7@xcite ) , they realizes the geometry intersection number . since they are isotopic in @xmath560 , they must be disjoint in @xmath560 . hence they must bounds an annulus in @xmath560 @xmath592 . so @xmath593 . however , by the assumption [ @xmath594 , @xmath595 and @xmath464 ] , @xmath596 , a contradiction . end ( claim 4.3 ) the waldhausen conjecture proved by johanson ( @xcite,@xcite ) and li @xcite implies that , for any positive integer @xmath4 , an atoroidal closed 3-manifold @xmath31 admits only finitely many heegaard splittings of genus @xmath4 up to homeomorphism . since @xmath542 admits a heegaard splitting with distance at least 4 , it is atoroidal for any @xmath597 , see @xcite and @xcite . now theorem 2 is immediately from claim 2 and the waldhausen conjecture . we rewrite the second part of theorem 1 as the following proposition : let @xmath32 be the compression body obtained by attaching a separating 2-handle to @xmath263 $ ] along @xmath264 , and @xmath33 be the compression body obtained by attaching a separating 2-handle to @xmath265 $ ] along @xmath604 . denote @xmath267 by @xmath9 too . then @xmath41 is a heegaard splitting . since @xmath32 contains only one essential disk @xmath123 with @xmath268 up to isotopy , and @xmath33 contains only one essential disk @xmath269 with @xmath605 up to isotopy , @xmath606 . let @xmath272 and @xmath273 be the components of @xmath274 , such that @xmath607 is homeomorphic to @xmath304 for @xmath292 . similarly , let @xmath277 and @xmath278 be the components of @xmath279 such that @xmath607 is homeomorphic to @xmath388 for @xmath383 . then both @xmath194 and @xmath280 are once - punctured tori , and @xmath272 and @xmath277 are two tori , see figure 2 . furthermore , both @xmath277 and @xmath278 have genus at least 2 . now @xmath123 cuts @xmath32 into two manifolds @xmath283 and @xmath284 , and @xmath269 cuts @xmath33 into two manifolds @xmath285 and @xmath286 . \(2 ) let @xmath22 be a full complex of @xmath614 such that @xmath615 , where @xmath174 is the handlebody obtained by attaching 2-handles to @xmath273 along the vertices of @xmath22 then 3-handles to capping off the spherical boundary components . let @xmath616 , and @xmath410 be the handlebody obtained by doing a surgery on @xmath617 along the slope @xmath394 on @xmath272 . by assumption 1 , @xmath391 , @xmath618 , @xmath410 is not a @xmath59-bundle over a compact surface with @xmath302 as a horizontal boundary for @xmath383 . by lemma 2.4 , @xmath619 for @xmath383 . \(2 ) let @xmath23 be a full complex of @xmath621 such that @xmath622 , where @xmath357 is the handlebody obtained by attaching 2-handles to @xmath278 along the vertices of @xmath23 then 3-handles to capping off the spherical boundary components . let @xmath623 , and @xmath403 be the handlebody obtained by doing a surgery on @xmath624 along the slope @xmath53 on @xmath277 . now both @xmath625 and @xmath405 are heegaard splittings . furthermore , we can prove that these two heegaard splittings have distance 2 by using lemma 2.2 to take place of lemma 2.3 in the proof of proposition 3.1 . * proof . * suppose , otherwise , that @xmath626 contains an essential torus @xmath211 . since the distance of @xmath627 is 2 , @xmath627 is strongly irreducible . by schultens s lemma , each component of @xmath629 is essential on both @xmath211 and @xmath9 . hence each component of @xmath630 and @xmath631 is an essential annulus in @xmath617 or @xmath624 . now @xmath617 contains a @xmath628-compressing disk @xmath635 of @xmath632 . by doing a surgery on @xmath632 along @xmath635 , we can get a disk @xmath493 in @xmath617 . since @xmath632 is essential , @xmath493 is essential . suppose that the two components of @xmath633 are isotopic to @xmath95 . since @xmath95 is non - separating on @xmath9 , @xmath510 bounds a once - punctured torus containing @xmath95 , see figure 8 . by assumption 1 , @xmath636 . since @xmath275 has genus @xmath637 , @xmath510 is not isotopic to @xmath638 . by a standard outermost argument , @xmath639 bounds an essential disk in @xmath174 . therefore @xmath640 . since @xmath641 , @xmath642 . hence @xmath643 . it contradicts assumption 2 . we assume that @xmath654 . by the above argument , @xmath493 is an essential disk in @xmath617 such that @xmath510 is disjoint from @xmath634 . furthermore , @xmath510 is not isotopic to @xmath94 . since @xmath123 cuts @xmath617 into @xmath283 and a handlebody @xmath353 such that @xmath655 , @xmath656 . furthermore , all outermost disks of @xmath501 on @xmath493 lie in @xmath353 . hence @xmath657 bounds an essential disk in @xmath353 . this means @xmath639 bounds an essential disk in @xmath174 . since @xmath644 is an essential annulus in @xmath624 , there is an essential disk @xmath661 obtained by doing boundary compression on @xmath644 in @xmath624 . even more @xmath662 . since @xmath269 cuts @xmath624 into @xmath285 and a handlebody @xmath663 containing @xmath357 , all outermost disks of @xmath664 in @xmath661 lies in @xmath663 . hence @xmath665 bounds an essential disk in @xmath357 . hence @xmath666 . since @xmath662 , by lemma 2.2 , @xmath667 . according to the definition of @xmath668 , @xmath669 . recall that the essential disk @xmath493 is obtained by doing a surgery on @xmath632 along a @xmath628-compressing disk in @xmath617 . since the distance of @xmath627 is two , @xmath670 . since @xmath391 and @xmath618 , @xmath617 is not a i - bundle of compact surface with @xmath281 as one horizontal boundary , by lemma 2.4 , @xmath671 . hence @xmath672 . since @xmath673 , @xmath674 . it means that * proof . * suppose , otherwise , that @xmath626 contains an essential torus @xmath111 . since the distance of @xmath625 is 2 , @xmath625 is strongly irreducible . by schultens s lemma , each component of @xmath676 is essential on both @xmath111 and @xmath9 . hence each component of @xmath677 and @xmath678 is either a spanning annulus or an essential annulus with two boundary components lying on @xmath9 . there are four cases : in this case , let @xmath632 be one component of @xmath677 and @xmath678 such that each of the two components of @xmath677 and @xmath678 incident to @xmath632 has its two boundary components lying on @xmath9 . by the proof of claim 1 , the claim holds . now @xmath111 intersects @xmath9 in an essential simple closed curve @xmath348 . furthermore , @xmath348 , together with an essential simple closed curve @xmath681 on @xmath272 , bounds a spanning annulus @xmath644 in @xmath617 , and @xmath348 , together with an essential simple closed curve @xmath682 on @xmath277 , bounds a spanning annulus @xmath683 in @xmath624 . hence @xmath684 in @xmath685 . since @xmath123 cuts @xmath617 into @xmath283 and a handlebody @xmath353 containing @xmath95 , @xmath686 in @xmath685 . hence @xmath348 is not isotopic to @xmath95 . there are three sub - cases : now @xmath694 . let @xmath72 be an outermost arc of @xmath695 on @xmath269 . this means that @xmath72 , together with a sub - arc @xmath203 of @xmath696 , bounds a disk @xmath697 in @xmath269 such that @xmath698 . now we can obtain a disk @xmath661 by doing a surgery on @xmath683 along @xmath697 , say @xmath661 . hence @xmath699 . furthermore , @xmath661 is an essential disk in @xmath624 . otherwise , we can reduce @xmath700 . since @xmath269 cuts @xmath624 into @xmath285 and a handlebody @xmath663 containing @xmath281 . hence @xmath701 . furthermore , one outermost disk of @xmath664 on @xmath661 lies in @xmath663 . otherwise , we can reduce @xmath700 . hence @xmath665 bounds an essential disk in @xmath357 . by the assumption , @xmath702 . hence @xmath703 . since the distance of @xmath627 is 2 , @xmath704 . now by lemma 2.2 and the above argument , let @xmath661 be as in case 2.2 . since @xmath707 , @xmath708 . let @xmath195 be an outermost arc of @xmath709 on @xmath123 . this means that @xmath195 , together with a sub - arc @xmath710 of @xmath94 , bounds a disk @xmath635 in @xmath123 such that @xmath711 . now we can obtain a disk @xmath493 by doing a surgery on @xmath644 along @xmath635 , say @xmath493 . hence @xmath712 . similarly , @xmath493 is an essential disk in @xmath617 . since the distance of @xmath627 is 2 , @xmath713 . by lemma 2.4 , @xmath714 . by lemma 2.2 , @xmath715 . by the argument in case 2.2 , @xmath716 . now we have now @xmath626 is a hyperbolic 3-manifold , @xmath625 is a distance 2 heegaard splitting of genus @xmath4 . furthermore , @xmath626 contains two toral boundary components @xmath272 and @xmath277 . by the main results in [ 1 ] and [ 16 ] , there are at most ten slopes @xmath394 on @xmath272 such that the manifold @xmath721 obtained by doing dehn filling on @xmath626 along @xmath394 is non - hyperbolic . by assumption 2 , there are infinitely many slopes @xmath394 so that @xmath721 has a distance 2 heegaard splitting of genus @xmath4 . hence there is at least one slope @xmath394 on @xmath272 such that @xmath721 is hyperbolic and @xmath721 admits a distance 2 heegaard splitting of genus @xmath4 . similarly , by assumption 3 , there is a hyperbolic closed manifold which admits a distance 2 heegaard splitting of genus @xmath4 . end j.johnson , _ non - uniquess of high distance heegaard splittings _ , personal communication . t. kobayashi and r. qiu , _ the amalgamation of high distance heegaard splittings is always efficient _ , math . 341(2008 ) , 707 - 715 .

gravitational wave ( gw ) signals from coalescing binary neutron stars may soon become detectable by laser - interferometer detectors . using our new post - newtonian ( pn ) smoothed particle hydrodynamics ( sph ) code , we have studied numerically the mergers of neutron star binaries with irrotational initial configurations . these are the most physically realistic initial conditions just prior to merger , since the neutron stars in these systems are expected to be spinning slowly at large separation , and the viscosity of neutron star matter is too small for tidal synchronization to be effective at small separation . however , the large shear that develops during the merger makes irrotational systems particularly difficult to study numerically in 3d . in addition , in pn gravity , accurate irrotational initial conditions are much more difficult to construct numerically than corotating initial conditions . here we describe a new method for constructing numerically accurate initial conditions for irrotational binary systems with circular orbits in pn gravity . we then compute the 3d hydrodynamic evolution of these systems until the two stars have completely merged , and we determine the corresponding gw signals . we present results for systems with different binary mass ratios , and for neutron stars represented by polytropes with or . compared to mergers of corotating binaries , we find that irrotational binary mergers produce similar peak gw luminosities , but they shed almost no mass at all to large distances . the dependence of the gw signal on numerical resolution for calculations performed with sph particles is extremely weak , and we find excellent agreement between runs utilizing and sph particles ( the largest sph calculation ever performed to study such irrotational binary mergers ) . we also compute gw energy spectra based on all calculations reported here and in our previous works . we find that pn effects lead to clearly identifiable features in the gw energy spectrum of binary neutron star mergers , which may yield important information about the nuclear equation of state at extreme densities .

gr-qc0201040

coalescing neutron star ( ns ) binaries are likely to be one of the most important sources of gravitational radiation for the ground - based laser - interferometer detectors in ligo @xcite , virgo @xcite , geo600 @xcite , and tama @xcite . these interferometers are most sensitive to gw signals in the frequency range from about @xmath5 to @xmath6 , which corresponds to the last several thousand orbits of the inspiral . during this period , the binary orbit is decaying very slowly , with the separation @xmath7 and phase @xmath8 following the standard theoretical treatment for the inspiral of two point masses ( see , e.g. , @xcite ) . theoretical templates for the corresponding quasi - periodic gw signals covering an appropriate range of values for the ns masses , as well as orbital phases and inclination angles , can be calculated to great precision ( see , e.g. , @xcite and references therein ) , and template matching techniques can therefore be used to extract signals from noisy interferometer data . when the binary separation @xmath7 has decreased all the way down to a few ns radii , the system becomes dynamically unstable @xcite and the two stars merge hydrodynamically in @xmath9ms . the characteristic gw frequency of the final burst - like signal is @xmath10 , outside the range accessible by current broadband detectors . these gw signals may become detectable , however , by the use of signal recycling techniques , which provide increased sensitivity in a narrow frequency band @xcite . these techniques are now being tested at geo600 , and will be used by the next generation of ground - based interferometers . point - mass inspiral templates break down during the final few orbits . instead , 3d numerical hydrodynamic calculations are required to describe the binary merger phase and predict theoretically the gw signals that will carry information about the ns equation of state ( eos ) . the first hydrodynamic calculations of binary ns mergers in newtonian gravity were performed by nakamura , oohara and collaborators using a grid - based , eulerian finite - difference code @xcite . rasio and shapiro ( @xcite , hereafter rs ) later used lagrangian sph calculations to study both the stability properties of close ns binaries and the evolution of dynamically unstable systems to complete coalescence . since then , several groups have performed increasingly sophisticated calculations in newtonian gravity , exploring the full parameter space of the problem with either sph @xcite or the eulerian , grid - based piecewise parabolic method ( ppm ) @xcite , and focusing on topics as diverse as the gw energy spectrum @xcite , the production of r - process elements @xcite , and the neutrino emission as a possible trigger for gamma - ray bursts @xcite . some of these newtonian calculations have included terms to model approximately the effects of the gravitational radiation reaction @xcite . the first calculations to include the lowest - order ( 1pn ) corrections to newtonian gravity , as well as the lowest - order dissipative effects of the gravitational radiation reaction ( 2.5pn ) were performed by shibata , oohara , and nakamura , using an eulerian grid - based method @xcite . more recently , the authors ( @xcite , hereafter paper 1 and paper 2 , respectively ) , as well as ayal et al.@xcite , have performed pn sph calculations , using the pn hydrodynamics formalism developed by blanchet , damour , and schfer ( @xcite , hereafter bds ) . these calculations have revealed that the addition of 1pn terms can have a significant effect on the results of hydrodynamic merger calculations , and on the theoretical predictions for gw signals . for example , paper 1 showed a comparison between two calculations for initially corotating , equal - mass binary ns systems . in the first calculation , radiation reaction effects were included , but no 1pn terms were used , whereas a complete set of both 1pn and 2.5pn terms were included in the second calculation . it was found that the inclusion of 1pn terms affected the evolution of the system both prior to merger and during the merger itself . the final inspiral rate of the pn binary just prior to merger was much more rapid , indicating that the orbit became dynamically unstable at a greater separation . additionally , the gw luminosity produced by the pn system showed a series of several peaks , absent from the newtonian calculation . most previous hydrodynamic calculations of binary ns mergers have assumed corotating initial conditions , and many modeled the stars as initially spherical . however , real binary ns are unlikely to be described well by such initial conditions . just prior to contact , tidal deformations can be quite large and the stars can have very nonspherical shapes @xcite . nevertheless , because of the very low viscosity of the ns fluid , the tidal synchronization timescale for coalescing ns binaries is expected to always be longer than the orbital decay timescale @xcite . therefore , a corotating state is unphysical . in addition , at large separation , the ns in these systems are expected to be spinning slowly ( see , e.g. , @xcite ; observed spin periods for radio pulsars in double ns systems are @xmath11ms , much longer than their final orbital periods ) . thus , the fluid in close ns binaries should remain nearly _ irrotational _ ( in the inertial frame ) and the stars in these systems can be described approximately by irrotational riemann ellipsoids @xcite . in paper 2 , we showed that nonsynchronized initial conditions can lead to significant differences in the hydrodynamic evolution of coalescing binaries , especially in the amount of mass ejected as a result of the rotational instability that develops during the merger . there are many difficulties associated with numerical calculations of binary mergers with irrotational initial conditions , especially in pn gravity . foremost of these is the difficulty in preparing the initial , quasi - equilibrium state of the binary system . in synchronized binaries , the two stars are at rest in a reference frame which corotates with the system , and thus relaxation techniques can be used to construct accurately the hydrostatic equilibrium initial state in this corotating frame ( see @xcite and paper 1 ) . for irrotational systems , no such frame exists in which the entire fluid would appear to be in hydrostatic equilibrium . instead , one must determine self - consistently the initial velocity field of the fluid in the inertial frame . otherwise ( e.g. , when simple spherical models are used ) , spurious oscillations caused by initial deviations from equilibrium can lead to numerical errors . this is especially of concern in pn gravity , where the strength of the gravitational force at any point in the ns contains terms proportional to the gravitational potential and pressure at 1pn order . another serious problem for hydrodynamic calculations with irrotational initial conditions is the issue of spatial resolution and numerical convergence . as was first pointed out in rs2 , in a frame corotating with the binary orbital motion , irrotational stars appear counterspinning so that , when they first make contact during the coalescence , a vortex sheet is formed along the interface . this tangential discontinuity is kelvin - helmholtz unstable at all wavelengths@xcite . the sheet is expected to break into a turbulent boundary layer which propagates into the fluid and generates vorticity through dissipation on small scales . how well this can be handled by 3d numerical calculations with limited spatial resolution is unclear . the outline of our paper is as follows . section ii presents a summary of our numerical methods , including a brief description of our pn sph code , and an explanation of the method used to construct irrotational initial conditions in pn gravity . additionally , we give the parameters and assumptions for all new calculations discussed in this paper . section iii presents our numerical results based on sph calculations for several representative binary systems , all with irrotational initial configurations , but varying mass ratios and ns eos . to test our numerical methods , we also study the effects of changing the initial binary separation and the numerical resolution . section iv presents gw energy spectra calculated from the runs in this and previous papers , as well as a discussion of how the measurement of spectral features could constrain the ns eos . a summary of our pn results and possible directions for further research , including the possibility of fully relativistic sph calculations , are presented in section v. all our calculations were performed using the post - newtonian ( pn ) sph code described in detail in papers 1 and 2 . it is a lagrangian , particle - based code , with a treatment of relativistic hydrodynamics and self - gravity adapted from the pn formalism of bds . as in our previous work , we use a hybrid 1pn/2.5pn formalism , in which radiation reaction effects are treated at full strength ( i.e. , corresponding to realistic ns parameters ) , but 1pn corrections are scaled down by a factor of about 3 to make them numerically tractable ( see below ) . we did not perform any new newtonian calculations for this paper . all the poisson - type field equations of the bds formalism are solved on grids of size @xmath12 , including the space for zero - padding , which yields the proper boundary conditions . shock heating , which is normally treated via an sph artificial viscosity , was ignored , since it plays a negligible role in binary coalescence , especially for fluids with a very stiff eos . all runs , except those used to study the effects of numerical resolution on our results , use @xmath13 sph particles per ns ( i.e. , the total number of particles @xmath3 ) , independent of the binary mass ratio @xmath14 . the number of sph particle neighbors is set to @xmath15 for all runs with @xmath3 . unless indicated otherwise , we use units such that @xmath16 , where @xmath17 and @xmath18 are the mass and radius of one ns . in calculations for unequal - mass binaries , we use the mass and radius of the primary . as in our previous papers , we compute the gravitational radiation reaction assuming that the speed of light @xmath19 in our units , which corresponds to a neutron star compactness @xmath20 . for a standard ns mass of @xmath21 this also corresponds to a radius @xmath22 km . the unit of frequency ( used throughout sec . iv ) is then @xmath23 in order to keep all 1pn terms sufficiently small with respect to newtonian quantities , we calculate them assuming a somewhat larger value of the speed of light , @xmath24 , which would correspond to ns with @xmath25 . as in our previous work , all calculations in this paper use a simple polytropic eos , i.e. , the pressure is given in terms of the mass - energy density by @xmath26 . we use @xmath1 to represent a typical stiff ns eos , and @xmath0 to represent a somewhat softer eos . the values of the polytropic constant @xmath27 for all our ns models ( determined by the mass - radius relation ) are the same as those used in paper 2 ( see table 1 therein ) . models for single ns are taken from papers 1 and 2 . it should be noted that models of ns with identical masses and polytropic constants but different numbers of sph particles may have slightly different effective radii , due to numerical resolution effects near the stellar surface . it is not easy to define a meaningful and accurate origin of time for merger calculations . the moment of first contact between the two stars is difficult to determine accurately , since it involves the smoothing lengths of particles located near the surface of each ns , where the method is least accurate . therefore , following the convention in our earlier papers , we set the absolute timescale for each run by defining the moment of peak gw luminosity to be at @xmath28 in our units . many runs therefore start at @xmath29 , but this is merely a matter of convention . to construct irrotational binary configurations in quasi - equilibrium , we start from our relaxed , equilibrium models for single ns ( paper 2 , sec . these models are transformed linearly into irrotational triaxial ellipsoids , with principal axes taken from the pn models of lombardi , rasio , and shapiro for equilibrium ns binaries @xcite . for example , for an equal - mass system containing two @xmath1 polytropes with an initial separation @xmath30 , we find from their table iii that the principal axes are given by @xmath31 , @xmath32 , and @xmath33 , where @xmath34 , @xmath35 , and @xmath36 are measured along the binary axis ( x - direction ) , the orbital motion ( y - direction ) , and the rotation axis ( z - direction ) , respectively . for an initial separation @xmath37 , we find @xmath38 , @xmath39 , and @xmath40 . for the initial velocity field of the fluid we adopt the simple form assumed for the internal fluid motion in irrotational riemann ellipsoids@xcite , @xmath41 where @xmath42 is the orbital angular velocity . it is easy to verify that this initial velocity field has zero vorticity in the inertial frame . calculating self - consistently the correct value of @xmath42 corresponding to a quasi - equilibrium circular orbit proves to be much more difficult in pn gravity than in newtonian gravity , especially for irrotational configurations . in newtonian gravity , the gravitational force between the two stars is independent of the magnitude of the tidal deformation of the bodies to lowest order . thus , even if the initial configuration of matter in the binary is slightly out of equilibrium , the orbit calculated for the two stars will still be almost perfectly circular . in pn gravity , the situation is quite different . from appendix a of paper 1 , we see that the gravitational acceleration , in our pn formalism , denoted there as @xmath43 , is calculated as @xmath44)\nabla u_*+\ldots,\ ] ] where @xmath45 is the the 1pn - adjusted velocity , and @xmath46 , @xmath47 , and @xmath48 are the pressure , rest - mass density , and gravitational potential , respectively , as defined in the bds formalism ( note that @xmath48 as defined here is a positive quantity ) . the quantity in brackets is the 1pn correction to the gravitational force , and depends explicitly on the initial thermodynamic state of the ns . small oscillations of each star about equilibrium result in errors when calculating the gravitational force felt by each component of the binary . in addition , the gravitational force is affected by the orbital velocity of the ns . thus , when we try to calculate the orbital velocity from the centripetal accelerations of the respective ns , as in eq . ( 1 ) of paper 2 , @xmath49 where @xmath50 and @xmath51 are the center - of - mass velocities of the primary ( located initially on the positive x - axis ) and the secondary ( located on the negative x - axis ) , respectively , we face the problem that the rhs of the equation is itself a function of @xmath42 . for an initially synchronized system , we solve this problem by relaxing the matter in a frame corotating with the binary , and in the process find the equilibrium value of @xmath42 self - consistently ( see paper 1 , appendix a ) . for an irrotational binary , this is not possible , since we can not describe a priori the proper velocity field toward which the matter must relax . we solve the problem instead by a more direct method . realizing that any simple approximation of the orbital velocity should be near the correct pn value , we alter the expression for @xmath42 to include a correction factor which , we hope , will remove the effect of any small deviations away from equilibrium on the gravitational accelerations felt by each ns . therefore , we introduce a parameter @xmath52 such that @xmath53 for each value of @xmath52 , we iterate eqs . [ eq : vx ] , [ eq : vy ] , and [ eq : kap ] over @xmath42 to determine a self - consistent velocity field at @xmath54 . we then perform trials for each value of @xmath52 , i.e. , we perform purely dynamical integrations ( with radiation reaction turned off ) for about one full orbit , until we find one which produces a nearly circular orbit ( such that @xmath7 changes by no more than @xmath55 ) . the results of one such set of calculations , for equal - mass ns with a @xmath1 eos and an initial separation @xmath30 , is shown in the top panel of fig . [ fig : rvstkap ] . we see that for @xmath56 , which works extremely well for all runs performed in newtonian gravity , the orbital angular velocity is too small , and the pericenter of the orbit lies within the dynamical stability limit , causing the system to merge ( artificially ) within the first orbital period . for @xmath57 , we see that the orbit is elliptical , with our initial state corresponding to pericenter . for a value of @xmath58 we obtain a very nearly circular orbit . a residual small - amplitude oscillation is shown in greater detail in the middle panel of fig . [ fig : rvstkap ] . we see that it occurs not on the orbital period timescale , but rather on the internal dynamical timescale of the stars . it is the direct result of small - amplitude pulsations of each ns about equilibrium . this is illustrated more clearly in the bottom panel of the figure , which shows the clear correlation between the radial acceleration of each ns ( shown as a fraction of the total inward gravitational acceleration ) and the central density of each ns . the numerical noise is an artifact of the small number of particles located near the very center of each star for the density curve , and of precision limits in the calculation for the acceleration curve . when averaged over time , we see nearly perfect correlation between the two quantities . also shown in the middle panel of fig . [ fig : rvstkap ] is a thick solid line depicting the dynamical evolution of a run with @xmath58 and radiation reaction included , to give a sense of the proper inspiral velocity in relation to the spurious radial velocities resulting from oscillations . we have performed several large - scale sph calculations of ns binary coalescence , assuming an irrotational initial condition for the binary system . table [ table : runs ] summarizes the relevant parameters of all runs performed , listing the adiabatic exponent @xmath59 , the mass ratio @xmath14 , the initial separation @xmath60 , and the number of sph particles . we continue to use the same nomenclature for our sph runs introduced in paper 2 , although we present here a new run e1 , using an improved initial configuration calculated by the method described in the previous section . run e1 is for a system with a @xmath1 eos , equal - mass ns , and an initial separation @xmath30 , and is similar in all respects to run b1 of paper 2 ( called the pn run in paper 1 ) , except that the initial condition is irrotational . it was continued until a quasi - stationary remnant configuration was reached . run e2 is for a system with the same @xmath1 eos , but a mass ratio of @xmath61 , and was started from a smaller initial separation @xmath37 , since binaries with smaller masses take longer to coalesce . in runs f1 and f2 , the eos has @xmath0 , but we use the same mass ratios and initial separations as runs e1 and e2 , respectively . additionally , to assess the numerical accuracy and convergence of calculations for irrotational binaries , we performed three runs which were primarily designed to test the dependence of the physical results on numerical parameters . first , we performed one run , labeled t2 , identical to e1 except for a smaller initial separation @xmath37 . this run was used to study how accurately the irrotational flow is maintained during the early stages of inspiral , and the effects of a small amount of spurious tidal synchronization on the gw signal . we then repeated run t2 using @xmath62 and @xmath63 sph particles per ns ( for a total of @xmath64 and @xmath4 particles , respectively ) , to study the effect of numerical resolution on calculations where we know small - scale instabilities will develop . the number of neighbors was adjusted in the two runs to be @xmath65 and @xmath66 , respectively . this choice is dictated by the convergence and consistency properties of our basic sph scheme : convergence toward a physically accurate solution is expected when _ both _ @xmath67 _ and _ @xmath68 , but with @xmath69 @xcite . the primary consideration behind the choice of the initial separation at @xmath37 ( rather than @xmath30 ) was the high computational cost of a run with @xmath4 . the coalescence process for binary ns systems is essentially the same qualitatively whether they are initially synchronized or irrotational . prior to merger , both ns show tidal elongation as well as the development of a tidal lag angle @xmath70 , as noted in papers 1 and 2 , created as the ns continuously try to maintain equilibrium while the coalescence timescale gets shorter and shorter . the inner edge of each ns rotates forward relative to the binary axis , and the outer edge of each ns rotates backward . we define @xmath70 to be the angle in the horizontal plane between the axis of the primary moment of inertia of each ns and the axis connecting the centers of mass of the respective ns . for equal - mass systems , the angle is the same for both ns . for binaries with @xmath71 we always find a larger lag angle for the secondary than the primary . mass shedding through the outer lagrange points is suppressed in irrotational binaries , but the formation of a differentially rotating remnant is quite similar . that said , it is important to understand the different computational challenges presented by irrotational systems . to demonstrate this , in fig . [ fig : dvirr ] we show the evolution of run e1 , with a @xmath1 eos , a mass ratio @xmath72 , and an initial separation @xmath30 . it is in all ways similar to run b1 , except that the ns start from an irrotational configuration . rather than plot sph particle positions , we instead show the density contours of the matter in the orbital plane , overlaying the velocity of the material in the corotating frame , defined by taking a particle - averaged tangential velocity , such that @xmath73_i}{\sum_im_i(r_{cyl})_i } , \label{eq : omegac}\ ] ] where the cylindrical radius is defined as @xmath74 . in synchronized binaries , the material maintains only a small velocity in the corotating frame prior to first contact . in contrast , for irrotational binaries material on the inner edge of each ns is counterspinning in the corotating frame , and thus we see a large discontinuity in the tangential velocity when first contact is made . this surface layer , initially at low density , is kelvin - helmholtz unstable to the formation of turbulent vortices on all length scales . meanwhile , material on the outer edge of each ns has less angular momentum in an irrotational binary configuration than in a synchronized one . thus , there is less total angular momentum in the system , and mass shedding is greatly suppressed , as we will discuss further in sec . [ sec : irrotq1 ] . at late times , though , the remnant rotates differentially , with the same profile seen in the synchronized case , namely @xmath75 attaining its maximum value at the center of the remnant , and decreasing as a function of radius . in table [ table : gw ] , we show some basic quantities pertaining our gw results , as well as the initial time for all runs , and the tidal lag angle @xmath70 which existed at the moment of first contact . the negative values of @xmath76 merely reflect the fact that our runs require more than 20 dynamical times before reaching peak gw luminosity . lag angles @xmath70 are listed for the primary and secondary , respectively , in systems with @xmath77 . each gw signal we compute typically shows an increasing gw luminosity as the stars approach contact , followed by a peak and then a decline as the ns merge together . most runs then show a second gw luminosity peak of smaller amplitude . for all of our runs we list the maximum gw luminosity @xmath78 and maximum gw amplitude @xmath79 , where @xmath80 , @xmath81 , and @xmath82 are defined by eqs . ( 23)-(25 ) of paper 1 ( see also eqs . [ eq : hpl ] and [ eq : hcr ] below ) . quantities referring to the first and second luminosity peaks are denoted `` 1 '' and `` 2 '' , respectively . in addition , we show the time @xmath83 at which the second peak occurs . we continued several of our runs to late times to study the full gw signal produced during the coalescence , as well as to study the properties of the merger remnants that may form in these situations . for each of these runs , we list several of the basic parameters of the merger remnant in table [ table : final ] , using the values computed for the remnant at @xmath84 . we identify the remnant mass @xmath85 , defining the edge of the remnant by a density cut @xmath86 , as well as the gravitational mass of the remnant in the pn runs , where the gravitational mass , which differs from the rest mass , is given by @xmath87 ( see paper 1 for more details ) . additionally , we list the kerr parameter @xmath88 , central and equatorial values of the angular velocity , @xmath89 and @xmath90 , the semi - major axis @xmath34 and ratios of the equatorial and vertical radii @xmath91 and @xmath92 , and the ratio of the principal moments of inertia @xmath93 . to study the effect of the eos on the evolution of irrotational ns binaries , we calculated mergers of both @xmath1 and @xmath0 polytropes ( runs e1 and f1 , respectively ) . a comparison of the binary separations as a function of time , along with the dimensionless gw luminosities and amplitudes are shown in fig . [ fig : gweos ] . immediately apparent is a difference in the location of the dynamical stability limit found in the two calculations . the orbit of the binary system containing ns with a softer eos ( run f1 ) remains stable at separations where the binary system with the stiffer eos has already begun to plunge inward dynamically . we see that , as in the synchronized case presented in paper 2 , the peak gw luminosity in dimensionless form is larger for the softer eos , but after a secondary luminosity peak the remnant relaxes toward a spheroidal , non - radiating configuration ( with essentially no emission whatsoever after @xmath94 ) . we also note that the secondary peak occurs sooner after the primary peak , by a factor of @xmath95 . even though they are presumed to be unphysical , calculations started from a synchronized initial condition make up much of the body of work performed to date on the binary ns coalescence problem . noting this , we compare our irrotational run e1 , with parameters detailed above , to one similar in every respect but started with a synchronized initial condition , our run b1 . in the top panel of fig . [ fig : gwspin ] , we see that the inspiral tracks do not align particularly well . synchronized binaries contain more total energy , and are thus less dynamically stable than irrotational ones , leading to a more rapid inspiral , even before the stability limit is reached . additionally , the binary separation `` hangs up '' earlier , at a separation @xmath96 , indicating the onset of mass shedding , as matter begins to expand radially outward from the system . the middle and bottom panels of the figure show the gw signals and luminosity , respectively , for the two runs . while the initial luminosity peaks are similar for both runs , both in amplitude and morphology , the secondary peaks are vastly different . the secondary peak for the synchronized system is of considerably greater magnitude than in the irrotational case , and delayed relative to it . we conclude that , while the adiabatic index seems to be the dominant factor in determining the gw signal during the merger itself , the initial velocity profiles of the ns play a key role in the evolution of the remnant , as well as affecting the orbital dynamics during inspiral . to better understand the features found in the gw signals of these calculations , particle plots for runs e1 , b1 , and f1 are shown in fig . [ fig : xyeos ] . comparing the leftmost panels , we see that at @xmath28 , when the gw luminosity peaks , the mass configurations are qualitatively similar , although more low - density material is seen at the edges of the newly forming remnant in the run with the @xmath0 eos , conforming to the general density profile expected for a softer eos . more subtle is the greater extension of the matter found in the synchronized run . since material on the outer edge of each ns has greater angular momentum when a synchronized initial condition is chosen , the calculation shows that such an initial condition allows for greater efficiency at channeling material outward during the final moments of inspiral . this difference is made abundantly clear by a comparison of the calculations at @xmath97 , shown in the center panels . we see extensive mass shedding from the synchronized run , much less from the irrotational runs . there is significantly more mass shedding from the run with the softer eos , but most of the material remains extremely close to the remnant . finally , by @xmath98 , we see that the softer eos produces a nearly spherical remnant , whereas the calculations with a stiffer choice of eos produce remnants which are clearly ellipsoidal , and will continue to radiate gws for some time , albeit at a much lower amplitude than at the peak . the strong influence of the choice of both eos and initial velocity profile on the final state of the remnant is shown in fig . [ fig : finaleos ] for the same three runs . in the top panel , we see the angular velocity profiles of the remnants at @xmath84 . we see that the choice of eos plays an important role near the center of the remnant , but at @xmath99 , the velocity profiles are essentially identical . the pattern holds as well for the mass profiles , which are shown in the bottom panel . the softer eos leads to a more centrally condensed remnant , as we would expect , but the remnants formed in both irrotational calculations contain virtually all the system mass within @xmath96 , with no more than @xmath55 escaping to larger radii . there is slightly more mass shedding past @xmath100 for the softer choice of eos , since more of the low - density material originally found at the edges of the ns is shed through the outer lagrange points of the system . by comparison , the synchronized run sheds almost @xmath101 of the total system mass past @xmath100 , even though the angular velocity profile at small radii is nearly the same as for the irrotational run with the same choice of eos . even though all well - measured ns masses in relativistic binary pulsars appear roughly consistent with a single ns mass @xmath102 @xcite , it is important to consider cases where the two ns have somewhat different masses . the roles played by the primary and secondary in unequal - mass binary mergers are remarkably different compared to the picture developed above for equal - mass systems . in paper 2 , we found that the primary ns in the system remains virtually undisturbed , simply settling at the center of the newly formed merger remnant . the secondary is tidally disrupted prior to merger , forming a single thick spiral arm . most of the material originally located in the secondary eventually forms the outer region of the merger remnant , but a significant amount of material is shed to form a thick torus around the central core . in fig . [ fig : xyq8 ] , we show particle plots for runs e2 and f2 , with a @xmath1 and @xmath0 eos , respectively , both with mass ratio @xmath61 . in the leftmost panels , at @xmath28 , we see that the secondary , located on the left , is tidally disrupted as it falls onto the primary . for the softer , @xmath0 eos ( run f2 ) , we see a greater extension of the secondary immediately prior to merger , as well as early mass shedding from the surface of the primary , as material from the secondary essentially blows it off the surface of the newly forming remnant . this process continues , so that by @xmath97 ( center panels ) , the secondary has begun to shed a considerable amount of mass in a single spiral arm which wraps around the system . much like in the equal - mass case , the spiral arm is much broader for the softer eos . mass loss from the primary is greatly reduced in the system with the stiffer eos , with only a scattering of particles originally located in the primary lifted off the surface . finally , by @xmath98 ( right panels ) , we see that the spiral arm has in both cases begun to dissipate , leaving a torus around the merger remnant containing approximately @xmath103 of the total system mass . although the merger process is significantly different for equal - mass and unequal - mass binaries , the details of the inspiral phase are reasonably similar . in particular , the evolution of the binary separation for runs with @xmath61 , shown in the top panel of fig . [ fig : gwq8 ] is roughly similar to what was found in fig . [ fig : gweos ] for binaries with @xmath72 . for both choices of the eos , the dynamical stability limit is located at approximately the same separation for both @xmath61 and @xmath72 binaries , although prior to the onset of instability the more massive equal - mass binaries show a more rapid stable inspiral . as we found before , the dynamical stability limit occurs further inward for the softer choice of the ns eos . in paper 2 , we found that the scaling of the maximum gw amplitude and luminosity as a function of the system mass ratio followed a steeper power law in synchronized binaries than would be predicted by newtonian point - mass estimates . newtonian physics predicts that @xmath104 and @xmath105 for merging binary systems . rs found the scaling in their numerical calculations to roughly follow empirical power law relationships given by @xmath106 and @xmath107 . the discrepancy results from the unequal role played by the two components during the final moments before plunge . the primary , which remains relatively undisturbed , contributes rather little to the gw signal , especially during the final moments before coalescence . thus , the gw power is reduced as the mass ratio is decreased . similar results were found in paper 2 for pn calculations of synchronized binaries , especially a steeper decrease in the gw luminosity as a function of the mass ratio for binaries with a soft eos . in the middle and bottom panels of fig . [ fig : gwq8 ] we show the gw signals and luminosity , respectively , for runs e2 and f2 . we find the same strong decrease in the gw power , especially for the softer @xmath0 eos . we expect that there should be a strong observational bias towards detecting mergers of equal - mass components , especially if the eos is softer . all our results presented above for equal - mass binaries used an initial binary separation @xmath30 , in part for the sake of comparison with previous calculations for synchronized binaries . this choice agrees with the standard approach of utilizing the largest possible separation for which the calculation can be performed using a reasonable amount of computational resources . this also has the advantage that any small deviations from equilibrium will generally be damped away before the ns actually make contact . it also allows for the best determination of the dynamical stability limit , however , a large initial separation can create problems in the case of irrotational binaries , since numerical shear viscosity inherently present in sph codes can lead to some degree of tidal synchronization of the ns during the inspiral phase . thus , by the time the merger takes place , the ns will no longer be completely irrotational . to study this effect , we calculated mergers for equal - mass ns with a @xmath1 eos starting at initial separations of both @xmath30 ( the aforementioned run e1 ) and @xmath37 ( run t2 ) . in the top panel of fig . [ fig : rvstsep0 ] , we show the binary separation as a function of time for both runs , noting the good agreement throughout . at the very end , during the merger itself , we do see the beginning of a slight discrepancy , attributable in large part to greater mass shedding in the calculation started at greater separation . this is similar to what was seen in sec . [ sec : irrotq1 ] , where run b1 , which had greater spin angular momentum , showed greater mass shedding , but the effect is greatly reduced in magnitude here since the ns in run t2 are nowhere near complete synchronization at the moment of first contact . in the bottom panel of the figure , we plot the ratio of the net spin angular momentum of the ns about their own centers of mass to the total angular momentum of the binary system , as a function of time . we see that the ns do gradually acquire a rotation pattern which corresponds to the direction of corotation , although there is nowhere near enough time to synchronize the binary . the effect is greatly enhanced immediately prior to merger in both cases , as the ns develop tidal lag angles and become distorted . by the time the binary initially started from @xmath30 reaches a separation of @xmath108 , the net angular momentum around each ns around its center of mass is equal to approximately @xmath109 the value we would expect should the binary be synchronized . this difference persists throughout the inspiral phase when the two calculations are compared . in fig . [ fig : gwsep0 ] , we compare the gw signals and luminosities for the two runs . we find excellent agreement between the two waveforms , both in amplitude and in phase . both runs show the modulated , damped gw luminosity which is characteristic of all runs we have computed using pn gravity . there is a slight difference in the amplitude of the signal during the second gw luminosity peak , but we expect such differences to be minor in light of such issues as the uncertainty in the equation of state and the larger problem of a proper relativistic treatment of gravitation . to focus on the effect of varying @xmath60 on the final results , we show the final mass and angular velocity profiles of the remnants for the two calculations in fig . [ fig : finsep0 ] . the results are in good agreement , although we see that the greater spin angular momentum of the run started at greater initial separation leads to approximately three times as much mass being deposited in a halo which surrounds the remnant while remaining gravitationally bound to it . in both cases , however , the total mass in the halo is less than @xmath55 of the total system mass . the inner region of the remnant in the run started from @xmath30 actually spins slightly slower than in the run started further inward , even though the ns have a greater spin angular momentum at the moment of first contact , but only because angular momentum transport outward was marginally more efficient in this case . we conclude that the choice of initial separation plays very little role in determining the results of our calculations , so long as we start from an initial separation @xmath110 . as discussed in sec . [ sec : summary ] , 3d calculations with limited spatial resolution could lead to gw signals which are dependent upon the number of particles used . to test this , gw signals computed from ns merger calculations , we performed runs t1 , t2 , and t3 , which all have equal - mass ns , start from the same initial separation @xmath37 , and use a @xmath1 eos , but vary by two orders of magnitude in the number of sph particles used , from run t1 with @xmath111 to run t3 with @xmath112 sph particles . although we could have used @xmath30 as an initial separation , we felt that the smaller initial separation was justified given the large computational overhead required to do a calculation with a million sph particles . to the best of our knowledge , run t3 is the largest and highest resolution sph calculation of irrotational binary ns coalescence to date . a comparison of the gw signals in both polarizations , as well as the gw luminosities , is shown in fig . [ fig : gwnp ] . we see that the lowest resolution run t1 produces a gw signal clearly different than higher resolution runs t2 and t3 . the difference is due in part to initial oscillations of the ns about quasi - equilibrium . such oscillations are greatly reduced by increasing the number of sph particles . since all three calculations started with the same approximate ellipsoidal models ( see sec . [ sec : initconf ] ) , we conclude that the amplitudes of the initial fluctuations result primarily from numerical noise . the two runs with higher resolution show good agreement from beginning to end , producing nearly identical gw luminosities . reassuringly , the gw signals also remain in phase throughout , indicating that calculations with @xmath2 can indeed model well these irrotational binary mergers . to estimate the precision which can be achieved for these numerical resolutions , we show in table [ table : gravnp ] a comparison of gw quantities computed at various times for each of the three runs . at @xmath113 , @xmath28 , and @xmath97 , we show the total gw strain @xmath114 , and the instantaneous frequency of the gws @xmath115 , defined by the relations @xmath116 we see that for the two high resolution runs , no quantity shown varies by more than about @xmath117 , whereas the difference is more than @xmath118 in the computed gw strains at late times between our highest and lowest resolution runs . the vortices forming at the surface of contact are shown in detail in figs . [ fig : dvnp1 ] and [ fig : dvnp2 ] . density contours in the orbital plane are overlaid with velocity vectors , which are plotted in the corotating frame of the binary , as defined by eq . [ eq : omegac ] . the upper left panels show the evolution of run t1 , with the bottom left and right panels representing runs t2 and t3 , respectively . in fig . [ fig : dvnp1 ] , we show the state of the three runs at @xmath28 . immediately apparent is that vortices have formed to the largest extent in the lowest resolution run , whereas in the higher resolution run there is little sign of particles mixing , except at a large distance along the vortex sheet from the center of the newly forming remnant . by @xmath119 , shown in fig . [ fig : dvnp2 ] , we see that there is a slight difference between the high resolution calculations with regard to the direction of the material flowing along the vortex sheet . in the highest resolution run , the streams of material flow nearly in a straight line from one vortex to the other , whereas in the middle and lowest resolution runs , there is a larger region of material which is accelerated toward the very center of the remnant . overall , though , there is excellent agreement between the two highest resolution calculations . the agreement of the gw signals calculated from runs t2 and t3 lead us to believe that the small - scale differences seen in the matter near the vortex sheet does not carry over into the bulk of the mass , responsible for the gw emission . essentially , the quadrupole moment at any given instant is most sensitively dependent upon the orientation of the densest regions at the cores of the respective ns , which are unaffected by the small - scale motion in the vortex sheet . the infall of the cores is driven by dynamical instability , leading them to plunge inward and merge , disrupting the vortex sheet and leading to the formation of the merger remnant . zhuge , centrella , and mcmillan first pointed out the importance of gw energy spectra for the interpretation of merger signals @xcite . in particular , on the basis of newtonian sph calculations , they showed how the observation of particular features in the spectra could directly constrain the ns radii and eos . since the detection of ns merger events will likely be made in narrow - band interferometers , it is especially important to understand the frequency dependence of the gw signals , and not just their time behavior ( as represented by waveforms ) . following the approach in @xcite , we calculate the gw energy spectrum from each of our calculations as follows . we first take the fourier transforms of both polarizations of the gw signal , @xmath120 and we insert them into the following expression giving the energy loss per unit frequency interval ( see , e.g. , @xcite ) , @xmath121 where @xmath122 represents the fourier transform of the second derivative of the traceless quadrupole tensor . for point - mass inspiral , the energy spectrum takes the power - law form @xmath123 @xcite , the slow _ decrease _ with increasing frequency coming from the acceleration of the orbital decay : although more energy is emitted per cycle , fewer cycles are spent in any particular frequency interval as the frequency sweeps up . near the final merger , large deviations from this simple power - law spectrum are expected . however , our initial binary configurations are still reasonably described by a point - mass model . therefore , to construct the complete gw signal , we attach a point - mass inspiral waveform ( hereafter referred to as the inspiral subcomponent ) onto the beginning of the signal calculated numerically with sph ( referred to as the merger subcomponent ) . the quadrupole tensor for the inspiral subcomponent is assumed to have the form @xmath124 where @xmath125 is the time _ before _ our dynamical calculation starts , and the amplitude and phase given by @xmath126 } \\ \phi(t')&=&-\int_{t'}^0 \omega(t ' ) dt'\\ \omega(t')&=&(1.0+\epsilon')\sqrt{\frac{m}{[r_0(1-\frac{t'}{t_{\rm mgr}})^{0.25}]^3}},\end{aligned}\ ] ] where @xmath60 is the initial binary separation and @xmath17 and @xmath127 are the total and reduced masses of the system , respectively . the time constant @xmath128 is given by the familiar expression @xmath129 with @xmath130 , these expressions correspond to the well - known quasi - newtonian point - mass inspiral results@xcite . here the correction factor @xmath131 is used to account for both finite - size and pn effects and is determined by matching the amplitude of the point - mass signal to the initial amplitude calculated from our sph initial condition . typically , it is no larger than about @xmath132 . the correction factor @xmath133 is used in a similar way to match the initial angular velocity of the system , and is of similar magnitude . with these corrections , [ eq : qxx]-[eq : qxz ] describe well the inspiral subcomponent at late times , but not , of course , at earlier times where one should have @xmath134 and @xmath135 as @xmath136 . the discrepancy , however , is no more than @xmath101 , and affects only the nearly featureless low - frequency part of the spectrum . in figs . [ fig : psn ] and [ fig : psp ] , we show a comparison between the energy spectra computed from a newtonian calculation with radiation reaction and from a pn calculation . the newtonian calculation is the n run from paper 1 ( referred to in paper 2 as run a1 ) , which was performed for equal - mass ns with @xmath1 eos , started at a separation @xmath137 in a synchronized initial state . the pn calculation is the run e1 described in sec . [ sec : sep0 ] , with equal - mass ns , a @xmath1 eos , and an irrotational initial condition with separation @xmath30 . in both figures , the dashed and dotted lines represent the merger and inspiral subcomponents , respectively , with the heavy solid line representing the total combined spectrum . we see immediately that there is a significant difference in the energy emitted between @xmath138 ( @xmath139 for our adopted ns parameters ; see eq . [ eq : fdyn ] ) . this is directly attributable to the earlier dynamical instability and faster inspiral rates found in pn calculations . since the binary system spends less time in this frequency interval , the energy emitted is greatly suppressed by the addition of 1pn effects . the characteristic `` cliff frequency , '' at which the energy spectrum plunges below the point - mass power - law , is the best indicator of the start of dynamical instability . a measurement of this frequency , combined with theoretical calculations such as those presented here , would lead directly to the determination of the ns radii , since all frequencies scale with @xmath140 and the destabilizing 1pn effects scale with @xmath141 . for the system considered in fig . [ fig : psp ] this `` cliff frequency '' is about @xmath142 ( @xmath143hz ) , almost within the reach of broad - band interferometers . had the 1pn effects been taken into account at full strength , this frequency would likely have been even lower , suggesting that deviations from point - mass behavior may even become measurable within the frequency range of current theoretical point - mass inspiral templates . two prominent features appear in the pn spectra at much higher frequencies . in general , the characteristic frequency of gw emission during mergers sweeps upward monotonically throughout the evolution . the sharp peaks at @xmath144 and @xmath145 ( @xmath146 and @xmath147 ) in fig . [ fig : psp ] are then clearly seen to result from emission at the time of the gw luminosity peak and during the remnant oscillation ( `` ring down '' ) phase , respectively . in contrast to some previous newtonian results @xcite , we find the amplitude of these peaks to be well below the point - mass power - law ( by a factor of almost 3 and 5 , respectively ) . no calculation we have performed with our pn formalism has ever produced energy above the point - mass power - law in any frequency range , confirming that the pn effects not only accelerate the dynamical instability of the system , but also cause a suppression of the total gw emission during the entire merger ( as we first showed in paper 1 ) . measurements of these peak frequencies would also provide independent constraints on the eos , but this will certainly require advanced detectors operated in narrow - band mode in order to beat the very high laser shot noise above @xmath9khz . we now examine the dependence of these calculated gw energy spectra on initial binary separation and numerical resolution . in the top panel of fig . [ fig : pssep0 ] , we compare the spectra computed from our @xmath1 eos , equal - mass calculations started at initial separations of @xmath37 ( run t2 ) and @xmath30 ( run e1 ) . we see excellent agreement above @xmath148 ( @xmath149 ) , but a slight discrepancy near the cliff frequency . at separations in the range @xmath150 , the binary does inspiral faster than a point - mass model would predict . since our code has been extensively tested and shown to reproduce newtonian results in this separation range ( see fig . 4 of paper 1 ) , we attribute the difference to the 1pn effects . assuming that a quasi - equilibrium description of the system is appropriate in this range of separations , the inspiral rate should be given by @xmath151 where @xmath152 is the energy loss rate to gravitational radiation , and @xmath153 is the rate of change of total energy as a function of separation along an equilibrium sequence of binary ns models @xcite . since the first factor is very insensitive to pn effects , we conclude that the slope of the equilibrium energy curve must be made smaller by the addition of 1pn corrections , in agreement with the results of pn equilibrium calculations @xcite . while there must exist a _ minimum _ in the equilibrium energy curve at some critical separation , formally representing the innermost stable circular orbit ( isco ) of the system @xcite , this may be less relevant to the onset of dynamical coalescence , which will occur earlier , as soon as the inspiral timescale becomes comparable to the orbital period . in the bottom panel of fig . [ fig : pssep0 ] we compare the energy spectra from runs t1 ( with @xmath64 particles ) and t2 ( @xmath3 ) , used to study the dependence of the gw signal on numerical resolution . we find that the spectra are nearly identical in the frequency range @xmath154 ( @xmath155 ) , characteristic of the peak emission and the following remnant oscillations , but disagree by a significant amount at lower frequencies . recall from fig . [ fig : gwnp ] that the lowest resolution run led to larger initial oscillations around equilibrium . while these ( spurious ) oscillations are at too high a frequency to appear in the energy spectrum , they do affect the initial inspiral rate significantly . the extra energy present in the system , combined with the tendency for the binary orbit to become slightly eccentric , leads to an artificially accelerated inspiral , and a further reduction of the energy above the cliff frequency . further comparisons are shown in fig . [ fig : pseos ] . in the top panel , we show the energy spectra for two different values of @xmath59 . these were computed from the two equal - mass runs started from @xmath30 , run e1 with @xmath1 , and run f1 with @xmath0 . note that these two different values of @xmath59 are compared here at constant @xmath17 and @xmath18 , which is of course not realistic . the dominant dependence of the gw emission on the eos is likely to be from the scaling of finite - size effects with @xmath18 , and from the scaling of pn effects with @xmath141 , which we can not study realistically given the limitations of our pn formalism ( sec . [ sec : formal ] ) . at low frequencies the two spectra are nearly identical . for the softer eos , the peak in the energy spectrum corresponding to maximum gw emission is of slightly lower amplitude and frequency . the peak corresponding to remnant oscillations is greatly suppressed , since the soft eos can not support a stable triaxial configuration ( see rs2 ) . in the bottom panel of fig . [ fig : pseos ] , we compare the energy spectra from synchronized binaries ( run b1 ) and irrotational binaries ( run e1 ) . we find a sharper `` cliff '' from the synchronized system , leading to lower energy in the frequency range up to about @xmath156 ( @xmath157 ) . at higher frequencies , however , the two spectra are remarkably similar . in particular , the spectral peak frequencies appear to be affected much more strongly by the eos than by the details of the initial velocity configuration . in fig . [ fig : psq8 ] , we compare gw spectra for systems with different binary mass ratios . in the top panel , we compare runs t2 and e2 , with mass ratios of @xmath72 and @xmath61 , respectively ( both with a @xmath1 eos and both started from an initial separation @xmath37 ) . we see a clear difference in the overall amplitude of the two spectra , but in general the low - frequency behavior of the spectra is not qualitatively different . in both cases , we see a smooth decline in the gw energy which indicates the onset of dynamical instability , and levels off as we reach the characteristic frequency of the maximum emission . a clear difference is the lack of a true peak in the spectrum for the @xmath61 binary , which reflects the suppression of gw emission in unequal - mass mergers , discussed in sec . [ sec : q8 ] . there is a well defined peak in the energy spectrum characteristic of emission from the remnant , nearly equal in amplitude to the first peak , as was found for the equal - mass binary with the same eos . the frequencies of both peaks in the @xmath61 spectrum are shifted lower relative to their location in the equal - mass case . in the bottom panel of fig . [ fig : psq8 ] , we show the same comparison for runs f1 and f2 with @xmath0 . the results are similar except that now the first peak for the binary with @xmath61 is completely absent . using the pn lagrangian sph code described in our previous papers , which is complete to 1pn order and includes radiation reaction effects , we have investigated a wide parameter space of binary ns mergers started from an irrotational initial condition . this initial configuration represents the most realistic approximation for ns binaries at separations of @xmath158 , corresponding to the onset of dynamical coalescence . based on these calculations , as well as those from our previous papers , we have calculated the energy spectrum of the gravitational wave emission for a variety of systems . the key result is the existence of a `` cliff frequency '' in all of the energy spectra we have studied , i.e. , a frequency above which the energy emitted dips dramatically beneath the point - mass approximation . we attribute this effect to the onset of dynamical instability in binary systems , which leads to a much more rapid inspiral than the point - mass formula predicts , an effect amplified when pn terms are taken into account . if the cliff frequencies of binary systems are even smaller than those found here ( typically around @xmath159 for our standard ns parameters ) when general relativity is treated consistently , they may lie within the frequency band accessible to broad - band laser interferometers . for example , proposals for ligo ii @xcite place its upper frequency limit around @xmath160 ( where the sensitivity in terms of a characteristic gw strain has been degraded by a factor @xmath161 due to photon shot noise ) , meaning that cliff frequencies should be seen if the true physical ns radius is sufficiently large , @xmath162 . our results also suggest that the high frequency features in the gw energy spectrum , which result from emission during the merger itself ( @xmath163 ) and from late - time oscillations of the remnant ( `` ring down '' @xmath164 ) , will be observable only by more advanced narrow - band detectors , but if observed , could place strong constraints on the ns eos . we also find from our calculations that initially irrotational binaries evolve in a qualitatively different way than do initially synchronized systems . regardless of the choice of eos , runs started from an irrotational configuration result in much less mass shedding than do synchronized runs , depositing no more than @xmath55 of the total system mass in an outer halo which remains bound to the merger remnant at the center of the system . additionally , there is a significant difference in the inspiral rate of such systems immediately prior to merger , with synchronized systems merging much more rapidly . thus , since real ns binaries should be essentially irrotational , it is important to exercise caution when interpreting the results drawn from calculations for initially synchronized systems , both before and after the merger occurs . the lack of mass shedding seen in our pn calculations of irrotational systems leads us to conclude that essentially no mass should be shed from realistic binary systems , since general relativistic effects further suppress the mass shedding instability @xcite . from a calculational standpoint , we find competing arguments for the ideal initial binary separation of an irrotational system . runs started from larger separations ( here @xmath165 ) require greater computational resources , and show more spurious synchronization because of the numerical viscosity of the sph method . such calculations do provide a better treatment of the deviations from point - mass inspiral prior to merger which result from both newtonian and pn finite - size effects . however , runs started from closer in ( @xmath166 ) are more reliable for drawing conclusions about the final state of the system , with regard to mass shedding as well as the rotational profile of the remnant . we find , reassuringly , that the phase and amplitude evolution of the gw signal during the primary luminosity peak is unaffected by the initial separation . we believe that our results are unaffected by the limited numerical resolution of 3d calculations . although merging binary ns systems develop small - scale instabilities , whose evolution we can not follow exactly , we see little effect on the gw signals we compute , so long as we use a sufficiently large number of sph particles . this is especially true for the phase of the gw signal . we conclude that numerical convergence for a given set of initial conditions and physical assumptions is possible without requiring excessive computational resources . based on these results , we believe that the fundamental limits of our method are not set by the numerical resolution or available computational resources , but rather by shortcomings in the pn formalism itself . all our pn calculations are limited by the magnitude of the 1pn terms that appear in the hydrodynamic equations . since we can not treat physically realistic ns models , we deal with the 1pn terms at reduced strength . this hybrid method does in some sense approximate the cancellations found in gr between 1pn and higher - order terms , but in the end can not fully model the non - linear nature of relativistic gravity . to do so in a more complete manner , two different approaches have been employed . the first is to attempt to integrate the hydrodynamics equations in full gr , as done recently by shibata and uryu @xcite . starting from quasi - equilibrium irrotational binary systems at the moment of first contact , they calculate the fully relativistic evolution of the system . such calculations represent a remarkable step forward , and can be considered the forefront of the current efforts to understand binary ns coalescence . they are limited only by the accuracy to which they can prepare their initial conditions and by the difficulty of extracting accurate gw signals from the boundary of 3d grids at a finite distance , usually well within the near zone of the source . as of yet no formalism has proven stable enough to handle a calculation which starts from the dynamically stable region and ends after the merger is complete . a second approach is to use an approximation of full gr which is known to be numerically stable , known as the _ conformally flat _ ( cf ) approximation @xcite . in this approximation to general relativity , assuming a specific form for the spatial part of the metric allows the equations of gr to be reduced to a set of linked non - linear _ elliptic _ equations . the drawback to the method is that the cf approximation is time - symmetric , and thus does not include the dissipative gravitational wave effects seen in full gr . such terms can be added externally to the formalism , though , to give the proper dynamical behavior @xcite . the cf approximation has been used successfully to compute quasi - equilibrium binary sequences to high accuracy for both synchronized @xcite and irrotational @xcite binaries , showing good agreement with fully relativistic calculations of synchronized @xcite and irrotational binaries @xcite , except for the case of extremely compact ns at small separations . additionally , a pn variant of the cf approximation has been used to study rapidly rotating single - star configurations @xcite , giving excellent agreement with fully gr calculations @xcite . in addition to equilibrium studies , dynamical sph calculations have recently been performed using relaxed , initially synchronized binary configurations @xcite . the authors are currently working on a code that will take as an initial condition fully relativistic irrotational , quasi - equilibrium models for binary ns systems , calculated using spectral methods @xcite . the system will then be evolved in the cf approximation . as the formalism is known to work for both binary configurations and rapidly rotating single - star configurations , reflective of our merger remnants , the hope is that such a method will allow us to calculate the evolution of a physically realistic binary system from the dynamically stable regime through merger and formation of a merger remnant , in a way that is consistent with gr throughout . the comparison of such calculations with those performed in full gr will serve as an important check of the shortcomings and successes of either approach . the authors wish to thank thomas baumgarte , joan centrella , roland oechslin , stu shapiro and kip thorne for useful discussions . this work was supported by nsf grants phy-0070918 and phy-0133425 . our computational work was supported by the national computational science alliance under grant ast980014n and utilized the ncsa sgi / cray origin2000 . h. lck _ et al . _ , class . quantum grav . * 14 * , 1471 ( 1997 ) ; k. danzmann , in _ relativistic astrophysics _ , proceedings of the 162nd w.e . heraeus seminar , edited by h. riffert _ et al . _ ( wiesbaden : vieweg verlag , 1998 ) , p.48 . k. kuroda _ et al . _ , in _ proceedings of the international conference on gravitational waves : sources and detectors _ , edited by i. ciufolini and f. fidecard ( world scientific , 1997 ) , p.100 ; h. tagoshi _ et al . _ , phys . d * 63 * , 062001 ( 2001 ) . meers , phys . d * 38 * , 2317 ( 1988 ) ; k.a . strain , and b.j . meers , b.j . , phys . lett . * 66 * , 1391 ( 1991 ) ; g. heinzel , _ et al . _ , phys . lett . * 81 * , 5493 ( 1998 ) ; a. freise _ et al . _ , phys . lett . a * 277 * , 135 ( 2000 ) . k. oohara and t. nakamura , prog . phys . * 82 * , 535 ( 1989 ) ; t. nakamura and k. oohara , _ ibid . _ * 82 * , 1066 ( 1989 ) ; k. oohara and t. nakamura , _ ibid . _ * 83 * , 906 ( 1990 ) ; t. nakamura and k. oohara , _ ibid . _ * 86 * , 73 ( 1991 ) . s. rosswog _ et al . _ , astron . astrophys . * 341 * , 499 ( 1999 ) ; s. rosswog , m.b . davies , f .- k . thielemann , and t.piran , astron . astrophys . * 360 * , 171 ( 2000 ) ; s. rosswog and m.b . davies , preprint , astro - ph/0110180 . m. ruffert , h .- th . janka , and g. schfer , astron . astrophys . * 311 * , 532 ( 1996 ) ; m. ruffert , h .- th . janka , k. takahashi , and g. schfer , astron . astrophys . * 319 * , 122 ( 1997 ) ; m. ruffert , m. rampp , and h .- th . janka , astron . astrophys . * 321 * , 991 ( 1997 ) . m. shibata and k. uryu , phys . d * 61 * , 064001 ( 2000 ) ; m. shibata and k. uryu in _ proceedings of the 20th texas symposium on relativistic astrophysics _ , edited by h. martel and j.c . wheeler , ( new york , aip press , 2001 ) , p. 717 . wilson , g.j . mathews , and p. marronetti , phys . d * 54 * , 1317 ( 1996 ) ; g.j . mathews , p. marronetti , and j.r . wilson , _ ibid . _ * 58 * , 043003 ( 1998 ) ; p. marronetti , g.j . mathews , and j.r . wilson , _ ibid . _ * 60 * , 087301 ( 1999 ) ; g.j . mathews and j.r . wilson _ ibid . _ * 61 * , 127304 ( 2000 ) . .input parameters for the runs described in this paper : @xmath59 is the adiabatic exponent , @xmath14 the binary mass ratio , and @xmath60 the initial separation . all runs used @xmath167 sph particles , except testing runs t1 and t3 , with @xmath111 and @xmath112 sph particles , respectively . testing run t2 differs from run e1 only in the initial separation . all runs in this paper were started from an irrotational initial condition . [ cols="^,^,^,^,^ " , ]

we consider domain walls ( dw s ) between single - mode and bimodal states that occur in coupled nonlinear diffusion ( nld ) , real ginzburg - landau ( rgl ) , and complex ginzburg - landau ( cgl ) equations with a spatially dependent coupling coefficient . group - velocity terms are added to the nld and rgl equations , which breaks the variational structure of these models . in the simplest case of two coupled nld equations , we reduce the description of stationary configurations to a single second - order ordinary differential equation . we demonstrate analytically that a necessary condition for existence of a stationary dw is that the group - velocity must be below a certain threshold value . above this threshold , dynamical behavior sets in , which we consider in detail . in the cgl equations , the dw may generate spatio - temporal chaos , depending on the nonlinear dispersion . a spatially dependent coupling coefficient as considered in this paper can be realized at least in two different convection systems : a rotating narrow annulus supporting two traveling - wave wall modes , and a large - aspect - ratio system with poor heat conductivity at the lateral boundaries , where the two phases separated by the dw are rolls and square cells .

patt-sol9609003

close to a pattern - forming transition , many hydrodynamic systems can be described by coupled one - dimensional ( 1d ) ginzburg - landau equations for some envelope functions ( order parameters ) @xmath0 and @xmath1 . these equations were first introduced in an explicit form ( with purely real coefficients , but also with the group - velocity terms ) by cross @xcite . in the presence of a small over - criticality parameter , and when rewritten in terms of two real amplitudes and two real phases , these equations for two complex order parameters are completely tantamount to a single fourth - order complex equation ( the `` complex swift - hohenberg '' equation ) introduced still earlier in ref . as is well known , the coefficient in front of the cross - coupling ( cc ) term is an essential parameter of these equations . for a small cc coefficient , homogenous bimodal states of the form @xmath2 occur , while for sufficiently large cc , this bimodal state is unstable , and the system evolves into a single - mode state in which either @xmath0 or @xmath1 is zero @xcite . in this work , we consider the case that the cc coefficient is a function of the spatial coordinate @xmath3 . in particular , we focus on the case that the cc coefficient is slightly above @xmath4 for @xmath5 and slightly below @xmath4 for @xmath6 . the simplest state that can occur in this case is a spatial juxtaposition of a single and a bimodal state , separated by a so - called domain - wall . we study this domain - wall in the nld , rgl , and cgl equations . we show that non - variational effects can destroy the stationary domain - wall and study the dynamical states that occur in this case . in particular we focus on the effect of the group - velocity terms in the coupled equations . this paper is organized as follows . first we introduce a simplified model for the occurrence of domain - walls . for stationary solutions of this model we derive a perturbation equation that allows us to predict the vanishing of stationary domain - walls when the group - velocity is sufficiently large . this is confirmed by numerical simulations of this model , and we describe the ensuing dynamical states . we study the relevance of this model for coupled rgl and cgl equations by performing numerical simulations . in particular for the cgl equations , a broad spectrum of dynamical behavior is observed , including spatio - temporal chaos . it is very difficult to characterize these states , and we restrict ourself to a limited exploration of the possible dynamical states . some of the essential features of the cgl equations that are important in the study of the domain - wall , i.e. , the stability of the single and bimodal states as a function of the cc can also be studied in simpler models . we will first consider two coupled nonlinear diffusion ( nld ) equations , to which we add group - velocity terms borrowed from the cgl equation that break the variational representation of the nld equations ( however , the group - velocity terms should not be added in the above - mentioned case of the rayleigh - bnard convection in the inhomogeneous large - aspect - ratio system @xcite ) . in the simplest approach , we take the amplitudes real - valued : @xmath7 these nld equations are the simplest set of equations that admit the dw s that we are interested in . these equations have no direct physical interpretation in terms of amplitude equations , but they have the advantage over more realistic models that the effect of the group - velocity terms on the domain - walls can be studied analytically . since the group - velocity terms have opposite sign for the @xmath0 and the @xmath1 equation of the coupled cgl equations ( [ cgl ] ) , we retain this property in the nld equations . without the cross - coupling , the group - velocity terms generate a counter propagation of the patterns in @xmath0 and @xmath1 . when the equations for @xmath0 and @xmath1 are coupled , the effect of the group - velocity terms can not be given in terms of a simple propagation rule . we will derive , in a certain limit , a single ordinary differential equation for the order parameter @xmath8 that describes stationary configurations . using this equation , we will be able to investigate the existence of a stationary dw analytically . in section [ varisec ] we consider the case that the group - velocity terms are absent . the equations are then variational and for the particular case that the cc coefficient , as a function of the spatial coordinate @xmath3 , is proportional to @xmath9 , an exact analytical solution for the dw can be obtained . next , we extend in section [ nldevsec ] the equation for @xmath10 to include the non - variational group - velocity terms , and when we take @xmath11 to be a step function , an analytical solution can be constructed . analyzing the latter solution , we find a transition from a regime where the dw is stationary to a regime where no stationary dw exists . this transition , which we show to have a nice geometrical interpretation , occurs when the group - velocity @xmath12 is equal to the critical value @xmath13 . we will show that this result remains valid for a broad class of inhomogeneous cc coefficients . direct simulations of the coupled nld equations , presented in section [ secfknum ] , confirm the analytical results and show that there is a value of the group - velocity , @xmath14 , where the stationary domain - wall looses its stability . this instability occurs because the position of the domain - wall diverges when @xmath12 approaches @xmath15 . the difference between @xmath15 and @xmath14 is of the order of a few percent . for a group - velocity just beyond @xmath14 , the dw is seen to perform small - amplitude chaotic oscillations around a mean position , without essentially disturbing the dw s shape . the study of the coupled cgl equations is of a more explorative nature . as a first step towards the study of the full coupled cgl equations , we allow the amplitudes @xmath0 and @xmath1 to be complex valued . these equations are reminiscent of the rgl equations , but it should be noted that in physical applications , the rgl equations have no group - velocity terms ; nevertheless we refer to this model as the coupled rgl equations . we explain that the main properties of the stationary domain - walls in this model can be described by the nld equations . simulations of coupled rgl equations , presented in section [ rglsec ] , show a transition between the stationary domain - walls and a dynamical state with the increase of the group - velocity that is similar to the nld - case . the role of wavenumbers of the initial conditions , which constitutes the main feature that is not present in the nld equations , is discussed briefly . for the coupled cgl equations , the numerical results presented in section [ cglsec ] show that the nonlinear dispersion terms may render the domain - walls unstable , even when the group - velocity terms are absent . spatio - temporal disordered states often occur . the oscillations of the dw can be understood qualitatively to arise from the large gradients of @xmath16 and @xmath17 around the domain - wall . it may be pertinent to note that both complex and real ginzburg - landau equations with the coefficients in front of the _ linear _ terms smoothly depending upon the spatial coordinate have been the subject of many studies @xcite . it was found that a parameter ramp in the real or complex gl equation can perform wavenumber selection @xcite . in the case of the full cgl equation , it can also render the single - mode traveling waves unstable , and can trap subcritical solitary pulses @xcite . however , as far as we know , there have been no studies aimed to consider effects of a spatial dependence in the coefficient in front of the nonlinear cc term . a spatially dependent cc can be realized in rayleigh - bnard convection in a rotating annulus of non - constant width @xcite . convection in rotating systems has recently been the focus of several studies @xcite , and quasi-1d traveling waves were shown to occur in these systems near the vertical side - walls of the annulus . in a rotating annulus , there are two so - called wall modes , localized , respectively , near its inner and outer side walls . note that there is only one wall - mode per side - wall ( instead of two ) ; this is due to a symmetry breaking that is induced by the rotation @xcite . the amplitude equations describing slow modulations of these modes in the co - rotating reference frame are two coupled cubic complex ginzburg - landau ( cgl ) equations @xcite . the strength of the cc between the two wall modes sensitively depends on the width of the annulus . when the annulus is not uniform but has a varying width , the corresponding cc can be made to vary across its critical value as a function of the longitudinal spatial coordinate ( which is going along the circumference of the rotating annulus ) . other coefficients of the amplitude equations describing such a system will also depend on this coordinate , but if we assume that these coefficients , in contrast to the cc coefficient , are not close to a critical value , their spatial dependence may be ignored , provided that it is smooth enough . a similar problem may be implemented in a related but different physical system , viz . , the rayleigh - bnard convection in a ( non - rotating ) large - aspect - ratio cell . in this case , the two modes are two orthogonal sets of parallel rolls . a single - mode state is stable provided that an effective cc coefficient between the orthogonal rolls is larger than a certain minimum value , while in the opposite case a square - lattice pattern ( the bimodal state obtained as a superposition of two orthogonal sets of rolls ) is stable @xcite . usually , the actual value of the cc coefficient is well above the above - mentioned minimum , so that the square lattice is unstable . however , in special cases , e.g. , for the convection between horizontal surfaces with poor heat conductivity , the cc coefficient may fall below the minimum @xcite . one may construct a suitable inhomogeneous system , for example , by means of a variable - thickness lid put on top of the convection layer , such that the local cc coefficient , being a function of the spatial coordinates , is passing through the minimum value . then one may expect a stationary dw separating the rolls and square lattice , which is impossible in the homogeneous system @xcite . in this section we will focus on the coupled nld and rgl equations , which are the simplest sets of equations where a dw between a single - mode and a bimodal state can occur . the dynamics of these systems without the group - velocity terms are relaxational in the sense that a lyapunov functional @xmath18 exists for each of them , allowing to predict final states by minimizing it . we derive a perturbation theory for a weak inhomogeneity in the cross - coupling coefficients of the coupled nld equations , that allows us to find a closed - form expression for the dw for a special choice of the inhomogeneity . this perturbation theory will be a starting point in the next subsection , where we will focus on the group - velocity terms that make the equations non - variational . the equations that we consider in this section are [ rglstat ] @xmath19 where @xmath0 and @xmath1 are real - valued in the nld case , and complex in the rgl case . the critical value of @xmath20 is @xmath4 : for @xmath21 the bimodal state @xmath22 is stable , whereas for @xmath23 this bimodal state loses its stability and the single - mode state with @xmath24 and @xmath25 ( or vice versa ) becomes stable @xcite . the cc coefficient @xmath11 is assumed to decrease monotonically from slightly above to slightly below its critical value as a function of the spatial coordinate @xmath3 . in what follows below , we will set @xmath26 , where @xmath27 is a small monotonically decreasing function of the spatial coordinate @xmath3 , such that @xmath27 is positive at @xmath28 and negative at @xmath29 . we assume that @xmath30 saturates at @xmath31 , i.e. , it assumes certain asymptotic values @xmath32 , and @xmath33 . the solution that one expects for this choice of @xmath11 is a stationary dw located around @xmath34 which matches the single - mode and bimodal states existing , respectively , for negative and positive @xmath3 . the coupled rgl equations can be derived from the lyapunov functional @xmath35 by setting @xmath36 and @xmath37 . as is commonly known , the lyapunov functional may only decrease in time , and , as it is bounded from below , a final state corresponds to a minimum of @xmath18 . the final state that corresponds to a global minimum of @xmath18 has a zero wave number , which suggests to consider also the particular case of real @xmath0 and @xmath1 . thus one obtains the coupled nld equations . for finite systems with periodic boundary conditions , the rgl equations may evolve to stationary states with nonzero wavenumber . this is discussed in section [ rglstatsec ] . we will now focus on the stationary solutions of the nld equations , and set the derivatives with respect to time equal to zero . the ensuing equations can be written as four coupled ordinary differential equations for the real - valued @xmath0 , @xmath38 , @xmath1 and @xmath39 . when we identify @xmath0 and @xmath1 with position coordinates and their derivatives with generalized momentum coordinates , the equations for a stationary domain - wall can be written as the _ hamilton equations _ corresponding to the hamiltonian @xmath40 where @xmath3 is the `` time '' coordinate . although the hamiltonian is similar to the lyapunov functional , the difference in the signs that occur between these expressions should be noted . in general we can not solve the hamilton equations for arbitrary @xmath27 . to proceed we will develop a perturbation expansion by taking @xmath41 and @xmath42 small . we will complement this condition by the assumption that @xmath43 and @xmath44 are slowly varying functions of @xmath3 , so that the diffusive terms in eqs . ( [ rglstat ] ) are small in comparison with the other terms below we will determine , in a self - consistent way , that the spatial scale @xmath45 over which @xmath0 and @xmath1 vary is of order @xmath46 . when the diffusive terms and the coupling term @xmath47 of the hamiltonian ( [ h ] ) are small , it follows that @xmath48 is almost constant @xcite . this suggests the following representation for @xmath43 and @xmath44 @xcite : @xmath49the single - mode states are those with @xmath10 being an integer multiple of @xmath50 , and bimodal states have @xmath51 ( throughout this paper , @xmath52 will represent an integer ) . when we substitute this representation into the hamiltonian ( [ h ] ) we obtain @xmath53 where a prime denotes differentiation with respect to @xmath3 . stationary solutions of the coupled nld equations are found by determining the minima of the hamiltonian ( [ h ] ) . since we assumed that @xmath0 and @xmath1 are slowly varying , and that @xmath54 is almost constant , @xmath55 and in the simplest non - trivial approximation we may set @xmath54 equal to one @xcite . differentiating the hamiltonian ( [ hchi ] ) with respect to @xmath3 and neglecting higher order terms , we obtain the following perturbation equation : @xmath56where the prime stand for @xmath57 . it follows from here that the relation between the small quantity @xmath58 and a large scale @xmath45 of variation of the function @xmath59 , which was implicitly assumed above , is @xmath60 @xcite . as an example , we may solve eq . ( [ chi ] ) exactly for the special choice @xmath61 . the solution is then @xmath62which can be verified by substitution ; note the scaling of @xmath3 with @xmath63 . this choice for @xmath27 and the corresponding solution are shown in fig . 1 for @xmath64 . we will proceed by investigating whether this type of the dw solutions exist for more general @xmath27 and when the group - velocity terms are included . in this section we will investigate what happens to the domain - walls when the group - velocity terms that are similar to those in the cgl equation are included in the nld equations . these terms destroy the variational structure of the nld equations , hence their final states need no longer be stationary . we will show that stationary dw s can not exist when the group - velocity is above a certain critical value . numerical simulations of the nld equations corroborate this prediction , and show that , beyond the threshold , time - periodic or disordered states occur . the nld equations with the group - velocity terms are [ fk ] @xmath7 as was demonstrated in ref . @xcite , one can use a _ balance equation _ for the hamiltonian ( [ h ] ) to treat effects of the small group - velocity terms on the stationary solutions . differentiating ( [ h ] ) and making use of the stationary version of eqs . ( [ fk ] ) , one finds @xmath65 ~. \label{eqwithc}\]]using the `` polar '' representation defined by eqs . ( [ chi ] ) yields an effective equation for @xmath10 , which is a generalization of eq . ( [ chi ] ) ( the same equation , albeit for constant @xmath58 , was obtained in ref . [ 13 ] ) : @xmath66this equation is the basis for the perturbative analysis presented below . it should be noted that this equation is invariant under a scale transformation @xmath67 , @xmath68 and @xmath69 . this freedom can in principle be used to scale out @xmath12 , but we will not do this ; this scale - invariance is reflected , however , in the formula for @xmath15 that is obtained below . it will be convenient to rewrite eq . ( [ balance ] ) in the form of a two - dimensional _ non - autonomous _ dynamical system : [ dynamical]@xmath70 fixed points of eqs . ( [ dynamical ] ) are @xmath71 , which correspond to single - mode stationary solutions of equations of eqs . ( [ fk ] ) at even @xmath52 , and to bimodal solutions at odd @xmath52 . hetero - clinic orbits going from one fixed point at pseudo - time @xmath72 to another fixed point at @xmath73 correspond to the dw s that we are interested in . since we have chosen @xmath27 to be a decreasing function of @xmath3 , these orbits go from a fixed point with even @xmath52 to one with odd @xmath52 . the system ( [ dynamical ] ) is invariant with respect to the following symmetry transformations : * @xmath74 , * @xmath75 , @xmath76 . * @xmath77 , @xmath78 , @xmath76 , @xmath79 . with regard to these symmetries , it is sufficient to consider only the hetero - clinic orbit that goes from the fixed point @xmath80 , to be referred to as fp0 , to the bimodal fixed point @xmath81 , which will be called fp1 ; the other dw s can be obtained by applying a combination of the transformations ( i ) , ( ii ) and ( iii ) to this hetero - clinic orbit . an essential feature of the dynamical system ( [ dynamical ] ) is that the direction of the phase - flow in the @xmath82 plane is not fixed , but depends on the current value of @xmath27 . therefore , even when the functional form of @xmath58 is fixed , the possible orbits of equation ( [ dynamical ] ) through a certain point form , in general , a one - parameter family . however , when we take @xmath27 to be a step function , i.e. , [ step ] @xmath83 there are only 2 orbits through a certain point ; one for @xmath84 ( @xmath28 ) and one for @xmath85 ( @xmath29 ) . notice that the lack of continuity of @xmath27 does not contradict the assumption that the stationary solution for @xmath10 is a smooth function of @xmath3 ; we will see below that solutions corresponding to this discontinuous @xmath30 are smooth indeed . in fact , the scaling properties of equation ( [ balance ] ) , that also hold for the dynamical system ( [ dynamical ] ) yield that when we rescale @xmath67 , with @xmath86 , the function @xmath27 becomes effectively steeper due to the rescaling of the spatial - coordinate ; when @xmath87 , @xmath27 becomes , in a sense , infinitely small and infinitely steep . we are now interested in the behavior of the hetero - clinic orbits as a function of @xmath12 . a hetero - clinic trajectory corresponding to the dw exists provided that the the outgoing ( unstable ) manifold of fp0 , which we will refer to as @xmath88 , intersects the ingoing ( stable ) manifold of fp1 , to be referred to as @xmath89 ( see figs . [ fig2 ] and [ fig2sub ] ) . obviously , @xmath90 and @xmath91 have to be calculated , respectively , for @xmath92 and @xmath93 . the question of existence of a domain - wall has thus been reduced to checking whether @xmath90 and @xmath91 intersect . the point of intersection then gives the values of @xmath10 and @xmath94 at @xmath34 . this is illustrated in the figs . [ fig2 ] and [ fig2sub ] , where we have plotted , for @xmath95 and @xmath96 , these two manifolds for a range of values of the group - velocity . the point of intersection has by definition @xmath34 , and so from these figs . we can read off the values of @xmath10 and @xmath97 at @xmath34 as a function of @xmath12 . the point of intersection of @xmath90 and @xmath91 shifts towards fp0 when @xmath12 is decreased . as a consequence , the location of the domain - wall , i.e , the value of @xmath3 where @xmath98 , shifts to larger and larger values . in fact , when @xmath12 approaches a certain value which we define as @xmath15 , the intersection of @xmath90 and @xmath91 approaches fp0 and the location of the domain - wall , that corresponds to the hetero - clinic orbit , diverges to infinity ( see fig . [ statfig ] ) . below we will find that @xmath99 . note that the square - root dependence follows from the scaling properties ; only the numerical factor @xmath100 needs to be determined . there are two complementary methods for the determination of the critical velocity . the first method involves analytically solving the flow - line equations of the dynamical system ( [ dynamical ] ) and the second method involves an inspection of the geometry of the flow - lines . after a description of these two methods we discuss the validity of @xmath15 for inhomogeneities that are not of the form ( [ step ] ) . the manifolds @xmath90 and @xmath91 can be found exactly when @xmath58 is piecewise constant , as in eqs . ( [ step ] ) , because we can solve the equation for the trajectories of the dynamical system ( [ dynamical ] ) explicitly when @xmath58 is a constant . this equation is obtained by dividing eq . ( [ dynamical]b ) by eq . ( [ dynamical]a ) which yields for constant @xmath58 : @xmath101to solve equation ( [ floweq ] ) , we first perform a coordinate transformation by introducing @xmath102 , which yields @xmath103eq . ( [ homo ] ) is homogeneous , and by defining @xmath104 and some rewriting , we find @xmath105 the manifold @xmath90 corresponds to the case that @xmath106 is a constant , which occurs for @xmath107 . by transforming this solution back to the @xmath108 coordinates we obtain for @xmath90 : @xmath109 where @xmath110 satisfies @xmath111 , and @xmath112 . the manifold @xmath89 is found by a straightforward integration of equation ( [ sepeq ] ) , which yields the unpleasant equation @xmath113 where @xmath114 is an arbitrary constant of integration , and @xmath85 . @xmath114 has to be chosen so that the trajectory is passing through the point fp1 , where @xmath115 . although we do not have explicit expressions for the manifolds in terms of the original variable @xmath10 , we can find from eqs . ( [ sol0 ] ) and ( [ sol1 ] ) under which conditions the manifolds @xmath88 and @xmath89 intersect . the intersection point can be found by substituting @xmath116 into eq . ( [ sol1 ] ) . simply solving the ensuing equation numerically yields that @xmath91 and @xmath90 intersect only ( apart from the intersection at fp0 itself ) when @xmath117where it should be noted that @xmath12 is assumed to be negative . this means that if the group - velocity @xmath12 , which is amenable for the non - variational effects in this model , is below the threshold @xmath118 , there can not be a stationary dw . it should be noted that we have focused here on the domain - wall that has a single - mode state for negative @xmath3 and a bimodal state for positive @xmath3 . when we reflect @xmath3 , and study a domain - wall of opposite chirality , i.e. , going from a bimodal state at @xmath119 to a single - mode state at @xmath120 , according to symmetry ( iii ) the critical velocity changes sign . when @xmath27 goes from negative values for @xmath28 to positive values for @xmath29 , the ensuing stationary domain - walls can not exist for sufficiently large _ positive _ @xmath12 . we will encounter domain - walls of both chiralities in the numerical simulations presented below . a simple geometric interpretation of the threshold is illustrated in the figs . [ fig2 ] and [ fig2sub ] . as is illustrated in the phase - portraits ( see figures [ fig2]a and [ fig2]b ) , @xmath91 intersects @xmath90 for all positive @xmath12 , and therefore we will concentrate now on the case of negative @xmath12 . the central point is that for negative @xmath12 and negative @xmath58 , fp0 is a spiral when @xmath121 , and a saddle when @xmath122 . therefore , the motion of @xmath91 , for which @xmath123 , in the neighborhood of fp0 is spiral - like when @xmath121 ; this is not visible on the scale of fig . [ fig2 ] but can be seen in fig . [ fig2sub ] . as long as @xmath90 spirals around fp0 , it has to intersect @xmath124 , which means that there is a hetero - clinic orbit and therefore a stationary domain - wall . when @xmath12 approaches @xmath15 , this spiraling motion becomes less prominent , and consequently , the intersection of @xmath90 and @xmath124 shifts to fp0 ( see fig . [ fig2sub ] ) . as a consequence , the position of the domain - wall shifts to large values of @xmath3 . when @xmath12 has crossed the critical value @xmath15 , fp0 is a saddle and it turns out that @xmath90 and @xmath124 only intersect in fp0 itself ( see figure [ fig2]d ) ; this corresponds to a domain - wall that is shifted to infinity . there are two different assumptions involved in the calculation of @xmath15 . first of all , @xmath58 needs to be small , in order for the perturbation equation ( [ balance ] ) to hold . secondly , we used for @xmath27 a step - function . we will now discuss the validity of @xmath15 when @xmath27 is of more general form . the latter assumption is , due to the scaling properties of equation ( [ balance ] ) , not crucial , as we have indicated above . when we rescale @xmath67 , with @xmath86 , we can compensate for this by rescaling the spatial coordinate ; effectively , the function @xmath27 becomes steeper then . when @xmath87 , @xmath27 becomes infinitely small and infinitely steep . therefore , the results that are obtained when @xmath58 is a step - function , are still valid when @xmath27 is an arbitrary monotonically decreasing function , provided that @xmath125 is small . it is tempting to try to extend the phase - space analysis to the non - perturbative case , i.e. , when @xmath58 and @xmath12 are not small . the stationary version of the coupled nld equations ( [ fk ] ) can then be written as a 4-dimensional dynamical system . the equations for the trajectories can not be solved in this case , but we can inspect the eigenvalues of the fixed points that correspond to the single - mode and bimodal states . the latter fixed point always has two positive and two negative eigenvalues . the fixed point corresponding to the single - mode state changed from a saddle to a spiral at @xmath126 in the perturbative case , and something similar happens in the 4-dimensional system . the eigenvalues of this fixed point are given by the expressions @xmath127 and @xmath128 . one can readily find that , when @xmath129 , the fixed points changes from being a saddle ( two negative and two positive eigenvalues ) to a fixed point with one positive , one negative and two complex conjugated eigenvalues . although this does not prove that the stationary dw disappears , this is at least an indication that the previously obtained value @xmath130 may be also valid when @xmath58 and @xmath12 are not small . to test the validity of the perturbation equation ( [ balance ] ) and to verify our predictions , we have performed numerical simulations of the coupled nld equations ( [ fk ] ) in a periodic system of size @xmath131 . because of the periodicity of the system , there are two domain - walls but we will , as before , focus attention on the domain - wall that connects a single - mode state @xmath132 to the left with a bimodal state to the right . the results reported here were obtained by using a pseudo - spectral code with the time step 0.05 and , typically , 256 modes ; runs with a higher number of modes up to 1024 were performed to check the results . the code was such that @xmath0 and @xmath1 could either be real or complex - valued in order to simplify the comparison between the results for the nld equations and the rgl and cgl equations later on . when @xmath133 , the intersection point of the manifolds approaches fp0 , and consequently the position of the dw shifts to @xmath134 . in this limiting case , a single - mode state intrudes into the domain where @xmath21 , where this state is unstable . therefore , one might expect that the dw is unstable when @xmath12 is still slightly above @xmath130 . we find below that this is the case indeed , and we will denote the value of the group - velocity where the domain - wall turns unstable by @xmath14 . the simulations carried out below focus on three items . first of all , we will verify that the stationary dw predicted by the phase - space analysis exists and is stable when @xmath12 is not too close to @xmath130 . subsequently , we determine , for various choices of the inhomogeneity , the value of @xmath12 where the domain - wall turns unstable , and compare it with @xmath15 . finally , we investigate the dynamical states that occur when @xmath12 is decreased beyond @xmath14 . in this section we present the results of the simulations of the nld equations for @xmath135 and step - like @xmath58 . most of the phase - space analysis presented above was based on the assumption that @xmath58 and @xmath12 were small and we therefore take @xmath136 at @xmath28 , and @xmath137 at @xmath29 ; in this case , the critical value of the group - velocity is @xmath138 . because of the periodicity of the system , there is also a step of @xmath58 and , hence , another dw at @xmath139 , but we will , as before , focus attention on the dw solution near @xmath34 that connects a single - mode state @xmath132 to the left with a bimodal state to the right . we have found that for @xmath140 the system relaxes to one of the stationary dw s that are shown in fig . [ statfig ] . these domain - walls were obtained by numerical simulations with @xmath141 modes . these numerical results clearly demonstrate that the stationary dw s are stable when @xmath12 is not very close to @xmath130 ; this fact can not be derived from the phase - space analysis alone . comparing the domain - walls obtained from direct simulations of the nld equations with those obtained by numerical integration of the ordinary differential equation @xmath142 along the analytically obtained flow - lines , we found that the shape of the domain - wall is predicted very well for all values of @xmath12 that we consider here . the largest deviations occur for @xmath143 , when the dw s obtained by the two aforementioned methods have a relative spatial shift @xmath144 . the quantity @xmath145 ( see eq . [ chi ] ) ) is equal to @xmath146 for the single - mode , and @xmath147 for the bimodal state ( note that from @xmath22 it follows that @xmath148 in the bimodal state ) . now that we have checked that the predications that follow from the perturbation theory are correct when @xmath12 is not too close to @xmath15 , we will investigate what happens when @xmath149 . the numerical simulations that we will present below reveal that the strength and the steepness of the inhomogeneity , the value of @xmath12 and the initial conditions all may influence the dynamics . to sample the parameter space without going into too much detail , we have eventually restricted ourselves to inhomogeneities of the form @xmath150 where @xmath151 stands for the steepness of the inhomogeneity and @xmath45 is the size of the system . the middle part of the system is where the bimodal state exists . we have shifted the inhomogeneities away from @xmath34 because this makes the pictures for the time evolution of @xmath0 and @xmath1 that are presented below more clear . we take values of the inhomogeneity strength @xmath152 of @xmath153 and @xmath154 in order to obtain information on the dependence of the dynamics on the strength of the inhomogeneity . for the steepness @xmath151 , two different values were selected : @xmath155 for a smooth inhomogeneity , and @xmath156 for a steep inhomogeneity ; for the numerical simulation presented below , this last value is practically equivalent to a step - like inhomogeneity . as can be seen in fig . [ statfig ] , the location of the domain - wall shifts to larger values of @xmath3 when @xmath12 approaches @xmath15 , in agreement with the phase - space analysis . in this case , a large patch of the unstable single - mode state intrudes in the @xmath21 domain . eventually , this renders the domain - wall state unstable , and we define @xmath14 as the group - velocity for which this instability occurs ( see fig . [ schemafig1 ] ) . we found that the primary instability indeed occurs when @xmath12 has approached @xmath15 within a few percent . the large patch of unstable single - mode state turns then convectively unstable , but the total domain - wall state is absolutely stable . this means that noise , which is mainly due to the discretization of space and time in the numerics , is amplified in the region where we have the unstable single - mode state . due to the group - velocity , these fluctuations are advected towards the stable bimodal state ( mode @xmath0 ) or the stable single - mode state ( mode @xmath1 ) , where they are dissipated . to monitor where this instability occurs , we have followed the time evolution of the norm @xmath157 as a function of the group - velocity . when the instability of the domain - wall state occurs , the norm starts to oscillate with well - defined frequency . we checked for hysteric effects but could find none , and we conclude that the domain - wall turns unstable via a forward hopf - bifurcation . however , the fact that in this case the system is still absolutely stable , yields that hidden line plots of the domain - wall hardly show a clear dynamic state . when @xmath12 is close the @xmath14 , one can observe the instability best in the norm @xmath157 . we have measured @xmath14 for @xmath158 and various values of @xmath152 , and the results are listed below . @xmath15 & @xmath159 + 0.05 & -0.43 & -0.447 & 0.96 + 0.1 & -0.62 & -0.632 & 0.98 + 0.2 & -0.87 & -0.894 & 0.97 + 0.3 & -1.06 & -1.095 & 0.97 + the error - bar on the measurements of @xmath14 is of order @xmath160 . the most important conclusion that we can draw from this measurements is that @xmath14 is very close to @xmath15 , and that this is independent of the value of @xmath58 . in this section we investigate the fate of the domain - wall when @xmath12 is decreased beyond @xmath14 . it is important to emphasize that , unlike many other transitions to dynamical behavior , there is , according to our phase - space analysis , no stationary albeit unstable state when @xmath161 . thus , the single - mode and bimodal states are stable , each in its own domain , but there is no _ stationary _ domain - wall to connect these two asymptotic states . we study again inhomogeneities of the form ( [ inhomoeq1 ] ) , and take the group - velocity @xmath162 , @xmath163 and @xmath164 times the critical value @xmath165 . as initial conditions , we always took @xmath166 which is close to the stationary state for @xmath167 . most of the dynamical states fall into two distinct types of behavior . for the weak inhomogeneities , the dw moves irregularly back and forth around an average position . for stronger inhomogeneities , we often found that traveling kinks connecting states with the opposite signs of the amplitude @xmath1 were generated in the bimodal regime . we will describe in detail the dynamics that are found for the inhomogeneity with the steepness @xmath158 , which is the value of @xmath151 that we will focus on ; the other value of @xmath151 will merely serve to explore the generality of the behavior found for @xmath158 . at @xmath168 , we observe a disorderedly moving dw , and this is also the case for @xmath169 . we will focus now on @xmath170 and @xmath169 . the dynamics of the amplitudes @xmath0 and @xmath1 and the quantity @xmath171 are shown in fig . [ figfkv1 ] . as can be clearly seen in this figure , an essential dynamical degree of freedom is the position of the dw that moves irregularly around a certain mean position . this mean position shifts to the right when @xmath12 decreases . by extending the simulations to longer time intervals we have checked that the motion remains disordered . to inspect whether the dynamics are chaotic , we have studied effects of small perturbations in the initial conditions . we have performed two runs , one ( the unperturbed run ) starting from initial conditions obtained as the final state of previous simulations in the disordered regime ; for the perturbed run , the initial profile of @xmath43 was unaltered , but the value of @xmath1 at @xmath34 was diminished by @xmath172 . the data pertaining to the perturbed run will be distinguished by a prime . we have plotted @xmath173 in fig . [ fkv2]b , and @xmath174 and @xmath175 in fig . [ fkv2]a . clearly , the dynamics are sensitive to the small perturbation . the perturbation of the initial data is seen to grow and spread out , but only an area around the dw is affected ( fig [ fkv2]b ) . when a perturbation far from the dw is initiated , the same effect is eventually observed , but after a longer transient time , which is presumably the time the perturbation needs to reach the region around the dw . for the motion to be chaotic , we need to find an _ exponential _ divergence between @xmath0 and @xmath176 for some time - interval . a close inspection of the dynamics for @xmath0 and @xmath176 reveals that this is _ not _ the case ; @xmath0 and @xmath176 diverge but not exponentially . therefore , we suspect the disordered motion of the domain - wall to be due to the convectively amplified discretization noise . this scenario is somewhat similar to the dynamics described in @xcite . it seems that the dynamics , at least at the lowest order , can be described by a disordered drift of the dw , which leaves its shape intact . to check this , we have collapsed 100 snapshots from the time evolution of a , plotting @xmath0 vs. @xmath177 . all the curves fall , with a good approximation , on top of each other , which indicates that , to the lowest order , the dynamics can be adequately modeled by the only degree of freedom , viz . , the position of the dw . when the strength of the inhomogeneity is increased , a different type of behavior is observed . as an example , we will consider what happens at @xmath178 , although qualitatively the same behavior occurs at larger @xmath152 . the motion that occurs at @xmath170 is shown in fig . [ perdyn ] . a close inspection of the data reveals that , to the left of the dw , where @xmath1 is very small , zeroes of the function @xmath44 are generated periodically . each of them is then convected and amplified , leading to the generation of a traveling kink . we have found that the frequency of the kink generation goes approximately linearly with @xmath12 , and that it is nonzero at @xmath179 . the velocity of the kinks is slightly smaller than the group - velocity . this is not surprising : associated with the kink in @xmath1 is a small `` bump '' in @xmath0 ( see fig [ fkv2 ] ) . without this bump , the kink would travel with precisely the group - velocity of @xmath1 , but since the group - velocity of @xmath0 is opposite to that of @xmath1 , the small bump somewhat lowers the velocity of the kink . taking a smoother inhomogeneity ( @xmath180 ) suppresses the generation of kinks , and if they are still generated , this process is mixed with the irregular motion of the dw , as is shown in fig [ mix ] . the generation of kinks can even become irregular . in conclusion , when @xmath12 is beyond @xmath130 , the dynamical state may be represented by the disordered dw , regularly generated and traveling kinks , or a mixture between the two . in this section we relax the condition that the order parameters @xmath0 and @xmath1 are real , and study the behavior of a dw in a system of two coupled rgl equations with the group - velocity terms : [ rgl ] @xmath181 the main question here is to see whether the results obtained for the nld equations are relevant for the rgl equations . the complex order parameter poses the question of wavenumber selection , and the possibility of nonzero wavenumbers constitutes the main difference with the nld equations . we have found that , for the stationary states , the wavenumbers often , but not always , relax to zero when the initial conditions have nonzero wavenumbers . when the wavenumbers of @xmath0 and @xmath1 are constant and equal , the nld equations can be used to describe the stationary states ; when the wavenumbers are more general , modified nld equations model the stationary domain - walls , as we will discuss below . in the dynamical regime , small wavenumbers may be generated starting from initial conditions with zero wavenumbers . note that the usual eckhaus band of stable wavenumbers for a single rgl equation @xcite will be affected by the coupling between the two rgl equations . in particular , since for @xmath182 both the single - modes and the bimodal states with zero wavenumber are only marginally stable , one may expect that the size of the band of stable wavenumbers is directly related to @xmath58 ; and that for @xmath182 the band of stable wavenumbers closes . we have not pursued this question further . the wavenumbers of @xmath0 and @xmath1 , which we refer to as @xmath183 and @xmath184 , also have an influence on the stability borders of the single and bimodal states . when @xmath185 , the crossover between single and bimodal states occurs at @xmath186 , just as when @xmath187 . this can be shown by substituting plane - wave solutions with equal wavenumbers in the rgl equations and performing a rescaling of @xmath3 , @xmath188 and the amplitudes , similar to the rescaling that is used to scale out the growth - rate @xmath189 . apart from this scale transformation , the equations for zero wavenumber and equal wavenumber are equivalent . when @xmath190 , the stability borders for both the homogeneous single - mode and bimodal states shift away from @xmath186 , and in particular there exists a tiny parameter regime @xmath191 , for which both the single - mode and bimodal states are linearly unstable . the differences between of @xmath192 and @xmath193 and @xmath4 are of order @xmath194 , and we will not go into this in detail . in this section we focus on the case @xmath135 and explore numerically the mechanisms by which the wavenumber relaxes . the values of @xmath14 that we found are very close to those found in the simulations of the nld equations ( we will come back to this below ) . the simulations of eqs . ( [ rgl ] ) were performed for a periodic system of the size 400 . the inhomogeneity was of the form given in eq . ( [ inhomoeq1 ] ) with @xmath158 . the initial conditions were chosen to be dw - like , similar to eq . ( [ initialcon ] ) , but the wavenumbers @xmath183 and @xmath184 of the complex variables @xmath43 and @xmath44 in the initial states were allowed to be nonzero but constant , to study the wavenumber relaxation . in the variational case @xmath195 the system ( [ rglsys ] ) has a lyaponav functional @xmath18 with a minimum at @xmath187 . however , such states can not always be reached , since the periodic boundary conditions lead to conservation of the total phase difference across the system , as long as the amplitude remains nonzero . so , the wavenumber of the , e.g. , mode @xmath43 can be relaxed through the so - called phase - slips occuring at the points where @xmath196 vanishes @xcite . when @xmath0 is not close to zero , generation of such a phase - slip may lead to a significant increase of @xmath18 , which is forbidden . it is then possible for the system to end up in a _ local _ instead of a _ global _ minimum of @xmath18 . but when @xmath0 is small in some region , phase - slips easily occur , and the wavenumber relaxes to zero . when @xmath197 , the rgl equations are no longer variational , but for the stationary domain - walls the tendency to evolve to states with zero wavenumber is still present . we found that , the wavenumber of the @xmath1 mode always relaxes to zero , while for the @xmath0 mode , the wavenumber occasionally remains nonzero . the @xmath1 mode is close to zero in the single - mode region . phase slips can easily occur in this regime , and we have observed that @xmath184 always relaxes to zero , both for @xmath198 and @xmath199 . at @xmath200 , the local wavenumber diffuses , and , at @xmath197 , the wavenumber is advected to the single - mode region , where it relaxes . when the initial @xmath184 is large , the phase slips of @xmath1 occasionally occur in the double - mode region , but only as a transient . for the @xmath0 mode , two different mechanisms for relaxation of @xmath183 are observed . in the simplest case , the phase slips occur close to the dw , until @xmath183 falls to a level @xmath201 . it may also occur that the phase slips lead to a single - mode state consisting of alternating patches with @xmath196 and @xmath25 . in that case , both @xmath183 and @xmath184 relax to zero . the kink that is generated between these two different single - mode states is not a global , but rather a local minimum of @xmath18 corresponding to a metastable state . the stationary domain - walls of the coupled rgl equations are very similar to those of the nld equations . when @xmath185 , we can show this by substituting plane - waves of the form @xmath202 and @xmath203 into the rgl equations , which yields that the rgl equations are similar to the nld equations , up to the coefficient @xmath204 in front of the linear term . by the aforementioned scaling we can scale this factors out , and this yields the nld equations for @xmath16 and @xmath17 . therefore , when @xmath185 , the stationary domain - walls produced by the rgl equations can be described by the nld equations . the critical value of the group - velocity where the position of the domain - wall diverges is then also precisely given by the value of @xmath15 obtained for the nld equations . the effect of nonzero wavenumber on @xmath14 can in principle be more subtle , but in practice we could not find a difference between the values of @xmath14 for the nld equations and the rgl equations with @xmath185 . when @xmath205 , we can not scale out the wavenumbers . substituting plane - waves of the form @xmath206 and @xmath207 into the rgl equations , yields two nld - like equations were the first terms of the right - hand side @xmath0 and @xmath1 are replaced by @xmath208 and @xmath209 . only one of these pre - factors can be scaled out , so the ordinary nld equations are not correct . however , since the difference of @xmath183 and @xmath184 that occurs in stationary domain - walls is of order @xmath155 at most , and the effect on the rgl equations is of order @xmath194 , the nld equations are still valuable to give a lowest order description of the ensuing domain - walls . in principle , one could carry out the geometrical analysis for the equations with the @xmath208 and @xmath209 terms ; we will not give all the details here , but it can be shown that @xmath15 is perturbed by terms of order @xmath194 . this is consistent with the numerically observed increase of the value of @xmath14 with @xmath183 ( when @xmath210 ) ; this value increases at most a few percent when @xmath211 . to investigate the dynamical solutions of the rgl equations , we used the same inhomogeneities and values of @xmath12 as for the nld equations . the same initial conditions were used , so we restrict ourselves to the zero initial wavenumbers . for a steepness @xmath158 , we have found for almost all @xmath12 beyond @xmath130 and @xmath152 a disordered fluctuating dw state . the mean position of the dw shifts to the right with decreasing @xmath12 ; this effect becomes weaker when @xmath152 becomes larger . a slightly different state was observed for @xmath169 and @xmath212 , when the oscillations of the dw showed a strong periodic component but are nevertheless disordered ; the reason for this is not clear . we found that when @xmath152 is increased , the fluctuations of the position of the domain - wall in general decrease . the phase of @xmath0 remains zero in the dynamical state . the phase of @xmath1 departs from zero , and @xmath1 slowly develops wavenumbers @xmath213 . at longer times , these wavenumbers often remain constant in space and time . a clear exception is the aforementioned state with @xmath169 and @xmath212 , where periodic modulations of the wavenumber persist . for smoother inhomogeneities ( s=0.1 ) and small @xmath152 , the wave number generation is suppressed , and a traveling kink state is observed . this state is similar to the states produced by the nld equations , in the sense that the phase difference across such a kink is exactly @xmath214 . however , the generation of kinks is now no longer periodic . when @xmath215 , these kinks are not produced , and the local wavenumber is generated instead , just as in the case of the steeper inhomogeneity . in conclusion , both the stationary states and the critical value of @xmath12 for the rgl equations are almost the same as for the nld equations , while the dynamical states are somewhat different . the main mechanisms that play a role for the domain - walls of the rgl equations can be described by the nld equations . in this section we will study the coupled cgl equations : [ cgl ] @xmath216 a ~,\label{cgleqa}\\ \partial_t b - c \partial_x b & = & b + ( 1 + i c_1 ) \partial_x^2 b - \left[(1-i c_3 ) |b|^2 + ( 1-i c_2 ) g(x ) |a|^2\right ] b ~. \label{cgleqb}\end{aligned}\ ] ] these are the generic amplitude equations for left- and right traveling waves , and the group - velocity terms appear here naturally . the coefficients @xmath217 and @xmath218 can be obtained from a systematic expansion of the underlying equations of motion and play an essential role for the dynamics of the cgl equations . the nonlinear dispersion coefficients @xmath219 and @xmath218 are , in general , not equal , and we will find below that their difference will be a crucial parameter for the dw s . the behavior of a single cgle is already incredibly rich @xcite , and the situation for the coupled cgle s is of course not simpler . as a function of the coefficients @xmath14 , typical states in homogeneous coupled cgle s include single and bimodal phase - winding solutions , periodic solutions and spatio - temporal chaotic solutions @xcite . for fixed values of the coefficients , different states can coexist . in the following we will restrict ourselves to describing some of the interesting states that occur in the numerical simulations of the coupled cgle s when there is a cc that passes through its critical value . the numerical simulations were carried out similarly to the rgl model , in a periodic system of the size 400 . the inhomogeneity was of the form given in eq . ( [ inhomoeq1 ] ) , and we focus on the case @xmath180 , although we have performed some runs with @xmath158 . it turns out that only the details of the dynamics are different for this larger value of @xmath151 . as initial conditions we use , as before , the stationary state that is obtained when @xmath220 this means in particular that the critical value for @xmath20 is @xmath4 ; it has been shown by sakaguchi @xcite , that for periodic or disordered states the transition between single and bimodal behavior can occur for values different from @xmath4 . even with these restrictions , the parameter - space of the cgle s is too large to warrant a complete overview of the dynamical behavior . the nonlinear dispersion is essential for the fate of the domain - wall , and to restrict the search in parameter - space , we restrict ourselves to three different cases : ( a ) @xmath221 , ( b ) @xmath222 , ( c ) @xmath223 . the role of the group - velocity and @xmath224 is discussed briefly for each case . we have found many interesting dynamical states ; in many cases , the homogeneous single and bimodal states appear to be invaded by more complicated states that grow from the initial dw . by carefully adjusting the various parameters , it seems possible to smoothly proceed from regular periodic to completely disordered states . below we will only sketch a few of the dynamical regimes that we have found . a more systematic exploration is left for a further work . it is relevant to note that , when the cc constant is far from its critical value 1 , we did not find new behavior . when the nonlinear dispersion coefficients @xmath219 and @xmath218 are equal , the nonlinear term of eq . ( [ cgleqa ] ) ( similar for ( [ cgleqb ] ) ) can be written as @xmath225 . just as in the rgl case , ( @xmath226 ) differs only slightly from the value 1 for a dw state , so the nonlinear dispersion acts similar to the linear dispersion . the main effect of the dispersion is then to shift the wavenumber of the @xmath1-mode , but apart from this , the behavior of the cgl s is qualitatively similar to that of the coupled rgl s . when @xmath227 , @xmath17 is smaller then @xmath16 in the bimodal regime , and as a result , the critical group - velocity is seen to increase ( see the discussion on the effect of nonzero wavenumber on @xmath228 for the rgl equations ) . for example , when @xmath229 and @xmath230 , the selected wavenumber of @xmath1 is close to @xmath155 , and @xmath17 is a few percent smaller than @xmath16 . the critical value of the group - velocity is then increased by a few percent . when both @xmath224 and @xmath231 are different from zero , the wavenumber of @xmath1 can be large enough to shift the critical group - velocity substantially . for @xmath232 and @xmath233 , the wavenumber of @xmath1 is close to @xmath234 , and the amplitude of @xmath1 is approximately 20 percent smaller than that of @xmath0 . the critical group - velocity is then found to be between @xmath235 and @xmath236 . although this is quite different from the case where the wavenumbers are zero , the essence of the transition to the dynamical dw s is still given by the analysis for the nld equations . when the difference between @xmath219 and @xmath218 is small , we find qualitatively the same behavior as for @xmath221 . when @xmath238 , the nonlinear dispersion is no longer spatially independent , and this leads to oscillatory , spatially periodic or chaotic behavior . when we take both the linear dispersion and the group - velocity equal to zero , we already find various types of behavior , as shown in fig . [ cglfig1 ] . the oscillatory behavior shown in fig . [ cglfig1]a and in more detail in fig . [ cglfig2 ] arises from a feedback mechanism between local wavenumbers and amplitudes . when @xmath237 , gradients of @xmath16 and @xmath17 generate local wavenumber ( see fig . [ cglfig2]b ) , and this local wavenumber suppresses , via the diffusive term , the amplitudes . as a result of this feedback mechanism , the dw becomes oscillatory . the strength of the aforementioned feedback mechanism grows with the nonlinear dispersion , and above a certain threshold , we find that the oscillations do not stay confined around the dw , but spread out into the single - mode region . for instance , for @xmath239 we found that a periodic state is generated ( fig . [ cglfig1]b ) , that is similar to the periodic state described recently by sakaguchi @xcite . the dw itself becomes disordered . it should be noted that for these values of the coefficients , there are asymptotic single and bimodal phase - winding solutions that are linearly stable . finally , the nonlinear dispersion can become so large that also the bimodal state becomes periodic . the disorder seen in fig . [ cglfig1]c may be either a transient behavior or an established state ; our simulations were not conclusive , and we leave this for further work . here the coefficients of the cgl equations are such , that homogenous phase - winding solutions are unstable , and that even in homogeneous systems , i.e. , for @xmath58 fixed at @xmath240 , periodic states arise . the domain - wall here is disordered , and so this state shows the competition between linearly stable periodic states and a disordered domain - wall . the linear dispersion has a damping effect on the dynamics as for @xmath241 similar behavior is observed for slightly higher values of the nonlinear dispersion . a nonzero group - velocity has a more complicated effect on the dynamics , as it breaks the reflection symmetry . the oscillations that occur for small nonlinear dispersion are damped , because the group - velocity terms advect the local wavenumbers away from the dw , and therefore they suppress the aforementioned feedback effect . for instance , for @xmath242 and @xmath243 , we have observed stationary states up to @xmath244 . on the other hand , the symmetry breaking can have a destabilizing effect on the dynamics : the periodic state shown in fig . [ cglfig1]b becomes disordered when @xmath12 is nonzero . when we move away from the line @xmath222 , we find that spatio - temporal chaos occurs quite easily . we have focused on the case @xmath245 . for @xmath246 , a periodic state similar to the one depicted in figure [ cglfig1]b occurs . when @xmath218 is increased , both the periodic and the bimodal states become gradually disordered ( figs . [ cglfig3]a and [ cglfig3]b ) . when @xmath200 , the chaotic state consists of more or less stationary , irregularly growing and decaying pulses , but when @xmath197 , this quasi - stationary character is destroyed . we have checked that in the disordered regime , two slightly different initial conditions diverge throughout the whole domain , which shows that these states are an example of spatio - temporal chaos . note that this occurs for values of @xmath217 and @xmath218 for which the homogeneous single and bimodal state are linearly stable . apparently the dw acts as a `` seed '' for the disorder , that then spreads out and completely destroys the plain - wave state . we have concentrated here on the case @xmath222 , although it should be stressed that when @xmath219 is not exactly @xmath247 , we observed similar behavior . the choice @xmath222 merely serves to limit ourselves in exploring the parameter - space . in this paper we have considered domain walls between single - mode and bimodal states for three types of coupled equations with a spatially dependent coupling coefficient . in the simplest case of two coupled nld equations with the group - velocity terms , we were able to reduce the description of stationary configurations to a single non - autonomous second - order ordinary differential equation , that was used to determine analytically a necessary condition for the existence of a stationary dw in terms of the group - velocity . we have found that our prediction for the destabilization of such a stationary dw is in good agreement with numerical simulations , and we have found chaotically oscillating dw s in the case when the group - velocity is beyond the corresponding threshold . for two coupled rgl equations we have found a similar scenario . finally , for the coupled cgl equations , we have found that , in most cases , the dw s are unstable , even when the group - velocity is zero , and spatio - temporal disordered states often occur in this model . in the future , it would be interesting to investigate the competition between the various states of the coupled cgl equations . in particular , not much is known about the periodic states that seem to play an important role here . possible research subjects include the development of analytical solutions , the development of counting arguments @xcite and the competition between periodic and phase - winding solutions . the effect of an inhomogeneity as studied in this chapter on the various states may be a valuable tool in probing the states that occur for constant @xmath20 . the effect of nonzero group - velocity on the domain - walls in the cgl equations is poorly understood ; it would be interesting to see whether the divergence and subsequent instability of the domain - wall , as observed for the nld equations and cgl equations , still has some relevance for the cgl equations when @xmath219 and @xmath218 are sufficiently different . we appreciate useful discussions with wim van saarloos , willem van de water and bert peletier .

we investigate two - way and one - way single - photon quantum key distribution ( qkd ) protocols in the presence of loss introduced by the quantum channel . our analysis is based on a simple precondition for secure qkd in each case . in particular , the legitimate users need to prove that there exists no separable state ( in the case of two - way qkd ) , or that there exists no quantum state having a symmetric extension ( one - way qkd ) , that is compatible with the available measurements results . we show that both criteria can be formulated as a convex optimisation problem known as a semidefinite program , which can be efficiently solved . moreover , we prove that the solution to the dual optimisation corresponds to the evaluation of an optimal witness operator that belongs to the minimal verification set of them for the given two - way ( or one - way ) qkd protocol . a positive expectation value of this optimal witness operator states that no secret key can be distilled from the available measurements results . we apply such analysis to several well - known single - photon qkd protocols under losses .

quant-ph0611151

quantum key distribution ( qkd ) protocols typically involve a two - step procedure in order to generate a secret key @xcite . first , the legitimate users ( alice and bob ) perform a set of measurements on effective bipartite quantum states that are distributed to them . as a result , they end up with a classical joint probability distribution , that we shall denote as @xmath0 , describing their outcomes . the second step consists of a classical post - processing of the data @xmath1 . it requires an authenticated classical channel , and it includes post - selection of data , error - correction to reconcile the data , and privacy amplification to decouple the data from a possible eavesdropper ( eve ) @xcite . in order to create the correlated data @xmath1 , qkd schemes usually require alice to prepare some non - orthogonal quantum states @xmath2 with a priori probabilities @xmath3 that are sent to bob . on the receiving side , bob measures each received signal with a _ positive operator value measure _ ( povm ) @xmath4 . generalising the ideas introduced by bennett _ _ in ref . @xcite , the signal preparation process in this kind of schemes can alternatively be thought of as follows : alice produces first bipartite states @xmath5 and , afterwards , she measures the first subsystem in the orthogonal basis @xmath6 corresponding to the measurement operators @xmath7 . this action generates the signal states @xmath2 with a priori probabilities @xmath3 . the reduced density matrix of alice , @xmath8 , depends only on the probabilities @xmath3 and on the overlap of the signals states @xmath2 . this means , in particular , that @xmath9 is always fixed by the preparation process and can not be modified by eve . in order to include this information in the measurement process one can add to the observables @xmath10 , measured by alice and bob , other observables @xmath11 such that the observables @xmath12 form a tomographic complete set of alice s hilbert space @xcite . from now on , we will consider that the data @xmath1 and the povm @xmath13 include also the observables @xmath14 . the classical post - processing of @xmath1 can involve either two - way or one - way classical communication . two - way classical communication protocols can tolerate a higher error rate than one - way communication techniques @xcite . on the other hand , one - way post - processing methods typically allow to derive simpler unconditional security proofs for qkd than those based on two - way communication @xcite . in this last paradigm , two different cases can be considered : _ reverse reconciliation _ ( rr ) refers to communication from bob to alice , and _ direct reconciliation _ ( dr ) permits only communication from alice to bob . ( see , for instance , refs . @xcite . ) an essential question in qkd is to determine whether the correlated data @xmath1 allow alice and bob to generate a secret key at all during the second phase of qkd . here we consider the so - called _ trusted device scenario _ , where eve can not modify the actual detection devices employed by alice and bob , as used in refs . we assume that the legitimate users have complete knowledge about their detection devices , which are fixed by the actual experiment . the case of two - way classical post - processing has been analysed in ref . @xcite , where it was proven that a necessary precondition for secure two - way qkd is the provable presence of quantum correlations in @xmath1 . that is , it must be possible to interprete @xmath1 , together with the knowledge of the corresponding observables @xmath13 , as coming _ exclusively _ from an entangled state . otherwise , no secret key can be distilled from @xmath1 . in order to deliver this entanglement proof any separability criteria ( see , for instance , ref . @xcite and references therein ) might be employed . the important question here is whether the chosen criterion can provide a necessary and sufficient condition to detect entanglement even when the knowledge about the quantum state is not tomographic complete . it was proven in ref . @xcite that entanglement witnesses ( ews ) fulfill this condition . an ew is an hermitian operator @xmath15 with a positive expectation value on all separable states @xcite . so , if a state @xmath16 obeys @xmath17 , the state @xmath16 must be entangled . with this separability criterion , refs . @xcite analysed three well - known qubit - based qkd schemes , and provided a compact description of a minimal verification set of ews ( _ i.e _ , one that does not contain any redundant ew ) for the four - state @xcite and the six - state @xcite qkd protocols , and a reduced verification set of ews ( _ i.e. _ , one which may still include some redundant ews ) for the two - state @xcite qkd scheme , respectively . these verification sets of ews allow a systematic search for quantum correlations in @xmath1 . one negative expectation value of one ew in the set suffices to detect entanglement . to guarantee that no verifiable entanglement is present in @xmath1 , however , it is necessary to test _ all _ the members of the set . unfortunately , to find a minimal verification set of ews , even for ideal qubit - based qkd schemes , is not always an easy task , and it seems to require a whole independent analysis for each protocol , let alone for higher dimensional qkd schemes @xcite . ( see also ref . @xcite . ) also , one would like to include in the analysis the attenuation introduced by the quantum channel , not considered in refs . @xcite , and which represents one of the main limitations for optical realisations of qkd . one central observation of this paper is very simple , yet potentially very useful : given _ any _ qubit - based two - way qkd scheme , one can search for quantum correlations in @xmath1 by just applying the _ positive partial transposition _ ( ppt ) criterion @xcite adapted to the case of a quantum state that can not be completely reconstructed . this criterion provides a necessary and sufficient entanglement verification condition for any qubit - based qkd protocol even in the presence of loss introduced by the channel , since , in this scenario , only _ nonpositive partial transposed _ ( npt ) entangled states exist . moreover , it is rather simple to evaluate in general since it can be cast into the form of a convex optimisation problem known as semidefinite program ( sdp ) @xcite . such instances of convex optimisation problems can be solved efficiently , for example by means of interior - point methods @xcite . this means , in particular , that this criterion can be applied to any qubit - based qkd scheme in a completely systematic way . one - way qkd schemes can be analysed as well with sdp techniques . it was shown in ref . @xcite that a necessary precondition for one - way qkd with rr ( dr ) is that alice and bob can prove that there exists no quantum state having a symmetric extension to two copies of system @xmath18 ( system @xmath19 ) that is compatible with the observed data @xmath1 . this kind of states ( with symmetric extensions ) have been analysed in detail in refs . @xcite , where it was proven that the search for symmetric extensions for a given quantum state can be stated as a sdp . ( see also refs . @xcite . ) here we complete the results contained in ref . @xcite , now presenting specifically the analysis for the case of a lossy channel . both qkd verification criteria mentioned above , based on sdp techniques , also provide a means to search for witness operators for a given two - way or one - way qkd protocol in a similar spirit as in refs . any sdp has an associated dual problem that represents also a sdp @xcite . this dual problem can be used to obtain a certificate of infeasibility whenever the primal problem is actually infeasible . most importantly , it can be proven that the solution to this dual problem corresponds to the evaluation of an optimal witness operator , that belongs to the minimal verification set of them for the given protocol , on the observed data @xmath1 . a positive expectation value of this optimal witness operator indicates that no secret key can be distilled from the observed data @xmath1 . the paper is organised as follows . in sec . ii we introduce the qkd verification criteria for two - way and one - way qkd in more detail , and we show how to cast them as primal sdps . then , in sec . iii , we present the dual problems associated to these primal sdps , and we show that the solution to these dual problems corresponds to evaluating an optimal witness operator on the observed data @xmath1 for the given protocol . these results are then illustrated in sec . iv , where we investigate in detail the two - state qkd protocol @xcite in the presence of loss . the analysis for other qubit - based qkd schemes is completely analogous , and we include very briefly the results of our investigations on other qkd protocols in an appendix . finally , sec . v concludes the paper with a summary . our starting point is the observed joint probability distribution @xmath1 obtained by alice and bob after their measurements @xmath13 . this probability distribution defines an equivalence class @xmath20 of quantum states that are compatible with it , @xmath21 by definition , every @xmath22 can represent the state shared by alice and bob before their measurements . in single - photon qkd schemes in the presence of loss , any state @xmath22 can be described on an hilbert space @xmath23 , with @xmath24 and @xmath25 denoting , respectively , alice s and bob s hilbert spaces , and where the subscript indicates the dimension of the corresponding hilbert space . to see this , we follow the signal preparation model introduced previously , where alice prepares states @xmath26 , and , afterwards , she measures the first subsystem in the orthogonal basis @xmath6 . using neumark s theorem @xcite , we can alternatively describe the preparation process as alice producing first bipartite states on @xmath27 and , afterwards , she measures the first subsystem with a povm @xmath28 . ( see also ref . @xcite . ) to include the loss of a photon in the quantum channel , we simply enlarge bob s hilbert space from @xmath29 to @xmath25 by adding the vacuum state @xmath30 . let us now consider two - way qkd protocols . whenever the observed joint probability distribution @xmath1 , together with the knowledge of the corresponding measurements performed by alice and bob , can be interpreted as coming from a separable state @xmath31 then no secret key can be distilled from the observed data @xcite . in @xmath23 only npt entangled states exist and , therefore , a simple necessary and sufficient criterion to detect entanglement in this scenario is given by the ppt criterion @xcite : a state @xmath32 is separable if and only if its partial transpose @xmath33 is a positive operator . partial transpose means a transpose with respect to one of the subsystems @xcite . such a result is generally not true in higher dimensions . _ observation 1 _ : consider a qubit - based qkd scheme in the presence of loss where alice and bob perform local measurements with povm elements @xmath34 and @xmath35 , respectively , to obtain the joint probability distribution of the outcomes @xmath1 . then , the correlations @xmath1 can originate from a separable state if and only if there exists @xmath22 such as @xmath36 . _ proof_. if @xmath1 can originate from a separable state , then there exists @xmath31 such as @xmath37 . moreover , we have that any separable state satisfies @xmath38 . to prove the other direction , note that if there exists @xmath39 such that @xmath36 then , since @xmath40 , we find that @xmath16 must be separable @xcite . @xmath41 to determine whether there exists @xmath22 such as @xmath36 can be solved by means of a primal semidefinite program ( sdp ) . this is a convex optimisation problem of the following form : @xmath42 where the vector @xmath43 represents the objective variable , the vector @xmath44 is fixed by the particular optimisation problem , and where the matrices @xmath45 and @xmath46 are hermitian matrices . the goal is to minimise the linear function @xmath47 subjected to the linear matrix inequality ( lmi ) constraint @xmath48 @xcite . if the vector @xmath49 , then the optimisation problem given by eq . ( [ primalsdp_marcos ] ) reduces to find whether the lmi constraint can be satisfied for some value of the vector @xmath50 or not . in this case , the sdp is called a _ feasibility problem_. remarkably , sdps can be solved with arbitrary accuracy in polynomial time , for example by means of interior - point methods @xcite . according to observation @xmath51 , we can find whether there exists a separable state that belongs to the equivalence class @xmath20 just by solving the following feasibility problem @xcite : @xmath52 where the objective variable @xmath50 is used to parametrise the density operators @xmath16 . the method used to parametrise @xmath16 is discussed in detail in sec . [ parametr ] . one - way rr ( dr ) qkd schemes require from alice and bob to show that there exists no quantum state @xmath53 with a symmetric extension to two copies of system @xmath18 ( system @xmath19 ) @xcite . a state @xmath16 is said to have a symmetric extension to two copies of system @xmath18 if and only if there exists a tripartite state @xmath54 , with @xmath55 , and where @xmath56 , such that @xcite : @xmath57 where the swap operator @xmath58 satisfies @xmath59 . this definition can be easily extended to cover also the case of symmetric extensions of @xmath16 to two copies of system @xmath19 , and also of extensions of @xmath16 to more than two copies of system @xmath18 or of system @xmath19 @xcite . to find whether @xmath60 has a symmetric extension to two copies of system @xmath18 can be solved with the following feasibility problem : @xmath61=\rho_{ab}({\bf x } ) , \\ \nonumber & & \rho_{aba'}({\bf x})\geq{}0.\end{aligned}\ ] ] note that this sdp does not include the constraint @xmath62 because non - negativity of the extension @xmath63 , together with the condition @xmath64=\rho_{ab}({\bf x})$ ] , already implies non - negativity of @xmath65 . the sdp for one - way qkd with dr can be obtained in a similar way . to actually implement the sdps given by eq . ( [ primalsdp_two_way ] ) and eq . ( [ primalsdp_one_way ] ) , one can parametrise @xmath16 and @xmath66 such that some constraints are automatically fulfilled . in particular , one can choose an operator basis of hermitian matrices @xmath67 for each hilbert space @xmath68 . these matrices @xmath69 can be taken such as they satisfy the following two conditions : @xmath70 , and @xmath71 . in the case of qubit systems , the pauli matrices @xmath72 can be selected , where the matrix @xmath73 denotes the identity operator @xmath74 . for systems on @xmath75 , we can use the gell - mann operators , that we shall denote as @xmath76 . with this representation , a general state @xmath40 can be written as @xmath77 where the operators @xmath78 , the coefficients @xmath79 are given by @xmath80 , and @xmath81 because of normalisation . ( [ eqnew ] ) allows us to describe any bipartite density operator in terms of a fixed number of real parameters @xmath79 . the knowledge of alice and bob s povms @xmath82 and @xmath83 , respectively , together with the observed probability distribution @xmath1 , determines the equivalence class of compatible states @xmath20 . each povm element @xmath34 and @xmath35 can also be expanded in the appropriate operator basis as @xmath84 , and @xmath85 , for some coefficients @xmath86 and @xmath87 , respectively . according to eq . ( [ eq_class ] ) , to guarantee that @xmath88 [ first constraint in eq . ( [ primalsdp_two_way ] ) and in eq . ( [ primalsdp_one_way ] ) ] , we obtain that the coefficients @xmath79 must satisfy the following conditions , @xmath89 that is , some coefficients @xmath79 are fixed by the known parameters @xmath86 , @xmath87 , and @xmath1 . any operator @xmath53 can then always be written in the following way , @xmath90 where @xmath91 corresponds to the part of @xmath92 that is completely determined by the parameters @xmath86 , @xmath87 , and @xmath1 . it can be expressed as @xcite @xmath93 where @xmath94 denotes a multi - index set labeling those combinations of the indexes @xmath95 and @xmath96 such that @xmath79 is fixed by eq . ( [ cond_coeff ] ) . ( see also ref . @xcite . ) with this representation for @xmath65 , the sdp given by eq . ( [ primalsdp_two_way ] ) can now be written as @xcite @xmath97 where the symbol @xmath98 denotes direct sum . let us compare the second part of eq . ( [ primalsdp_marcos ] ) with the second part of eq . ( [ primalsdp_two_way_b ] ) . the objective variables @xmath99 are now given by the coefficients @xmath79 of @xmath65 , with @xmath100 , the matrix @xmath45 is given by @xmath101 , and the matrices @xmath46 are those operators @xmath102 with @xmath100 . in the sdp given by eq . ( [ primalsdp_one_way ] ) we need to parametrise as well the quantum state @xmath103 . the second constraint in eq . ( [ primalsdp_one_way ] ) imposes that @xmath103 must remain invariant under permutation of systems @xmath18 and @xmath104 . this can be done with the following parametrisation @xcite : @xmath105 with @xmath106 and @xmath107 . to guarantee that @xmath108 [ third constraint in eq . ( [ primalsdp_one_way ] ) ] , the state coefficients of @xmath16 and @xmath103 need to fulfill the following conditions , @xmath109 that is , some of the state parameters of @xmath103 are already fixed by the coefficients of @xmath16 . to simplify the notation used later on , we shall collect the objective variables of the sdp given by eq . ( [ primalsdp_one_way ] ) within two different groups of them : the vector @xmath110 contains those coefficients @xmath79 of @xmath16 not fixed by eq . ( [ cond_coeff ] ) , and the vector @xmath111 contains those coefficients @xmath112 of @xmath103 not fixed by eq . ( [ last_ap2 ] ) . with this parametrisation , the first three constraints in eq . ( [ primalsdp_one_way ] ) are fulfilled automatically and the sdp given by eq . ( [ primalsdp_one_way ] ) can be reduced to solve the following one @xmath113 in this section we show how to rephrase the qkd verification criteria introduced in the previous section into a search for appropriate witness operators . in order to do this , we use the dual problems associated with the primal sdps given by eq . ( [ primalsdp_two_way_b ] ) and eq . ( [ primalsdp_one_way_b ] ) , respectively . in particular , we prove that the solutions to these dual problems correspond to the evaluation of an optimal witness operator , that belongs to the minimal verification set of them for the given two - way or one - way qkd protocol , on the observed data @xmath1 . a positive expectation value of this optimal witness operator states that no secret key can be distilled from the observed data @xmath1 . this approach has already been considered for the symmetric extension case in ref . @xcite , and also for a slightly different scenario in ref . our main motivation here is to show specifically that this relationship still holds even if we restrict ourselves to partial information about the quantum state . a detailed discussion on some duality properties that guarantee that the solution to these dual problems can actually be associated with a witness operator is included in appendix [ ap_b ] . let us first introduce the dual problem associated to the primal sdp given by eq . ( [ primalsdp_marcos ] ) . it has the following form @xcite : @xmath114 where the hermitian matrix @xmath115 is now the objective variable . this matrix is positive semidefinite @xmath116 and is subjected to several linear constraints of the form @xmath117 . in this section we show that the solution to the dual problem associated with the sdp given by eq . ( [ primalsdp_two_way_b ] ) corresponds to the evaluation of an optimal _ decomposable _ ews ( dews ) @xcite on the observed data @xmath1 . ( see also ref . @xcite . ) an ew @xmath15 is called decomposable if and only if there exist two positive operators @xmath118 , and a real parameter @xmath119 $ ] , such that @xmath120 @xcite . in @xmath23 all ews are dews . in what follows , we establish this connection explicitly via the dual problem . the sdp given by eq . ( [ primalsdp_two_way_b ] ) can be transformed into a slightly different , but completely equivalent , form as follows ( see appendix [ ap_b ] ) , @xmath121 where @xmath122 denotes an auxiliary objective variable . according to eq . ( [ dual_marcos ] ) , the dual problem associated with eq . ( [ ptprimal ] ) can be written as @xmath123 \\ \nonumber \text{subject to } & & z \geq 0 \\ \nonumber & & \text{tr}(z)= 1 \\ & & \text{tr}[(s_{kl } \oplus s_{kl}^{\gamma } ) z]=0\ ; \forall kl \not \in i.\end{aligned}\ ] ] the structure of all the matrices which appear in this dual problem is the direct sum of two different matrices . then , without loss of generality , we can assume that the same block structure is satisfied for @xmath115 , _ i.e. _ , @xmath124 . this means , in particular , that the objective function in eq . ( [ ptdual ] ) can now be re - expressed as @xmath125 & = & \text{tr}[(z_1 + z_2^{\gamma})\rho_{\text{fix}}]\\ \label{decomp_wit } & \equiv & \text{tr}(w\rho_{\text{fix}}),\end{aligned}\ ] ] where we have used the property @xmath126 and , in the last equality , we defined the operator @xmath127 . next we show that @xmath15 is a dews . for that , note that the semidefinite constraint @xmath128 implies @xmath129 . moreover , the witness is normalised , since @xmath130 implies @xmath131 . to conclude , we use the remaining equality constraints , @xmath132=0\ \forall kl \not \in i$ ] , to show that to evaluate the expectation value of @xmath15 one only needs to consider @xmath91 . that is , @xmath15 belongs to the minimal verification set of ews for the given qkd protocol , and its expectation value can be obtained from the observed data @xmath1 only @xcite . using the ansatz @xmath133 , and @xmath134 , the equality constraints impose @xmath135 . hence , the dew @xmath15 has the following structure @xmath136 with @xmath137 . combining eq . ( [ rho_class ] ) and eq . ( [ wit_structure ] ) , we obtain @xmath138=\text{tr}(w \rho_{\text{fix}})=\sum_{kl \in i } w_{kl } x_{kl}$ ] . whenever the solution to the dual problem given by eq . ( [ ptdual ] ) delivers @xmath139\equiv{}\text{tr}(w \rho_{\text{fix}})\geq{}0 $ ] then no secret key can be distilled from the observed data @xmath1 with two - way classical communication . to see this , note that , by definition , eq . ( [ ptdual ] ) guarantees that there exists no other dew @xmath140 , that belongs to a verification set of them for the given qkd protocol , such that @xmath141 . in this part we use the dual problem associated with the sdp given by eq . ( [ primalsdp_one_way_b ] ) to show that its solution corresponds to the evaluation of an optimal witness operator for the case of states with symmetric extensions . we shall follow the method introduced in ref . @xcite , but now we will consider specifically the case of partial knowledge about the quantum state . ( see also ref . @xcite . ) like in the previous section , the feasibility problem given by eq . ( [ primalsdp_one_way_b ] ) can be transformed as follows ( see appendix [ ap_b ] ) , @xmath142 with @xmath143 , _ e.g. _ , in our case @xmath144 . the inclusion of the factor @xmath145 in eq . ( [ primalext ] ) does not alter its result and , as we will see at the end of this section , it gives the correct normalisation for the witnesses . for convenience , we will express the state @xmath146 in terms of a map @xmath147 that takes an arbitrary hermitian operator @xmath148 , with @xmath149 , to the hermitian operator @xmath150,\end{aligned}\ ] ] let @xmath91 be again the part of @xmath151 that is fixed by the parameters @xmath86 , @xmath87 , and @xmath1 . without loss of generality , we consider the following structure for @xmath91 : @xmath152 , where the multi - index @xmath94 has the same meaning as before , _ i.e. _ , it labels those combinations of the indexes @xmath153 and @xmath154 such that @xmath79 is fixed by eq . ( [ cond_coeff ] ) . using eq . ( [ map ] ) , we can rewrite @xmath146 in the following compact way , @xmath155 where the hermitian matrices @xmath156 can be grouped into two different sets , @xmath157 and where the multi - index @xmath158 is used to label both different combinations of the indices @xmath159 , and @xmath160 . the dual problem associated with eq . ( [ primalext ] ) can now be written as @xmath161 \\ \nonumber \text{subject to } & & z \geq 0 \\ \nonumber & & \text{tr}(z)=d_a \\ \nonumber & & \text{tr}[z \lambda(s_{kl})]= 0\ \;\forall kl \not \in i \\ \nonumber & & \text{tr}(z g_j)=0\ \;\forall j.\end{aligned}\ ] ] next we search for the most general form of a possible solution @xmath115 for this dual problem . it will enable us to extract the most compact form of a witness operator for the symmetric extendibility problem . all the linear constraints on the operator @xmath115 contained in eq . ( [ dualext ] ) , as well as the objective function itself , are invariant under the swap operator @xmath58 , which exchanges the first and the third subsystem . moreover , the positive semidefinite constraint @xmath162 is also satisfied since @xmath58 is a unitary operator , _ i.e. _ , @xmath163 . this means that , if @xmath115 is a solution for the dual problem , the operator @xmath164 is also a possible solution for it , since it fulfills all the constraints and it gives exactly the same expectation value . following a similar argumentation , also the equal mixture of @xmath115 and @xmath164 , _ i.e. _ , @xmath165 , is as well a possible solution . therefore , without loss of generality , we can consider that @xmath115 is invariant under the swap operator @xmath58 . under this assumption , it turns out that @xmath115 can be decomposed as follows @xmath166.\nonumber\end{aligned}\ ] ] let us now analyse in more detail the linear constraints on @xmath115 given in eq . ( [ primalext ] ) . each linear constraint cancels one of the coefficients @xmath167 . for instance , the constraint @xmath168 imposes @xmath169 . then , we can remove all these linear constraints , except the normalisation condition @xmath170 , by just making the proper coefficients @xmath167 in eq . ( [ new_m ] ) equal to zero . this way we arrive at the following form for the variable @xmath115 , which we shall denote by @xmath171 , @xmath172.\nonumber\end{aligned}\ ] ] that is , if we assume the form @xmath171 for the variable @xmath115 in the dual problem given by eq . ( [ dualext ] ) , then the linear constraints are fulfilled automatically . substituting the variable @xmath115 with @xmath171 in eq . ( [ dualext ] ) we obtain the following shorter form for the dual problem , @xmath173 \\ \nonumber \text{subject to } & & z^ * \geq 0 \\ \nonumber & & \text{tr}(z^*)=d_a.\end{aligned}\ ] ] now , in order to extract a witness operator from the hermitian operator @xmath171 , we follow the method proposed in ref . @xcite . in particular , every map @xmath147 has associated an adjoint map @xmath174 defined as @xmath175=\text{tr}[\lambda^{\dag}(u)\ v]$ ] for any hermitian operators @xmath176 , and @xmath177 . with this definition , we can rewrite the objective function in eq . ( [ dualext2 ] ) as @xmath178 = \text{tr}[\lambda^{\dag}(z^*)\rho_{\text{fix } } ] \equiv \text{tr}(w_{\text{sym } } \rho_{\text{fix}}),\ ] ] where we defined @xmath179 as the desired witness operator for the symmetric extendibility problem . in the remaining part of this section , we obtain the general structure of the witness operator @xmath180 . for that , we simply apply the adjoint map @xmath181 to the operator @xmath171 , set the resulting operator equal to an operator of arbitrary form @xmath182 , solve the equality constraint , and , finally , formulate the dual problem in terms of this new operator @xmath182 . the map @xmath181 can be written as @xcite , @xmath183.\ ] ] setting @xmath184 equal to an arbitrary hermitian operator @xmath185 we obtain the following equality constraint @xmath186 since we have expressed every operator in terms of an operator basis , the equality constraint can only be fulfilled if the coefficients @xmath187 of @xmath171 , and the coefficients @xmath188 of @xmath182 , are related via : @xmath189 now , instead of considering the matrix @xmath171 as the objective variable of the dual problem , we can equivalently consider the matrix @xmath182 as the free variable . in order to do so , we only need to translate the positive semidefinite constraint @xmath190 together with the normalisation condition @xmath191 included in eq . ( [ dualext2 ] ) into new constraints on @xmath182 . this can be done by using eq . ( [ connectionzwsym ] ) . this way , we arrive at the following form for the dual problem : @xmath192 where the variable @xmath182 represents a witness operator for the symmetric extendibility problem . moreover , from eq . ( [ connectionzwsym ] ) we obtain that @xmath182 can always be expressed as @xmath193 that is , @xmath182 belongs to the minimal verification set of witnesses for the given one - way ( rr ) qkd protocol . like in the previous section , whenever the solution to the dual problem given by eq . ( [ dualext3 ] ) delivers @xmath194 then no secret key can be distilled from the observed data @xmath1 with one - way rr . the case of one - way qkd with dr can be analysed in a similar way . in this section we study the two - state qkd protocol @xcite , both for the case of two - way and one - way classical communication . the analysis for other qubit - based qkd schemes is completely analogous , and we include very briefly the results of our investigations on other well - known qkd protocols in appendix [ other_qkd ] . we refer here to single - photon implementations of the qubit . the state of the qubit is described , for instance , by some degree of freedom in the polarisation of the photon . in our calculations we follow the approach introduced in sec . [ verif_criteria ] , although similar results could also be obtained using the witness approach presented in sec . [ sec2a ] . the numerical evaluations are performed with the freely - available sdp solver sdpt3 - 3.02 @xcite , together with the input tool yalmip @xcite . we shall consider that the observed joint probability distribution @xmath1 originates from alice and bob measuring the following quantum state @xmath195+p\rho_a\otimes{\mbox{$ | vac \rangle $ } } _ b{\mbox{$ \langle vac | $ } } , \end{aligned}\ ] ] where @xmath196 $ ] denotes the probability that bob receives the vacuum state @xmath30 , @xmath197 $ ] represents an error parameter ( or depolarising rate ) of the channel , @xmath198 is the identity operator on alice s hilbert space , @xmath199 represents a unitary operator acting on bob s system , @xmath200 denotes the effective bipartite state initially prepared by alice in the given qkd protocol , @xmath9 represents alice s reduced density matrix ( _ i.e. _ , @xmath201 ) , and the operator @xmath202 is given by @xmath203 . the quantum state given by eq . ( [ channel_eq ] ) defines one possible eavesdropping interaction . but our analysis can straightforwardly be applied to other quantum channels , as it depends only on the probability distribution @xmath1 that characterises the results of alice s and bob s measurements . we include the operator @xmath199 in eq . ( [ channel_eq ] ) to model the collective noise ( or correlated noise ) introduced by the quantum channel ( _ e.g. _ , optical fiber ) @xcite . this noise arises from the fluctuation of the birefringence of the optical fiber which alters the polarisation state of the photons . when this fluctuation is slow in time , its effect can be described with a unitary operation @xcite . for simplicity , we shall consider that @xmath199 is parametrised only with one real parameter @xmath204 . in particular , we choose @xmath205 with @xmath206 $ ] . if @xmath207 no collective noise is present and eq . ( [ channel_eq ] ) describes a depolarising channel with loss . in order to illustrate our results , we calculate an upper bound on the tolerable depolarising rate @xmath208 as a function of the photon loss probability @xmath196 $ ] . moreover , for simplicity , we take only two different values of the angle @xmath204 . for instance , we choose @xmath207 and @xmath209 . these three parameters , @xmath208 , @xmath210 , and @xmath204 , allow us to evaluate the performance of a qkd protocol when the quantum channel is described by eq . ( [ channel_eq ] ) . one could also select other figures of merit in order to evaluate a protocol , such as the quantum bit error rate ( qber ) . this is the rate of events where alice and bob obtain different results . it refers to the sifted key , _ i.e _ , it considers only those events where the signal preparation and detection methods employ the same polarisation basis . we include as well an analytic expression for the qber for the given qkd protocol . the two - state protocol @xcite is one of the simplest qkd protocols . it is based on the random transmission of only two nonorthogonal states , @xmath211 and @xmath212 . alice chooses , at random and independently every time , a bit value @xmath213 , and prepares a qubit in the state @xmath214 , with @xmath215 and @xmath216 , that is sent it to bob . on the receiving side , bob measures the qubit he receives in a basis chosen at random within the set @xmath217 . the loss of a photon corresponds to a projection onto the vacuum state @xmath218 . bob could also employ a different detection method defined by a povm with the following operators : @xmath219 with @xmath220 , @xmath221 , and @xmath222 . in this last case , alice s bit value @xmath213 is associated with the operator @xmath223 , while the operator @xmath224 represents an inconclusive result . this is the approach that we shall consider here . the preparation process can be thought of as alice prepares first the bipartite signal state @xmath225 with @xmath226 . the fact that the reduced density matrix of alice is fixed and can not be modified by eve is vital to guarantee the security of this scheme . otherwise , the joint probability distribution @xmath1 alone does not allow alice and bob to distinguish between the entangled state @xmath227 and the separable one @xmath228 and @xmath229 such as alice has complete tomographic knowledge of @xmath9 . following the approach introduced in sec . [ verif_criteria ] , in fig . [ two - state - fig ] we present an upper bound on the tolerable depolarising rate @xmath208 as a function of the photon loss probability @xmath210 for two different values of the parameter @xmath230 . it states that no secret key can be distilled from the correlations established by the users . in this example , the results obtained coincide when @xmath207 and @xmath209 . to obtain an upper bound on the tolerable qber one can use the following expression @xmath231\}},\ ] ] with @xmath232 . in particular , for given values of the parameters @xmath230 , @xmath233 , and @xmath204 , one only needs to substitute in eq . ( [ qber_2 ] ) the value of @xmath208 given in fig . [ two - state - fig ] as a function of the parameter @xmath210 . remarkably , the cut - off point for two - way qkd presented in fig . [ two - state - fig ] , _ i.e. _ , the value of the photon loss probability @xmath210 that makes @xmath234 and also @xmath235 , coincides with the limit imposed by the unambiguous state discrimination attack @xcite . in the two - state protocol this limit is given by @xmath236 . ( see also ref . [ two - state - fig ] also shows a difference between one - way classical post - processing with rr and with dr as a function of the parameter @xmath210 . the reason behind this effect is beyond the scope of this paper and needs further investigations . a fundamental question in quantum key distribution ( qkd ) is to determine whether the legitimate users of the system can use their available measurement results to generate a secret key via two - way or one - way classical post - processing of the observed data . in this paper we have investigated single - photon qkd protocols in the presence of loss introduced by the quantum channel . our results are based on a simple precondition for secure qkd for two - way and one - way classical communication . in particular , the legitimate users need to prove that there exists no separable state ( in the case of two - way qkd ) , or that there exists no quantum state having a symmetric extension ( one - way qkd ) , that is compatible with the available measurements results . we have shown that both criteria can be formulated as a convex optimisation problem known as a primal semidefinite program ( sdp ) . such instances of convex optimisation problems can be solved efficiently , for example by means of interior - point methods . moreover , these sdp techniques allow us to evaluate these criteria for any single - photon qkd protocol in a completely systematic way . a similar approach was already used in ref . @xcite for the case of one - way qkd without losses . here we complete these results , now presenting specifically the analysis for the case of a lossy channel . furthermore , we have shown that these qkd verification criteria based on sdp provide also a means to search for witness operators for a given two - way or one - way qkd protocol . any sdp has an associated dual problem that represents also a sdp . we have demonstrated that the solution to the this dual problem corresponds to the evaluation of an optimal witness operator that belongs to the minimal verification set of them for the given two - way ( or one - way ) qkd protocol . most importantly , a positive expectation value of this optimal witness operator guarantees that no secret key can be distilled from the available measurements results . finally , we have illustrated our results by analysing the performance of several well - known qubit - based qkd protocols for a given channel model . we gladly acknowledge stimulating discussions with k. tamaki , o. ghne , c .- h . f. fung , and x. ma . we specially thank h .- k . lo and n. ltkenhaus for their critical discussion of this article , and f. just for help on the numerics . financial support from nserc , cipi , crc program , cfi , oit , ciar , prea , dfg under the emmy noether programme , and the european commission ( integrated project secoqc ) are gratefully acknowledged . m.c . also thanks the financial support from a post - doctoral grant from the spanish ministry of science ( mec ) . in this appendix we present some duality properties of a general sdp @xcite that guarantee that the solution to the dual problems introduced in sec . [ sec2a ] can be associated with a witness operator . the primal problem given by eq . ( [ primalsdp_marcos ] ) is called feasible ( strictly feasible ) if there exists @xmath110 such as @xmath237 ( @xmath238 ) . similarly , the dual problem given by eq . ( [ dual_marcos ] ) is called _ feasible _ ( _ strictly feasible _ ) if there exists a matrix @xmath128 ( @xmath239 ) which fulfills all the desired constraints . the _ weak duality condition _ , illustrated in eq . ( [ weakduality ] ) , allows to derive simple upper and lower bounds for the solution of either the primal or dual problem . in particular , for every feasible solution @xmath240 of the primal problem and for every feasible solution @xmath115 of the dual problem , the following relation holds : @xmath241 the _ strong duality condition _ certifies whether the optimal solution to the primal and dual problem , that we shall denote as @xmath242 and @xmath243 , respectively , are equal . more precisely , @xmath244 if ( 1 ) the primal problem is strictly feasible , or ( 2 ) the dual problem is strictly feasible . moreover , if both conditions are satisfied simultaneously then it is guaranteed that there is a feasible pair @xmath245 achieving the optimal values @xmath244 . this last condition is known as the _ complementary slackness condition_. the sdp given by eq . ( [ primalsdp_marcos ] ) , when @xmath49 , can always be transformed as follows @xcite @xmath246 this sdp is always strictly feasible . to see this , note that if @xmath247 and @xmath248 , where @xmath249 denote the eigenvalues of the matrix @xmath45 , then @xmath250 . moreover , it can be shown that eq . ( [ primalsdp ] ) is equivalent to the original sdp . let @xmath251 be the solution to eq . ( [ primalsdp ] ) . if @xmath252 , the original problem is infeasible since @xmath253 . on the other hand , if @xmath254 there exists @xmath255 such that @xmath256 , stating that the original problem is feasible . that is , the solution @xmath251 of the sdp given by eq . ( [ primalsdp ] ) certifies whether the original problem is indeed feasible or not . the dual problem associated with eq . ( [ primalsdp ] ) is given by : @xmath257 if all the matrices @xmath46 are traceless , _ i.e. _ , @xmath258 , this dual problem is also always strictly feasible . a trivial strictly feasible solution to this problem is given by @xmath259 , where @xmath260 denotes the dimension of @xmath46 . if we apply the three duality conditions mentioned above to the sdps given by eq . ( [ primalsdp ] ) and eq . ( [ dualsdp ] ) we find that , if eq . ( [ primalsdp ] ) delivers an infeasible solution @xmath261 , @xmath262 this arises from the fact that the strong duality relation guarantees that @xmath263 , and the complementary slackness condition certifies that there is a @xmath264 that achieves the optimal value @xmath243 . when @xmath265 the weak duality condition assures that @xmath266 for every feasible solution @xmath115 of the dual problem . both results together show that the solution to the dual problem given by eq . ( [ dualsdp ] ) can be associated to a witness operator @xmath15 . in particular , the ability of @xmath15 to detect , at least , one state , _ i.e. _ , @xmath267 such as @xmath268 , can be related with eq . ( [ violation ] ) . on the other hand , the requirement that @xmath15 is positive on all states belonging to a given set of them can be related with eq . ( [ requirement ] ) . to achieve the desired equivalence , however , the dual problem must be strictly feasible , otherwise the complementary slackness condition does not hold and the existence of an appropriate witness is not guaranteed . it turns out that all the linear constraints included in the dual problems considered in sec . [ sec2a ] have traceless matrices @xmath46 , such that these dual problems are always strictly feasible . in this appendix we include very briefly the results of our investigations on other well - known qubit - based qkd protocols . like in sec . [ section_example ] , we shall consider that the observed data @xmath1 are generated via measurements onto the state given by eq . ( [ channel_eq ] ) . in this scheme , alice prepares a qubit in one of the following six quantum states : @xmath269 , and sends it to bob @xcite . on the receiving side , bob measures each incoming signal by projecting it onto one of the three possible bases . the loss of a photon in the channel is characterised by a projection onto the vacuum state @xmath218 . the qber is given by @xmath270.\ ] ] for @xmath207 we find , as expected , that whenever @xmath271 ( corresponding to a value of @xmath272 ) no secret key can be distilled by two - way classical post - processing @xcite . in the case of one - way classical post - processing ( both for rr and dr ) , and assuming @xmath207 and @xmath273 , we obtain that secure qkd might only be possible for a @xmath274 ( @xmath275 ) . a possible eavesdropping strategy to attain this cut - off point is , for instance , to use an universal cloning machine to clone every signal sent by alice such as the fidelities of eve s and bob s clones coincide @xcite . ( see also ref . @xcite . ) the four - state protocol @xcite is similar to the six - state protocol , but now alice sends one of four possible signal states instead of one of six . in particular , she chooses one state within the set @xmath276 and sends it bob . each received signal is projected by bob onto one of the two possible bases , together with a projection onto the vacuum state @xmath218 corresponding to the loss of a photon . the qber is now given by @xmath277 if @xmath207 we obtain the well - known result stating that whenever @xmath278 ( corresponding to a value of @xmath279 ) no secret key can be distilled by two - way classical post - processing @xcite . similarly , for the case of one - way classical post - processing , and assuming @xmath207 and @xmath273 , we find that the @xmath280 must be lower than @xmath281 ( @xmath282 ) . this last result coincides with the value of the qber produced by an eavesdropping strategy where eve and bob shannon information are equal @xcite . this scheme can be seen as a combination of two two - state qkd protocols @xcite . more precisely , alice selects , at random and independently each time , one of the following four signal states , @xmath283 , and sends it to bob . on the receiving side , bob measures each incoming signal by choosing , at random and independently for each signal , one of two possible povms . each povm corresponds to the one used in the two - state protocol ( see sec . [ sec_two - state ] ) for the signal states @xmath284 , with @xmath285 , and @xmath286 , with @xmath285 , respectively . the qber is given by @xmath289\sin^2\theta}{2\big[e+(1-e ) [ 2(\alpha^4+\beta^4)\sin^2\theta+4\alpha^2\beta^2\cos^2\theta]\big]}.\ ] ] in the case of two - way classical post - processing , the maximum tolerable value of @xmath208 shown in fig . [ fptwo - state - fig ] starts decreasing as the losses in the channel increase , and , at some point , it becomes constant independently of @xmath210 . interestingly , the value of @xmath210 where this inflexion occurs , corresponds to the point where eve can discriminate unambiguously between the two states in the set @xmath290 , with @xmath285 , or between those states in the set @xmath291 , with @xmath285 . this happens when @xmath236 . this qkd scheme requires alice sending to bob one of the following three quantum states @xmath292 @xcite . on the receiving side , bob projects each incoming signal onto one of the two possible bases used in the four - state protocol ( see sec . [ four_sec ] ) , together with a projection onto the vacuum state @xmath218 . the qber has now the following form @xmath293.\ ] ] for the quantum channel given by eq . ( [ channel_eq ] ) , and assuming @xmath207 or @xmath209 , the maximum value of @xmath208 tolerated by the three - state protocol coincides with the four - state protocol for the cases of two - way and one - way post - processing with dr . in the trine protocol @xcite , alice selects , at random and independently each time , a qubit in one of the following three states : @xmath294 , and @xmath295 , and sends it to bob . each received signal is measured by bob with a povm defined by the following operators : @xmath296 , @xmath297 , with @xmath298 , and where @xmath299 , and @xmath222 . for the quantum channel given by eq . ( [ channel_eq ] ) , and assuming @xmath207 or @xmath209 , it turns out that the maximum value of @xmath208 tolerated by this scheme , both for two - way and one - way post - processing , coincides with the four - state protocol ( see fig . [ four - state - fig ] ) . the qber , however , is given by @xmath300 in this scheme , alice sends to bob one of the following six states : @xmath301 , and @xmath302 @xcite . on the receiving side , bob measures each incoming signal with one of two possible measurements corresponding to the bases @xmath303 with @xmath304 , that he selects at random and independently for each signal , together with a projection onto the vacuum state @xmath218 . when @xmath207 , the maximum value of @xmath208 tolerated by this protocol , both for two - way and one - way post - processing , coincides with the four - state protocol ( see fig . [ four - state - fig ] ) . the case @xmath209 is illustrated in fig . [ acin1-state - fig ] . 99 n. gisin , g. ribordy , w. tittel , and h. zbinden , rev . phys . * 74 * , 145 ( 2002 ) . m. duek , n. ltkenhaus , m. hendrych , to appear in progress in optics * 49 * , edt . e. wolf ( elsevier ) . c. h. bennett , g. brassard , and n. d. mermin , phys . * 68 * , 557 ( 1992 ) . m. curty , o. ghne , m. lewenstein , and n. ltkenhaus , phys . a * 71 * , 022306 ( 2005 ) . d. gottesman , and h .- k . lo , ieee trans . inf . theory * 49 * , 457 ( 2003 ) . d. mayers , in _ advances in cryptology proceedings of crypto96 _ ( springer , berlin , 1996 ) , pp . 343 - 357 , available at quant - ph/9606003 . p. w. shor , and j. preskill , phys . . lett . * 85 * , 441 ( 2000 ) . lo , qic * 1 * , 81 ( 2001 ) . k. tamaki , m. koashi , and n. imoto , phys . lett . * 90 * , 167904 ( 2003 ) . f. grosshans , g. van assche , j. wenger , r. brouri , n. cerf , and p. grangier , nature * 421 * , 238 ( 2003 ) . m. heid , and n. ltkenhaus , phys . rev . a * 73 * , 052316 ( 2006 ) . m. curty , and n. ltkenhaus , phys . a * 69 * , 042321 ( 2004 ) . m. curty , m. lewenstein , and n. ltkenhaus , phys . rev . lett . * 92 * , 217903 ( 2004 ) . m. horodecki , p. horodecki , and r. horodecki , in _ quantum information : an introduction to basic theoretical concepts and experiments _ , ed.g . alber et al . ( springer , heidelberg , 2001 ) , pp . 151 ; k.eckert , o. ghne , f. hulpke , p. hyllus , j. korbicz , j.mompart , d. bru , m. lewenstein , and a. sanpera , in _ quantum information processing _ , ed . g. leuchs and t. beth , ( wiley - vch , verlag , 2003 ) , pp . 79 . see also second edition 2005 . m. horodecki , p. horodecki , and r. horodecki , phys . a * 223 * , 1 ( 1996 ) . b. m. terhal , phys . a * 271 * , 319 ( 2000 ) . m. lewenstein , b. kraus , j. i. cirac , and p. horodecki , phys . a * 62 * , 052310 ( 2000 ) . c. h. bennett and g. brassard , proc . conference on computers , systems and signal processing , bangalore , india ( ieee press , new york , 1984 ) , 175 . d. bru , phys . lett . * 81 * , 3018 ( 1998 ) . c. h. bennett , phys . lett . * 68 * , 3121 ( 1992 ) . j. eisert , p. hyllus , o. ghne , and m. curty , phys . a * 70 * , 062317 ( 2004 ) . a. peres , phys . lett . * 77 * , 1413 ( 1996 ) . l. vandenberghe , and s. boyd , siam review * 38 * , 49 ( 1996 ) . s. boyd , and l. vandenberghe , _ convex optimization _ ( cambridge university press , 2004 ) . t. moroder , m. curty , and n. ltkenhaus , phys . a * 74 * , 052301 ( 2006 ) . a. c. doherty , p. a. parrilo and f. m. spedalieri , phys . * 88 * , 187904 ( 2002 ) . a. c. doherty , p. a. parrilo and f. m. spedalieri , phys . a * 69 * , 022308 ( 2004 ) . a. c. doherty , p. a. parrilo and f. m. spedalieri , phys . a * 71 * , 032333 ( 2005 ) . b. m. terhal , a. c. doherty , and d. schwab , phys . lett . * 90 * , 157903 ( 2003 ) . p. horodecki , and m. nowakowski , quant - ph/0503070 . a. peres , found . * 12 * , 1441 ( 1990 ) . a. peres , _ quantum theory : concepts and methods _ , ( kluwer academic publishers , 1993 ) . using the schmidt decomposition we have that the state @xmath306 can be written as @xmath307 for some orthogonal states @xmath308 and where @xmath309 . the orthogonal basis @xmath6 can always be selected such as @xmath310 for @xmath311 , and @xmath312 for @xmath313 , with @xmath314 and @xmath315 denoting two auxiliary systems under alice s control , and where @xmath316 satisfy @xmath317 . on the other hand , the states @xmath318 can always be obtained from the orthogonal states @xmath319 by means of a unitary transformation @xmath320 , _ i.e. _ , @xmath321 . with this notation , we can write the initial state prepared by alice as @xmath322 with @xmath323 . the measurement process in the basis @xmath6 can then be described as : alice first prepares the signal state @xmath324 , and then she measures her subsystem @xmath314 with the povm operators @xmath325 with @xmath326 . given an operator @xmath327 and an orthonormal basis @xmath328 , with @xmath329 , the partial transposed of @xmath16 with respect to subsystem @xmath19 in that basis is defined as @xmath330 in the same way , one can also define the partial transposed of @xmath16 with respect to subsystem @xmath18 . note that given @xmath331 lmis constraints @xmath332 , we can always combine them to a single new lmi constraint as : @xmath333 the operator @xmath91 can always be expressed as @xmath334 because of the following reason : suppose , for instance , that a given set of measurement operators @xmath335 do not form a set of operator basis elements . then , one can find the minimal set of linear independent measurement operators and then apply the gram - schmidt orthogonalization to these elements in order to obtain a minimal set of linear independent operator basis elements . moreover , from the original data @xmath1 , it is straightforward to compute the observed probability distribution for these new operators . as a result , one finds a fixed part @xmath91 of the form @xmath336 . alternatively to this method , the equality constraints given by eq . ( [ cond_coeff ] ) can also be included directly in the lmi constraint of the sdp . each of these constraint can be represented by means of two inequality constraints as follows : @xmath337 , and @xmath338 . this approach , however , increases the number of objective variables to be considered . p. hyllus , and j. eisert , new journal of physics * 8 * , 51 ( 2006 ) . s. l. woronowicz , rev . * 10 * , 165 ( 1976 ) . t. moroder , diploma thesis , institut fr theoretische physik i , universitt erlangen - nrnberg ( germany ) , 2005 . k. c. toh , r. h. tutuncu , and m. j. todd , optimization methods and software * 11 * , 545 ( 1999 ) , available from http://www.math.nus.edu.sg/mattohkc/sdpt3.html . j. lfberg , in _ proceedings of the cacsd conference _ ( taipei , taiwan , 2004 ) , pp . 284 - 289 , available from http://control.ee.ethz.ch/ joloef / yalmip.php . t. yamamoto , j. shimamura , s. k. zdemir , m. koashi , and n. imoto , phys . * 95 * , 040503 ( 2005 ) . boileau , r. laflamme , m. laforest , and c. r. myers , phys . lett . * 93 * , 220501 ( 2004 ) . a. chefles , contemporary physics 41 , 401 ( 2000 ) . i. d. ivanovic , phys . a * 123 * , 257 ( 1987 ) . d. dieks , phys . a * 126 * , 303 ( 1988 ) . a. peres , phys . lett . a * 128 * , 19 ( 1988 ) . m. duek , m. jahma , and n. ltkenhaus , phys . a * 62 * , 022306 ( 2000 ) . k. tamaki , m. koashi , and n. imoto , phys . a * 67 * , 032310 ( 2003 ) . b. huttner , n. imoto , n. gisin , and t. mor , phys . a * 51 * , 1863 ( 1995 ) . s. n. molotkov , and s. s. nazin , j. of experimental and theoretical physics letters * 63 * , 924 ( 1996 ) . s. n. molotkov , j. of experimental and theoretical physics letters * 87 * , 288 ( 1998 ) . b .- s . shi , y .- k . jiang , and g .- c . guo , appl . b * 70 * , 415 ( 2000 ) . f. fung , and h .- k . lo , phys . a * 74 * , 042342 ( 2006 ) . j. m. renes , phys . a * 70 * , 052314 ( 2004 ) . a. acn , s. massar , and s. pironio , new j. phys . * 8 * , 126 ( 2006 ) . b. huttner , and a. ekert , j. mod . opt . * 41 * , 2455 ( 1994 ) . h. bechmann - pasquinucci , and n. gisin , phys . a * 59 * , 4238 ( 1999 ) . c. a. fuchs , n. gisin , r. b. griffiths , c .- s . niu , and a. peres , phys . a * 56 * , 1163 ( 1997 ) . j. i. cirac , and n. gisin , phys . a * 229 * , 1 ( 1997 ) . in its original proposal , the four - plus - two - state protocol was meant to operate with weak coherent pulses as signal states @xcite . here we refer to the single - photon version of this protocol . the main motivation behind this protocol is that it can be proven to be secure also against an hypothetical eve only limited by the no - signalling principe @xcite . in its original proposal , alice and bob share a noisy quantum channel that distributes pairs of qubits in the maximally entangled state @xmath339 . these pairs are then measured by alice and bob . here , however , we consider a prepare and measure version of this scheme with alice being the source of the signal states that are sent to bob .

nonparametric regression models with locally stationary covariates have received increasing interest in recent years . as a nice relief of `` curse of dimensionality '' induced by large dimension of covariates , additive regression model is commonly used . however , in locally stationary context , to catch the dynamic nature of regression function , we adopt a flexible varying - coefficient additive model where the regression function has the form for this model , we propose a three - step spline estimation method for each univariate nonparametric function , and show its consistency and rate of convergence . furthermore , based upon the three - step estimators , we develop a two - stage penalty procedure to identify pure additive terms and varying - coefficient terms in varying - coefficient additive model . as expected , we demonstrate that the proposed identification procedure is consistent , and the penalized estimators achieve the same rate of convergence as the polynomial spline estimators . simulation studies are presented to illustrate the finite sample performance of the proposed three - step spline estimation method and two - stage model selection procedure . locally stationary process , varying - coefficient additive regression model , b - spline , scad , penalized least squares

1612.00136

modelling nonparametric time series has received increasing interest among scholars for a few decades , see , for example , @xcite . in classical time series analysis , the stationarity of time series is a fundamental assumption . yet , it may be violated on some occasions in such the fields as finance , sound analysis and neuroscience , especially when the time span of observations tends to infinity . so , it is necessary to generalize the stationary process to the nonstationary process . priestley ( 1965 ) @xcite first introduced a stochastic process with evolutionary spectra , which locally displays an approximately stationary behavior . but in his framework , it is impossible to establish an asymptotic statistical inference . dahlhaus ( 1997 ) proposed a new generalization of stationarity , called locally stationary process , and investigated its statistical inference . more details can refer to @xcite . in essence , the locally stationary process is locally close to a stationary process over short periods of time , but its second order characteristic is gradually changing as time evolves . a formal description of locally stationary process can refer to assumption ( a1 ) in the appendix . in parametric context , the statistical inference of locally stationary process has been studied extensively by @xcite . in nonparametric context , vogt ( 2012 ) @xcite considered the time - varying nonlinear autoregressive ( tvnar ) models including its general form and estimated the time - varying multivariate regression function using the kernel - type method . however , it still suffers the `` curse of dimensionality '' problem when the dimension of covariates is high . in order to solve the aforementioned problem , a familiar way is to adopt the additive nonparametric regression model suggested by @xcite . it not only remedies the `` curse of dimensionality '' , but also has an independent interest in practical applications due to its flexibility and interpretability . there exists abound research findings about the additive regression model in the literature . in the case of iid observations , the additive nonparametric component functions can be estimated by kernel - based methods : the classic backfitting estimators of @xcite , the marginal integration estimators of @xcite , the smoothing backfitting estimators of @xcite , and two - stage estimators of @xcite . in the stationary time series context , there are kernel estimators via marginal integration of @xcite , spline estimators of @xcite , and the spline - backfitted kernel ( sbk ) estimators which borrow the strength of both kernel estimation and spline estimation , see @xcite . vogt @xcite considered the locally stationary additive model and proposed smooth backfitting method to estimate bivariate additive component functions . on the other hand , the varying - coefficient model is a natural extension of linear model which allows the coefficients to change over certain common covariates instead of being invariant . this model succeeds to relax the parameter limitation of linear model and may have practical as well as theoretical significance . for this model , there are three types of estimation methods : local polynomial smoothing method @xcite , polynomial spline estimation method @xcite and smoothing spline method @xcite . zhang and wang @xcite proposed a so - called varying - coefficient additive model to catch the evolutionary nature of time - varying regression function in the analysis of functional data . their model assumes the evolutionary regression function has the form @xmath2 which is more flexible in the sense that it covers both varying - coefficient model and additive model as special cases . specifically speaking , it reduces to an additive model when @xmath3 are all constants , and a varying - coefficient model if @xmath4 are all linear functions . extracting the special meaning of time in functional data analysis , one can generalize time to some other common covariates . in this paper , we model locally stationary time series . to concreteness , let @xmath5 be a length-@xmath6 realization of @xmath7 dimension locally stationary time series , and assume that the data is generated by varying - coefficient additive model as follows @xmath8 where @xmath9 s are i.i.d , @xmath10 is the varying - coefficient component function , @xmath11 is the additive component function and @xmath12 is a bivariate nonparametric function , which allows the heteroscedasticity case . without loss of generality , we require @xmath13^{p}},$ ] where the superscript ` @xmath14 ' means transposition of vector or matrix . in order to identify these multiplied component functions , we require that @xmath15 where @xmath16 is the @xmath1 norm of any function @xmath17 defined on @xmath18 $ ] such that @xmath19 for functional data , zhang and wang @xcite proposed a two - step spline estimation procedure . in the first step , sorting the data within each subject in ascending order of time and averaging the response for each subject using trapezoidal rule to fit an additive model , then , in the second step , fitting a varying - coefficient model by substituting the estimated additive function into varying - coefficient additive model . his estimation methodology works since there are dense observation for every subject and covariates is independent of observation time within subject . however , for some other practical problems , such as longitudinal data with finite observertion time , time series data , such an assumption fails . to circumvent this problem , under mild assumptions , we derive an initial estimation of additive component by employing a segmentation technique . then we can fit a varying - coefficient model and an additive model , respectively , to get spline estimators of varying - coefficient function and additive function . as expected , we show that the proposed estimators of @xmath10 and @xmath11 are consistent and present the corresponding @xmath1 rate of convergence . on the other hand , the product term in may simply reduce to a varying - coefficient term or an additive term in the case of @xmath11 being linear function or @xmath10 being constant . so , in the parsimony sense , identifying additive terms and varying - coefficient terms in are of interest . to this end , we propose a two - stage penalized least squares estimator based on scad penalty function , and , furthermore , show that our model identification strategy is consistent , i.e. , the additive term and the varying - coefficient term are correctly selected with probability approaching to 1 . meantime , @xmath1 rate of convergence of penalized spline estimator of each component function achieves the rate of the spline estimator of univariate nonparametric function . the rest of this paper is organized as follows . we propose a three - step spline estimation method in section 2 and a two - stage model identification procedure in section 3 . section 4 describes the smoothing parameter selection strategies . section 5 establishes the asymptotic properties of the proposed model estimation and identification methods . simulation studies are illustrated in section 6 . the main technical proofs are presented in the appendix . lemmas and other similar proofs are given in the supplementary . in this section , we propose a three - step spline estimation method for the proposed locally stationary varying - coefficient additive model . * step i : segment the rescaled time @xmath20 into several groups , and approximate each varying - coefficient function @xmath10 within the same group by a local constant . thus , model can be approximated by an additive model , and a scaled - version of the additive component functions @xmath11 can be obtained using spline - based method . * step ii : substitute the initial estimates of the scaled additive component functions into model to yield an approximated varying - coefficient model , and then obtain spline estimators of varying - coefficient component functions @xmath21 * step iii : plug - in spline estimators of varying - coefficient component functions @xmath10 into model to yield an approximated additive model , and then update the spline estimation of additive component functions @xmath11 . we first present with some notations before detailing our proposed estimation method . let @xmath22 be @xmath23 order b - spline basis with @xmath24 interior knots and @xmath25 is the number of b - spline functions estimating additive component function @xmath26 similarly , we denote @xmath27 as @xmath28 order b - spline basis with @xmath29 interior knots , and @xmath30 is the number of b - spline functions estimating varying - coefficient component function @xmath21 here , ` @xmath31 ' and ` @xmath32 ' in the subscript of b - spline functions and knots number mean that is for the additive component function and varying - coefficient function , respectively . denote @xmath33 and @xmath34 the nice properties of scaled b - spline basis are listed in the appendix . + * step i : initial estimators of scaled additive component functions * we segment the sample @xmath35 in ascending order of time into @xmath36 groups with @xmath37 observations in each group , where @xmath38 hinges on the sample size @xmath6 and @xmath39 . then approximate @xmath40 in the @xmath41th group , i.e. @xmath42 by a constant @xmath43 , where @xmath43 is some constant dependent on @xmath44 and @xmath41 such that @xmath45 for the sake of convenient presentation , we suppress the triangular array index in locally stationary time series , and represent time index in the @xmath41th group as @xmath46 for given @xmath47 then one can approximate model as @xmath48 where @xmath49 if @xmath50 are all known , one can easily construct the spline estimator of @xmath26 suppose that @xmath51 minimizes @xmath52 ^ 2,\ ] ] then @xmath53 where @xmath54 however , @xmath43 s are unknown . we instead rewrite as an additive model , @xmath55 where @xmath56 for each given @xmath57 let @xmath58 minimize @xmath59 ^ 2.\ ] ] by and , it is easy to see that @xmath60 which implies @xmath61 in a word , although the additive component function in can not be estimated directly , the scaled additive component function @xmath62 with @xmath63 is estimable using the proposed segmentation techniques . + * step ii : spline estimators of varying - coefficient component functions * define @xmath64 and @xmath65 , @xmath66 . by , substituting @xmath67 into yields @xmath68 where @xmath69.$ ] model can be viewed as a varying - coefficient model , and the spline estimators of varying - coefficient functions @xmath70 are easily obtained . suppose that @xmath71 minimizes @xmath72 ^ 2.\ ] ] then , @xmath73 by the definition of @xmath74 and identifiability conditions for @xmath75 we have the spline estimators of varying - coefficient functions @xmath76 s in model as @xmath77 * step iii : spline estimators of additive component functions * substituting into model yields @xmath78 where @xmath79\beta_{k}(x_{t , t}^{\left(k\right)}).$ ] model can be viewed as a varying - coefficient model . suppose that @xmath80 minimizes @xmath81 ^ 2,\ ] ] then , spline estimators of additive component functions in are given by @xmath82 * remark 1 : * the spline estimators @xmath83 and @xmath84 can be updated by iterating step ii and step iii . however , one step estimation is enough and there is no great improvement through iteration procedure . * remark 2 : * one may employ different b - spline basis functions in step i and step iii for estimating the additive component functions @xmath85 . yet , we do nt distinguish them in symbols for the sake of simplicity . the proposed varying - coefficient additive model is more general and flexible than either varying - coefficient model or additive model , and covers them as special cases . but , in practice , a parsimonious model is always one s preference when there exist several potential options . hence , it is of great interest to explore whether the varying - coefficient component function is truly varying and whether the additive component function degenerates to simply linear function . in this paper , we decompose varying - coefficient additive terms into additive terms @xmath11 and varying - coefficient terms @xmath10 , and , motivated by @xcite , propose a two - stage penalized least squares ( pls ) model identification procedure to identify the term that @xmath11 is constant ( @xmath86 ) or / and @xmath10 is linear ( @xmath87 ) . * stage i : plug - in the spline estimators of additive component functions @xmath11 obtained in the estimation stage into model , and penalize @xmath88 to identify linear additive terms . * stage ii : given the penalized spline estimators of additive component functions @xmath11 obtained in stage i of the model identification process , penalize @xmath89 to select constant varying - coefficient terms . we first introduce some notations . let @xmath90 and @xmath91 denote @xmath92 and @xmath93 * stage i : identifying linear additive terms * by substituting the additive component functions @xmath94 by their spline estimates @xmath95 obtained in the estimation stage , model becomes @xmath96 where @xmath97.$ ] let @xmath98 with @xmath99 , and assume @xmath100 is determined by @xmath101 ^ 2\\ & + t\sum_{k=1}^{p}p_{\lambda_{t}}\left(k_{c}^{-3/2}\parallel\alpha'_{k}\parallel_{l_{2}}\right ) , \end{aligned}\ ] ] where @xmath102 and @xmath103 is a penalty function with a tuning parameter @xmath104 then the penalized spline estimators of @xmath76 are given by @xmath105 here the superscript ` @xmath106 ' denotes the penalized spline estimation . + * stage ii : identifying constant varying - coefficient terms * by replacing the varying - coefficient function @xmath76 with their penalized spline estimates @xmath107 obtained in stage i , model becomes @xmath108 where @xmath109\beta_{k}\left(x_{t , t}^{\left(k\right)}\right).$ ] let @xmath110 with @xmath111 , and assume @xmath112 is given by @xmath113 ^ 2\\ & + t\sum_{k=1}^{p}p_{\mu_{t}}\left(k_{a}^{-3/2}\parallel\beta''_{k}\parallel_{l_{2}}\right ) , \end{aligned}\ ] ] where @xmath114 and @xmath115 is a penalty function with a tuning parameter @xmath116 therefore , the penalized spline estimators of @xmath117 are given by @xmath118,\ k=1,\cdots , p.\ ] ] in this section , we discuss various implementation issues for the proposed model estimation and identification procedures . we predetermine the degree of polynomial spline . usual options are 0 , 1 or 2 , that is to choose linear , quadratic or cubic spline functions . it is known that , when sufficient number of knots is used , the spline approximation method is quite stable . therefore , we suggest to use the same number of interior knots @xmath119 for all component functions and @xmath120th order b - spline basis functions in the @xmath121th step estimation to facilitate the computation . by experience , it is reasonable to choose @xmath122 in addition , in order to solve the least squares problem in each group in step i estimation , we require @xmath123 under this constraint , we choose the optimal @xmath119 and @xmath37 by bic @xmath124 where @xmath125 ^ 2 $ ] and @xmath126 is the number of b - spline basis functions used in step iii estimation . various penalty functions @xcite can be used in practice . we choose the scad penalty function proposed by @xcite , which is defined by its first derivative @xmath127 for some @xmath128 and @xmath129 where symbol @xmath130 it is well - known that the scad penalty function has nice properties such as unbiasedness , sparsity and continuous . meantime , it is singular at the origin , and have no continuous second order derivatives . yet it can be locally approximated by a quadratic function . specifically speaking , given an initial estimate @xmath131 or equivalently @xmath132 if @xmath133 then one can locally approximate @xmath134 by @xmath135 this implies that the objective function in , denoted by @xmath136 can be approximated , up to a constant , by @xmath137 where @xmath138 @xmath139 with @xmath140 and @xmath141 and @xmath142 therefore , we can find the solution of by iteratively computing the following ridge regression estimator @xmath143 until convergence . in the same vein , we can iteratively solve the optimization problem . let @xmath144 @xmath145 with @xmath146 and @xmath147 and @xmath148 therefore , we can iteratively compute the following ridge regression estimator @xmath149 until convergence . based on the optimal segment length @xmath150 and the optimal number of interior knots @xmath151 , we then select tuning parameters @xmath152 and @xmath153 for the proposed two - stage model identification procedure . following @xcite , we take @xmath154 and find optimal tuning parameters @xmath152 and @xmath153 by bic in two steps . first , to select optimal @xmath152 , we define @xmath155 where @xmath156^{2}$ ] , @xmath157 is the number of b - spline basis functions adopted in the second step estimation and @xmath158 is the number of linear additive terms , i.e. , @xmath159 is sufficiently small , say , no larger than @xmath160 second , to select optimal @xmath153 , we define @xmath161 where @xmath162^{2}$ ] , and @xmath163 is the number of constant varying - coefficient terms , i.e. , @xmath164 is sufficiently small . thus , we select the optimal tuning parameters @xmath165 + in this section , we demonstrate that the proposed three - step spline estimation method is consistent under regularity conditions and show that the proposed two stage model identification procedure can correctly select additive terms and varying - coefficient terms with probability approaching one . furthermore , we conclude that @xmath1 rate of convergence of each component function achieves the optimal rate of the spline estimator of univariate nonparametric function stated in @xcite . the regularity conditions and assumptions are given in the appendix . + let @xmath166 and @xmath167 . we introduce @xmath168}|\beta_{k}\left(x\right)-\mu_{k}\left(x\right)|,\ \ \rho_{c}=\sum_{k=1}^{p}\inf_{\nu_{k}\in h_{k}}\sup_{x\in[0,1]}|\alpha_{k}\left(x\right)-\nu_{k}\left(x\right)|\ ] ] to measure the degree of spline approximation of varying - coefficient component function and additive component function . proposition 1 establishes @xmath1 rate of convergence of initial estimators of scaled additive component functions @xmath169 [ prop1 ] under assumptions ( a1 ) , ( a2 ) , ( a4 ) , ( a5 ) , ( a7 ) , ( a8 ) and ( a9 ) , if @xmath170 as @xmath171 @xmath172 @xmath173^{2 } = o_p\left(\frac{\rho_{a}^{2}}{n_{t}}+\frac{k_{a}}{t}\right),\ ] ] * remark 3 : * in comparison with the convergence of the spline estimation of univariate nonparametric function , we notice that the bias term in @xmath1 rate of convergence of initial estimators is smaller when the number of groups @xmath36 is larger than 1 . the larger number of segmentation groups , the smaller the bias , given the number of observations in each group is at least larger than the number of parameters in spline approximation of @xmath174 based on the result of proposition 1 , one can construct @xmath1 rate of convergence of the spline estimation of varying - coefficient component function @xmath10 as follows . [ alpha ] under assumptions ( a1 ) ( a10 ) , if @xmath175 as @xmath171 @xmath176 @xmath177^{2 } = o_p\left(\frac{\rho_{a}^{2}}{n_{t}}+\rho_{c}^{2}+\frac{k_{a}\vee k_{c}}{t}\right),\ ] ] where ` @xmath178 ' denotes the maximum of @xmath179 and @xmath180 * remark 4 : * theorem [ alpha ] shows that there exists an additional bias term @xmath181 in comparison with the convergence of the spline estimation of univariate nonparametric function . this term happens to be the rate of convergence obtained in proposition 1 and reflects the error of the initial estimator of scaled additive function @xmath174 next theorem presents @xmath1 rate of convergence of the spline estimation of additive component function @xmath26 [ beta ] under the assumptions of theorem 1 , @xmath182 * remark 5 : * similarly , in comparison with the rate of convergence of spline estimation for univariate nonparametric function , theorem [ beta ] also has an additional bias term @xmath183 . the reason this term exists is because the estimation of additive component function @xmath11 in step iii is based on the spline estimates of varying - coefficient function @xmath10 obtained in step ii . as expected , the convergence of the updated spline estimation of @xmath11 doe not depend on the number of segmentation groups in step i of the initial estimation of rescaled additive function @xmath174 we , here , respectively , demonstrate the consistency of selecting additive terms and varying - coefficient terms , and present the @xmath1 rate of convergence of penalized spline estimators of @xmath10 and @xmath26 [ cadditive ] suppose that @xmath184 given @xmath185 @xmath186 and @xmath187 with @xmath188 , then , under assumptions ( a1 ) ( a10 ) , as @xmath171 * with probability approaching to 1 , @xmath189 is constant a function a.s . for @xmath190 * @xmath1 rate of convergence for penalized spline estimator of @xmath10 is given by @xmath191 for @xmath192 where @xmath193 and @xmath194 [ cvarying ] suppose that @xmath195 given @xmath196 @xmath197 and @xmath198 with @xmath199 then , under assumptions ( a1 ) ( a10 ) , as @xmath171 * with probability approaching to 1 , @xmath200 is a linear function a.s . for @xmath201 * @xmath1 rate of convergence for penalized spline estimator of @xmath11 is given by @xmath202 for @xmath203 where @xmath204 and @xmath205 * remark 6 : * theorems [ cadditive ] and [ cvarying ] show that the penalized spline estimators of varying - coefficient component function @xmath10 and additive component function @xmath11 both have the same @xmath1 rate of convergence as that of the spline estimator of univariate nonparametric function . we consider two simulation examples to illustrate the finite sample performance of the proposed three - step spline estimation method and two - stage model selection procedure , respectively . the data are generated from the varying - coefficient additive model as follows @xmath206 where @xmath207 @xmath208 and @xmath209 are iid standard normal variables and @xmath210 to appraise the performance of the proposed three - step spline estimators , we use the mean integrated squared error ( mise ) based on @xmath211 monte carlo replications , that is @xmath212^{2}\mathrm{d}u,\ \ k=0,1,2,\ ] ] @xmath213^{2}\mathrm{d}x_{k},\ \ k=1,2,\ ] ] where @xmath214 and @xmath215 are estimators of @xmath76 and @xmath216 respectively , in the @xmath217-th monto carlo sample . the univariate nonparametric functions are approximated by b - spline of order of three or the quadratic splines . we consider the sample size @xmath218 the number of interior knots @xmath219 and the segmentation length @xmath220 in step i estimation . we run the simulation for 500 times , and find out the mise of three - step estimators decreases as the sample size increases , regardless the values of @xmath119 and @xmath221 . table [ tag:1 ] only gives the results of @xmath222 for different combinations of @xmath221 and @xmath6 . in addition , we list the mise of oracle estimators , which refer to the spline estimator of @xmath10 given all additive component functions are known in advance , or correspondingly , the spline estimator of @xmath11 given all varying - coefficient component functions are known in advance . as expected , mise of oracle estimators for varying - coefficient components and additive components are better than those of the proposed three - step spline estimators . the last two columns in table [ tag:1 ] depict the mise of spline estimators for nonparametric component functions in misspecified varying - coefficient model and misspecified additive model . we note that they are discernibly larger than three - step spline estimators in varying - coefficient additive model . .comparison of mise of different estimators in example 1 ( @xmath222 ) [ cols="^,^,^,^,^,^,^",options="header " , ] let @xmath223 be the euclidean norm of a real valued vector @xmath224 denote the space of @xmath225-order smooth functions defined on @xmath18 $ ] as @xmath226=\{m|m^{\left(l\right)}\in c[0,1]\}$ ] and the class of lipschitz continuous functions for some fixed constant @xmath227 as @xmath228,c\right)=\{m||m\left(x\right)-m\left(x'\right)|\leq c|x - x'|,\forall x , x'\in[0,1]\}.$ ] + the necessary conditions to prove asymptotic properties are listed as below . * the process @xmath229 is stationary locally in time , that is , for each rescaled time point @xmath230,$ ] there exists a strictly stationary process @xmath231 such that @xmath232 a.s . with @xmath233\leq c$ ] for some @xmath234 and @xmath235 independent of @xmath236 and @xmath237 * at each rescaled time point @xmath230,$ ] the joint density function @xmath238 of the stationary approximation process @xmath239 is bounded below and above uniformly on @xmath230:$ ] @xmath240^{p}}f_{u}\left(\mathbf{x}\right)\leq\sup_{\mathbf{x}\in[0,1]^{p}}f_{u}\left(\mathbf{x}\right)\leq c_{f } \ \text{uniformly on}\ u\in[0,1].\ ] ] meantime , @xmath241 has density function with respective to certain measure . * @xmath242 iid , and @xmath243 @xmath244 given @xmath245 @xmath246 * there exists positive constants @xmath247 and @xmath248 such that @xmath249 for all @xmath250 where @xmath251 is the @xmath252-mixing coefficients for process @xmath253 and defined as latexmath:[\[\alpha\left(k\right)=\sup_{b\in\sigma\{\mathcal{z}_s , s\leq t\ } , c\in \sigma\{\mathcal{z}_s , s\ge t+k\ } } * the conditional standard deviation function @xmath12 is bounded below and above uniformly on @xmath230,$ ] i.e. , @xmath255^{p}}\sigma\left(u,\mathbf{x}\right ) \leq \sup_{\mathbf{x}\in[0,1]^{p}}\sigma\left(u,\mathbf{x}\right)\leq c_{\sigma}\ \text{uniformly on}\ u\in[0,1]\ ] ] for some positive constants @xmath256 and @xmath257 * @xmath258 $ ] and @xmath259,c_{1,k})$ ] where @xmath260 is an integer such that @xmath261 * @xmath262 $ ] and @xmath263,c_{2,k})$ ] where @xmath264 is an integer such that @xmath265 * the knots for @xmath266 @xmath267 and the knots for @xmath268 @xmath269 has bounded mesh ratio : @xmath270 @xmath271 * @xmath272 , @xmath273 @xmath274 and @xmath275 * @xmath276 and @xmath277 * remark 7 : * assumption ( a1 ) specifies a data generating process ( gdp ) . assumptions ( a2 ) - ( a5 ) are standard in time - series context , see @xcite . assumptions ( a6 ) - ( a8 ) are common in typical spline approximation literature , for instance , @xcite . firstly , we need some lemmas before proving main theorems . let @xmath278 and @xmath279 be any two - vector valued function . define empirical inner product @xmath280 and theoretical inner product @xmath281\mathrm{d}u,\ ] ] denote the induced norm by @xmath282 and @xmath283 as @xmath284 and @xmath285 respectively . in addition to , @xmath286 given sequences of positive numbers @xmath287 and @xmath288 @xmath289 means @xmath290 is bounded and @xmath291 means @xmath289 and @xmath292 hold . [ inducenorm ] let @xmath293 and @xmath294 for @xmath295 denote @xmath296 and @xmath297 then @xmath298 holds under assumption ( a2 ) . in combination with assumption ( a2 ) and the property ( iv ) of @xmath299 we have @xmath300\mathrm{d}u = \sum_{k=1}^{p}\int_{0}^{1}e\big[\sum_{l=1}^{j_{k , a}}\gamma_{kl}\psi_{kl}(x_{t}^{\left(k\right)}\left(u\right))\big]^{2}\mathrm{d}u\\ \asymp&\sum_{k=1}^{p}\int_{0}^{1}\parallel g_{k}\parallel^{2}_{l_{2}}\mathrm{d}u=\parallel\mathbf{g}\parallel_{l_{2}}^{2 } \asymp\sum_{k=1}^{p}|\gamma_{k}|^{2}=|\mathbf{\gamma}|^{2 } \end{aligned}\ ] ] [ inner ] let @xmath301 be the collection of vector valued functions @xmath278 with @xmath302 such that @xmath303 for @xmath304 where @xmath305 is defined in section 5.1 . then under assumption ( a1 ) , ( a2 ) , ( a4 ) and ( a9 ) , as @xmath171 @xmath306 for any @xmath307 there exists coefficients @xmath308 and @xmath309 @xmath310 such that @xmath311 for @xmath295 it is not difficult to see that @xmath312 for any given @xmath304 let @xmath313 if the intersection of the supports of @xmath314 and @xmath315 contains an open interval . that is , @xmath316 if @xmath317 moreover , it is known @xmath318 for some constant @xmath32 and all @xmath319 moveover , for any given @xmath320 and @xmath321 we have @xmath322\mathrm{d}u|\\ \leq & t^{-1}\sum_{t=1}^{t}|\psi_{kl}(x_{t , t}^{\left(k\right)})\psi_{kl'}(x_{t , t}^{\left(k\right ) } ) -\psi_{kl}(x_{t}^{\left(k\right)}(t / t))\psi_{kl'}(x_{t}^{\left(k\right)}(t / t))|\\ + & t^{-1}\sum_{t=1}^{t}|\psi_{kl}(x_{t}^{\left(k\right)}(t / t))\psi_{kl'}(x_{t}^{\left(k\right)}(t / t ) ) -e[\psi_{kl}(x_{t}^{\left(k\right)}(t / t))\psi_{kl'}(x_{t}^{\left(k\right)}(t / t))]|\\ + & |t^{-1}\sum_{t=1}^{t}e[\psi_{kl}(x_{t}^{\left(k\right)}\left(t / t\right))\psi_{kl'}(x_{t}^{\left(k\right)}\left(t / t\right ) ) ] -\int_{0}^{1}e[\psi_{kl}(x_{t}^{\left(k\right)}\left(u\right))\psi_{kl'}(x_{t}^{\left(k\right)}\left(u\right))]\mathrm{d}u| . \end{aligned}\ ] ] by assumption ( a1 ) and the boundness of b - spline , the first term above is bounded by @xmath323 employing berstein s inequality , the second term is bounded by @xmath324 the last term is bounded by @xmath325 from the integral theory . therefore , using cauchy - schwartz inequality and assumption ( a9 ) , we obtain that @xmath326 where @xmath327 and @xmath328 denote the vectors with entries @xmath308 and @xmath309 respectively . by lemma [ inducenorm ] , we see @xmath329 which completes the proof . [ eigenvalue ] under assumption ( a1 ) , ( a2 ) ( a4 ) and ( a9 ) , as @xmath171 * for each @xmath330 @xmath331 has eigenvalues bounded away from 0 and @xmath332 with probability tending to one ; * @xmath333 has eigenvalues bounded away from 0 and @xmath334 with probability tending to one . we only show ( ii ) , the proof of ( i ) is similar . for any given vector @xmath296 with @xmath335 @xmath336 let @xmath293 and @xmath337 where @xmath338 denote @xmath339 then by lemma [ inducenorm ] and [ inner ] , @xmath340 which implies @xmath341 has eigenvalues bounded away from 0 and @xmath342 therefore , @xmath343 also has eigenvalues bounded away from 0 and @xmath342 * proof for proposition 1 . * + let @xmath344 then @xmath345 define @xmath346 @xmath347 and @xmath348 denote @xmath349 and @xmath350 @xmath351 note that @xmath352 by cauchy - schwartz inequality , we have @xmath353 it suffices to deal with the approximation error terms @xmath354 and stochastic error terms @xmath355 * approximate error terms : * we will show the rate of approximation error @xmath356 we have @xmath357\parallel^2_{l_{2}}\\ & \preceq \frac{1}{n_{t}^2}\sum_{s=1}^{n_{t } } \parallel\tilde{\beta}_{k}^{\left(s\right)}\left(x_{k}\right)-\beta_{k}^{\left(s\right)}\left(x_{k}\right)\parallel^2_{l_{2}}. \end{aligned}\ ] ] by the definition of @xmath358 there exists @xmath359 and @xmath360 such that @xmath361}\big|\breve{\beta}_{k}^{\left(s\right)}\left(x_{k}\right)-\beta_{k}^{\left(s\right)}\left(x_{k}\right)\big|=o\left(\rho_{a}\right).$ ] therefore , @xmath362 let @xmath363 then @xmath364 on the one hand , @xmath365 ^ 2 \asymp|\mathbf{\tilde{\alpha}}^{\left(s\right)}-\mathbf{\breve{\alpha}}^{\left(s\right)}|^2 $ ] by lemma [ eigenvalue ] . on the other hand , @xmath366 ensures that @xmath367^{2 } \leq \frac{1}{i_{t}}\sum_{j=1}^{i_{t}}(\omega_{s , j}-v_{sj}^{\tau}\mathbf{\breve{\alpha}}^{\left(s\right)})^2\\ = & \frac{1}{i_{t}}\sum_{j=1}^{i_{t}}\big\{\sum_{k=1}^{p}\big[\beta_{k}^{\left(s\right)}(x_{t_{sj},t}^{\left(k\right ) } ) -\breve{\beta}_{k}^{\left(s\right)}(x_{t_{sj},t}^{\left(k\right)})\big]\big\}^{2}=o\left(\rho_{a}^{2}\right ) . \end{aligned}\ ] ] thus , @xmath368 which means @xmath369 * stochastic error terms : * we next show the rate of stochastic errors : @xmath370 note that @xmath371 under assumption ( a5 ) , we obtain that @xmath372 \varepsilon_{t_{sj}}^{2}\big\}\\ & + c_{\sigma}e\big\{\sum_{\substack{j=1\\j\neq j'}}^{i_{t}}\big[1+\sum_{k=1}^{p}\sum_{l=1}^{j_{k , a}}\psi_{kl}(x_{t_{sj},t}^{\left(k\right ) } ) \psi_{kl}(x_{t_{sj'},t}^{\left(k\right)})\big]\varepsilon_{t_{sj}}\varepsilon_{t_{sj'}}\big\}. \end{aligned}\ ] ] assumption ( a3 ) makes the second term be zero , and the first term is bounded by @xmath373^{2}.$ ] however , @xmath374^{2 } \preceq & \sum_{k=1}^{p}\sum_{l=1}^{j_{k , a}}e\big[\psi_{kl}(x_{t_{sj},t}^{\left(k\right ) } ) -\psi_{kl}(x_{t_{sj}}^{\left(k\right)}\left(t_{sj}/t\right))\big]^{2}\\ & + \sum_{k=1}^{p}\sum_{l=1}^{j_{k , a}}e[\psi_{kl}(x_{t_{sj}}^{\left(k\right)}\left(t_{sj}/t\right))]^{2}. \end{aligned}\ ] ] by assumption ( a2 ) and the properties of b - spline , @xmath375 ^ 2 \asymp\sum_{k=1}^{p}j_{k , a}=o\left(k_{a}\right ) . \ ] ] on the other hand , @xmath376^{2 } = j_{k , a}e[b_{kl , a}(x_{t , t}^{\left(k\right)})-b_{kl , a}(x_{t}^{\left(k\right)}\left(t / t\right))]^{2}\\ \leq & cj_{k , a}e[|\mathbf{x}_{t , t}-\mathbf{x}_{t}\left(t / t\right)|^2 ] \leq cj_{k , a}\frac{1}{t^{2}}e[u_{t , t}^{2}\left(t / t\right ) ] . \end{aligned}\ ] ] therefore , @xmath377^{2 } = o\left(k_{a}i_{t}\right)+\frac{i_{t}}{t^{2}}\sum_{k=1}^{p}j_{k , a}^{2}. \end{aligned}\ ] ] and in turn @xmath378 which completes the proof of and hence the first half of theorem 1 . the rest is direct from lemma [ eigenvalue ] . + * proof for theorem [ alpha ] . * let @xmath379 @xmath380 where @xmath381 denote @xmath382 @xmath383 denote @xmath384 where @xmath385 @xmath386 and @xmath387 then @xmath388 furthermore , assuming that @xmath389 with @xmath390 is given by @xmath391 and @xmath392 for @xmath393 by cauchy - schwartz inequality and identifiable condition @xmath394 we get @xmath395 * approximate error term : * we show the rate of approximate error term as follows @xmath396 by the definition of @xmath397 there exists @xmath398 such that @xmath399 satisfying @xmath400}|\breve{\alpha}_{k}\left(u\right)-\alpha_{k}\left(u\right)|=o\left(\rho_{c}\right)\ ] ] for @xmath393 thus @xmath401 note that @xmath402 and the normal equation @xmath403 yields @xmath404 according to proposition 1 and boundness of @xmath405 @xmath406^{2 } = o_{p}\left(\frac{\rho_{a}^{2}}{n_{t}}+\frac{k_{a}}{t}\right ) . \end{aligned}\ ] ] on the other hand , from assumption ( a7 ) @xmath407^{2 } + \frac{1}{t}\sum_{t=1}^{t}\sum_{k=1}^{p}\hat{\gamma}_{k}^{2}(x_{t , t}^{\left(k\right ) } ) [ \delta_{k}\left(t / t\right)-\breve{\delta}_{k}\left(t / t\right)]^{2}\\ \preceq&\rho_{c}^{2}+\rho_{c}^{2}\frac{1}{t}\sum_{k=1}^{p}\sum_{t=1}^{t}\hat{\gamma}_{k}^{2}(x_{t , t}^{\left(k\right)})\\ = & o\left(\rho_{c}^{2}\right ) . \end{aligned}\ ] ] * stochastic error terms : * we next show the following rate of stochastic error term @xmath408 it is easy to see @xmath409 based on assumption ( a5 ) , it is sufficient to bound @xmath410 let @xmath411 and @xmath412 then @xmath413 obviously , under assumption ( a3 ) , @xmath414 . $ ] from assumption ( a1 ) and assumption ( a7 ) , @xmath415^{2 } \preceq & e[\gamma_{k}(x_{t , t}^{\left(k\right)})-\gamma_{k}(x_{t}^{\left(k\right)}(t / t))]^{2 } + e[\gamma_{k}(x_{t}^{\left(k\right)}(t / t))]^{2}\\ \preceq & e|\mathbf{x}_{t , t}-\mathbf{x}_{t}\left(t / t\right)|^{2}+o\left(1\right)\\ = & o\left(t^{-2}\right)+o\left(1\right)=o\left(1\right ) , \end{aligned}\ ] ] where @xmath416 is the @xmath44-th component of stationary approximation process @xmath231 of locally stationary process @xmath229 at rescaled time @xmath417 in combination with @xmath418 we have @xmath419\preceq & \sum_{l=1}^{j_{0,c}}\varphi_{0l}^{2}\left(t / t\right)+\sum_{k=1}^{p}\sum_{l=1}^{j_{k , c}}\varphi_{kl}^{2}\left(t / t\right ) = o\left(k_{c}\right ) , \end{aligned}\ ] ] which means @xmath420 meantime , @xmath421^{2}. \end{aligned}\ ] ] by cauchy - schwartz inequality , @xmath422^{2 } \leq 3 e[\hat{\gamma}_{k}(x_{t , t}^{\left(k\right)})-\hat{\gamma}_{k}(x_{t}^{\left(k\right)}\left(t / t\right))]^{2}\\ & + 3e[\hat{\gamma}_{k}(x_{t}^{\left(k\right)}\left(t / t\right))-\gamma_{k}(x_{t}^{\left(k\right)}\left(t / t\right))]^{2 } + 3e[\gamma_{k}(x_{t}^{\left(k\right)}\left(t / t\right))-\gamma_{k}\big(x_{t , t}^{\left(k\right)})]^{2}. \end{aligned}\ ] ] the third term is bounded by @xmath423 from assumption ( a1 ) and ( a7 ) . assumption ( a2 ) and proposition 1 ensure that the second term is bounded by @xmath424 for the first term , we note that @xmath425^{2}\\ = & e\big[n_{t}^{-1}\sum_{s=1}^{n_{t}}\sum_{l=1}^{j_{k , a}}\hat{h}^{\left(s\right)}_{kl}j_{k , a}^{1/2 } \{b_{kl , a}(x_{t , t}^{\left(k\right)})-b_{kl , a}(x_{t}^{\left(k\right)}\left(t / t\right))\}\big]^{2}\\ \preceq & j_{k , a}n_{t}^{-2}\sum_{s=1}^{n_{t}}\sum_{l=1}^{j_{k , a}}(\hat{h}^{\left(s\right)}_{kl})^{2 } = & o_p\left(\frac{k_{a}^{2}}{t^{2}n_{t}}\right ) , \end{aligned}\ ] ] and thus @xmath426^{2 } o_{p}\left(\frac{\rho_{a}^{2}}{n_{t}}+\frac{k_{a}}{t}\right ) = o_p\left(\frac{k_{c}}{t}\big\{\frac{\rho_{a}^{2}}{n_{t}}+\frac{k_{a}}{t}\big\}\right ) , \end{aligned}\ ] ] which shows . + * proof for theorem [ beta ] . * let @xmath427 @xmath428 and @xmath429 @xmath430 define @xmath431 where @xmath432 @xmath433 and @xmath434 for @xmath435 suppose @xmath436 is given by @xmath437 and @xmath438 analogously , represent @xmath439 with @xmath440 obviously , @xmath441 * approximation error term : * the rate of approximation error term is given by @xmath442 on the one hand , by the definition of @xmath358 there exists @xmath443 and @xmath444 such that @xmath445}|\beta^*\left(x_{k}\right)-\beta_{k}\left(x_{k}\right)|=o\left(\rho_{a}\right),\ ] ] which means @xmath446 on the other hand , @xmath447 furthermore , @xmath448 since @xmath449 according to theorem 1 , @xmath450^{2}+ \frac{1}{t}\sum_{k=1}^{p}[\alpha_{k}\left(t / t\right)-\hat{\alpha}_{k}\left(t / t\right)]^{2}\beta_{k}^{2}(x_{t , t}^{\left(k\right)})\\ = & o_{p}\left(\frac{\rho_{a}^{2}}{n_{t}}+\rho_{c}^{2}+\frac{k_{a}\vee k_{c } } { t}\right ) . \end{aligned}\ ] ] finally , we note that @xmath451\big|^{2}\\ \preceq&\frac{1}{t}\sum_{t=1}^{t}\sum_{k=1}^{p } \hat{\alpha}_{k}^{2}\left(t / t\right)[\beta_{k}(x_{t , t}^{\left(k\right ) } ) -\beta^*_{k}(x_{t , t}^{\left(k\right)})]^{2 } = o\left(\rho_{a}^{2}\right ) \end{aligned}\ ] ] since for each @xmath304 @xmath452^{2 } + \frac{1}{t}\sum_{t=1}^{t}\alpha_{k}^{2}\left(t / t\right)\\ = & \int_{0}^{1}\alpha_{k}^{2}\left(u\right)\mathrm{d}u+o\left(1\right)=o\left(1\right ) . \end{aligned}\ ] ] therefore , holds . + * stochastic error term : * we will show the rate of stochastic error term : @xmath453 firstly , @xmath454 however , @xmath455 which implies @xmath456 because of assumption ( a5 ) . similar the counterpart in the proof of theorem 1 , we have @xmath457\\ = & \sum_{k=1}^p[\hat{\alpha}_{k}\left(t / t\right)-\alpha_{k}\left(t / t\right)]^{2}\sum_{l=1}^{j_{k , a}}e[\psi_{kl}(x_{t , t}^{\left(k\right)})]^2 . \end{aligned}\ ] ] furthermore , @xmath458 ^ 2\preceq \sum_{l=1}^{j_{k , a}}e[\psi_{kl}(x_{t , t}^{\left(k\right ) } ) -\psi_{kl}(x_{t}^{\left(k\right)}\left(t / t\right))]^{2 } + \sum_{l=1}^{j_{k , a}}e[\psi_{kl}(x_{t}^{\left(k\right)}\left(t / t\right))]^2 . \end{aligned}\ ] ] by assumption ( a1 ) , @xmath459^{2 } = & j_{k , a}e[b_{kl , a}(x_{t , t}^{\left(k\right)})-b_{kl , a}(x_{t}^{\left(k\right)}\left(t / t\right))]^{2}\\ \preceq & j_{k , a}e|x_{t , t}^{\left(k\right)}-x_{t}^{\left(k\right)}\left(t / t\right)|^{2}=o\left(j_{k , a}/t^{2}\right ) \end{aligned}\ ] ] assumption ( a2 ) leads to @xmath460 ^ 2 \asymp \sum_{l=1}^{j_{k , a}}\int\psi_{kl}^2\left(z\right)\mathrm{d}z \leq \int\big[\sum_{l=1}^{j_{k , a}}\psi_{kl}\left(z\right)\big]^{2}\mathrm{d}z = j_{k , a}. \end{aligned}\ ] ] therefore , @xmath461^{2 } = o\left(j_{k , a}+j_{k , a}^{2}/t^{2}\right)=o\left(k_{a}+k_{a}^{2}/t^{2}\right),$ ] which yields @xmath462 similarly , @xmath463\\ = & \frac{1}{t^{2}}\sum_{t=1}^{t}\delta\left(t / t\right)^{\tau } e[\mathbf{\psi}\left(\mathbf{x}_{t , t}\right)\mathbf{\psi}\left(\mathbf{x}_{t , t}\right)^{\tau}]\delta\left(t / t\right ) . \end{aligned}\ ] ] note that @xmath464 = & e\big[\sum_{l=1}^{j_{k , a}}\psi_{kl}^{2}(x_{t , t}^{\left(k\right)})\big]\leq\int \big\{\sum_{l=1}^{j_{k , a}}\psi_{kl}\left(z\right)\big\}^{2}f_{x_{t , t}}^{\left(k\right ) } \left(z\right)\mathrm{d}z\\=&j_{k , a } , \end{aligned}\ ] ] where @xmath465 is the marginal density of @xmath44-th component of @xmath466 so , @xmath467=diag\left(j_{k , a}\right)_{k=1}^{p}=k_{a}i_{p}\ ] ] and @xmath468 which completes the proof of . + * proof for theorem [ cadditive ] : * * without loss of generality , we assume the true model is @xmath469 let @xmath470 and @xmath471 as the collection of all functions having form @xmath472 it is sufficient to show @xmath473 for any @xmath474 such that @xmath475 and for any @xmath476 where @xmath477 such that @xmath478 + for the sake of convenient presentation , we also denote @xmath479 as @xmath480 if @xmath481 let @xmath482 and @xmath483 then @xmath484 furthermore , @xmath485 where @xmath486 is the inner product of vector @xmath179 and @xmath180 + let @xmath487 we have @xmath488 and @xmath489 where the last step holds since assumption ( a7 ) and theorem [ beta ] . + therefore , @xmath490 in which @xmath491 lies between 0 and @xmath492 and @xmath493 the proof of part(i ) is completed since @xmath494 @xmath495 and @xmath496 * according to theorem 6 ( p149 ) of @xcite , under assumption ( a6 ) , there exists @xmath497 and @xmath498 such that @xmath499 let @xmath500 and @xmath501 with @xmath502 next , we will show that for any given @xmath503 there is a sufficiently large @xmath32 such that @xmath504 according to lemma 3 of @xcite , @xmath505 has eigenvalues bounded away from 0 and @xmath332 with probability tending to one as @xmath506 therefore , @xmath507 where we use the fact that @xmath508 and @xmath509 for @xmath510 notice that the first term @xmath511 we also may choose the sufficiently large @xmath32 such that the third term can be dominated by the first term uniformly on @xmath512 finally , we observe that the @xmath513-th element of @xmath514 is given by @xmath515 which is bounded by @xmath516 thus , the second is bounded by @xmath517 which is also dominated by the first term . in combination with the nonnegativity of the first term , we show , which implies with probability at least @xmath518 that there exists a local minimizer in the ball @xmath519 i.e. , @xmath520 again by the property of b - spline , we have that @xmath521 the proof is finished in combination with @xmath499 the proofs for theorem [ cvarying ] is very similar to theorem [ cadditive ] , and thus omitted here . 9 cai , z. , fan , j. and yao , q. ( 2000 ) . functional - coefficient regression models for nonlinear time series . _ j. amer . assoc . _ * 95 * , 941956 . cai , z. and xu , x. ( 2008 ) . nonparametric quantile estimations for dynamic smooth coefficient models . _ j. amer . assoc . _ * 103 * , 15951607 . chiang , c .- t . , rice , j. a. andwu , c. o. ( 2001 ) . smoothing spline estimation for varying coefficient models with repeatedly measured dependent variables . _ j. amer . assoc . _ * 96 * , 605619 . mr1946428 dahlhaus , r. ( 1996a ) . asymptotic statistical inference for nonstationary processes with evolutionary spectra , _ in : athens conference on applied probability and time series analysis , springer . _ 145259 . dahlhaus , r. ( 1996b ) . on the kullback - leibler information divergence of locally stationary processes . _ stochastic process . appl . _ * 62 * , 139168 . dahlhaus , r. ( 1997 ) . fitting time series models to nonstationary processes . statist . _ * 25 * , 137 . dahlhaus , r. , neumann , m. h. , and von sachs . nonlinear wavelet estimation of time - varying autoregressive processes . _ bernoulli . _ * 5 * , 873906 . dahlhaus , r. and rao , s. s. ( 2006 ) . statistical inference for time - varying arch processes . statist . _ * 34 * , 10751114 . de boor , c. ( 1978 ) . _ a practical guide to splines_. springer - verlag , new york . fan , j. , hrdle , w. and mammen , e. ( 1998 ) . direct estimation of low dimensional components in additive models . statist . _ * 26 * , 943971 . fan , j. and li , r. ( 2001 ) . variable selection via nonconcave penalized likelihood and its oracle properties . _ j. amer . assoc . _ * 96 * , 13481360 . fan , j. , yao , q. , and cai , z. ( 2002 ) . adaptive varying - coefficient linear models . _ j. r. stat . methodol . _ * 65 * , 5780 . fan , j. and zhang , w. ( 1999 ) . statistical estimation in varying coefficient models . _ ann . statist . _ * 27 * 14911518 . mr1742497 fryzlewicz , p. , sapatinas , t. and rao , s. s. ( 2008 ) . normalized least - squares estimation in time - varying arch models._ann . statist . _ * 36 * , 742786 . fu , w. ( 1998 ) . penalized regression : the bridge versus the lasso . _ j. comput . graph . statist . _ * 7 * , 397416 . hafner , c. m. and linton , o. ( 2010 ) . efficient estimation of a multivariate multiplicative volatility model . _ j. econometrics . _ * 159 * , 5573 . hastie , t. j. and tibshirani , r. j. ( 1990 ) . _ generalized additive models_. crc press . hastie , t. j. and tibshirani , r. j. ( 1993 ) varying - coefficient models . b. _ * 55 * , 757796 . mr1229881 hoover , d. r. , rice , j. a. , wu , c. o. and yang , l .- p . nonparametric smoothing estimates of time - varying coefficient models with longitudinal data . _ biometrika . _ * 85 * , 809822 . mr1666699 horowitz , j. , klemel , j. and mammen , e. ( 2006 ) . optimal estimation in additive regression models . _ bernoulli . _ * 12 * , 271298 . huang , j. z. ( 1998 ) . projection estimation in multiple regression with application to functional anova models _ * 26 * , 242272 . huang , j. z. , wu , c. o. and zhou , l. ( 2002 ) . varying - coefficient models and basis function approximations for the analysis of repeated measurements . _ biometrika . _ * 89 * , 111128 . mr1888349 huang , j. z. and shen , h. ( 2004 ) . functional coefficient regression models for nonlinear time series : a polynomial spline approach . j. stat . _ * 31 * , 515534 . mr2101537 huang , j. z. and yang , l. ( 2004 ) . identification of non - linear additive autoregressive models . _ j. r. stat . methodol . _ * 66 * , 463477 . huang , j. z.,wu , c. o. and zhou , l. ( 2004 ) . polynomial spline estimation and inference for varying coefficient models with longitudinal data . _ statist . * 14 * 763788 . mr2087972 huang , j. , horowitz , j. l. and wei f. ( 2010 ) . variable selection in nonparametric additive models . statist . _ * 38 * , 22822313 . kim , w. , linton , o. b. , and hengartner , n. w. ( 1999 ) . a computationally efficient oracle estimator for additive nonparametric regression with bootstrap confidence intervals . _ j. comput . graph . statist . _ * 8 * , 278297 . linton , o. and nielsen , j. p. ( 1995 ) . a kernel method of estimating structured nonparametric regression based on marginal integration . _ biometrika . linton , o. ( 1997 ) . efficient estimation of additive nonparametric regression models . _ biometrika . _ * 84 * , 469473 . r. , yang . l. and hrdle , w. k. ( 2013 ) . oraclly efficient two - step estimation of generalized aditive model . _ j. amer . assoc . _ * 108 * , 619631 . koo , b. and linton , o. ( 2012 ) . semiparametric estimation of locally stationary diffusion models . _ j. econometrics . _ * 170 * , 210233 . mammen , e. , linton , o. and nielsen , j. ( 1999 ) . the existence and asymptotic properties of a backfitting projection algorithm under weak conditions . _ statist . _ * 27 * , 14431490 . priestley , m. b. ( 1965 ) . evolutionary spectra and non - stationary process . _ j. r. stat . methodol . _ * 27 * , 204237 . stone , c. j. ( 1982 ) . optimal global rates of convergence for nonparametric regression . statist . _ * 10 * , 10401053 . mr0673642 stone , c. j. ( 1985 ) . additive regression and other nonparametric models . statist . _ * 13 * , 689705 . stone , c. j. ( 1994 ) . the use of polynomial splines and their tensor products in multivariate function estimation . statist . _ * 22 * , 118171 . tibshirani , r. ( 1996 ) . regression shrinkage and selection vis the lasso . _ j. r. stat . soc . ser . methodol . _ * 58 * , 267288 . tjstheim , d. and auestad , b. h. ( 1994 ) . nonparametric identification of nonlinear time series : projections . vogt , m. ( 2012 ) . nonparametric regression for locally stationary time series . _ * 40 * , 26012633 . wang , l. and yang , l. ( 2007 ) . spline - backfitted kernel smoothing of nonlinear additive autoregression model . _ * 35 * , 24742503 . wang , j. and yang , l. ( 2009 ) . efficient and fast spline - backfitted kernel smoothing of additive regression model . _ * 61 * , 663690 . wu , c. o. , chiang , c. t. and hoover , d. r. ( 1998 ) . asymptotic confidence regions for kernel smoothing of a varying - coefficient model with longitudinal data . _ j. amer . _ * 93 * 13881402 . mr1666635 xue , l. and yang , l.(2006 ) . estimation of semi - parametric additive coefficient model . _ j. statist . plann . inference . _ * 136 * , 25062534 . xue , lan . consistent variable selection in additive models . _ _ statist . sinica.__**19 * * , 12811296 . yang , l. , hardle , w. and nielsen , j. ( 1999 ) . nonparametric autoregression with multiplicative volatility and additive mean . _ j. time series . _ * 20 * , 579604 . yuan , m. and lin , y. ( 2006 ) . model selection and estimation in regression with grouped variables . _ j. r. stat . . b. stat . _ * 68 * , 4967 . zhang , x. k. , park , b.u . and wang , j. l. time - varying additive models for longitudinal data . _ j. amer . * 108 * , 983998 . zhang , x. k , and wang , j. l. ( 2013 ) . varying - coefficient additive models for functional data . _ biometrika . _ * 102 * , 1532 . zou , h. ( 2006 ) . the adaptive lasso and its oracle properties . _ j. amer . * 101 * , 14181429 .

we present _ ab initio _ plus cumulant - expansion calculations for an organic compound ( tmtsf)pf and a transition - metal oxide srvo . these materials exhibit characteristic low - energy band structures around the fermi level , which bring about interesting low - energy properties ; the low - energy bands near the fermi level are isolated from the other bands and , in the isolated bands , unusually low - energy plasmon excitations occur . to study the effect of this low - energy - plasmon fluctuation on the electronic structure , we calculate spectral functions and photoemission spectra using the _ ab initio _ cumulant expansion of the green s function based on the self - energy . we found that the low - energy plasmon fluctuation leads to an appreciable renormalization of the low - energy bands and a transfer of the spectral weight into the incoherent part , thus resulting in an agreement with experimental photoemission data .

1511.00218

the understanding of low - energy electronic structures and excitations in real materials is an important subject of condensed - matter physics and material science . interesting phenomena such as a non - fermi - liquid behavior and unconventional superconductivity are caused by the instability of electronic structures near the fermi level . a common feature is often found in the band structures showing such phenomena ; isolated bands appear near the fermi level . the width of these isolated bands is typically the order of 1 ev , which is comparable to local electronic interactions . thus , in these isolated bands , the kinetic and potential energies compete with each other , and the competition is often discussed within a local - interaction approximation , as in the hubbard model . in real materials , however , there exist various elementally excitations not described by the local electronic interaction . the plasmon in metallic systems or the exciton in insulating systems are well known examples of such nonlocal excitations , which result from the long - range coulomb interaction . in the above - mentioned isolated - band systems , the plasmon excitation can occur in this band , and its energy scale can be very small ( of the order of 1 ev ) , which is comparable to the bandwidth and the size of the local coulomb interaction . in this study , we investigate the effect of the low - energy - plasmon fluctuation on the electronic structure of real isolated - band systems from first principles . for this purpose , we choose two materials , a quasi - one dimensional organic conductor ( tmtsf)@xmath1pf@xmath2 ( ref . ) , where tmtsf stands for tetramethyltetraselenafulvalene , and a three - dimensional perovskite transition - metal oxide srvo@xmath3 ( ref . ) . these materials are typical isolated - band systems and are studied as benchmark materials of the correlated metal , where the local coulomb - interaction effect on the electronic properties is investigated with much interest . @xcite in the present work , we focus on the low - energy plasmon effect on the electronic structure . @xcite through the comparison between theoretical and experimental results on the plasmon - related properties and spectral functions , we verify low - energy plasmon effects on the electronic structure of the real system . the organic conductor ( tmtsf)@xmath1pf@xmath2 is a representative quasi - one - dimensional material , @xcite and basically behaves as a good metallic conductor . @xcite at low temperature ( around 12 k ) , it undergoes a transition to a spin - density - wave phase . @xcite in the high - temperature metallic region , photoemission spectroscopy has observed small spectral weight near the fermi level . @xcite from this observation and the quasi - one - dimensional nature , the origin of the renormalization has been discussed in view of the tomonaga - luttinger liquid . @xcite on the other hand , this material exhibits clear low - energy plasma edges around 0.1 - 1 ev in the reflectance spectra . @xcite therefore , this plasmon excitation would also be a prominent renormalization factor of the electronic structure . the transition - metal oxide srvo@xmath3 is another well known correlated metal . @xcite many high - resolution photoemission measurements @xcite including the bulk - sensitive version @xcite were performed and clarified a strong renormalization of the low - energy isolated band and the satellite peak just below this band . the origin has actively been discussed in terms of the local electronic correlation . @xcite on the other hand , the reflectance and electron - energy - loss spectra have clarified low - energy plasmon excitations around 1.4 ev . @xcite in density - functional band structure , srvo@xmath3 has isolated bands of the @xmath4 orbitals around the fermi level , whose bandwidth is about 2.7 ev . @xcite also , the constrained random phase approximation gives an estimate of the local electronic interaction of @xmath52 - 3 ev . @xcite thus , the energy scale of the experimentally observed plasmon excitation is comparable to the bandwidth and the local coulomb interaction . hence , the low - energy plasmon fluctuation would certainly be relevant to the low - energy properties . to study how the plasmon excitation affects the electronic structure , we perform _ ab initio _ calculations based on the @xmath0 approximation . @xcite the @xmath0 calculation considers the self - energy effect due to the plasmon fluctuation and describes properly quasiparticle energies in the valence region . on the other hand , the description for the plasmon satellite in the spectral function is known to be less accurate . @xcite in order to improve this deficiency , _ ab initio _ @xmath0 plus cumulant ( @xmath0+@xmath6 ) calculations have recently been performed . @xcite the accuracy of the @xmath0+@xmath6 method has been verified in bulk silicon @xcite and simple metals , @xcite where the satellite property is satisfactorily improved . in the present study , we apply the @xmath0+@xmath6 method to the study of the above mentioned isolated - band systems , and show that the low - energy plasmon fluctuations modify substantially the low - energy electronic structure . the present paper is organized as follows . in sec . ii , we describe the @xmath0 and @xmath0+@xmath6 methods to calculate dielectric and spectral properties . computational details and results for ( tmtsf)@xmath1pf@xmath2 and srvo@xmath3 are given in sec . we also discuss the comparison between theory and experiment , focusing on the renormalization of the electronic structure due to the plasmon excitation . the summary is given in sec . in this section , we describe _ ab initio _ @xmath0 and @xmath7 methods . the latter is a post-@xmath0 treatment and uses the self - energies calculated with the @xmath0 approximation . below , we first describe details of the @xmath0 calculation . the non - interacting green s function is written as @xmath8 where @xmath9 and @xmath10 are wavevector and frequency , respectively . @xmath11 and @xmath12 are the kohn - sham ( ks ) wavefunction and its energy , respectively , and @xmath13 is the fermi level of the ks system . the @xmath14 parameter is chosen to be a small positive value to stabilize numerical calculations . a polarization function of a type @xmath15 is written in a matrix form in the plane wave basis as @xmath16 with @xmath17 and @xmath18 where @xmath19 is a reciprocal lattice vector . with this polarization function , the symmetrized dielectric matrix in reciprocal space is defined by @xmath20 with @xmath21 being the crystal volume . in the @xmath22 limit , the dielectric matrix in eq . ( [ symeps ] ) is expressed by @xcite @xmath23 \displaystyle -\frac{4\pi}{v } 2\sum_{{\bf k}}\sum^{unocc}_{\alpha}\sum^{occ}_{\beta } \biggl(\frac{p^{\mu}_{\alpha\beta}({\bf k } ) } { \epsilon_{\alpha{\bf k}}-\epsilon_{\beta{\bf k}}}\biggr ) \frac{m_{\alpha\beta}^{{\bf g'}}({\bf k , k})^*}{|{\bf g'}| } x_{\alpha{\bf k},\beta{\bf k}}(\omega ) & \mbox{(${\bf g=0},{\bf g'\ne 0}$ ) , } \\[+20pt ] \displaystyle -\frac{4\pi}{v } 2\sum_{{\bf k}}\sum^{unocc}_{\alpha}\sum^{occ}_{\beta } \frac{m_{\alpha\beta}^{{\bf g}}({\bf k , k})}{|{\bf g}| } \biggl(\frac{p^{\mu}_{\alpha\beta}({\bf k } ) } { \epsilon_{\alpha{\bf k}}-\epsilon_{\beta{\bf k}}}\biggr)^ * x_{\alpha{\bf k},\beta{\bf k}}(\omega ) & \mbox{($ { \bf g \ne 0 } , { \bf g'=0}$ ) , } \\[+20pt ] \displaystyle \delta_{{\bf gg'}}-\frac{4\pi}{v } \frac{1}{|{\bf g}| } \chi_{{\bf gg'}}({\bf 0},\omega)\frac{1}{|{\bf g'}| } & \mbox{($otherwise$ ) } \end{array } \right . \label{eq0 } \end{aligned}\ ] ] where @xmath24 approaches zero along the cartesian @xmath13 direction and @xmath25 is a matrix element of a momentum as @xmath26|\phi_{\beta{\bf k}}\rangle , \label{p_ij } \end{aligned}\ ] ] with @xmath27 being the nonlocal part of the pseudopotential . in the first line of eq . ( [ eq0 ] ) , the last term on the right hand side is the drude term of the intraband transitions around the fermi level , @xcite where @xmath28 is the bare plasma frequency . the other terms in eq . ( [ eq0 ] ) result from the interband transitions . we next describe the calculation of the self - energy . the operator of the exchange self - energy is defined by @xmath29 where @xmath30 is the bare coulomb interaction . in practice , we use an attenuation coulomb interaction @xmath31 with a cutoff @xmath32 instead of @xmath33 to treat the integrable singularities in the bare coulomb interaction . @xcite the matrix element of the exchange self - energy is thus @xmath34 where @xmath35 and @xmath36 are the `` body '' and `` head '' components of the exchange self - energy , respectively . the former body matrix element is expressed as @xmath37 with @xmath38 the prime in the sum of eq . ( [ sx ] ) represents the summation excluding the contribution of the head term of @xmath39 . corresponding to the replacement of @xmath33 with @xmath40 , the related fourier transform is modified from @xmath41 to @xmath42 in eq . ( [ sx ] ) . accordingly , the body matrix element in eq . ( [ sx ] ) is supplemented by the head component in eq . ( [ sxb+sxh ] ) @xmath43 the operator of the correlation self - energy is defined by @xmath44 where @xmath45 is the correlation part of the symmetrized screened coulomb interaction @xmath46 . here , @xmath47 is the inverse dielectric function , which is calculated by inverting the symmetrized dielectric matrix in eqs . ( [ symeps ] ) and ( [ eq0 ] ) . for a practical calculation of the matrix element of @xmath48 in eq . ( [ gw_sigma ] ) , we introduce the following model screened interaction @xcite @xmath49 with @xmath50 in eq . ( [ modelw ] ) , the real frequency @xmath10 is discretized into @xmath51 , and @xmath52 and @xmath53 are the pole and amplitude of the model interactions , respectively . the matrix element @xmath54 comprises a square matrix [ see sec . iii(a ) ] . since the frequency - dependent part @xmath55 is decoupled from the amplitude one , the frequency integral in @xmath56 can be analytically performed . the matrix element of @xmath57 consists of the body and head components as follows : @xmath58 the body matrix element in the above is given by @xmath59 where the numerator is given by @xmath60 with @xmath61 note that , in the practical calculation , the frequency @xmath10 for the self - energy in eq . ( [ eq : sgm ] ) is distinguished from the frequency @xmath51 for the screened interaction in eq . ( [ modelw ] ) . the body matrix element @xmath62 in eq . ( [ eq : sgm ] ) is supplemented with the head component in eq . ( [ scb+sch ] ) @xmath63 with @xmath64 with these ingredients , the spectral function is calculated via the wannier - interpolation method ( see below ) . the spectral function at an arbitrary @xmath9 is @xmath65 where @xmath66 is obtained by diagonalizing non - symmetric complex matrix in the wannier basis @xmath67 where @xmath68 is the fourier transform of the ks hamiltonian matrix in the wannier basis as @xmath69 with @xmath70 here , @xmath71 is a @xmath72 point in the regular mesh and @xmath73 is the total number of the @xmath72 points in the regular mesh . also , @xmath74 is the @xmath75th wannier orbital at the lattice point @xmath76 , and the transform @xmath77 is obtained in the wannier - function - generation routine . @xmath78 in eq . ( [ hij ] ) is the fourier transform of the self - energy in the wannier basis as @xmath79 with @xmath80 the matrix element @xmath81 is defined by @xmath82 with @xmath83 being the exchange - correlation potential in density - functional theory . in eq . ( [ akw ] ) , the energy shift @xmath84 is introduced to correct the mismatch of the fermi level between the ks and one - shot @xmath0 systems . this parameter is determined from the equation on the spectral norm @xmath85 where @xmath86 is the total number of electrons in the system and @xmath87 is the total number of sampling @xmath72 points after the interpolation . note that @xmath13 is set to the fermi level for the ks system . the flow of the calculation is as follows : we first perform density - functional calculations to obtain the band structures and the wannier functions for bands associated with the self - energy calculations . we then calculate the self - energies for the irreducible @xmath72-points @xmath88 in a regular mesh , including band off - diagonal terms as @xmath89 in the selected energy region . the self - energies at a @xmath90 point symmetrically equivalent to @xmath91 are the same as @xmath92 , but , in the time - reversal symmetry case , @xmath81 is obtained by @xmath93 . then , we transform @xmath81 in eq . ( [ sigma - od ] ) to the wannier representation @xmath94 with eq . ( [ sijrw ] ) . with these data , we evaluate the self - energy at an arbitrary @xmath95 via eq . ( [ sijkw ] ) . note that the calculated spectral function includes band - off - diagonal effects , which is discussed in appendix a. finally , we calculate the spectral function of eq . ( [ akw ] ) considering the energy shift @xmath84 in eqs . ( [ akw ] ) and ( [ norm ] ) . the @xmath7 approach is a theory beyond the @xmath0 approximation , @xcite which is based on systematic diagrammatic expansions . this approach is suitable for dealing with long - range correlations , i.e. , various types of the plasmon - fluctuation diagrams not included in the usual @xmath0 diagram . in the initial stage of the study , it was applied to a system of core electrons interacting with a plasmon field . @xcite currently , _ ab initio _ @xmath0+@xmath6 calculations have been known to give a better description for satellites due to the plasmon excitation . @xcite the green s function with the cumulant expansion is defined in the time domain by @xcite @xmath96 where @xmath97 for the first term on the right hand side and @xmath98 for the second term . the @xmath99 and @xmath100 are the cumulants for the hole and particle states , respectively . the spectral function is calculated by the fourier transform as @xmath101 which consists of the hole @xmath102 and particle @xmath103 contributions . the spectral function for the hole part is written as @xmath104 where the band sum is taken over the occupied states . to the lowest order in the screened interaction @xmath105 , the cumulant is obtained by @xcite @xmath106 where @xmath107 . in the present study , the cumulant is expanded around the quasiparticle energy @xmath108 , @xcite which is a solution of @xmath109 where @xmath84 is an energy shift to correct a mismatch of the fermi level between the ks and one - shot @xmath0+@xmath6 systems . by considering the fourier transform of @xmath110 in eq . ( [ chdef ] ) and the spectral representation of @xmath111 , and after some manipulation , @xcite the expression of the cumulant is obtained , which consists of the quasiparticle and satellite parts as @xmath112 with @xmath113 and @xmath114 note that @xmath115 is negative for the hole part . to show the expansion point explicitly , we add @xmath108 as an index in the cumulant eqs . ( [ ctoth ] ) , ( [ cqph ] ) , and ( [ csh ] ) . within the one - shot calculation , the position of the cumulant expansion may be taken at the non - interacting energy @xmath12 . @xcite by taking the expansion point at @xmath108 , the results may include some sort of the self - consistency effect . the hole self - energy in eq . ( [ cqph ] ) is defined by @xmath116 it should be noted that the derivative @xmath117 in eq . ( [ cqph ] ) is related to the @xmath115=0 component of the satellite cumulant in eq . ( [ csh ] ) as @xcite @xmath118 this expression is practically used for the evaluation of the derivative ; we avoid the numerical calculation of the derivative by a finite difference and use the integral expression on the right hand side of eq . ( [ del - eq - c ] ) for the derivative , since we find that the latter treatment is numerically more stable . the stable calculation of the derivative is important in keeping the sum rule on the @xmath7 spectrum . in the practical calculation , we divide @xmath119 in eq . ( [ ahkwgwc ] ) into the two parts for stable calculations as @xcite @xmath120 the first quasiparticle term on the right hand side can be calculated analytically as @xmath121 where @xmath122 the latter convolution term in eq . ( [ ahkw ] ) is calculated via the numerical integration as @xmath123 where the integrand in the right - hand side decays to zero rapidly . the particle part of the spectral function in eq . ( [ akwgwc ] ) is given by @xmath124 with @xmath125 in this case @xmath115 is positive . the band sum in eq . ( [ apkwgwc ] ) runs over the unoccupied states . the @xmath126 and @xmath127 are given by @xmath128 and @xmath129 respectively . the @xmath130 in eq . ( [ cqpp ] ) is the particle self - energy as @xmath131 note that the self - energies in the above equations are defined as the causal one , so that the imaginary part of the self - energy is positive for @xmath132 and negative for @xmath133 . in the particle part , @xmath134 holds similarly to eq . ( [ del - eq - c ] ) . the contribution from the quasiparticle part to @xmath135 is @xmath136 with @xmath137 and the convolution - part contribution is @xmath138 density - functional calculations were performed with @xcite with plane - wave basis sets , where we employed norm - conserving pseudopotentials @xcite and generalized gradient approximation ( gga ) for the exchange - correlation potential . @xcite maximally localized wannier functions @xcite were used for the interpolation of the self - energy . for the atomic coordinates of ( tmtsf)@xmath1pf@xmath2 , the experimental structure obtained by a neutron measurement @xcite at 20 k was adopted . the cutoff energies in wavefunction and in charge densities are 36 ry and 144 ry , respectively , and a 15@xmath13915@xmath1393 @xmath72-point sampling was employed . the cutoff for the polarization function in eq . ( [ eq : chi ] ) was set to be 3 ry , and 200 bands were considered , which covers an energy range from the bottom of the occupied states near @xmath14030 ev to the top of the unoccupied states near 15 ev , where 0 ev is the fermi level . the frequency grid of the polarization was taken up to @xmath141 = 86 ev in a double - logarithmic form , for which we sampled 99 energy - points for [ 0.1 ev : 43 ev ] and 10 points for [ 43 ev : 86 ev ] with initial grid set to be 0.0 ev . the @xmath72 sum over bz in eqs . ( [ eq : chi ] ) , ( [ eq0 ] ) , and ( [ wpl ] ) was evaluated by the generalized tetrahedron method . @xcite the drude term of @xmath142 is evaluated at the slightly shifted frequency @xmath143 ( au ) . the self - energy in eq . ( [ eq : sgm ] ) was calculated for the frequency range [ @xmath144 ev : @xmath145 ev ] , where @xmath145 was set to be 100 ev . in the practical calculation , we sampled 450 points for [ @xmath140100 ev : @xmath14010 ev ] with the interval of 0.2 ev , 1000 points for [ @xmath14010 ev : 10 ev ] with the 0.02 ev interval , and 450 points for [ 10 ev : 100 ev ] with the 0.2 ev interval . with this energy range , the high - frequency tail of the self - energy is sufficiently small . this convergence is important to preserve the norm of the spectral function . for srvo@xmath3 , band calculations were performed for the idealized simple cubic structure , where the lattice parameter was set to be @xmath146=3.84 . the cutoff energies for wavefunction and charge densities are 49 ry and 196 ry , respectively , and an 11@xmath13911@xmath13911 @xmath72-point sampling was employed . the cutoff energy for the polarization function was set to be 10 ry and 130 bands were considered , which cover from the bottom of the occupied states near @xmath14020 ev to the top of the unoccupied states near 90 ev . the frequency range of the polarization function was taken to be @xmath147 ev , where finer logarithmic sampling was applied to [ 0.1 ev : 110 ev ] with 189 points and coarser sampling was done for [ 110 ev : 220 ev ] with 10 points , with initial grid set to be 0.0 ev . in the self - energy calculation , @xmath145 was set to be 200 ev , thus , the frequency dependence of the self - energy was calculated for [ @xmath140200 ev : 200 ev ] , where we sampled 200 points for [ @xmath140200 ev : @xmath14040 ev ] with the 0.8 ev interval , 1600 points for [ @xmath14040 ev : 40 ev ] with the 0.05 ev interval , and 200 points for [ 40 ev : 200 ev ] with the 0.8 ev interval . in the fitting of the model screened interaction [ eqs . ( [ modelw ] ) and ( [ pole - z ] ) ] , the positions of the poles in the model interaction are set as follows : @xcite @xmath148 with @xmath149 . the total number of \{@xmath150 } is the same as that of the frequency grid \{@xmath51 } for the polarization function . we checked that the _ ab initio _ screened interaction @xmath151 are satisfactorily reproduced by this model function @xmath152 . the broadening @xmath14 in eqs . ( [ eq : chi ] ) , ( [ theta ] ) , ( [ eq : sgm ] ) , ( [ csh ] ) , ( [ imsgmh ] ) , ( [ csp ] ) , and ( [ imsgmp ] ) was set to be 0.02 ev for ( tmtsf)@xmath1pf@xmath2 and 0.05 ev for srvo@xmath3 . the value of @xmath14 is desirable to be small enough , but the lower - bound in the practical calculation is determined by the resolution of the band dispersion , depending primary on the @xmath72-mesh density . also , the cutoff @xmath32 in the attenuation potential [ eqs . ( [ sx ] ) and ( [ wc ] ) ] is 28.27 for ( tmtsf)@xmath1pf@xmath2 and 11.53 for srvo@xmath3 , respectively . for ( tmtsf)@xmath1pf@xmath2 , the shift @xmath84 in the spectral function @xmath153 in eq . ( [ akw ] ) was found to be 1.06 ev for the @xmath0 calculation and 0.90 ev for the @xmath0+@xmath6 one , respectively . for srvo@xmath3 , @xmath84 = 2.1 ev for the @xmath0 and 2.35 ev for the @xmath0+@xmath6 . in @xmath0+@xmath6 , two numerical integrals on time and frequency appear [ eqs . ( [ ash ] ) and ( [ asp ] ) for the time integral and eqs . ( [ csh ] ) and ( [ csp ] ) for frequency integral ] , which must be treated carefully . we performed time integrals in eq . ( [ ash ] ) numerically for the range [ @xmath154 au : 0 au ] and those in eq . ( [ asp ] ) for [ 0 au : @xmath155 au ] , where @xmath155 is 50 ( au ) . the total number of the time grid @xmath156 is 50000 , with the interval @xmath157 au . note that @xmath158 is necessary to reproduce the norm of the spectral weight correctly . the frequency integral in eqs . ( [ csh ] ) and ( [ csp ] ) was numerically evaluated with the simpson s formula for the interval @xmath159 divided into 21 subintervals , which is also important in obtaining the correct time dependence of the satellite cumulant . also , in the @xmath72 integration to obtain the @xmath7 density of states @xmath160 , the random @xmath72-point sampling was performed to improve the statistical average , where the wannier interpolation technique was efficiently applied . with this condition , we obtained well converged spectra . figure [ ggaband ] shows calculated gga band structures of ( tmtsf)@xmath1pf@xmath2 [ panel ( a ) ] and srvo@xmath3 [ ( b ) ] . in both systems , isolated bands are found around the fermi level . in ( tmtsf)@xmath1pf@xmath2 , the isolated bands consist of highest - occupied molecular orbital ( homo ) of two molecules in the unit cell . [ for detailed atomic geometry of ( tmtsf)@xmath1pf@xmath2 , refer to refs . and . ] we call them the homo " bands which are shown with the green - dotted curves . in addition , in this figure , homo@xmath1401 " and homo@xmath1402 " bands are shown by blue- and black - dotted curves , respectively . @xcite in srvo@xmath3 , the isolated bands ( green - dotted curves ) around the fermi level are formed by the @xmath4 orbitals of the vanadium atom and bands around [ @xmath1407 ev : @xmath1402 ev ] come from the oxygen-@xmath161 orbitals ( blue - dotted curves ) . @xcite pf@xmath2 and ( b ) srvo@xmath3 . the fermi level is at zero energy . in the panel ( a ) , the green - dotted , blue - dotted , and black - dotted curves denote the homo , homo@xmath162 , and homo@xmath163 bands , respectively . in ( b ) , the green - dotted and blue - dotted curves correspond to the @xmath4 and o@xmath164 bands , respectively.,scaledwidth=35.0% ] to confirm the low - energy plasmon excitation in the above isolated - band systems , we calculated the reflectance spectra with the random - phase approximation @xmath165 where @xmath166 is obtained by inverse of the dielectric matrix in eq . ( [ eq0 ] ) . figure [ reflectance ] ( a ) is the result for ( tmtsf)@xmath1pf@xmath2 , where dark - red and light - green colors represent the results in the light polarization of @xmath167 and @xmath168 , respectively , and the @xmath146 axis ( @xmath146@xmath169@xmath170 ) is the one - dimensional conducting axis . the calculated results ( solid curves ) are compared with experimental results ( circles ) . we see that the theoretical plasma edges satisfactorily reproduce the experimental ones around 0.8 ev for @xmath167 and 0.1 - 0.2 ev for @xmath168 , which indicates that the present scheme correctly captures the low - energy plasmon excitation . the panel ( b ) shows the result for srvo@xmath3 . we again see a reasonable agreement between the theory and experiment for the plasma edge ( 1.8 ev for theory and 1.4 ev for the experiment ) . ( solid curve ) based on the random - phase approximation and experimental one ( open circles ) of ( a ) ( tmtsf)@xmath1pf@xmath2 measured at 25 k and ( b ) srvo@xmath3 measured at room temperature . the experimental data are taken from ref . for ( tmtsf)@xmath1pf@xmath2 and ref . for srvo@xmath3 . in ( tmtsf)@xmath1pf@xmath2 , the results for @xmath167 and @xmath168 are displayed by dark red and light green , respectively.,scaledwidth=40.0% ] table [ param - wwu ] summarizes parameters characterizing low - energy electronic structures of the two isolated - band systems , i.e. , the homo bands of ( tmtsf)@xmath1pf@xmath2 and the @xmath4 bands of srvo@xmath3 . the table includes the calculated bare plasma frequency @xmath171 in eq . ( [ wpl ] ) , bandwidth @xmath105 , and effective local - interaction parameter @xmath172 with @xmath173 and @xmath21 being onsite and nearest - neighbor interactions , respectively , calculated with the constrained random phase approximation . @xcite we note that @xmath171 is by definition different from the plasma edge in the reflectance spectra in fig . [ reflectance ] ; the latter energies are lowered from the bare @xmath171 by the presence of the individual electronic excitations . we see that @xmath171 has the same size as @xmath105 and @xmath172 , indicating that energy scale of the plasmon excitation would compete with those of kinetic and local electronic - interaction energies of electrons in the isolated band . the long - range interaction and related plasmon fluctuations would clearly be important for electronic structure . c@ c@ c@ c@ c@ c + [ -8pt ] & @xmath171 & @xmath105 & @xmath173 & @xmath21 & @xmath172 + [ 3pt ] + [ -5pt ] & 1.25 ( @xmath167 ) & & & & + & 0.20 ( @xmath168 ) & & & & + [ 4pt ] srvo@xmath3 & 3.54 & 2.55 & 3.48 & 0.79 & 2.69 + [ 2pt ] [ param - wwu ] to study effects of the low - energy plasmon excitation in the isolated bands on the electronic structure , we calculated a spectral function @xmath174 for the homo bands of ( tmtsf)@xmath1pf@xmath2 . figure [ akw - tmtsf ] displays the calculated spectra , where the panels ( a ) and ( b ) are the @xmath0 result via eq . ( [ akw ] ) and the @xmath0+@xmath6 one via eq . ( [ akwgwc ] ) , respectively . for comparison , the gga band structure is superposed with blue - solid curves . in the @xmath0 spectrum , clear incoherent peaks appear ; along the y-@xmath175 line , the spectral intensities of plasmaron states @xcite emerge about 1 ev above ( below ) the unoccupied ( occupied ) part of the homo bands . @xcite also , along the x - m line , the spectra are more broadened and spread in the range from @xmath1401.5 to 0 ev . interestingly , these sharp plasmaron peaks do not appear in the @xmath0+@xmath6 spectrum . instead , the @xmath0+@xmath6 spectrum exhibits a broad incoherent structure throughout bz . this is because the @xmath0+@xmath6 treatment further takes into account the long - range correlation effect @xcite or various types of the self - energy diagram involving the plasmon fluctuation , which is not included in the standard @xmath0 calculations . in the panel ( c ) , density of states @xmath176 is shown , where the @xmath0+@xmath6 , @xmath0 , and gga results are plotted by red - solid , blue - dotted , and thin - black - solid curves , respectively . compared to the gga spectrum , the @xmath0 and @xmath7 spectra show an appreciable band renormalization around the fermi level by the plasmon excitation . we again confirm that the distinct plasmon satellite ( plasmaron ) around @xmath1402 ev and + 1 ev in the @xmath0 spectrum disappears in the @xmath0+@xmath6 spectrum . pf@xmath2 calculated with ( a ) @xmath0 approximation and ( b ) @xmath7 method . blue - solid curves are the gga results . the fermi level is at zero energy . the colorbar is in linear scale . ( c ) comparison of the density of states among the @xmath7 ( red - solid curve ) , @xmath0 ( blue - dotted curve ) , and gga ( black - thin curve ) results . , scaledwidth=45.0% ] figure [ pes - tmtsf ] is the comparison between theoretical photo - emission spectra @xmath177 and experimental one ( green open circles ) obtained with heii radiation ( @xmath178=40.8 ev ) at 50 k ( ref . ) . thick - red - solid , blue - dotted , and thin - black - solid curves are the @xmath0+@xmath6 , @xmath0 , and gga results , respectively . the spectra were calculated for the homo , homo@xmath162 , and homo@xmath163 bands to cover the energy range measured in the experiment . we see that the gga spectrum around the fermi level is largely reduced in the @xmath0 and @xmath0+@xmath6 spectra by the self - energy effect due to the plasmon excitation . the @xmath0+@xmath6 spectrum agrees with the overall profile of the experimental spectrum better than the @xmath0 result ; the broader spectrum is obtained in the @xmath7 result around @xmath1793 ev than in the @xmath0 one . the discrepancy in the level position in this region between the theory and experiment arises probably from the level underestimation of the flat gga homo@xmath1401 band [ see fig . [ ggaband ] ( a ) ] . , and homo@xmath163 bands . red - thick - solid , blue - dotted , and black - thin - solid curves denote the @xmath0+@xmath6 , @xmath0 , and gga results , respectively . a lorentzian broadening of 0.02 ev is applied to the calculated spectra.,scaledwidth=40.0% ] next , we consider the low - energy plasmon - fluctuation effect on the electronic structure of the transition - metal oxide srvo@xmath3 . figure [ akw - svo ] shows the spectral function calculated for the @xmath4 and o@xmath180 bands , where the panels ( a ) and ( b ) show the @xmath0 and @xmath0+@xmath6 results , respectively . the gga bands are depicted with blue - solid curves . the low - energy plasmon satellite emerges around 1 ev above ( below ) the unoccupied ( occupied ) part of the @xmath4 bands , but the intensity is weaker than that of ( tmtsf)@xmath1pf@xmath2 . similarly to the tmtsf case , the @xmath0+@xmath6 for srvo@xmath3 makes the plasmon satellite broader than the @xmath0 result . on the o@xmath180 bands , the self - energy effect is appreciable ; the imaginary part of the self - energy , which is related to the lifetime of the quasiparticle states , becomes large as the binding energy is apart from the fermi level . thus , the spectrum of the o@xmath180 band becomes rather broad . the panel ( c ) shows the comparison of calculated density of states . we see that the gga spectrum in the @xmath4 and o@xmath180 bands is largely renormalized in the @xmath0 and @xmath7 spectra . the shape of the @xmath0 spectrum resembles the @xmath0+@xmath6 one , though the latter is somewhat more broadened . and o@xmath180 bands of srvo@xmath3 calculated with ( a ) @xmath0 approximation and ( b ) @xmath7 method . blue - solid curves are the gga results . the fermi level is at zero energy . the colorbar is in linear scale . ( c ) comparison of the density of states among @xmath0+@xmath6 ( red - solid curve ) , @xmath0 ( blue - dotted curve ) , and gga ( black - thin curve ) results . , scaledwidth=45.0% ] figure [ pes - svo ] compares theoretical photo - emission spectra with experimental ones taken from refs . and . light - green circles and black dots are the experimental photoemission spectra with the photon energies @xmath181 60 ev ( ref . ) and @xmath181 900 ev ( ref . ) , respectively . on the overall profile , the _ ab initio _ @xmath0+@xmath6 spectrum reasonably reproduces the experimental results such as the relative intensity between the @xmath4 and o@xmath180 bands . @xcite on the other hand , we see that the theoretical incoherent intensity around @xmath1822 ev is weaker than the experimental one . + @xmath6 , @xmath0 , and gga results , respectively . in the experiment , light - green circles and black dots denote the photoemission spectra with the photon energies @xmath181 60 ev ( ref . ) and @xmath181 900 ev ( ref . ) , respectively . a lorentzian broadening of 0.05 ev is applied to the calculated spectra.,scaledwidth=40.0% ] in fig . [ xas - svo ] , we compare theoretical spectra of an unoccupied region in the @xmath4 bands @xmath183 with experimental one taken from the soft x - ray absorption spectrum ( green circles ) ( ref . ) . the effect of the low - energy plasmon excitation on the electronic structure appears to be stronger in the unoccupied region than in the occupied one . @xcite the _ ab initio _ @xmath0+@xmath6 spectrum resembles the experimental spectrum more closely than the @xmath0 result ; the cumulant expansion reduces the quasiparticle intensity and shifts the plasmon satellite to a lower energy . we note that the theoretical spectra do not include the contribution from the @xmath184 states . also , in srvo@xmath3 , since the local - interaction effect competes with the plasmon excitation , the view of the competition of the several factors would be important for quantitative understanding , which remains to be explored . bands . the experimental data are taken from the soft x - ray absorption spectrum ( ref . ) . red - thick - solid , blue - dotted , and black - thin - solid curves are the @xmath0+@xmath6 , @xmath0 , and gga results , respectively . a lorentzian broadening of 0.05 ev is applied to the calculated spectra.,scaledwidth=40.0% ] we have performed _ ab initio _ @xmath0 plus cumulant - expansion calculations for an organic conductor ( tmtsf)@xmath1pf@xmath2 and a transition - metal oxide srvo@xmath3 to study the low - energy plasmon - fluctuation effect on the electronic structure . the bands around the fermi level of these materials are isolated from the other bands , and the low - energy plasmon excitations derived from these isolated bands exist . our calculated reflectance spectra well identify the experimental low - energy plasmon peaks . by calculating the cumulant - expanded green s function based on the @xmath0 approximation to the self - energy , we have simulated spectral functions and compare them with photoemission data . we have found agreements between them , indicating that the low - energy plasmon excitation certainly affects the low - energy electronic structure ; it reduces the quasiparticle spectral weight around the fermi level and leads to the weight transfer to the satellite parts . this effect was found to be more or less appreciable in ( tmtsf)@xmath1pf@xmath2 than in srvo@xmath3 . in particular , in ( tmtsf)@xmath1pf@xmath2 , the spectrum at the standard @xmath0 level exhibits a clear plasmaron state , but considering the plasmon - fluctuation effects not treated in the standard @xmath0 calculation leads to the disappearance of the state in the @xmath7 . since the low - energy isolated - band structure is commonly found in various materials , the low - energy plasmon effect pursued in the present work can provide a basis for understanding the electronic structure of real systems . the recent progress in the photoemission experiment for the correlated materials ( refs . ) requires a concomitant progress on the theoretical side and the present _ ab initio _ many - body calculations would provide firm basis for them . in the present study , we have focused on the long - range correlation and treated it effectively with the cumulant - expansion method , while the short - ranged correlation , which is appropriately described by the @xmath185-matrix framework , is neglected . this is a future challenge which remains to be explored . in addition , recent photoemission spectroscopy reveals appreciable differences in the electronic structure between the bulk and thin film systems . @xcite the electronic structure of the surface is very sensitive to the atomic configurations at the surface @xcite and therefore careful analyses for the atomic structure and its effect on electronic structure are required . the _ ab initio _ calculations for the surface effect are clearly important for the deep understanding of the spectroscopy of the real materials . this is also a future challenge . we would like to thank norikazu tomita and teppei yoshida for useful discussions . calculations were done at supercomputer center at institute for solid state physics , university of tokyo . this work was supported by grants - in - aid for scientific research ( no . 22740215 , 22104010 , 23110708 , 23340095 , 23510120 , 25800200 ) from mext , japan , and consolidator grant correlmat of the european research council ( project number 617196 ) . we briefly describe effects of band - off - diagonal terms of the self - energy on the spectral functions . with neglecting the band - off - diagonal terms , the spectral function is calculated as @xmath186 where the matrix element of the self - energy with respect to the ks state @xmath187 @xmath188 is the diagonal term of eq . ( [ sigma - od ] ) . the @xmath84 in eq . ( [ akwdiag ] ) is the energy shift to correct the mismatch of the fermi level between the initial and final states [ see eq . ( [ norm ] ) ] . figure [ akw - svo - diag ] displays the gw spectral function of srvo@xmath3 ( a ) without and ( b ) with band off - diagonal terms of self - energy . also , the panel ( c ) shows the comparison between the calculated density of states . we do not see discernible difference between the two ; the band off - diagonal terms of the self - energy is negligible . this is because , in srvo@xmath3 , the @xmath189 orbitals of the v atom are well localized and the hybridization with o@xmath180 orbital is small . similarly to srvo@xmath3 , the band - off - diagonal effect on the spectral function is found to be very small in ( tmtsf)@xmath1pf@xmath2 . and o@xmath180 bands of srvo@xmath3 ( a ) without and ( b ) with the band off - diagonal matrix elements of the self - energy . blue - solid curves are the gga result . the fermi level is at zero energy . the colorbar is in linear scale . ( c ) comparison of the density of states without ( dark - red - solid curve ) and with ( light - green - dotted curve ) the off - diagonal terms.,scaledwidth=45.0% ] d. jerome , a. mazaud , m. ribault and k. bechgaard , j. phys . ( paris ) * 41 * , l95 ( 1980 ) . for a review , see , e.g , t. ishiguro , k. yamaji , and g. saito , _ organic superconductors _ ( springer - verlag , heidelberg , 1998 ) . for a review , see , e.g , k. kuroki , j. phys . . jpn . * 75 * , 051013 ( 2006 ) . s. ishibashi , sci . technol . adv . mater . * 10 * , 024311 ( 2009 ) . e. pavarini , s. biermann , a. poteryaev , a. i. lichtenstein , a. georges , and o. k. andersen , phys . lett . , * 92 * , 176403 ( 2004 ) . i. a. nekrasov , k. held , g. keller , d. e. kondakov , th . pruschke , m. kollar , o. k. andersen , v. i. anisimov , and d. vollhardt , phys . b * 73 * , 155112 ( 2006 ) . f. lechermann , a. georges , a. poteryaev , s. biermann , m. posternak , a. yamasaki , and o. k. andersen , phys . b * 74 * , 125120 ( 2006 ) . b. gallois , j. gaultier , c. hauw , t .- d . lamcharfi and a. filhol , acta cryst . * b42 * , 564 ( 1986 ) . martin dressel , isrn condensed matter physics , * 2012 * 732973 ( 2012 ) . c. s. jacobsen , d. b. tanner , and k. bechgaard , phys . 46 * 1142 ( 1981 ) . d. malterre , b. dardel , m. grioni , p. weibel , y. baer , journal de physique iv , * 3 * , 97 ( 1993 ) . b. dardel , d. malterre , m. grioni , p. weibel , y. baer , j. voit and d. jrme , europhys . * 24 * , 687 ( 1993 ) . f. zwick , s. brown , g. margaritondo , c. merlic , m. onellion , j. voit , and m. grioni , phys . , * 79 * , 3982 ( 1997 ) . f. zwick , m. grionia , g. margaritondoa , v. vescolib , l. degiorgib , b. alavic , g. grnerc , solid state commun . * 113 * , 179 ( 2000 ) . v. vescoli , f. zwick , w. henderson , l. degiorgi , m. grioni , g. gruner , and l.k . montgomery , eur . j. b * 13 * , 503 ( 2000 ) . i. h. inoue , o. goto , h. makino , n. e. hussey , m. ishikawa , phys . b * 58 * , 4372 ( 1998 ) . h. makino , i. h. inoue , m. j. rozenberg , i. hase , and y. aiura , and s. onari , phys . b * 58 * , 4384 ( 1998 ) . k. morikawa , t. mizokawa , k. kobayashi , a. fujimori , h. eisaki and s. uchida , f. iga , and y. nishihara , phys . b * 52 * , 13711 ( 1995 ) . a. sekiyama , h. fujiwara , s. imada , s. suga , h. eisaki , s. i. uchida , k. takegahara , h. harima , y. saitoh , i. a. nekrasov , g. keller , d. e. kondakov , a.v . kozhevnikov , th . pruschke , k. held , d. vollhardt , and v. i. anisimov , phys . lett . , * 93 * , 156402 ( 2004 ) . k. maiti , u. manju , s. ray , p. mahadevan , i. h. inoue , c. carbone , and d. d. sarma , phys . b * 73 * , 052508 ( 2006 ) . g. kotliar , s. y. savrasov , k. haule , v. s. oudovenko , o. parcollet , and c. a. marianetti , rev . phys . , * 78 * 865 ( 2006 ) . m karolak , t o wehling , f lechermann and a i lichtenstein , j. phys . : condens . matter * 23 * , 085601 ( 2011 ) . l. huang and y. wang , europhys . lett , * 99 * 67003 ( 2012 ) . m. casula , a. rubtsov , and s. biermann , phys . b * 85 * , 035115 ( 2012 ) . h. lee , k. foyevtsova , j. ferber , m. aichhorn , h. o. jeschke , and r. valent , phys . b * 85 * , 165103 ( 2012 ) . s. biermann , f. aryasetiawan , and a. georges , phys . , * 90 * , 086402 ( 2003 ) . j. m. tomczak , m. casula , t. miyake , f. aryasetiawan and s. biermann , europhys . * 100 * , 67001 ( 2012 ) . c. taranto , m. kaltak , n. parragh , g. sangiovanni , g. kresse , a. toschi , and k. held , phys . b * 88 * , 165119 ( 2013 ) . r. sakuma , p. werner , and f. aryasetiawan , phys . b * 88 * , 235110 ( 2013 ) . r. sakuma , c. martins , t. miyake , and f. aryasetiawan , phys . b * 89 * , 235119 ( 2014 ) . j. m. tomczak , m. casula , t. miyake , and s. biermann , phys . rev . b * 90 * , 165138 ( 2014 ) . l. hedin , phys . rev . * 139 * , a796 ( 1965 ) . m. s. hybertsen and s. g. louie , phys . rev . b * 34 * , 5390 ( 1986 ) . g. onida , l. reining , a. rubio , rev . phys , * 74 * , 601 ( 2002 ) . f. aryasetiawan and o. gunnarsson , rep . . phys . * 61 * , 237 ( 1998 ) . m. s. hybertsen and s. g. louie , phys . lett . * 55 * , 1418 ( 1985 ) . r. w. godby , m. schlter , and l. j. sham , phys . lett . * 56 * , 2415 ( 1986 ) . r. w. godby , m. schlter , and l. j. sham , phys . b * 35 * , 4170 ( 1987 ) . j. e. northrup , m. s. hybertsen , and s. g. louie , phys . lett . * 59 * , 819 ( 1987 ) . n. hamada , m. hwang , and a. j. freeman , phys . b * 41 * , 3620 ( 1990 ) . f. aryasetiawan , phys . b * 46 * , 13051 ( 1992 ) . f. aryasetiawan and o. gunnarsson , phys . lett . * 74 * , 3221 ( 1995 ) . w. g. aulbur , l. joensson , and j. w. wilkins , solid state phys . * 54 * , 1 ( 2000 ) . b. arnaud and m. alouani , phys . b * 62 * , 4464 ( 2000 ) . w. ku and a. g. eguiluz , phys . lett . * 89 * , 126401 ( 2002 ) . m. usuda , n. hamada , t. kotani , and m. van schilfgaarde , phys . b * 66 * , 125101 ( 2002 ) . t. kotani and m. van schilfgaarde , solid state commun . * 121 * , 461 ( 2002 ) . a. yamasaki and t. fujiwara , phys . b * 66 * , 245108 ( 2002 ) . s. lebgue , b. arnaud , m. alouani , and p. e. bloechl , phys . b * 67 * , 155208 ( 2003 ) . s. v. faleev , m. van schilfgaarde , and t. kotani , phys . lett . * 93 * , 126406 ( 2004 ) . m. shishkin and g. kresse , phys . b * 74 * , 035101 ( 2006 ) . t. kotani , m. van schilfgaarde , and s. v. faleev , phys . b * 76 * , 165106 ( 2007 ) c. friedrich , s. blgel , and a. schindlmayr , phys . b * 81 * , 125102 ( 2010 ) . r. sakuma , t. miyake , and f. aryasetiawan , phys . b * 86 * , 245126 ( 2012 ) . t. miyake , c. martins , r. sakuma , and f. aryasetiawan , phys . b * 87 * , 115110 ( 2013 ) . y. nohara , s. yamamoto , and takeo fujiwara , phys . b * 79 * , 195110 ( 2009 ) . l. hedin , physica scripta . * 21 * , 477 ( 1980 ) . f. aryasetiawan , l. hedin , and k. karlsson , phys . lett . * 77 * , 2268 ( 1996 ) . m. guzzo , g. lani , f. sottile , p. romaniello , m. gatti , j. j. kas , j. j. rehr , m. g. silly , f. sirotti , and l. reining , phys . lett . * 107 * , 166401 ( 2011 ) . j. lischner , d. vigil - fowler , and s. g. louie , phys . lett . * 110 * , 146801 ( 2013 ) . j. lischner , d. vigil - fowler , and s. g. louie , phys . b * 89 * , 125430 ( 2014 ) . j. j. kas , j. j. rehr , and l. reining , phys . b * 90 * , 085112 ( 2014 ) . j. lischner , g. k. plsson , d. vigil - fowler , s. nemsak , j. avila , m. c. asensio , c. s. fadley , and s. g. louie , phys . rev . b bf 91 , 205113 ( 2015 ) . m. s. hybertsen and s. g. louie , phys . b * 35 * , 5585 ( 1987 ) . p. puschnig and c. ambrosch - draxl , phys . b * 66 * , 165105 ( 2002 ) . c. ambrosch - draxl and j. o. sofo , computer physics communications * 175 * , 1 ( 2006 ) j. spencer and a. alavi , phys . b * 77 * , 193110 ( 2008 ) . the hole green s function based the cumulant expansion is written in the time domain as @xmath190 where @xmath191 and @xmath192 are the non - interacting green s function and the cumulant , respectively , and we adopt the compressed notation @xmath193 , etc . the green s function based on the dyson equation is @xmath194 in the lowest order , the following integral equation for the cumulant holds : @xmath195 from this equation , the time dependence of the cumulant is determined . in the practical calculation , the @xmath0 approximation is employed in the self - energy in the right hand side in eq . ( [ ch12 ] ) . j. yamauchi , m. tsukada , s. watanabe , and o. sugino , phys . b * 54 * , 5586 ( 1996 ) . n. troullier and j. l. martins , phys . rev . b * 43 * 1993 ( 1991 ) . l. kleinman and d. m. bylander , phys . . lett . * 48 * 1425 ( 1982 ) . j. p. perdew , k. burke , and m. ernzerhof , phys . lett . * 77 * , 3865 ( 1996 ) . n. marzari and d. vanderbilt , phys . b * 56 * , 12847 ( 1997 ) . i. souza , n. marzari , and d. vanderbilt , phys . b * 65 * , 035109 ( 2001 ) . t. fujiwara , s. yamamoto , and y. ishii , j. phys . * 72 * , 777 ( 2003 ) . for ( tmtsf)@xmath1pf@xmath2 , we constructed the maximally localized wannier orbitals for three separated energy windows , which contains homo , homo@xmath1401 , and homo@xmath1402 bands respectively . as an initial guess for the construction of the homo wannier orbital in ( tmtsf)@xmath1pf@xmath2 , we used the @xmath161-type gaussians with the width 1.67 , which is placed on the center of the central c bond of the tmtsf molecule , in a direction perpendicular to the bond . also , an initial guess for the homo@xmath162 wannier orbital is a superposition of the @xmath161-type gaussians with width 1.18 put on the c atoms forming the center bond , in a direction perpendicular to the bond . in the homo@xmath163 band , the wannier orbital is found to be localized in only one of the two molecules in the unit cell . therefore , we take as an initial guess for the homo@xmath163 wannier orbital a superposition of the @xmath161-type gaussians with width 1.18 put on the se atoms in the molecule , in a direction perpendicular to the molecular plane . for srvo@xmath3 , we constructed wannier functions from the single energy window containing v 3d and o 2p orbitals . as the initial guess , @xmath189- and @xmath161- gaussians are set on the v and o sites , respectively . k. nakamura , r. arita , and m. imada , j. phys . . jpn . * 77 * , 093711 ( 2008 ) . k. nakamura , y. yoshimoto , t. kosugi , r. arita , and m. imada , j. phys . . jpn . * 78 * , 083710 ( 2009 ) . k. nakamura , y. yoshimoto , and m. imada , phys . b * 86 * , 205117 ( 2012 ) . t. kiss , a. chainani , h. m. yamamoto , t. miyazaki , t. akimoto , t. shimojima , k. ishizaka , s. watanabe , c .- t . chen , a. fukaya , r. kato and s. shin , nature communications , doi:10.1038/ncomms2079 ( 2012 ) . f. zwick , s. brown , g. margaritondo , c. merlic , m. onellion , j. voit , and m. grioni , phys . lett . * 79 * , 3982 ( 1997 ) . r. claessen , m. sing , u. schwingenschlgl , p. blaha , m. dressel , and c. s. jacobsen , phys . * 88 * , 096402 ( 2002 ) . m. takizawa , m. minohara , h. kumigashira , d. toyota , m. oshima , h. wadati , t. yoshida , a. fujimori , m. lippmaa , m. kawasaki , h. koinuma , g. sordi , and m. rozenberg , phys . b * 80 * , 235104 ( 2009 ) . t. yoshida , m. hashimoto , t. takizawa , a. fujimori , m. kubota , k. ono , and h. eisaki , phys . rev . b * 82 * , 085119 ( 2010 ) . b. dardel , d. malterre , m. grioni , p. weibel , and y. baer , f. levy , phys . lett . , * 67 * , 3144 ( 1991 ) . a. sekiyama , a. fujimori , s. aonuma , h. sawa , and r. kato , phys . b * 51 * 13899(r ) ( 1995 ) . h. ando , t. yokoya , k. ishizaka , s. tsuda , t. kiss , s. shin1 , t. eguchi , m. nohara , and h. takagi , j. phys . : condens . matter * 17 * 4935 ( 2005 ) . m. grioni , s. pons , and e. frantzeskakis , j. phys . : condens . matter * 21 * , 023201 ( 2009 ) k. koizumi , k. ishizaka , t. kiss , m. okawa , r. kato , and s. shin , j. phys . jpn , * 82 * 025004 ( 2013 ) .

* cafeas exhibits collapsed tetragonal ( ct ) structure and varied exotic behavior under pressure at low temperatures that led to debate on linking the structural changes to its exceptional electronic properties like superconductivity , magnetism , etc . here , we investigate the electronic structure of cafeas forming in different structures employing density functional theory . the results indicate better stability of the ct phase with enhancement in hybridization induced effects and shift of the energy bands towards lower energies . the fermi surface centered around point gradually vanishes with the increase in pressure . consequently , the nesting between the hole and electron fermi surfaces associated to the spin density wave state disappears indicating a pathway to achieve the proximity to quantum fluctuations . the magnetic moment at the fe sites diminishes in the ct phase consistent with the magnetic susceptibility results . notably , the hybridization of ca 4 states ( ca - layer may be treated as a charge reservoir layer akin to those in cuprate superconductors ) is significantly enhanced in the ct phase revealing its relevance in its interesting electronic properties . *

1703.07040

cafe@xmath0as@xmath0 exhibits a rich temperature - pressure phase diagram and provides an ideal platform to study the interplay between magnetism and superconductivity in fe - based compounds , where the superconductivity is believed to be driven by magnetic fluctuations . the crystal structure of cafe@xmath0as@xmath0 at room temperature and ambient pressure is tetragonal possessing @xmath3 space group with lattice parameters @xmath4 = 3.8915(2) , @xmath5 = 11.690(1) , and @xmath6 = 0.372(1)@xcite . upon cooling at ambient pressure , it undergoes a concomitant transition to spin density wave ( sdw ) state and orthorhombic ( o ) structure from its room temperature paramagnetic tetragonal ( t ) structure at about 170 k @xcite . the crystal structures in ambient conditions are shown in fig . [ fig1_struct ] . on application of pressure ( @xmath7 ) , _ both _ the structural and magnetic transitions get suppressed , and a collapsed tetragonal ( ct ) phase emerges with no magnetic long range order @xcite . superconductivity can be achieved in this material at an intermediate pressure range or via suitable electron or hole doping @xcite . ( a ) double unit cell of the tetragonal structure and ( b ) single unit cell of the orthorhombic structure . ] electronic structure calculations based on first principle approaches were employed extensively to address the ground state properties , fermi surface and pressure induced effects @xcite . while ligand @xmath8 states play major role in the electronic properties of cuprate superconductors ( the most studied unconventional superconductors ) , the main contribution at the fermi level , @xmath9 in the fe - based materials arises from the fe 3@xmath10 orbitals . these systems are more complex than cuprates due to the contribution of multiple bands at @xmath9 and drastic changes in the fermi surface topology arising from magnetic and structural transitions . for example , the two dimensional hole pockets at @xmath1-point in the tetragonal phase exhibit a transition to three dimensional topology at low temperatures @xcite and in ct phase , they disappear due to the shift of the energy bands . the disappearance of the fermi surface in ct phase rules out the possibility of fermi surface nesting necessary for the transition to the sdw state . the fe - based systems , in general , are found to be moderately correlated and inclusion of electron correlation strength , @xmath11 in the calculations renormalizes the energy bands without drastic change in the ground state properties @xcite . jean diehl _ et al._@xcite showed that the hund s coupling , @xmath12 ( @xmath130.8 ev ) influences the electronic structure more strongly than the changes found due to @xmath11 ( @xmath13 4 ev ) . they also showed that upon transition from t to ct phase , the @xmath14 bands become the least renormalized orbital from the highest renormalized one , which was attributed to increased hybridization of fe 3@xmath14 orbital with as 4@xmath15 and 4@xmath16 orbitals . the appearance of the ct phase under pressure and/or on suitable doping @xcite has been attributed to the increased hybridization among as 4@xmath8 orbitals along the compressed @xmath5 axis . _ @xcite found that the fe spin state strongly depends on the @xmath5 axis reduction in ct phase . based on fixed moment calculation , he showed that reduction by about half of the fe magnetic moment corresponds to the @xmath5 axis collapse to the experimental value . thus , there is a finite magnetic moment present at the fe sites in the ct phase when it is not superconducting . evidently , understanding of the importance of fe moment in superconductivity is still at its infancy and needs further investigation . these results indicate that the electronic structure is very sensitive to the distance of as sites from the fe plane , which is termed as _ ` pnictogen height'_. in the antiferromagnetic ( afm ) orthorhombic ( o ) phase , the spins show antiparallel coupling along the longer axis , @xmath4 and parallel coupling along shorter axis , @xmath17 . this has been theoretically found by many groups @xcite but a satisfactory resolution of the preferred spin coupling along @xmath4 and @xmath17 axis is not yet found . first principle calculations under hydrostatic@xcite and uniaxial@xcite pressure conditions reproduced the first order transition to ct phase from afm o / t phase accompanied by quenching of fe magnetic moments and also rule out the possibility of an intermediate tetragonal phase in cafe@xmath0as@xmath0 in contrary to various experiments @xcite , which is different from the behavior in bafe@xmath0as@xmath0 at intermediate pressure range @xcite . the dotted line represents notional phase boundary between various structural and magnetic phases . ] in the experimental front , the pressure induced effects in afe@xmath0as@xmath0 ( a = ca , ba , sr ) has been studied by many groups resulting into conflicting conclusions as discussed below . superconductivity in bafe@xmath0as@xmath0 & srfe@xmath0as@xmath0 containing alkaline earth elements larger than ca was found at a reasonably high pressure ( @xmath7 @xmath18 4 gpa and 3.2 gpa , respectively@xcite ) with @xmath19 of about 29 k and 27 k @xcite . on the other hand , cafe@xmath0as@xmath0 shows superconductivity with @xmath19 upto 12 k in a low pressure range from 0.2 to 0.9 gpa @xcite although the smaller atomic size of ca is expected to employ less strain into the system . a typical temperature - pressure phase diagram shown in fig . [ fig2_schem ] exhibits the experimental results obtained by torikachvili _ @xcite , which is consistent with the results in other studies @xcite . in fig . [ fig2_schem ] , we observe that the transition to orthorhombic afm phase is suppressed upon increasing pressure and disappears above @xmath7 = 0.35 gpa with the emergence of the ct phase . the transition temperature from t to ct phase gradually increases with pressure . the superconductivity appears at an intermediate @xmath7 range from 0.3 gpa to 0.9 gpa . in fact , superconductivity with @xmath19 @xmath18 10 k at 0.69 gpa pressure has been detected by several groups @xcite . subsequent experiments , however , did not find superconductivity although ct phase could be reached on application of pressure . it was suggested that the earlier measurements carried out using convectional liquid medium clamp pressure cell might have made the pressure quasi hydrostatic due to freezing of the medium . these later results obtained using helium as a pressure medium question the relation of superconductivity to ct phase and in line with other views suggesting non - magnetic behavior in ct phase that rules out spin - fluctuation . evidently , the observance / non - observance of superconductivity under pressure , non - observance of other structural phases under pressure , no superconductivity in ambient conditions , _ etc . _ are curious and remains to be an outstanding puzzle in this field . a study of the reported resistivity data exhibiting superconductivity reveal an interesting scenario ; the transitions in the resistivity data at @xmath7 = 0.23 gpa are significantly broad . on the other hand , the results exhibiting absence of superconductivity show sharp transitions throughout the pressure range involving the transitions from t to o , o to ct and t to ct phases . moreover , the superconductivity always found to appear at a pressure where the system is in the proximity of the transition to ct phase or some volume fraction is already in ct phase . thus , the origin of superconductivity in this material appears to have relation to the multi - crystallographic phases within the intermediate pressure regime . nuclear magnetic resonance @xcite , @xmath20 spin rotation @xcite , and neutron diffraction studies @xcite support the possibilities of coexisting o , ct and t phases . some studies indicated signature of an unidentified third phase appearing in the intermediate pressure regime @xcite . while the magnetic order in single or multiple phase scenario can be attributed to the o phase , it is difficult to delineate the phase responsible for superconductivity . clearly , superconductivity does not appear to be related to pure ct or o phase under pressure @xcite . then the question is ; how the superconductivity sets in cafe@xmath0as@xmath0 under pressure ? how non hydrostatic pressure helps to get supeconductivity while hydrostatic pressure does not ? superconductivity also appears via doping of charge carriers in these material - how different is that mechanism from the pressure induced one ? here , we studied the detailed electronic structure of cafe@xmath0as@xmath0 employing density functional theoretical ( dft ) methods at varied conditions with an emphasis on the tendency of the system to evolve to a certain electronic structure due to change in structural parameters induced by the pressure . we observe interesting spectral weight transfer with the application of pressure leading to complex change in fermiology of the system . the covalency between ca , as and fe states appear to play the key role in deriving the electronic properties of this system . the electronic structure of a material is very sensitive to the change in crystal structure as the hybridization of various electronic states depends on the bond length , distortion in the lattice structure ( leading to a change in bond angle ) etc . often it is found that covalency induces key interactions leading to distortion in the crystal structure @xcite that derives the ground state properties . in fig . [ fig : tdos ] , we show the calculated density of states ( dos ) of cafe@xmath0as@xmath0 in tetragonal phase . there are two distinctly separated energy regions in the figure contributed by as 4@xmath2 shallow core levels ( -14 to -10 ev ) and the valence electronic states above -6 ev . in the valence band regime , we observe signatures of hybridization between fe 3@xmath10 and as 4@xmath8 states . major contribution from as 4@xmath8 states appear in the bonding energy bands between -6 to -2 ev energies . the contribution for antibonding states appear above -2 ev and possess dominant fe 3@xmath10 character . the valence electrons in the vicinity of the fermi level , @xmath9 consist of essentially fe 3@xmath10 states . the fe - as hybridization leads to finite contribution from as 4@xmath8 states ( @xmath13 4% of the total contribution at @xmath9 ) . interestingly , the contribution of the ca 3@xmath10 states are also found to be significant ( @xmath13 2.4% ) . this suggests that although fe 3@xmath10 states play the dominant role in the electronic properties of cafe@xmath0as@xmath0 as also found in earlier studies , the role of as 4@xmath8 and ca 3@xmath10 states are non - zero due to covalency between these states with fe 3@xmath10 states . the contribution from ca 4@xmath2 states , however , is almost negligible ( @xmath13 0.2% ) in the vicinity of @xmath9 . . ] at ambient pressure , cafe@xmath0as@xmath0 undergoes a magneto - structural transition from paramagnetic tetragonal to antiferromagnetic orthorhombic phase at 170 k. in fig . [ fig4-strcomp ] , we show the calculated dos for tetragonal and orthorhombic structures to investigate the effect in the electronic structure due to the above mentioned transitions . while the dos at @xmath9 in the tetragonal structure is large and has a decreasing slope with the increase in energy , a _ pseudogap _ like feature appears in the magnetically ordered phase as shown in fig . [ fig4-strcomp](b ) . @xmath9 is situated at the middle of the _ pseudogap _ exhibiting particle - hole symmetry in its close proximity . the position and width of the pseudogap observed here matches well with the lda+dmft results at ambient pressure@xcite ( here , lda stands for local density approximation and dmft for dynamical mean field theory ) . while there is no apparent change in overall fe 3@xmath10 bandwidth due to the transitions , significant change is observed in the details , such as the shift of the fe 3@xmath10 spectral weight away from @xmath9 leading to significant broadening of the sharp feature above @xmath9 . a new feature appear around -2 ev energy . the center of mass of the as 4@xmath8 contributions shifts to lower energy along with an enhanced contribution near @xmath9 indicating a stronger mixing of the fe 3@xmath10 and as 4@xmath8 electronic states . such enhanced mixing / covalency between fe 3@xmath10 and as 4@xmath8 states might be the driving force for the magnetic and structural transitions in the ground state . orbitals in tetragonal ( t)-phase ( solid lines ) and collapsed tetragonal ( ct ) phase ( dashed lines ) . ( a ) 3@xmath21 + 3@xmath22 and 3@xmath14 are shown by thin black line and thick red line , respectively . ( b ) 3@xmath23 and 3@xmath24 states are shown by thick maroon line and thin blue line , respectively . ] in fig . [ fig5tct - fe3d ] , we superimpose the fe 3@xmath10 partial density of states ( pdos ) for collapsed tetragonal and tetragonal phases to have a comparative study . fe atoms in tetragonal structure experience tetrahedral crystal field created by the as atoms . a regular tetrahedral crystal field splits the fe 3@xmath10 levels into a doubly degenerate @xmath25 band and a triply degenerate @xmath26 band . the distortions in the crystal structure of cafe@xmath0as@xmath0 leads to a deviation from the regular tetrahedral field and the degeneracy of the crystal field split bands are lifted with @xmath14 band slightly shifted above the doubly degenerate @xmath21 & @xmath22 bands . the contributing energy range and width derived by the hopping interaction terms remain very similar for all these bands as manifested in the figure . the @xmath26 band consisting of @xmath21 , @xmath22 and @xmath14 orbitals lies close to @xmath9 and are the major contributors to the fermi surface . the orbitals @xmath24 and @xmath23 possessing @xmath25 symmetry are situated far below the fermi level . in the ct phase , the as - fe - as angle (= 106.13@xmath27 ) deviates from the regular tetrahedral angle at ambient condition ( @xmath18 @xmath28 ) and gives rise to additional distortions in the feas@xmath29 tetrahedra that enhances further the crystal filed splitting @xcite . in fig . [ fig5tct - fe3d](a ) and ( b ) , it appears that all the energy bands in ct phase represented by dashed lines are shifted towards lower energies relative to those in t phase . the energy bandwidth of @xmath21 + @xmath22 bands increases slightly in ct phase compared to that in t phase while the other @xmath26 and @xmath25 bandwidths remain unchanged . this implies that in ct phase , the hybridization between fe - as along @xmath30 direction enhances as expected due to the compression along @xmath5-axis ; the in - plane orbitals remain almost unaffected . in t phase , major contribution at @xmath9 comes from ( @xmath21 + @xmath22 ) and @xmath14 electronic states . the @xmath14 band shifts below @xmath9 in ct phase and thus , contribution at @xmath9 in ct phase arises primarily from degenerate ( @xmath21 + @xmath22 ) orbitals . evidently , the carrier density at @xmath9 is significantly less in the ct phase due to the shift of the energy bands , which is in contrast to the transport data indicating a decrease in electrical resistivity in the ct phase compared to the tetragonal phase @xcite . while more studies are required to understand this phenomena , the decrease in resistivity may be attributed to the enhancement of the mobility of charge carriers ; the increase in hybridization of fe ( @xmath21 + @xmath22 ) and as 4@xmath8 states facilitates higher degree of the hopping of charge carriers . the compression along @xmath5 axis in the ct phase influences the hybridization among as 4@xmath8 states too @xcite leading to notable changes in their density of states . in fig . [ fig6tct - as4p](a ) and ( b ) , we show the pdos corresponding to as 4@xmath31 and the degenerate as ( 4@xmath15 + 4@xmath16 ) energy bands . the results corresponding to the orthorhombic phase shown in the figure were calculated without considering magnetic order to study the effect on electronic structure due to the change in structural parameters only . from the figure , it is clear that all the as 4@xmath8 bands in o and t phases are essentially identical while the @xmath8 bands in ct phase exhibit a shift towards lower energies relative to those in t and o phases - a trend similar to the findings in fe 3@xmath10 cases . in addition , there are several subtle changes observed for various @xmath8 orbitals . for example , the overall width of the as 4@xmath31 is somewhat larger in the ct phase compared to t phase due to enhancement of dos near the band edges suggesting an effective increase in hybridization among the @xmath31 orbitals . the energy distribution of the density of states corresponding to the @xmath15 and @xmath16 electronic states exhibit spectral weight shift near the top of the as 4@xmath8 bands ( see around -2 ev in the figure ) . the unoccupied part appears to be close to each other in all the three cases studied . the changes in the bonding bands between -6 to -2 ev indicate signature of enhancement of covalency in the system in the ct phase . the results for the ca 3@xmath10 and 4@xmath2 pdos is shown in the fig . [ fig6tct - as4p](c ) and ( d ) , respectively . ca 3@xmath10 states contribute primarily above @xmath9 ( energy range 3 to 6 ev ) with small contributions at and below @xmath9 . the intensity of ca 4@xmath2 states is distributed over larger energy range . while the major contribution of 4@xmath2 states appear between 3 to 8 ev above @xmath9 , there is finite contribution below -3 ev energy . interestingly , the contributions below @xmath9 exhibit a shift towards lower energies in the ct phase as found in fe and as cases keeping the pdos above @xmath9 within almost the same energy range in all the structural phases . in the crystal structure , the ca layers are sandwiched between two as layers and the changes in ca states below @xmath9 can be attributed to the hybridization of ca 4@xmath2 states with the electronic states associated to as in the valence band . the compression @xcite along @xmath5-axis due to the transition to the ct phase reduces the separation between the ca and as atomic layers and the hybridization between ca 4@xmath2 and as 4@xmath8 will be enhanced . it is clear that although the ca contribution in the valence band regime is small , it is difficult to ignore the role of ca in the structural changes and their electronic structure as well @xcite . almost all the high temperature ( unconventional ) superconductors exhibit effective two dimensional electronic structure , where the conduction sheet is separated by an insulating layer , often called _ reservoir _ layer . it is believed that the cations with @xmath2 electrons in the charge reservoir layer plays an important role in the pairing interactions , renormalizing disorder induced effects , etc @xcite . in the cuprate superconductors , elements like la , hg , pb , bi form the charge reservoir layers and the cu - o layers form the two dimensional conduction layers . in the present case , ca layer sandwiched between conducting feas layers appears to play similar role and hence enhanced interaction of ca 4@xmath2 states in the ct phase appears to have great implication for the onset of superconductivity in these systems . partial density of states in tetragonal ( t ; thin solid line ) and collapsed tetragonal structures at pressures , @xmath7 = 0.47 gpa ( ct@xmath32 ; dashed green line ) and @xmath7 = 0.63 gpa ( ct@xmath33 ; thick solid line ) in ( a ) non - magnetic and ( b ) magnetically ordered phases . the inset in ( b ) shows the total fe 3@xmath10 pdos at different structures . the definition of lines is kept similar to the main part of the figure.,scaledwidth=60.0% ] the results for ct phase discussed so far correspond to the lattice constants found experimentally for the pressure of 0.63 gpa . now , we study the density of states at different pressures utilizing the structural parameters obtained experimentally . the dos calculated for tetragonal , collapsed tetragonal phases at @xmath7 = 0.47 gpa ( ct@xmath32 ) and @xmath7 = 0.63 gpa ( ct@xmath33 ) are superimposed in fig . [ fig7-pres ] . the fe 3@xmath10 pdos in the non - magnetic phase shown in fig . [ fig7-pres](a ) exhibit significant intensity at @xmath9 in the tetragonal phase . with the application of pressure , fe 3@xmath10 pdos shift to lower energies reducing the intensity at @xmath9 to almost zero . interestingly , the results for ct@xmath32 and ct@xmath33 appear identical indicating less dependence of the non magnetic phase on pressure once the _ pseudogap _ like feature at @xmath9 is present in the electronic structure . the overall bandwidth appears to be similar in every case . the magnetic solution shown in fig . [ fig7-pres](b ) exhibits large separation between the up and down spin bands in the tetragonal phase with an exchange splitting of the order of 2 ev . with the application of pressure , up and down spin fe 3@xmath10 pdos shift towards each other reducing the exchange splitting with the increase in pressure . this change , however , is not monotonic ; the data corresponding to 0.47 gpa shows much larger shift and smaller exchange splitting compared the case for @xmath7 = 0.63 gpa . since the bottom of the valence band remains almost same in every case , the change at the top of the band also leads to an increase in bandwidth . thus , the width of the valence band for @xmath7 = 0.47 gpa appears to be the largest . the transition to the magnetically ordered phase in tetragonal structure leads to a large dip ( @xmath34 ) in dos at @xmath9 . interestingly , the dip in the magnetically ordered phase reduces with the application of pressure in ct phase as shown in the inset of fig . [ fig7-pres](b ) , where sum of the up and down spin fe 3@xmath10 contributions are plotted . this indicates poorer degree of stabilization of the magnetic phase in ct structures and the electronic properties at @xmath7 = 0.47 gpa is closer to the proximity of instability in magnetic order relative to the other cases . in order to study the difference between magnetic and non - magnetic solutions in more detail , we superimpose the results of magnetic and non - magnetic calculations for fe 3@xmath10 and as 4@xmath8 pdos in fig . [ fig8-nmmagtetra ] for tetragonal structure fe 3@xmath23 spin - orbitals exhibit two features separated by about 2 ev in the non - magnetic phase . fe 3@xmath24 exhibit similar structures with somewhat larger separation ( @xmath13 2.5 ev ) . both these bands do not contribute at @xmath9 . fe 3@xmath14 and ( 3@xmath21 + 3@xmath22 ) bands possess one intense feature in each case and appear in the vicinity of @xmath9 providing dominant contribution at @xmath9 . in the magnetically ordered phase , each band splits into up and down spin bands due to the exchange interactions . such splitting is found to be about 2 ev in 3@xmath23 and 3@xmath35 bands , while it is about 2.5 ev for all other cases . all the up spin bands appear below @xmath9 and the down spin states having 3@xmath14 , 3@xmath21 and 3@xmath22 symmetries primarily contribute at the fermi surface indicating an essentially half - metallic scenario . as 4@xmath8 states exhibit interesting scenario in magnetically ordered phase . in addition to a small enhancement of as 4@xmath31 contributions at @xmath9 , these states move towards lower energies in the magnetically ordered phase similar to the scenario in fe 3@xmath10 case and exhibit small exchange splitting induced by fe 3@xmath10 moments . in sharp contrast , the contribution of as ( 4@xmath15 + 4@xmath16 ) at @xmath9 in the magnetic phase is almost vanished , the energy bands move in opposite direction ( towards higher energies as shown in fig . [ fig8-nmmagtetra](f ) ) and does not show exchange splitting . absence of exchange splitting may be apprehended due to the absence of hybridization of these states with the fe 3@xmath10 states ( as layers are situated above and below the fe layers ) . however , energy shift in opposite direction is curious . the scenario is somewhat different in ct phase - the results for ct@xmath33 phase is shown in fig . [ fig9-nmmagctp2 ] . while the overall structure of the pdos appear quite similar to the tetragonal case , there are significant differences . for examples , fe 3@xmath23 and 3@xmath14 pdos exhibit exchange splitting of about 1 ev , which is much smaller than that in t phase . the behavior of 3@xmath35 and 3@xmath21 + 3@xmath22 pdos , however , appear similar to the tetragonal case . interestingly , the contrasting behavior of as 4@xmath31 and ( 4@xmath15 + 4@xmath16 ) bands is not present in this case - @xmath36 the bands show small exchange splitting and an energy shift towards lower energies . a compression along the @xmath5 axis in the ct phase brings fe and as atomic layers closer to each other and hence , the electronic states within as layer ( @xmath37-plane ) also experiences fe 3@xmath10 moment induced interactions . here , the contribution of up spin ( 4@xmath15 + 4@xmath16 ) bands at @xmath9 becomes finite , while it was absent in the non - magnetic case . it is to note here that the down spin contributions corresponding to 3@xmath14 and 3@xmath21 + 3@xmath22 states are essentially unoccupied with @xmath9 pinning at the bottom of the conduction band . however , these bands are significantly occupied in ct phase with @xmath9 appearing close to the peak of the bands . ( @xmath13 0.353 ) ; for better clarity in the demonstration of the fermi sheets , the individual fermi sheets are shown in ( b ) -(e ) . ] (= 0.353 ) . in the primitive tetragonal unit cell , the @xmath38 point corresponds to the @xmath39 point of body centered tetragonal unit cell . ] in order to understand the fermiology and its link to different crystal structures , the fermi surfaces ( fs ) of cafe@xmath0as@xmath0 are investigated below . fig.[fig10_tetra_rlx_fs ] depicts the fermi surface in the tetragonal structure exhibiting quasi 2@xmath40 symmetry with the cylindrical axis as @xmath41-axis . in these calculations , @xmath6 is to optimize the structural parameters . it consists of four distinct fermi sheets - two of them are wrapped around the @xmath41 axis at the brillouin zone center , @xmath1 and represent the hole states . two fermi sheets centered at the corner of the brillouin zone are the electron pockets . the spin density wave type magnetic order observed in cafe@xmath0as@xmath0 is associated to the fermi surface nesting ; a hole pocket centered around @xmath1-point is nested with an electron pocket around @xmath38-point . the dimensionality of the fermi surface plays an important role in fs nesting ; the 2@xmath40 behavior enhances the probability of nesting because parallel hole and electron sheets results in particle hole excitation and manifests as a fermi surface nesting . the nesting wave vector @xmath42 along @xmath43 direction in the primitive tetragonal unit cell is shown in fig . [ fig11_fs_2d ] . since , the nesting vector @xmath42 is almost equal for both the hole sheet , both contribute in magnetic ordering . fixed to its experimentally observed value of 0.372 . ( b ) -(f ) shows the individual fermi sheets . ] the dimensionality and the topology of the fermi surface depends on @xmath6 that is linked to the _ pnictogen height_. the fermi surface calculated with the experimental parameters ( without employing relaxation in the structure ) is shown in fig . [ fig12_tetra_fs ] . in this case , the fermi surface consists of five sheets ; the additional hole sheet appears at the zone center due to lifting of the degeneracy of one of the hole fermi sheets in the relaxed structure . one of the electron sheet spreads around the perimeter of the brillouin zone , the other remains unchanged . the change in the shape of the fermi sheets at the zone corners and the zone center leads to the disappearance of the nesting in one of the hole sheets . from various calculations , we observe that the electronic structure corresponding to the experimentally found crystal structure is significantly different from the one obtained for the optimised structure within the model calculation via relaxing as positions . the relaxed structure favors smaller @xmath6 value and hence , smaller _ pnictogen height _ compared to the experimentally observed values and larger degree of fermi surface nesting , thereby enhancing the degree of magnetic order . the real material , however , forms with relatively larger pnictogen height and one nesting vector as also observed in arpes measurements of various materials in similar class . one reason for such difference could be related to the underestimation of correlation induced effects in the density functional calculations - the system prefers to reduce the hybridization and hence itineracy via increasing the pnictogen height . the other possibility could be a proximity of the ground state crystal structure of cafe@xmath0as@xmath0 to a metastable state and thus , the system gets arrested to different strained crystal structure during sample preparation instead of going to the ground state configuration . the finding of the change of volume fraction of different structures and stains in srfe@xmath0as@xmath0 with preparation conditions @xcite could be a realization of this conjecture . the experimental findings of contrasting behavior in cafe@xmath0as@xmath0 is also in line with this view . as@xmath0 in ( a ) tetragonal , ( b ) orthorhombic ( c ) collapsed tetragonal ( ct phase at @xmath7 = 0.67 gpa ) and ( d ) antiferromagnetic orthorhombic structures . ( e ) and ( f ) show the brillouin zone of the body centered tetragonal and tetragonal structures , respectively . ] in fig . [ fig13_fs ] , we investigate the fermi surfaces in all the structural phases realized in cafe@xmath0as@xmath0 by varying parameters like temperature , pressure etc . in the o phase , without any magnetic order , the fermi surface looks very similar to the t phase . the difference in the definition of @xmath44 vectors appears due to their brillouin zone orientation . it is evident from the comparison of figs . [ fig13_fs](a ) and [ fig13_fs](b ) that the fermi sheets in the orthorhombic phase are shrinked compared to those in the tetragonal phase as evident from the larger separation between electron and hole sheets . the orthorhombic phase emerges at low temperature presumably due to thermal compression leading to a small distortion of the lattice in the @xmath37 plane and small ( @xmath13 5% ) compression in the @xmath5 axis . thus , the pnictogen height becomes smaller as reflected in the fermiology . fermi sheets in the o phase with afm ordering are shown in fig . [ fig13_fs](d ) . here , the fermi sheets look three dimensional and very different from all other cases due to the opening of energy gap at @xmath9 induced by the nesting of the electron sheets at the zone corner with the hole sheets at zone center . notable difference appears in the ct structure ; the fermi surface shown here correspond to the structural parameters at @xmath7 = 0.67 gpa with @xmath6 value slightly smaller than the case shown in fig . [ fig10_tetra_rlx_fs ] . we chose these parameters corresponding to slightly higher pressure than the previous cases to demonstrate the changes in fermi surface topology with greater clarity . the 2@xmath40 nature of the hole fermi surfaces around the @xmath30-axis is lost completely in the results for the ct phase . here , all the three hole pockets around the @xmath1-point disappeared . only one hole sheet survives around the @xmath45 point making it a closed one with larger radius . the electron sheets are not affected significantly . such drastic change in the hole fermi surface around @xmath41-axis rules out the electron - hole nesting responsible for magnetic long range order and is consistent with the experimental finding of absence of magnetic order in this phase . as@xmath0 in the tetragonal structure highlighting ( a ) fe 3@xmath10 and ( b ) as 4@xmath8 character ; the thickness of lines represents the weight of the character . as 4@xmath8 contributions are rescaled by a factor of 2 to enhance the visibility of the contributions . ] we now turn to the investigation of the properties of the energy bands . while the density of states discussed above provide information about the contribution of various electronic states as a function of energy , these results will help to understand the @xmath44-resolved information , which are often helpful to understand the electronic properties and can also be observed in angle resolved photoemission measurements . in fig . [ fig14_bonding ] , we show the energy bands of cafe@xmath0as@xmath0 for the tetragonal structure ; the thickness of the line represents the orbital contributions . it is evident that the fe 3@xmath10 symmetry adapted bonding bands appearing between -6 to -2 ev possess dominant as 4@xmath8 character . the bands close to the fermi level are primarily constituted by fe 3@xmath10 states with notable contributions from as 4@xmath8 states . interestingly , while the bands around @xmath1 point forming the hole pocket in the fermi surface possess essentially fe 3@xmath10 character , one of the electron pocket around @xmath39-point possess large as 4@xmath8 character . the detailed orbital symmetries near @xmath9 are shown in fig . [ fig15_t_band_char ] . at @xmath1 , three energy bands are observed to cross @xmath9 - these bands are denoted by @xmath46 , @xmath47 and @xmath48 as shown separately in fig . [ fig15_t_band_char](g ) for better clarity . these three energy bands create three hole pocket around @xmath1 @xcite . there are two electron pockets around the high symmetry point , @xmath39 . it is clear from fig . [ fig15_t_band_char](c ) and fig . [ fig15_t_band_char](d ) that the @xmath48 band possesses mainly @xmath14 character and the @xmath46 & @xmath47 bands have contribution from degenerate @xmath21 and @xmath22 orbitals @xcite . these three bands possess @xmath26 symmetry . the @xmath25 bands constituted by @xmath24 and @xmath23 spin - orbitals appear far away from the fermi level and hence do not play significant role in the electronic properties of the system . overall , the bands near the fermi level has very less as 4@xmath8 character except one electron pocket at @xmath39 , which has as 4@xmath31 character . this is in line with the expectation from the fact that as layers appear above and below fe layers and hence @xmath31 orbital is expected to play a major role in the fe 3@xmath10-as 4@xmath8 hybridization . in fig . [ fig16_band_xgz ] , we show the change in band structure in different phases along @xmath49 symmetry lines . in collapsed tetragonal phase , the bands at @xmath1 shifts below the fermi level and the hole pockets around @xmath1 disappear . the nonmagnetic calculation of the orthorhombic structure gives rise to similar band structure like tetragonal phase , but if we consider the antiferromagnetic configuration then the band structure becomes complex . the hole pocket at @xmath1 nesting with an electron pocket at @xmath39 gives rise to an energy gap at @xmath9 via folding of the energy bands at the zone boundary created by the magnetic supercell . from the above results , it is clear that electronic structure of cafe@xmath0as@xmath0 is very sensitive to the crystal structure and _ pnictogen height _ is an important parameter in deriving the ground state properties . the system appears to favor smaller _ pnictogen height _ although the experiments found structures with somewhat higher values and significant dependence of the structure on preparation conditions . on application of pressure , the tetragonal symmetry gets arrested till the low temperatures and the electronic properties become significantly different from magnetically ordered phase observed in ambient conditions . in ct phase , the @xmath5-axis collapse reduces fe moment and fe - as hybridization along @xmath30 direction enhances @xcite . in fact , the change in fe - as hybridization and consequent change in the magnetic moment of fe turn out to be the key factors in deciding the magnetic and structural properties of cafe@xmath0as@xmath0 . to get a clear idea about this we have calculated the local magnetic moment at fe - sites . the calculation are done for tetragonal phase and collapsed tetragonal phase . in the later case , we choose the lattice parameters corresponding to two different applied pressures . the phase realized under 0.47 gpa pressure is termed as ct@xmath32 and the ct phase with @xmath7 = 0.63 gpa is called ct@xmath33 . experimentally , the magnetization in the ct phase has been studied extensively and no signature of long range magnetism was found . there are controversy on the magnetic moment though - while ma long _ et al . _ @xcite found signature of quenching of fe moment on transition to ct phase induced by pr substitution at ca sites , magnetic susceptibility data by w. wu _ et al . _ @xcite indicate finite moment of fe in the ct phase though depleted from its value in ambient tetragonal phase . moreover , the fe magnetic moment appears to be sample dependent . we calculated fe moments assuming parallel alignment of fe - moments and compare the results with their values corresponding to the non magnetic phase in order to investigate the trend of the change of converged energies with the change in external parameters , which is often captured well employing dft methodologies @xcite . the table below gives the list of lattice parameters used for the calculations along with the calculated magnetic moment and ground state energies . .the magnetic moment of fe and ground state energy at various structural configuration of cafe@xmath0as@xmath0 , the experimental latice parameters are taken from @xcite [ cols="^,^,^,^,^,^,^,^,^,^ " , ] in the tetragonal structure , the converged energy for the magnetically ordered configuration is about 150 mev lower than that for the nonmagnetic ( nm ) configuration suggesting a preference for the magnetic order in the ground state . the magnetic moment of fe atoms is found to be 2.37 @xmath50 in this structure . the ct structure manifests a contrasting scenario exhibiting significantly higher energy for magnetically ordered phase relative to the nm phase . application of pressure leads to a gradual reduction of the lattice constant , @xmath5 and as - as bondlength ( @xmath51 ) . the magnitude of the lattice constant , @xmath4 and @xmath6 exhibit anomalous scenario with the largest and smallest values at @xmath7 = 0.47 gpa , respectively . with the application of pressure , the nonmagnetic phase appears to be favored more . the ground state energy of the non - magnetic phase at ct@xmath32 structure is found to be the lowest relative to all the calculations we have done . the fe moment appears to diminish with the application of pressure . interestingly , the lowest moment of 1.95 @xmath50 is found for the pressure of 0.47 gpa ( ct@xmath32 phase ) and then the fe moment increases with further increase in pressure although the compression of @xmath5 is more at ct@xmath33 phase compared to ct@xmath32 phase . the value of @xmath6 , however , is smaller in the ct@xmath32 phase indicating a direct link of the fe - moment to @xmath6 value . it is still an open question whether the magnetic moment at fe sites actually vanishes or there is a small magnetic moment on fe atom , but they do nt give any long range order @xcite . all our calculations indicate that the magnetic configuration in the ct structure has higher energy compared to the non - magnetic phase . the magnetic moment decreases as the fe - as bond length decreases . this can be understood considering the fact that decrease in fe - as bondlength makes the conduction electrons more itinerant and hence less local moment . it is puzzling that the fe - as bondlength is smallest in ct@xmath32 structure and a small decrease in fe - as bond length from 2.3405 in ct@xmath33 structure to 2.338 in ct@xmath32 structure changes the magnetic moment from @xmath52 to @xmath53 . the above results establish that the fe spin state is highly sensitive to the fe - as hybridization , which is linked to the _ pnictogen height _ too . the shorter as - as distance leads to increased hybridization within the as - plane and higher degree of as @xmath8 electron itineracy . however , the local character of the fe @xmath10 electrons remain less affected . on the other hand , higher fe - as hybridization leads to larger covalent splitting of the fe - as bonding and antibonding bands leading to a shift of the valence bands . some of the bands shifts below @xmath9 and become completely filled . the energy bands crossing the fermi level will have higher itineracy and hence less local character , which is presumably one of the reasons for the reduction of fe moment and the resistivity . the calculated fe - as bond length in the afm ground state with orthorhombic structure is 2.3654 , which is larger than that in ct phase and shorter than the one in t phase . the fe spin moment is found to be 1.87 @xmath50 , which is much larger than the experimental value of 0.80 @xmath50 . various theoretical studies tried to capture the experimental moment by varying @xmath6 ( experimental value = 0.3664 ) . it is observed that the spin moment becomes closer to the experimental value for the @xmath6 values of 0.3567 @xcite . it is clear that @xmath6 and hence the _ pnictogen height _ is an important parameter leading to varied electronic properties in this system . moreover , these results establish that the ct phase possesses finite magnetic moment as also found experimentally in magnetic susceptibility measurements for almost all kinds of phases prepared so far . the absence of long range order appears to be due to its proximity to the critical point of loosing fermi surface nesting & no magnetic order @xcite . pdos ( area plot ) is compared with as 4@xmath8 ( thick solid line ) and fe 3@xmath10 thin solid line ) for ( a ) orthorhombic , ( b ) tetragonal and ( c ) collapsed tetragonal structures . the as 4@xmath8 contributions are rescaled by 4 times for better clarity . ] interestingly , coupling of ca 4@xmath2 states to the conduction layer becomes stronger in the ct phase as discussed in fig . [ fig6tct - as4p ] and redrawn here to emphasize the scenario . in fig . [ fig17_chreserv ] , we show the ca 4@xmath2 pdos along with fe 3@xmath10 and as 4@xmath8 contributions for different structures . ca 4@xmath2 contributions in orthorhombic and tetragonal phases look almost similar with major intensities above @xmath9 , a reasonably intense feature between -6 to -3 ev and almost no intensity in the vicinity of @xmath9 . while the intensity above @xmath9 remains dominant in the ct phase too like other cases , there is significant redistribution of pdos in the lower energy regime . most importantly the contribution in the vicinity of @xmath9 becomes almost double of the contributions in other two structures . the contributions in the energy range -3 to -6 ev appear due to the fe - as bonding bands with dominant as 4@xmath8 character and -2.5 to 0 ev is contributed by anti - bonding bands with dominant fe 3@xmath10 character . the intense feature between -2.5 to -4 ev in as 4@xmath8 pdos are the non - bonding as 4@xmath8 states . the features in ca 4@xmath2 pdos resemble well with structures in the as 4@xmath8 pdos as these states emerges due to the hybridization of ca 4@xmath2 states with as 4@xmath8 states . thus , the enhancement of 4@xmath2 contributions below @xmath9 can be attributed to enhanced hybridization of ca and as states due to compression of @xmath5 axis in the ct phase . charge reservoir layers , generally insulating in nature , are believed to play an important role in superconductivity @xcite as these layers help to preserve two dimensionality of the system and can protect the conduction layers via screening of substitutional and/or other type of disorder induced effects . effective two dimensional electronic structure seem to be essential to achieve high superconducting transition temperature and almost all the superconductors ( conventional and unconventional ) having high transition temperature possess two dimensional electronic structure . thus , the higher degree of ca - as hybridization in collapsed tetragonal phase and the observation of pressure induced superconductivity @xcite appears to be linked . in summary , we present here the detailed electronic structure of cafe@xmath0as@xmath0 for different crystallographic structures using density functional theory . the calculated density of states indicate strong covalency induced effect in the electronic structure , which increases with the compression of @xmath5 axis observed in collapsed tetragonal phase . while the energy distribution of density of states appear similar in tetragonal and orthorhombic phase in ambient conditions , a significant shift towards lower energies is observed in collapsed tetragonal phase , which reduces the total energy of the system . the hole pockets at the @xmath1 point eventually vanishes at higher pressure ruling out the fermi surface nesting related to the formation of spin density wave states . the electron pockets centered around the high symmetry points @xmath38/@xmath39/@xmath45 evolves significantly in the ct phase . the enhancement of fe - as hybridization in the ct phase presumably leads to higher degree of itineracy relevant for the changes in electrical resistivity in this system . interestingly , the changes in ca 4@xmath2 partial density of states suggest enhancement of ca - as hybridization with the decrease in @xmath5 , which is significant in the context of superconductivity as it is believed that such layers act as charge reservoir layer in various high temperature superconductors . we observe that the changes in the fermiology and energy band shifts due to application of pressure depletes the magnetic moment at fe - sites although they do not show signature of magnetic order indicating the proximity of such strained condition to the quantum fluctuations . electronic structure calculations were carried out using full potential linearized augmented plane wave method ( flapw ) captured in the wien2k software @xcite . convergence was achieved by fixing the energy convergence criteria to 0.0001 rydberg ( 1 mev ) . for nonmagnetic calculations , we have used 10@xmath5410@xmath5410 @xmath44-points in the brillouin zone and for fermi surface calculations 39@xmath5439@xmath5410 @xmath44-points were used . we have used the perdew - burke - ernzerhof generalized gradient approximation ( gga)@xcite for our density functional theoretical calculation @xcite . the fermi surfaces were calculated using xcrysden@xcite . the collapsed tetragonal phase of cafe@xmath0as@xmath0 achieved via application of pressure / varying annealing conditions , possesses the same space group but a reduced @xmath5 axis and slightly increased @xmath4 axis . the volume of the unit cell reduces effectively . the lattice parameters at @xmath7 = 0.63 gpa are @xmath4 = 3.9780(1) , @xmath5 = 10.6073(7) , and @xmath6 = 0.372(1)@xcite . orthorhombic structure of cafe@xmath0as@xmath0 appearing at low temperature and ambient pressure has @xmath55 space group with the lattice parameters @xmath4 = 5.506(2) , @xmath17 = 5.450(2) , @xmath5 = 11.664(6) , and @xmath6 = 0.36642(5)@xcite . for antiferromagnetic calculations , we used 10@xmath5410@xmath5410 @xmath44-points in the brillouin zone and for fermi surface 23@xmath5423@xmath5410 @xmath44-points . 99 ni , n. _ et al . _ . b _ * 78 , * 014523 ( 2008 ) . tompsett , d. a. & lonzarich , g. g. dimensionality and magnetic interactions in cafe@xmath56as@xmath56 : an ab initio study . _ physica b : condensed matter _ * 405 , * 2440 - 2443 ( 2010 ) . kreyssig , a. _ et al . _ pressure - induced volume - collapsed tetragonal phase of cafe@xmath56as@xmath56 as seen via neutron scattering . b _ * 78 , * 184517 ( 2008 ) . pratt , d. k. _ et al . _ suppression of antiferromagnetic spin fluctuations in the collapsed phase of cafe@xmath56as@xmath56 . b _ * 79 , * 060510 ( 2009 ) . kumar , m. _ et al . _ effect of pressure on the magnetostructural transition in srfe@xmath56as@xmath56 . b _ * 78 , * 184516 ( 2008 ) . goldman , a. i. _ et al . _ lattice and magnetic instabilities in cafe@xmath56as@xmath56 : a single - crystal neutron diffraction study . b _ * 78 , * 100506(r ) ( 2008 ) . torikachvili , m. _ et al . _ pressure induced superconductivity in cafe@xmath56as@xmath56 . lett . _ * 101 , * 057006 ( 2008 ) . saha , s. r. _ et al . _ structural collapse and superconductivity in rare - earth - doped cafe@xmath57as@xmath57 . b _ * 85 , * 024525 ( 2012 ) . yildirim , t. strong coupling of the fe - spin state and the as - as hybridization in iron - pnictide superconductors from first - principle calculations . lett . _ * 102 , * 037003 ( 2009 ) . adhikary g. _ et al . _ electronic structure of eufe@xmath0as@xmath0 . _ j. phys . : condens . matter _ * 25 , * 225701 ( 2013 ) . colonna , n. & profeta , g. & continenza , a. & massidda , s. structural and magnetic properties of cafe@xmath57as@xmath57 and bafe@xmath57as@xmath57 from first - principles density functional theory . b _ * 83 , * 094529 ( 2011 ) . zhang , yu - zhong _ et al . _ microscopic origin of pressure - induced phase transitions in the iron pnictide superconductors afe@xmath56as@xmath56 : an _ ab initio _ molecular dynamics study . b _ * 80 , * 094530 ( 2009 ) . liu c. _ et al._. three- to two - dimensional transition of the electronic structure in cafe@xmath0as@xmath0 : a parent compound for an iron arsenic high - temperature superconductor . lett . _ * 102 * , 167004 ( 2009 ) . adhikary g. _ et al . _ complex temperature evolution of the electronic structure of cafe@xmath0as@xmath0 . _ j. appl . phys . _ * 115 , * 123901 ( 2014 ) . diehl , j. & backes , s. & guterding , d. & jeschke , harald o. & valent , r. correlation effects in the tetragonal and collapsed - tetragonal phase of @xmath58 . _ b _ * 90 , * 085110 ( 2014 ) . kurmaev , e. z. _ et al . _ contribution of fe 3_d _ states to the fermi level of cafe@xmath59 . b _ * 80 , * 054508 ( 2009 ) . adhikary g. _ et al . _ importance of ligands in the electronic properties of fete@xmath60se@xmath61 . _ j. appl . phys . _ * 114 , * 163906 ( 2013 ) . d. v. evtushinsky , d. v. _ et al . _ direct observation of dispersive lower hubbard band in iron - based superconductor fese . _ _ arxiv:__1612.02313 aichhorn , m. , biermann , s. , miyake , t. , georges , a. & imada , m. theoretical evidence for strong correlations and incoherent metallic state in fese . b _ * 82 * , 064504 ( 2010 ) sanna , a. & profeta , g. & massidda , s. & gross , e. k. u. first - principles study of rare - earth - doped superconducting cafe@xmath57as@xmath57 . b _ * 86 , * 014507 ( 2012 ) . kashinathan , d. _ et al . _ afe@xmath0@xmath62 ( a = ca , sr , ba , eu ) and srfe@xmath63tm@xmath64as@xmath65 ( tm = mn , co , ni ) : crystal structure , charge doping , magnetism and superconductivity . _ new j. phys . _ * 11 , * 025023 ( 2009 ) . widom , m. & quader , k. first - principles study of cafe@xmath57as@xmath57 under pressure . b _ * 88 , * 045117 ( 2013 ) . tomi , m. & valent , r. & jeschke , harald o. uniaxial versus hydrostatic pressure - induced phase transitions in cafe@xmath57as@xmath57 and bafe@xmath57as@xmath57 . b _ * 85 , * 094105 ( 2012 ) . park , t. _ et al . _ pressure - induced superconductivity in cafe@xmath0as@xmath0 . _ j. phys . : condens * 20 , * 322204 ( 2008 ) . goldman , a. i. _ et al . _ lattice collapse and quenching of magnetism in cafe@xmath56as@xmath56 under pressure : a single - crystal neutron and x - ray diffraction investigation . b _ * 79 , * 024513 ( 2009 ) . alireza , p. l. _ et al . _ superconductivity up to 29 k in srfe@xmath0as@xmath0 and bafe@xmath0as@xmath0 at high pressures . _ j. phys . : condens . matter _ * 21 , * 012208 ( 2009 ) . mani , a. _ et al . _ pressure - induced superconductivity in bafe@xmath0as@xmath0 single crystal . _ europhys . letts . _ * 87 , * 17004 ( 2009 ) . kotegawa , h. _ et al . _ effect of uniaxial stress for pressure - induced superconductor srfe@xmath0as@xmath0 . _ j. phys . jpn . _ * 78 , * 083702 ( 2009 ) . colombier , e. & budko , s. l. & ni , n. & canfield , p. c. complete pressure - dependent phase diagrams for srfe@xmath56as@xmath56 and bafe@xmath56as@xmath56 . b _ * 79 , * 224518 ( 2009 ) . kotegawa , h. _ et al . _ abrupt emergence of pressure - induced superconductivity of 34 k in srfe@xmath0as@xmath0 : a resistivity study under pressure . _ j. phys . jpn . _ * 78 , * 013709 ( 2009 ) . igawa , k. _ et al . _ pressure - induced superconductivity in iron pnictide compound srfe@xmath0as@xmath0 . _ j. phys . soc . jpn . _ * 78 , * 025001 ( 2009 ) . kumar , m. _ et al . _ effect of pressure on the magnetostructural transition in srfe@xmath56as@xmath56 . b _ * 78 , * 184516 ( 2008 ) . torikachvili , m. _ et al . _ effect of pressure on transport and magnetotransport properties in cafe@xmath56as@xmath56 single crystals . b _ * 80 , * 014521 ( 2009 ) . lee , h. _ et al . _ pressure - induced superconducting state of antiferromagnetic cafe@xmath56as@xmath56 . b _ * 80 , * 024519 ( 2009 ) . baek , s .- h . _ et al . _ nmr investigation of superconductivity and antiferromagnetism in @xmath58 under pressure . lett . _ * 102 , * 227601 ( 2009 ) . kawasaki , s. _ et al . _ pressure - induced unconventional superconductivity near a quantum critical point in cafe@xmath0as@xmath0 . _ superconductor science and technology _ * 23 , * 054004 ( 2010 ) . goko , t. _ et al . _ superconducting state coexisting with a phase - separated static magnetic order in ( ba , k)fe@xmath56as@xmath56 , ( sr , na)fe@xmath56as@xmath56 , and cafe@xmath56as@xmath56 . b _ * 80 , * 024508 ( 2009 ) . proke , k. _ et al . _ evidence from neutron diffraction for superconductivity in the stabilized tetragonal phase of cafe@xmath56as@xmath56 under uniaxial pressure . _ phys . rev . b _ * 81 , * 180506 ( 2010 ) . yu , w. _ et al . _ absence of superconductivity in single - phase cafe@xmath56as@xmath56 under hydrostatic pressure . b _ * 79 , * 020511(r ) ( 2009 ) . pavarini , e. _ et al . _ mott transition and suppression of orbital fluctuations in orthorhombic 3@xmath66 perovskites . _ * 92 * , 176403 ( 2004 ) . maiti , k. role of covalency in the ground - state properties of perovskite ruthenates : a first - principles study using local spin density approximations . b _ * 73 , * 235110 ( 2006 ) . maiti , k. electronic structure of bairo@xmath67 : a first - principles study using the local spin density approximation . b _ * 73 * , 115119 ( 2006 ) , z. p. & haule , k. & kotliar , g. kinetic frustration and the nature of the magnetic and paramagnetic states in iron pnictides and iron chalcogenides . _ nature materials _ * 10 , * 932 - 935 ( 2011 ) . , b. _ et al . _ complex structures of different cafe@xmath56as@xmath56 samples . _ scintific reports ( nature ) _ * 4 , * 4120 ; 10.1038/srep04120 ( 2014 ) . roekeghem , v. _ et al . _ tetragonal and collapsed - tetragonal phases of @xmath58 : a view from angle - resolved photoemission and dynamical mean - field theory . b _ * 93 , * 245139 ( 2016 ) . , r. s. & maiti , k. manifestation of screening effects and a - o covalency in the core level spectra of a site elements in the abo@xmath67 structure of ca@xmath68sr@xmath69ruo@xmath67 _ phys b _ * 76 , * 085102 ( 2007 ) . , k and singh , r. s. complexity in the core level spectra of ca@xmath68sr@xmath69ruo@xmath67 _ sci . lett . _ * 3 , * 55 ( 2014 ) . , t. h. & mayzhes , b. y. & dickinson , p. h. reservoir layers in high tc mercury cuprates . _ arxiv _ ; cond - mat/9904435 ( 1999 ) . , t. h. , koster , g. what t@xmath70 can teach about superconductivity . 325 - 344 , ( _ handbook of high temperrature superconductivity _ , springer new york , 2007 ) , s. r. , butch , n. p. , kirshenbaum , k. , paglione , j. & zavalij , p. y. superconducting and ferromagnetic phases induced by lattice distortions in stoichiometric srfe@xmath0as@xmath0 single crystals . _ * 103 * , 037005 ( 2009 ) . , c. _ et al . _ three- to two - dimensional transition of the electronic structure in cafe@xmath0as@xmath0 : a parent compound for an iron arsenic high - temperature superconductor . lett_. * 102 * , 167004 ( 2009 ) . long , m. _ et al . _ quenched fe moment in the collapsed tetragonal phase of ca@xmath71pr@xmath69fe@xmath0as@xmath0 . _ chin . phys . b _ * 22 , * 057401 ( 2013 ) . adhikary , g. , bindu , r. , pandey , s. k. & maiti , k. electronic structure near the quantum critical point in v - doped cr a high - resolution photoemission study . _ europhys . _ , * 99 , * 37009 ( 2012 ) . , g. , korneta , o. , chikara , s. , delong , l. e. , & schlottmann , p. non - fermi - liquid behavior in single - crystal caruo@xmath67 : comparison to ferromagnetic srruo@xmath67 . _ solid state communications _ * 148 * , 305 ( 2008 ) . maiti , k. , singh , r. s. & medicherla , v. r. r. observation of particle hole asymmetry and phonon excitations in non - fermi - liquid systems : a high - resolution photoemission study of ruthenates . _ europhys . _ * 78 , * 17002 ( 2007 ) . lv , b. _ et al . _ unusual superconducting state at 49 k in electron - doped cafe@xmath56as@xmath56 at ambient pressure . _ * 108 , * 15705 - 15709 ( 2011 ) . jeffries , j. r. _ et al . _ suppression of magnetism and development of superconductivity within the collapsed tetragonal phase of @xmath72 under pressure . b _ * 85 * , 0184501 ( 2012 ) . kumar , n. _ et al . _ anisotropic magnetic and superconducting properties of cafe@xmath73co@xmath74as@xmath56 @xmath75 single crystals . b _ * 79 , * 012504 ( 2009 ) . kawasaki , s. _ et al . _ pressure - induced unconventional superconductivity near a quantum critical point in cafe@xmath0as@xmath0 . _ superconductor science and technology _ * 23 , * 054004 ( 2010 ) . blaha , p. and schwarz , k. and madsen , g. k. h. and kvasnicka , d. and luitz , j. _ wien2k , an augmented plane wave + local orbitals program for calculating crystal properties_. ( karlheinz schwarz , techn . universitt wien , austria ) perdew , john p. & burke , k. & ernzerhof , m. generalized gradient approximation made simple . lett . _ * 77 , * 3865 - 3868 ( 1996 ) . , p. & kohn , w. inhomogeneous electron gas . rev . _ * 136 , * 864 - 871 ( 1964 ) . kokalj , a. computer graphics and graphical user interfaces as tools in simulations of matter at the atomic scale . _ computational materials science _ * 28 , * 155 - 168 ( 2003 ) . k. m. acknowledges financial assistance from the department of science and technology , govt . of india under j.c . bose fellowship program and department of atomic energy , govt . of india . k.a . carried out all the calculations . k.m . initiated the study and supervised the project . k.a . and k.m . jointly analyzed the data and prepared the manuscript . the authors declare no competing financial interests .

a clique colouring of a graph is a colouring of the vertices so that no maximal clique is monochromatic ( ignoring isolated vertices ) . the smallest number of colours in such a colouring is the clique chromatic number . in this paper , we study the asymptotic behaviour of the clique chromatic number of the random graph for a wide range of edge - probabilities . we see that the typical clique chromatic number , as a function of the average degree , forms an intriguing step function . [ section ] [ theorem]lemma [ theorem]definition [ theorem]conjecture [ theorem]proposition [ theorem]algorithm [ theorem]corollary [ theorem]observation [ theorem]open problem [ theorem]remark

1611.01782

a _ proper colouring _ of a graph is a labeling of its vertices with colours such that no two vertices sharing the same edge have the same colour . the smallest number of colours in a proper colouring of a graph @xmath2 is called its _ chromatic number _ , and it is denoted by @xmath3 . in this paper we are concerned with another notion of vertex colouring . a _ clique _ @xmath4 is a subset of the vertex set such that any pair of vertices in @xmath5 is connected by an edge . moreover , a clique @xmath5 is _ maximal _ if there is no vertex in @xmath6 connected by an edge to every vertex in @xmath5 ( in other words , @xmath5 is not a proper subset of another clique ) . a _ clique colouring _ of a graph @xmath7 is a colouring of the vertices so that no maximal clique is monochromatic , ignoring isolated vertices . the smallest number of colours in such a colouring is called the _ clique chromatic number _ of @xmath7 , denoted by @xmath8 . clearly , @xmath9 but it is possible that @xmath8 is much smaller than @xmath3 . for example , for any @xmath10 we have @xmath11 but @xmath12 . note that if @xmath7 is triangle - free then @xmath13 . the problem has received considerable attention for deterministic graphs : in @xcite it was shown that planar graphs satisfy @xmath14 . in @xcite a necessary and sufficient condition for @xmath15 on line graphs was given . moreover , several graph classes are known to satisfy @xmath16 : claw - free perfect graphs @xcite , co - diamond free graphs @xcite , claw - free planar graphs @xcite , powers of cycles ( other than odd cycles longer than three that need three colours ) @xcite , and also claw - free graphs with maximum degree at most @xmath17 ( except for odd cycles longer than @xmath18 ) @xcite . also , circular - arc graphs are known to have @xmath19 ( see @xcite ) . further results about other classes of graphs can also be found in @xcite , and the clique chromatic number of graphs without having long paths was studied in @xcite . on the algorithmic side , it is known that testing whether @xmath20 for a planar graph can be performed in polynomial time @xcite , but deciding whether @xmath20 is @xmath21-complete for @xmath18-chromatic perfect graphs @xcite and for graphs with maximum degree @xmath18 @xcite . the clique chromatic number for geometric graphs ( in particular , random geometric graphs ) is analysed in the accompanying paper @xcite . let us recall the classic model of random graphs that we study in this paper . the _ binomial random graph _ @xmath0 is the random graph @xmath7 with vertex set @xmath22 $ ] in which every pair @xmath23}{2}$ ] appears independently as an edge in @xmath7 with probability @xmath24 . note that @xmath1 may ( and usually does ) tend to zero as @xmath25 tends to infinity . the behaviour of many colouring problems has been investigated for @xmath0 : the classic chromatic number has been intensively studied , see @xcite and the references therein ; the list chromatic number ( known also as the choice number ) was studied among others in @xcite , and other variants were analysed recently in @xcite and @xcite . all asymptotics throughout are as @xmath26 ( we emphasize that the notations @xmath27 and @xmath28 refer to functions of @xmath25 , not necessarily positive , whose growth is bounded ) . we use the notations @xmath29 for @xmath30 and @xmath31 for @xmath32 . we also write @xmath33 if @xmath34 as @xmath35 ( that is , when @xmath36 ) . we say that events @xmath37 in a probability space hold _ with high probability _ ( or _ whp _ ) , if the probability that @xmath37 holds tends to @xmath38 as @xmath25 goes to infinity . since we aim for results that hold whp , we will always assume that @xmath25 is large enough . we often write @xmath0 when we mean a graph drawn from the distribution @xmath0 . finally , we use @xmath39 to denote natural logarithms . here is our main result . we consider the edge probability @xmath40 ranging from the sparse case when @xmath41 arbitrarily slowly , to the dense case when @xmath42 for an arbitrarily small @xmath43 , and we break this range into 8 parts . after the theorem we give a corollary which is less precise but easier to read . [ thm : gnp ] let @xmath44 be a constant ( arbitrarily small ) . let @xmath45 be a function tending to infinity with @xmath25 ( arbitrarily slowly ) , and suppose that @xmath46 . let @xmath47 for some @xmath1 . then , the following holds whp : * if @xmath48 and @xmath49 , then @xmath50 * if @xmath51 , then @xmath52 and @xmath53 * if @xmath54 , then @xmath55 * if @xmath56 , then @xmath57 * if @xmath58 , then @xmath59 and @xmath60 * if @xmath61 and @xmath62 , then @xmath63 * if @xmath64 , then @xmath59 and @xmath60 * if @xmath65 and @xmath66 , then @xmath67 note that @xmath68 in places a slightly tighter result than the one given in theorem [ thm : gnp ] can be obtained . we are interested in the ratio between the upper and lower bounds on @xmath8 for each @xmath24 , but in order to keep the statement reasonably simple , we spelled out results only for large intervals of @xmath24 . for instance , as cases ( d ) and ( f ) are concerned with @xmath69 and @xmath61 respectively , we treated these cases in the statement of theorem [ thm : gnp ] less fully . for the reader interested in more precise bounds on this ratio , we refer to the proofs in the sections below for more details . in order to understand better the behaviour of the clique chromatic number , let us introduce the following function @xmath70 : @xmath71 this function is depicted in figure [ fig1 ] . we get immediately the following corollary . related to the clique chromatic number of @xmath72,width=192,height=153 ] [ cor.main ] let @xmath70 be defined as above . let @xmath47 for some @xmath1 . if @xmath73 for some @xmath74 then whp @xmath75 . figure [ fig1 ] shows that there are dramatic transitions in corollary [ cor.main ] when @xmath24 is about @xmath76 and @xmath77 , and there is some problem at about @xmath78 in theorem [ thm : gnp ] . what is happening at these points ? here is the rough story . for @xmath79 , whp most edges are not in triangles , and @xmath8 is close to @xmath3 . for @xmath80 , whp each edge is in many triangles , and we may use triangle - free sets as colour classes ; or we may find relatively small dominating sets which yield an upper bound on @xmath8 ( see lemma [ lem : deterministic_upperbound ] ) . for @xmath81 , the first bound ( which increases with @xmath24 ) is stronger , whereas from then on the second bound ( which decreases with @xmath24 ) gives better results . for lower bounds on @xmath8 , depending on the range of @xmath24 , we find a value of @xmath82 , so that whp there are many @xmath82-cliques and most of them are not contained inside @xmath83-cliques . ( in the central range in figure [ fig1 ] , for @xmath84 , we use @xmath85 . ) for @xmath86 whp most triangles are not contained in @xmath87-cliques ; and for @xmath88 whp each triangle is in many @xmath87-cliques . in this section , let us recall a few inequalities that are well known and can be found , for example , in @xcite . we will use the following version of _ chernoff s bound_. suppose that @xmath89 is a binomial random variable with expectation @xmath90 . if @xmath91 , then @xmath92 \le \exp \left ( -\frac{\delta^2 \mu}{2 } \right),\ ] ] and if @xmath93 , @xmath94 \le \exp\left(-\frac{\delta^2 \mu}{2+\delta}\right).\ ] ] we will use the following version of _ janson s inequality _ , adapted to our setting . [ thm : janson ] let @xmath95 , @xmath4 , and @xmath96 . for @xmath97 with @xmath98 , let @xmath99 be the indicator variable which is @xmath38 if the vertices in @xmath100 form a clique of size @xmath82 , and @xmath101 otherwise . let @xmath102 be the random variable counting copies of @xmath103 in @xmath5 , @xmath104 $ ] , and @xmath105.$ ] then , for @xmath106 , @xmath107 where @xmath108 . ( let us note that indicator random variables @xmath99 and @xmath109 are independent if @xmath110 ; see the definition of @xmath111 above . ) we will use the following result of vu to obtain bounds on the upper tail of certain subgraphs ( see @xcite ) . denote by @xmath112 ( @xmath113 , respectively ) the number of edges ( vertices , respectively ) of a graph @xmath114 . we say that a graph @xmath114 is _ balanced _ if for every subgraph @xmath115 with @xmath116 , we have @xmath117 . [ lem : vu ] let @xmath114 be a fixed balanced graph on @xmath82 vertices , and let @xmath118 be a constant . then there are constants @xmath119 and @xmath120 such that the following holds . let @xmath121 , let @xmath122 denote the number of appearances of @xmath114 in the graph @xmath7 , and let @xmath123 $ ] . if @xmath124 and @xmath125 , then @xmath126 ( let us remark that stronger bounds on the upper tail are known ; see , for example , @xcite . for our purposes , however , the current one is sufficient . ) we will also make use of the following version of _ harris s inequality _ ( see @xcite ) . let @xmath47 , with vertex set @xmath127 $ ] and random edge set @xmath128 . let @xmath129 so @xmath130 is a set of possible edges . let @xmath131 be the collection of all subsets of the set @xmath130 . for @xmath132 let @xmath133 mean @xmath134 . for @xmath132 , we say that @xmath100 is a _ downset _ if @xmath135 and @xmath136 implies @xmath137 , and @xmath100 is an _ upset _ if @xmath135 and @xmath138 implies @xmath137 . we also may refer to the event @xmath139 as a _ down - event _ or an _ up - event _ in these two cases . [ lem.harris ] with notation as above , let @xmath100 and @xmath140 be upsets and let @xmath141 and @xmath142 be downsets . then @xmath100 and @xmath140 are positively correlated , i.e. , @xmath143 , and similarly @xmath141 and @xmath142 are positively correlated ; and @xmath100 and @xmath141 are negatively correlated , i.e. , @xmath144 . the final lemma here is a simplified version of lemma 2.3 of @xcite . it says roughly that a random walk which is biased upwards when needed tends to move upwards as we should expect . we shall use it in the proof of theorem [ thm.ubforc ] ( to analyse a greedy procedure to find triangle - free sets ) . [ lem.bin2 ] let @xmath145 be a filter ( in a probability space ) . let @xmath146 be binary random variables such that each @xmath147 is @xmath148-measurable , and let @xmath149 . let @xmath150 be events where @xmath151 for @xmath152 . let @xmath153 and let @xmath154 be a positive integer . suppose that for each @xmath155 @xmath156 then @xmath157 part ( a ) and ( b ) of theorem [ thm : gnp ] follow immediately from the fact that @xmath9 for each graph . it is well known that whp@xmath158 provided that @xmath41 as @xmath35 , and @xmath159 ( see @xcite ) . for @xmath24 as here , we shall see that whp each edge is in a triangle , so we can use triangle - free sets of vertices as colour classes . we repeatedly use a natural greedy algorithm to construct maximal triangle - free sets and remove them . while many vertices remain , with high probability each set constructed has size at least about @xmath160 , so we will not remove too many of them . later phases require a negligible number of further colours . in the proof , in order to show that our greedy triangle - free sets are large , we need first to show that they do not tend to contain more edges than a typical set of the same size . [ thm.ubforc ] suppose that @xmath161 is such that @xmath162 . let @xmath163 be a constant . then whp for @xmath47 we have @xmath164 first let us check that whp each edge is in a triangle , for then we can colour using triangle - free sets . the expected number of edges that are not part of any triangle is @xmath165 since @xmath166 and @xmath159 . the desired property now follows from markov s inequality . we will repeatedly use the following greedy algorithm to construct a triangle - free set @xmath100 of vertices in a graph @xmath7 with vertex set a subset of the positive integers . consider the vertices in increasing order . initially set @xmath167 and @xmath168 . while @xmath169 is non - empty : remove the first member @xmath170 from @xmath169 , and if @xmath171 is triangle - free then add @xmath170 to @xmath100 . note that this procedure examines only potential edges with at least one end in @xmath100 . how large is the final triangle - free set @xmath100 for @xmath47 ? the first thing to consider is the number of edges of @xmath7 within @xmath100 when @xmath172 . we shall see in part ( 1 ) of the proof below that this is stochastically at most @xmath173 . in part ( 2 ) of the proof we shall define appropriate integers @xmath174 and @xmath175 such that whp we find a triangle - free set of size @xmath174 within @xmath175 steps ; and we complete the proof in part ( 3 ) . _ part ( 1)_. first we consider the number of edges within the set @xmath100 . let @xmath25 be a ( fixed ) integer , with corresponding @xmath1 , and embed the random graph @xmath7 in an infinite random graph @xmath176 on @xmath177 with the same edge probability @xmath24 . let @xmath178 be the random edge - set of @xmath176 . for @xmath179 let @xmath180 be the set @xmath100 when @xmath181 ( which is well defined for each @xmath182 since we are working in @xmath176 and @xmath183 ) . let @xmath184 be the number of edges within @xmath180 ( and set @xmath185 ) , and let @xmath186 . we * claim * that @xmath187 . this is clearly true for @xmath188 since @xmath189 . let @xmath190 and suppose it is true for @xmath191 . let @xmath192 and let @xmath193 be the event that @xmath194 . let @xmath140 be a set of @xmath191 vertices , let @xmath195 be a set of @xmath196 edges within @xmath140 , and let @xmath170 be an element of @xmath197 after all of @xmath140 . let @xmath198 be the event that @xmath199 and @xmath195 is the set of edges of @xmath176 within @xmath140 . let @xmath200 be the event that @xmath170 is the vertex added to @xmath201 to form @xmath180 . let @xmath202 be the set of @xmath191 possible edges between @xmath170 and @xmath140 . let @xmath203 , the random set of edges present in @xmath202 . note that @xmath204 . for each real @xmath175 , let @xmath205 : and observe that @xmath206 is an up - event . let @xmath207 and observe that @xmath208 is a down - event . hence , for each @xmath175 , by harris s inequality , lemma [ lem.harris ] , @xmath209 since this holds for each possible choice of @xmath140 , @xmath195 and @xmath170 specifying events @xmath198 and @xmath200 , we obtain @xmath210 now fix @xmath211 , and let @xmath212 . by the last inequality , since @xmath213 , @xmath214.\end{aligned}\ ] ] but @xmath215 by the induction hypothesis , and @xmath216 is non - decreasing , so @xmath217 & \leq & { \mathbb e}[g({\mathrm{bin}}(\binom{j\!-\!1}{2},p ) ) ] \\ & = & \sum_x { \mathbb{p}}({\mathrm{bin}}(\binom{j\!-\!1}{2},p)=x ) \ , { \mathbb{p}}({\mathrm{bin}}(j\!-\!1,p ) \geq t\!-\!x)\\ & = & { \mathbb{p}}({\mathrm{bin}}(\binom{j}{2},p ) \geq t).\end{aligned}\ ] ] thus @xmath218 and we have established the claim that @xmath187 for each @xmath182 . this completes part ( 1 ) of the proof . _ part ( 2)_. still in the infinite case , let @xmath219 be the number of vertices @xmath170 tested until we find one to add to @xmath201 . for @xmath220 the first available vertex must be added . fix @xmath221 . for each positive integer @xmath222 , let @xmath223 be the @xmath222th vertex in @xmath169 after the largest vertex in @xmath201 ( the @xmath222th candidate to add to @xmath201 ) . let @xmath196 , @xmath140 , @xmath130 , @xmath170 , @xmath193 , @xmath198 and @xmath224 be as before . observe that @xmath224 is an intersection of @xmath196 down - sets in @xmath225 , and for each of these downsets @xmath226 we have @xmath227 . by harris s inequality ( lemma [ lem.harris ] ) again , @xmath228 since this holds for each possible choice of @xmath140 , @xmath130 and @xmath170 , it follows that for each positive integer @xmath222 we have @xmath229 hence @xmath230 and since the upper bound is nondecreasing in @xmath196 , it follows that @xmath231 let @xmath232 , and let @xmath233 . note that @xmath234 , and @xmath235 . observe that @xmath236 ; and let @xmath237 . let @xmath238 ; and let @xmath239 . then for each @xmath240 , by the chernoff bound ( [ eq : chern2 ] ) @xmath241 where the final step follows since @xmath159 . let @xmath242 , so @xmath243 and @xmath244 . then @xmath245 by the above . but since @xmath159 , @xmath246 and so @xmath247 thus @xmath248 in other words , the probability that the greedy algorithm fails to find a triangle - free set of size @xmath174 within the first @xmath249 available vertices is @xmath250 . now let us return to the finite case @xmath47 , and note that the last statement still applies . this completes part ( 2 ) of the proof . _ part ( 3)_. let us repeatedly seek a maximal triangle - free set of size at least @xmath174 , and remove it , as long as at least @xmath175 vertices remain . let @xmath251 be the set of vertices remaining at some step ( when we are about to seek a new triangle - free set ) . each time , we have @xmath252 ( that is , we start with at least @xmath175 vertices ) , and the potential edges have not been examined before . hence , each time , the probability we fail is @xmath250 , and so the probability we ever fail is @xmath253 . this whole process uses at most @xmath254 colours . observe for later that this number is @xmath255 . by this stage , @xmath256 . while @xmath257 , we continue to seek maximal triangle - free sets using the same greedy algorithm , but now we will be content with smaller sets . any remaining vertices can each be given a new colour , adding at most @xmath258 colours . we need new notation . let @xmath259 , and let @xmath260 . list the vertices in @xmath251 in their natural increasing order as @xmath261 . for @xmath262 let @xmath148 be the @xmath263-field generated by the appearance or not of the edges amongst @xmath264 ; let @xmath265 be the random set of edges amongst @xmath264 which appear ; and let @xmath147 be the indicator that the @xmath222th vertex @xmath266 is accepted into @xmath100 . let @xmath267 . then @xmath268 , with expected value at most about @xmath269 . let @xmath270 . then @xmath271 by the chernoff bound ( [ eq : chern2 ] ) . given that @xmath272 , the event that @xmath273 is the union over the edges @xmath274 of the event that @xmath266 is adjacent to both end vertices of @xmath275 . thus , for each event @xmath276 @xmath277 ( assuming that the conditioning event has positive probability ) . hence @xmath278 let @xmath279 be the value of @xmath280 after trying to add @xmath281 ( starting from @xmath167 ) . thus @xmath282 . hence by lemma [ lem.bin2 ] and the chernoff bound ( [ eq : chern2 ] ) @xmath283 and so @xmath284 throughout the time when @xmath285 , whp each triangle - free set @xmath100 found has size at least @xmath286 ; so the number of colours used is at most @xmath287 . throughout the time when @xmath288 , whp each triangle - free set @xmath100 found has size at least @xmath289 ; so the number of colours used is @xmath290 . as we noted before , the final phase uses @xmath291 colours . thus the total number of colours used after the first phase is @xmath292 , which is negligible compared with @xmath293 . before we state a useful observation that holds for deterministic graphs , let us introduce a few definitions . independent set _ ( or a _ stable set _ ) of a graph @xmath294 is a set of vertices in @xmath7 , no two of which are adjacent . a _ dominating set _ in @xmath7 is a subset @xmath142 of @xmath197 such that every vertex not in @xmath142 is adjacent to at least one member of @xmath142 . of course , a set could be both independent and dominating ; this situation happens if and only if the set is a maximal independent set . finally , for @xmath295 let @xmath296 denote the set of neighbours of @xmath297 . the following lemma is part of theorem 3 of @xcite : we give a short proof here for completeness . [ lem : deterministic_upperbound ] let @xmath298 be a graph , and suppose that @xmath299 is a dominating set . then @xmath300 . let @xmath301 . for @xmath302 in turn , assign colour @xmath222 to each uncoloured neighbour of @xmath303 . any vertices remaining uncoloured must form an independent subset of @xmath100 : give each such vertex colour 0 . this gives a proper clique - colouring , since any vertices coloured 0 form an independent set ; and for each @xmath302 the set of vertices coloured @xmath222 is dominated by @xmath303 , which is not coloured @xmath222 . from lemma [ lem : deterministic_upperbound ] , we quickly get the following bound for binomial random graphs . this proves the upper bounds in parts ( d ) to ( g ) of theorem [ thm : gnp ] . we say that @xmath1 is _ bounded below 1 _ if there is some constant @xmath304 such that @xmath305 for all @xmath25 . [ lem.upperbound_sparse_gnp ] suppose that @xmath1 is bounded below 1 , and @xmath306 as @xmath35 . let @xmath307 . then for @xmath47 whp @xmath308 we construct a maximal independent set in the usual greedy way , by fixing an arbitrary order of the vertices @xmath309 and adding vertices ( in this order , one by one ) to form a maximal independent set . observe that at the beginning of a given step @xmath175 of this process , no edge emanating from @xmath310 has been exposed so far . note that whp the process terminates after at most @xmath82 vertices are added to the independent set : indeed , the probability that a larger set is created during this greedy process is the probability that a set of size @xmath82 is created at some point of the process , and this set does not dominate the remaining vertices ; and ( by considering just the second part ) this probability is at most @xmath311 . the result now follows by lemma [ lem : deterministic_upperbound ] . we will now show that for dense random graphs , we can improve the bound from part ( d ) by a factor of about @xmath312 . this then proves part ( h ) of theorem [ thm : gnp ] . ( for sparser graphs , the improvement might also be possible but the argument would be more tedious . ) suppose that @xmath1 is bounded below 1 , and satisfies @xmath65 . then , for @xmath47 whp @xmath313 first , let us start with a few simple properties of @xmath0 . _ claim 1 _ whp there is no clique of size @xmath314 . indeed , noting that @xmath315 we get that the expected number of cliques of size @xmath82 is @xmath316 the claim holds by markov s inequality . _ claim 2 _ whp there is no maximal clique of size at most @xmath317 indeed , the expected number of sets of size @xmath82 ( not necessarily inducing cliques ) for which there is no vertex adjacent to all of them is @xmath318 as before , the claim holds by markov s inequality . the rest of the proof uses ideas similar to the ones used in the proofs of lemmas [ lem : deterministic_upperbound ] and [ lem.upperbound_sparse_gnp ] . let @xmath319 and note that @xmath320 as @xmath35 . as before , we fix an arbitrary order of the vertices and construct an independent set @xmath100 greedily , but we stop once its size is equal to @xmath321 . as before , we use @xmath322 colours to create colour classes that do not contain any maximal cliques . let @xmath323 be the random number of vertices not in @xmath100 and not dominated by @xmath100 ( and so not coloured yet ) . then @xmath323 is stochastically at most @xmath324 , and so by ( [ eq : chern1 ] ) whp @xmath325 , where @xmath326 . we will show that whp one additional colour is enough to finish the job . we may assume that @xmath325 . as this part of the random graph is not exposed yet , this subgraph is distributed as a subgraph of @xmath327 and claim 1 implies that whp the maximum size of a clique in the additional colour is at most @xmath328 on the other hand , claim 2 implies that whp no clique of such size is maximal in @xmath7 . the proofs of the upper bounds of theorem [ thm : gnp ] are finished . given a graph @xmath7 , we say that a set @xmath5 of vertices is _ maximal - clique - free _ if it contains no clique of size at least 2 which is a maximal clique in @xmath7 . we let @xmath329 denote the maximum size of a maximal - clique - free set of vertices . observe that if @xmath7 has @xmath25 vertices then @xmath330 our lower bounds on @xmath8 are usually obtained from upper bounds on @xmath329 by using this inequality . let us first consider very sparse graphs . observe that for any graph @xmath114 with @xmath25 vertices and at most @xmath175 vertices in triangles , we have @xmath331 . let @xmath47 . if @xmath41 and @xmath159 , then @xmath332 whp ( see , for example , @xcite ) . for @xmath333 the expected number of triangles is @xmath334 , so whp the number of vertices in triangles is @xmath335 . thus for @xmath336 with @xmath333 @xmath337 and for such values of @xmath338 the lower bound of part ( a ) is proven . for the remaining range of @xmath338 to be considered in part ( a ) we may therefore clearly assume @xmath339 . ( in fact , we may assume @xmath340 for some function @xmath341 as @xmath35 but it would not simplify the argument . ) let @xmath47 , @xmath342 be a function tending to infinity with @xmath25 arbitrarily slowly , and let @xmath343 ( observe that @xmath344 . ) we will start with the following elementary observation . [ lem : triangles ] let @xmath345 be defined as above , and let @xmath346 . let @xmath47 . then whp no edge of @xmath7 belongs to more than @xmath347 triangles . consider a pair @xmath348 of distinct vertices . the number of common neighbours of @xmath349 and @xmath297 is a random variable @xmath350 with distribution @xmath351 , which is stochastically bounded from above by @xmath352 . we first observe that @xmath353 since @xmath354 and @xmath355 . hence , @xmath356 since there are @xmath357 pairs to consider , the result holds by a union bound . the next lemma is less straightforward . [ lem : lower_bound1 ] let @xmath345 be defined as above . suppose that @xmath358 . let @xmath359 and @xmath360 . then @xmath361 moreover , let @xmath362 . then , whp every set @xmath363 of size @xmath364 has the following properties : * the graph induced by @xmath365 , @xmath366 $ ] , has at least one edge , * there is no set of edge - disjoint triangles , each with one vertex in @xmath367 , which contains at least a @xmath368-fraction of the edges in @xmath366 $ ] . first , recall that @xmath369 and so @xmath370 is an increasing function in @xmath24 ( if we think of @xmath24 as a variable ) . hence , writing @xmath371 as the value of @xmath372 when @xmath24 attains its maximum value @xmath373 , @xmath374 and so also @xmath375 , and the first assertion follows . on the other hand , it follows immediately from the definition of @xmath376 that @xmath377 . for the second assertion , part ( a ) is standard . the probability @xmath378 that the desired property is not satisfied can be estimated as follows : @xmath379 where the second last step holds for @xmath25 sufficiently large . for part ( b ) , for a given @xmath380 such that @xmath381 , we are going to estimate the probability @xmath382 that there exists a set @xmath365 of size @xmath82 with precisely @xmath380 edges in @xmath366 $ ] and with at least @xmath383 edges that belong to edge - disjoint triangles with the third vertex in @xmath367 . we have @xmath384 where @xmath385 + p \right\ } \\ & = & k \left ( \log ( pn ) + o(1 ) - \log \log ( pn ) - \frac { 2+{\varepsilon}'}{2 } \log(pn ) \right ) \\ & & + \ \ m \left\ { \log \left [ \left ( \frac { e(2+{\varepsilon}')^2 \log^2 ( pn)}{2mp } \right ) \left ( \frac { 2ec(pn)^2}{n } \right)^{1/(2c ) } \right ] + p \right\}.\end{aligned}\ ] ] hence , @xmath386 , where @xmath387 + p \right\}.\end{aligned}\ ] ] let @xmath388 first , note that if @xmath389 , then @xmath390 for @xmath391 , we get @xmath392 finally , for @xmath393 for any @xmath394 , since @xmath395 , we get @xmath396 - o(1 ) \ge 1.\end{aligned}\ ] ] as a result , @xmath397 and so @xmath398 . the result holds . the last two lemmas give us easily the following lower bound on @xmath399 that matches ( asymptotically ) the upper bound . this proves part ( a ) of theorem [ thm : gnp ] . let @xmath400 be a function tending to infinity with @xmath25 arbitrarily slowly . suppose that @xmath1 is such that @xmath401 let @xmath402 . then , whp @xmath403 , and so @xmath404 by ( [ eqn.sparselb ] ) , we may assume that @xmath405 ( or indeed @xmath406 for some @xmath407 ) . let @xmath408 be defined as in lemma [ lem : lower_bound1 ] . as we aim for a statement that holds whp , we may assume that all properties in lemma [ lem : lower_bound1 ] and lemma [ lem : triangles ] hold deterministically . consider any colouring that yields @xmath8 . we will show that no colour class has size at least @xmath82 which will finish the proof . for a contradiction , suppose that the vertices of some set @xmath365 of size @xmath82 are of the same colour . it follows from lemma [ lem : lower_bound1](a ) that @xmath366 $ ] contains at least one edge . as no maximal clique is monochromatic , each edge of @xmath366 $ ] belongs to a triangle with the third vertex in @xmath367 . as , by lemma [ lem : triangles ] , no edge belongs to more than @xmath347 triangles , we may greedily pick a set of edge - disjoint such triangles containing at least a @xmath368-fraction of the edges in @xmath366 $ ] . but this contradicts lemma [ lem : lower_bound1](b ) . we start with the following observation . we will apply the result for @xmath85 but we state it in a more general case as the proof of this generalization is exactly the same . [ lem : lowerboundgeneral ] let @xmath409 be a fixed integer , and let @xmath410 be a fixed arbitrarily small positive constant . suppose that @xmath1 satisfies @xmath411 and let @xmath412 . let @xmath362 . then , whp the following property holds : every set @xmath4 of @xmath413 vertices contains at least @xmath414 copies of @xmath103 . ( note that the lower bound for @xmath338 is there only to make sure that @xmath415 . ) for a fixed set @xmath4 of size @xmath413 , let @xmath350 be the random variable counting the number of copies of @xmath103 in @xmath5 . clearly , @xmath416=\binom{\lceil s \rceil}{k}p^{\binom{k}{2}}$ ] . using the notation of lemma [ thm : janson ] , we may deal separately with the cases @xmath417 with @xmath418 to get @xmath419 let us observe that it follows from the definition of @xmath420 that the @xmath222-th term of the sum is of order @xmath421 . since @xmath422 , the logarithmic factor is negligible and it is easy to see that the sum is dominated by the last term ( @xmath423 ) . therefore , @xmath424 applying lemma [ thm : janson ] with @xmath425/2 $ ] we get @xmath426}{2 } \right ) & \le & \exp \big ( - ( 1+o(1 ) ) \varphi(-1/2 ) \mu \big ) \le \exp \left(- \frac { 0.15 s^k}{k ! } p^{\binom{k}{2 } } \right ) \\ & = & \exp \left(- \frac { 0.15 ( 3k)^{k-1 } s \log n}{k ! } \right ) \le \exp ( -2 s \log n).\end{aligned}\ ] ] taking a union bound over all @xmath427 sets of size @xmath428 , the desired property holds . we will also need the following property . [ lem : nodensegeneral ] let @xmath429 be a fixed integer , and let @xmath430 satisfy @xmath431 . suppose @xmath1 satisfies @xmath432 , and let @xmath433 . then whp no edge in @xmath7 belongs to as many as @xmath434 copies of @xmath435 . let the graph @xmath114 be formed from an edge @xmath436 together with some cliques @xmath435 containing @xmath275 ( at least one such clique ) . we will show that we have @xmath437 to see this , we may use induction on the number of cliques in a construction of @xmath114 . the inequality holds ( at equality ) if @xmath114 is a single @xmath83-clique . suppose that there are at least two cliques in the construction , and the last one added @xmath438 new vertices and @xmath439 new edges . if @xmath440 , then the inequality still holds after adding the last clique , so we may assume that @xmath441 then @xmath442 where the second inequality follows since @xmath443 . thus @xmath444 , and the induction step follows , since from @xmath445 for positive @xmath446 and @xmath447 , we may conclude that @xmath448 . the number of copies of @xmath435 on @xmath449 vertices that contain the edge @xmath450 is at most @xmath451 for @xmath452 ( since @xmath453 ) ; and @xmath454 , so the result holds also for @xmath85 . thus if @xmath114 contains at least @xmath455 such cliques we must have @xmath456 . further @xmath114 must then have a subgraph on at most @xmath457 vertices which contains at least @xmath455 such cliques . call the set @xmath169 of vertices in @xmath7 _ dense _ if @xmath458)}{|w|-2 } \geq \frac{k+2}{2}.\ ] ] let @xmath459 be the expected number of dense sets of size @xmath222 in @xmath7 . it will suffice for us to show that @xmath460 now @xmath461 but @xmath462 . thus for @xmath463 , @xmath464 as @xmath409 . since there are at most @xmath465 terms in the sum in ( [ eqn.showai ] ) , the desired conclusion there follows . now , finally we are able to prove the lower bounds in parts ( b ) and ( c ) of theorem [ thm : gnp ] . [ thm : lower_bound_k ] let @xmath429 be a fixed integer , and let @xmath466 . suppose that @xmath1 satisfies @xmath467 and let @xmath468 . let @xmath362 . then , whp @xmath469 and so @xmath470 for simplicity we shall ignore the fact that certain numbers should be integers : this never matters here . let @xmath5 be a fixed set of size @xmath420 . we shall show that @xmath471 is very small . if @xmath5 is maximal - clique - free , then it must be possible to extend each copy @xmath472 of @xmath103 inside @xmath5 to @xmath435 by adding some vertex @xmath473 which is complete to @xmath472 , that is , @xmath297 is connected by an edge to each of the @xmath82 vertices of @xmath472 . unfortunately ( for our proof ) these edges can be reused in extensions for different copies of @xmath103 inside @xmath5 . our plan is to show that , if @xmath5 is maximal - clique - free , then with very high probability there must be a large collection of copies of @xmath103 inside @xmath5 that are extended to @xmath6 using each edge between @xmath5 and @xmath6 at most once ; and this is very unlikely to happen . let @xmath474 , the expected number of copies of @xmath103 inside @xmath5 . let @xmath475 be the event that @xmath5 contains at least @xmath476 copies of @xmath103 . let @xmath477 be the event that each edge appears in at most @xmath478 copies of @xmath435 . let @xmath479 be the event that each edge @xmath450 with @xmath480 and @xmath473 appears in at most @xmath478 copies of @xmath435 such that each vertex other than @xmath297 is in @xmath5 . let @xmath481 be any set of edges within @xmath5 such that @xmath482 holds , and condition on @xmath483 . here @xmath484 denotes the set of edges within @xmath5 in the random edge set @xmath128 of @xmath7 . observe that , since @xmath485 is fixed , @xmath479 becomes a down - event in the lattice @xmath131 of subsets of @xmath486 . now let @xmath487 , and consider a possible ` edge - disjoint extension of length @xmath175 ' . suppose that @xmath488 are distinct copies of @xmath103 inside @xmath5 , and @xmath489 are vertices in @xmath6 ( not necessarily distinct ) , such that each possible edge appears at most once as @xmath490 for some vertex @xmath349 in @xmath491 . then the event that @xmath303 is complete to @xmath491 for each @xmath492 $ ] is the up - event in @xmath131 that each of the @xmath493 relevant edges is present . hence by lemma [ lem.harris ] ( harris s inequality ) @xmath494 \ , | \ , ( e|_s = e_0 ) \land { { \mathcal f}}_s ) \leq p^{kt}.\ ] ] keep the conditioning on @xmath483 , condition also on @xmath479 , and suppose that @xmath5 is maximal - clique - free . list the copies of @xmath103 inside @xmath5 in say lexicographic order , discarding any after the first @xmath476 ; and consider the vertices in @xmath6 in their natural order . consider the first copy @xmath472 of @xmath103 on the list : test the vertices in @xmath6 one by one until we find a vertex @xmath297 complete to @xmath472 ( we must succeed since @xmath5 is maximal - clique - free ) . eliminate from the list the clique @xmath472 and all other copies @xmath495 of @xmath103 such that @xmath495 and @xmath472 have at least one vertex in common and @xmath297 is complete to @xmath495 ( so the extensions by @xmath297 have an edge in common ) . since we conditioned on @xmath479 , at most @xmath496 copies of @xmath103 are eliminated from the list in this round , and we move to the next copy of @xmath103 that is left in the list and continue . since we also conditioned on @xmath483 , there are at least @xmath497 copies of @xmath103 in @xmath5 , and so the process must last for at least @xmath175 rounds , where @xmath498 now we lower bound @xmath175 : we have @xmath499 as @xmath500 . we get @xmath501 as @xmath502 for @xmath429 . the number of sets of @xmath175 distinct copies @xmath503 of @xmath103 inside @xmath5 which are on the list , and vertices @xmath504 yielding possible edge - disjoint extensions as in ( [ eqn.disjext ] ) , is at most @xmath505 . thus , by ( [ eqn.disjext ] ) and a union bound , @xmath506 since this holds for every choice @xmath481 of edges within @xmath5 such that @xmath482 holds , and since @xmath507 , it follows that @xmath508 we have @xmath509 which is @xmath510 since @xmath511 . let us rename the fixed set @xmath5 as @xmath512 . then , with the union below being over all @xmath420-subsets @xmath5 of vertices , @xmath513 finally , let @xmath514 . then @xmath515 but each of the three terms in the upper bound here is @xmath253 . we have just seen this for the first term : the second and third terms are @xmath253 by lemmas [ lem : lowerboundgeneral ] and [ lem : nodensegeneral ] respectively . the proof of the theorem is complete . let us start with the following lemma , which is proved in a similar way to lemma [ lem : lowerboundgeneral ] . [ lem : lowerboundgeneral2 ] let @xmath429 be a fixed integer , and let @xmath516 be a fixed arbitrarily small positive constant . let @xmath517 be such that @xmath518 . suppose that @xmath1 satisfies @xmath519 let @xmath520 , and let @xmath362 . then , whp the following property holds : every set @xmath4 of @xmath413 vertices contains at least @xmath521 copies of @xmath103 . as before , for simplicity we shall ignore the fact that certain numbers should be integers . for a fixed set @xmath4 of size @xmath413 , let @xmath350 be the random variable counting the number of copies of @xmath103 in @xmath5 . clearly , @xmath416=\binom{\lceil s \rceil}{k}p^{\binom{k}{2}}$ ] . using the notation of lemma [ thm : janson ] , we may independently deal with the cases @xmath417 with @xmath418 to get @xmath522 for @xmath523 , let @xmath524 and note that the @xmath222th term in the last sum is @xmath525 . the ratio @xmath526 , which is increasing with @xmath222 . it follows that @xmath527 is a unimodal sequence , with maximum either @xmath528 or @xmath529 . moreover , since @xmath530 , at most two consecutive terms can be of the same order . as a result , the sum is of order the larger of the term @xmath531 and the term @xmath423 . ( note that we did not rule our the possibility that the @xmath532-st term is of order of the @xmath82-th term yet . ) more precisely , comparing the two terms , the sum is of order of the term @xmath423 if @xmath533 , and of order of the term @xmath531 otherwise . moreover , in the case @xmath533 , by comparing the @xmath532-st and the @xmath82-th term , we see that the @xmath82-th term dominates the @xmath532-st term if @xmath534 , and since @xmath535 for @xmath429 , this condition is satisfied . hence , if @xmath533 , the sum is in fact asymptotic to the term @xmath423 . suppose first that @xmath533 . as we already mentioned , in this case @xmath536 , and so applying lemma [ thm : janson ] with @xmath425/2 $ ] we get @xmath426}{2 } \right ) & \le & \exp \big ( - ( 1+o(1 ) ) \varphi(-1/2 ) \mu \big ) \le \exp \left(- \frac { 0.15 ( cpn)^k}{k ! } p^{\binom{k}{2 } } \right ) \\ & \leq & \exp \left(- \frac { 0.15 ( 3k)^{k-1 } ( cpn ) \log n}{k ! } \right ) \le \exp ( -2 s \log n),\end{aligned}\ ] ] by our assumption that @xmath537 . taking a union bound over all @xmath427 sets of size @xmath428 , the desired property holds . assume now that @xmath538 . in this case we get @xmath539 therefore , applying lemma [ thm : janson ] with @xmath425/2 $ ] we get @xmath426}{2 } \right ) & \le & \exp \big ( - \omega(\mu^2/\overline{\delta } ) \big ) = \exp \left(- \omega \left ( \frac{s c ( pn)^2}{n}\right)\right ) \le \exp ( -2 s \log n ) , \end{aligned}\ ] ] as , by our assumption on @xmath24 and @xmath141 , we have @xmath540 . taking a union bound over all @xmath427 sets of size @xmath428 , the desired property holds . we can now prove the following theorem that implies the lower bounds in parts ( e ) and ( g ) of theorem [ thm : gnp ] . [ thm : lower_bound_k2 ] let @xmath429 be a fixed integer . suppose that @xmath1 satisfies @xmath541 let @xmath542 , where @xmath543 let @xmath362 . then , whp @xmath469 , and so @xmath544 before we move to the proof of this result , let us make a few comments . first , obviously , if there are several values of @xmath82 that satisfy the assumptions on @xmath338 , one should consider the one that gives the best lower bound for @xmath8 . in particular , the case @xmath85 covers values of @xmath338 between @xmath545 and @xmath546 , and the case @xmath547 covers the range of @xmath338 between @xmath548 and @xmath549 . the case @xmath550 can be applied already if @xmath338 is at least @xmath551 , and so for values of @xmath338 in the interval @xmath552 $ ] both @xmath547 and @xmath550 satisfy the conditions of the theorem . similarly , if we fix @xmath43 ( arbitrarily small ) , all values of @xmath338 that belong to the interval @xmath553 $ ] are covered by at least two values of @xmath82 at most the constant @xmath554 ( since @xmath555 , for @xmath556 we are concerned only with @xmath82 such that @xmath557 , that is @xmath558 ) . if @xmath24 is in the interval for @xmath559 then @xmath560 , for some @xmath561 . thus for at least one of the relevant values @xmath82 ( any one except the smallest ) we have @xmath562 , and so @xmath563 . this in turn implies that @xmath564 and so @xmath565 . the only range not covered by this theorem is when @xmath69 or @xmath61 which will be done separately later on . the proof is similar to the proof of theorem [ thm : lower_bound_k ] . however , one additional idea is needed : informally speaking , we now have too many copies of @xmath103 , but still only @xmath566 of them are going to be extended to @xmath435 using disjoint edge sets . the union bound over all choices of these would be too large to be successfully applied , and so we focus on a randomly chosen subset of the @xmath103 s . to this end , define @xmath567 now , in order to obtain the desired smaller subset of cliques , we do the following , independently of the random graph @xmath7 : for each set of @xmath82 vertices , we independently colour it _ red _ with probability @xmath568 . our first goal is to show that no edge @xmath450 belongs to too many cliques of size @xmath322 which after removal of @xmath297 form a red @xmath82-set . let @xmath569 , say @xmath570 . for each pair of distinct vertices @xmath349 and @xmath297 , let @xmath571 be the event that @xmath572 let @xmath573 be the event that @xmath574 holds for each pair @xmath575 . since @xmath576 , by the chernoff bounds and , @xmath577 ; and so by a union bound , whp @xmath573 holds . fix any ordered pair of distinct vertices @xmath578 , and any set @xmath579 with @xmath580 . let @xmath581 if @xmath582 induces a clique , @xmath349 and @xmath297 are both complete to @xmath582 ( we do not require @xmath349 and @xmath297 to be adjacent ) . finally , let @xmath583 . let @xmath584 . let @xmath279 be the number of @xmath532-cliques in the random graph with vertex set @xmath585 $ ] and edge - probability @xmath40 . clearly , conditional on @xmath574,@xmath586 is stochastically at most @xmath279 . we have @xmath587 hence , by lemma [ lem : vu ] , once @xmath25 is sufficiently large that @xmath588 , @xmath589 thus @xmath590 by taking a union bound over all ordered pairs of distinct vertices , we see that whp for each @xmath575 we have @xmath591 . now let @xmath592 be defined as @xmath593 with the additional condition that @xmath594 is red , and let @xmath595 . clearly , conditional on @xmath591 , @xmath596 is stochastically bounded above by the binomial random variable @xmath597 with expectation @xmath598 . it is a straightforward application of ( together with a union bound over all ordered pairs @xmath348 ) to see that whp for each pair @xmath599 we have @xmath600 . let @xmath601 be the event that this property holds , so that @xmath226 holds whp . this completes the first part of the proof . for the rest of the proof , we argue very much as in the proof of theorem [ thm : lower_bound_k ] . as before , we ignore rounding issues for @xmath420 , since this does not matter . let @xmath602 and let @xmath5 be a set of size @xmath420 . let @xmath603 be the event that each edge @xmath450 with @xmath480 and @xmath473 appears in at most @xmath604 copies of @xmath435 which are such that the @xmath82 vertices other than @xmath297 form a red @xmath82-set in @xmath5 . note that @xmath605 . let @xmath474 , the expected number of @xmath82-cliques in @xmath5 ; so @xmath606 is the expected number of red @xmath82-cliques in @xmath5 . let @xmath607 be the event that @xmath5 contains at least @xmath608 red cliques of size @xmath82 ; and let @xmath609 . ( we shall check later that @xmath610 holds whp ) denote the random colouring of the @xmath82-sets of vertices by @xmath611 ; and let @xmath612 denote its restriction to the @xmath82-sets in @xmath5 . let @xmath481 be any set of edges within @xmath5 and let @xmath613 be any colouring of the @xmath82-sets within @xmath5 such that @xmath607 holds when @xmath483 and @xmath614 . condition on @xmath483 and @xmath614 . let @xmath487 , and consider a possible ` edge - disjoint extension of length @xmath175 ' . suppose that @xmath488 are distinct red cliques of size @xmath82 inside @xmath5 , and @xmath489 are vertices in @xmath6 ( not necessarily distinct ) , such that each possible edge appears at most once as @xmath490 for some vertex @xmath349 in @xmath491 . then , as in ( [ eqn.disjext]),@xmath615 \ , | \ , ( e|_s = e_0 ) \land ( r|_s = r_0 ) \land { { \mathcal d}}_s ) \leq p^{kt}.\ ] ] also , as before , if @xmath5 is maximal - clique - free , then by considering a list of red @xmath82-cliques of length @xmath616 , there must be a disjoint extension of length @xmath617 . hence , by a union bound , @xmath618 since this holds for each choice of @xmath481 and @xmath613 such that @xmath607 holds , @xmath619 note that @xmath620 plugging in the value of @xmath141 , we see that this expectation is at most @xmath621 . hence @xmath622 but , letting @xmath623 , @xmath624 we have just seen that the first term in the upper bound is @xmath253 , and we noted earlier that @xmath226 holds whp ; so it remains to show that @xmath625 . let @xmath512 be a fixed set of @xmath420 vertices , and let @xmath5 now be an arbitrary such set . let @xmath626 be the event that @xmath5 contains at least @xmath627 copies of @xmath103 ; and let @xmath628 . by lemma [ lem : lowerboundgeneral2 ] we have @xmath629 . also , by the chernoff bound , @xmath630 and so , by a union bound , @xmath631 but @xmath632 and so @xmath625 , as required . this completes the proof of the theorem . finally , we deal with the missing gaps when @xmath69 or @xmath61 which will finish the lower bounds in parts ( d ) and ( f ) of theorem [ thm : gnp ] . suppose that @xmath1 satisfies * @xmath633 , or * @xmath61 and @xmath634 . let @xmath362 . then , whp , @xmath635 let us focus on part ( a ) first . since the proof of this theorem is almost identical to the one of theorem [ thm : lower_bound_k2 ] ( with @xmath85 ) , we only mention two technical issues that are relatively easy to deal with . first , note that in order to apply lemma [ lem : lowerboundgeneral2 ] , we can not keep @xmath141 as defined in ( [ eq : def_c ] ) ; this time , we set @xmath636 we do not necessarily have @xmath637 anymore but still it is the case that @xmath638 . ( recall that @xmath85 in the proof of part ( a ) . ) this time there is no need to reduce the number of cliques for the union bound to work , so we may keep @xmath639 , which simplifies the argument slightly . the last difference is with the application of lemma [ lem : vu ] . this time the expected value of @xmath640 ( see ( [ eq : expectation_for_vu ] ) ) is not large enough for the lemma to be directly applied . however , as standard in such situations , one can increase the probability @xmath24 to some value @xmath641 for which the expected value of @xmath642 is of order greater than @xmath643 ( which equals @xmath39 in the case @xmath85 we deal with in part(a ) ) as required . for such a value @xmath644 , it follows that whp for any pair @xmath348 we have @xmath645 , and by standard coupling arguments the same holds for @xmath24 . the rest of the proof is not affected . exactly the same adjustments are required for part ( b ) , this time with @xmath547 . let us pick up two points for further thought . * we investigated the clique colouring number @xmath8 for random graphs @xmath47 , and in theorem [ thm : gnp ] we obtained fairly good estimates for values of @xmath1 , other than @xmath646 where @xmath399 drops dramatically as @xmath24 increases . by parts ( a ) and ( c ) , for suitable @xmath647 ( going to 0 slowly ) , if @xmath648 then @xmath649 whp ; whereas if @xmath650 then @xmath651 whp . for intermediate values of @xmath24 , all we say in part ( b ) is that @xmath8 lies whpbetween the values @xmath652 and @xmath653 . it would be interesting to learn more about this jump . * a second natural question for random graphs @xmath47 concerns the dense case , when @xmath24 is a constant with @xmath654 . we have seen that @xmath655 is @xmath656 whp but what about a lower bound ? cerioli , a.l . korenchendler , clique - coloring circular - arc graphs . _ electron . notes discret . _ * 35 * , 2009 , 287292 . d. dfossez , clique - coloring some classes of odd - hole - free graphs , _ j. graph theory _ * 53 * , 2006 , 233249 . j. kratochvl , zs . tuza , on the complexity of bicoloring clique hypergraphs of graphs , _ j. algorithms _ , * 45 * , 2002 , 4054 . m. krivelevich , b. sudakov , v.h . vu , n.c . wormald , on the probability of independent sets in random graphs , _ random structures & algorithms _ * 22 * ( 1 ) , 2003 , 114 . z. liang , e. shan , l kang , clique - coloring claw - free graphs , _ graphs and combinatorics _ , doi10.1007/s00373 - 015 - 1657 - 8 , 2015 , 116 . t. uczak , the chromatic number of random graphs , _ combinatorica _ * 11*(1 ) , 1991 , 4554 .

we present a sample of 2865 emission line galaxies with strong nebular emissions in sloan digital sky survey data release 7 and use this sample to investigate the origin of this line in star - forming galaxies . we show that star - forming galaxies and galaxies dominated by an active galactic nucleus form clearly separated branches in the versus\:\lambda6584}/{\ensuremath{\mathrm{h}\alpha}}$ ] diagnostic diagram and derive an empirical classification scheme which separates the two classes . we also present an analysis of the physical properties of 189 star forming galaxies with strong emissions . these star - forming galaxies provide constraints on the hard ionizing continuum of massive stars . to make a quantitative comparison with observation we use photoionization models and examine how different stellar population models affect the predicted emission . we confirm previous findings that the models can predict emission only for instantaneous bursts of 20% solar metallicity or higher , and only for ages of myr , the period when the extreme - ultraviolet continuum is dominated by emission from wolf - rayet stars . we find however that 83 of the star - forming galaxies ( 40% ) in our sample do not have wolf - rayet features in their spectra despite showing strong nebular emission . we discuss possible reasons for this and possible mechanisms for the emission in these galaxies . stars : wolf - rayet galaxies : evolution galaxies : stellar content galaxies : star formation

1201.1290

the ionizing continuum of stars at @xmath4 is of major importance for interpreting emission line observations of galaxies because many of the strong lines observed in the spectra of galaxies , such as @xmath5\:\lambda5007}$ ] , @xmath6\:\lambda3869}$ ] and @xmath0 , have ionization potentials in excess of 13.6 ev . despite this importance we are severely limited by interstellar absorption in observing stellar spectra in this spectral window directly ( e.g. * ? ? ? although we can get direct information at slightly longer wavelengths with space - based uv spectroscopy ( e.g. * ? ? ? * ) , most of our knowledge about the @xmath7 region is based on indirect evidence , even for solar metallicity . a promising way to indirectly obtain information on the stellar ionizing continuum is to compare emission line properties ( e.g. flux , equivalent width ) to predictions from photoionization codes such as cloudy @xcite or mappings iii @xcite . in practice these kinds of studies provide modest constraints on stellar atmosphere models ( e.g. * ? ? ? . however where predictions of models differ significantly , this approach can yield useful information . this is the approach we will adopt in this work , where we will make use of the @xmath0 nebular emission line to place constraints on stellar models and in particular on the ionization mechanism for this line . the presence of a nebular @xmath0 line in the integrated spectrum of a galaxy indicates the existence of sources of hard ionizing radiation as the ionization energy for he@xmath8 is 54.4 ev ( @xmath9 ) . this hard radiation can of course be produced by an active galactic nucleus ( agn ) , and most sources with luminous @xmath10ii emission , in a flux limited sample , are indeed galaxies with an agn . however the required hard radiation can also be provided by stellar sources and @xmath0 emission is frequently seen in @xmath11ii - galaxies . the line appears to be associated with young stellar populations ; for instance , @xcite proposed o__f _ _ stars as the sources of @xmath0 emission in dwarf galaxies . subsequent discussion has mostly focused on wolf - rayet ( wr ) stars , although the distinction between these two classes is rather blurred ( e.g. * ? ? ? * ) . * see also @xcite ) showed that the hard radiation field of wr stars could provide a good explanation of the nebular @xmath12 seen in @xmath11ii - galaxies . @xcite tried to test this in a careful study of @xmath11ii - galaxies with prominent wr features . they were however unable to find wr features in 12 out of the 30 galaxies with nebular @xmath13 emission . the same lack of wr features in metal poor blue compact dwarf ( bcd ) galaxies was pointed out by @xcite . they proposed that fast radiative shocks could be responsible for this emission ( see also @xcite ) . similar results were reported by ( * ? ? ? * hereafter b08 ) , who analysed a sample of strong emission line galaxies in the sloan digital sky survey @xcite with @xmath0 emission . they showed that at least at metallicities of @xmath14 , there appeared to be a close correlation between wr features in galaxies and the presence of @xmath15 emission , but this appeared not to be so clear - cut at lower metallicities . this apparent lack of connection of @xmath10ii emission with the hard uv radiation from wr stars has also been seen in spatially resolved spectroscopic studies . @xcite performed an integral field spectroscopy study for the @xmath11ii galaxy iizw70 and found that the region associated with nebular @xmath0 emission was a few arcsec offset from the region with detected wr features . more recently @xcite and @xcite have presented studies of @xmath15 emission in m33 . both studies find some regions with nebular @xmath15 emission that are not associated with wr stars ( see also @xcite , @xcite and @xcite ) . thus a series of studies have shown that while @xmath0 emission frequently is found in association with wr stars , it appears not to be so in all cases , particularly at low metallicity . as mentioned above , possible additional sources of high energy photons could be x - ray binaries @xcite , strong shocks @xcite , low - level agn activity and alternative models for stellar evolution @xcite . however the existing studies do not show clear trends that allow us to distinguish between these scenarios . crucially the samples in most of the previous studies have not been selected specifically to study @xmath10ii emission lines . to make progress in understanding this puzzle it is important to have as large as possible sample of @xmath10ii emitting galaxies to allow one to study the relationship between @xmath10ii emission and other physical properties . to this end we present here an analysis of emission line galaxies with strong @xmath0 emission in sdss data release 7 ( dr7 , * ? ? ? * ) . in section [ sec : data ] we discuss the sample selection and carefully account for agn emission . the physical properties of the @xmath10ii emitting galaxies are discussed in section [ sec : physical ] and the observed @xmath16 ratios are compared to model predictions in section [ sec : prediction ] . in section [ sec : origin ] we test these model predictions and investigate whether the presence of @xmath15 is associated with wr features . we find that low metallicity systems frequently do not show signs of wr stars . we discuss possible explanations for this finding in section [ sec : explanation ] and conclude in section [ sec : conclusion ] . our sample is based on galaxy spectra from sdss dr7 which cover a wavelength range of 3800 - 9200 . the spectra were analysed using the methodology discussed in @xcite ( see also brinchmann et al . 2004 ) to provide accurate continuum subtraction . all emission line sources were additionally analysed using the pipeline discussed in b08 to measure a wider gamut of emission lines . for each galaxy we measure 40 lines , these lines and the number of spectra that show these lines with @xmath17 are summarised in table [ table0 ] . measured line & number of spectra & fraction + @xmath18\:\lambda 3726,3729}$ ] & 243977 & 16.51% + @xmath6\:\lambda 3869}$ ] & 39143 & 2.65% + @xmath19 & 92230 & 6.24% + @xmath6\:\lambda 3967}$ ] & 18367 & 1.24% + @xmath20 & 64120 & 4.34% + @xmath21 & 6368 & 0.43% + @xmath22\:\lambda 4069}$ ] & 1503 & 0.10% + @xmath23 & 163534 & 11.07% + @xmath24 & 295528 & 20.00% + @xmath5\:\lambda 4363}$ ] & 7886 & 0.53% + @xmath25 & 15563 & 1.05% + @xmath26\:\lambda 4658}$ ] & 2626 & 0.18% + @xmath27 & 4034 & 0.27% + @xmath28\:\lambda 4711}$ ] & 2893 & 0.20% + @xmath28\:\lambda 4740}$ ] & 1074 & 0.07% + @xmath29 & 458324 & 31.02% + @xmath5\:\lambda 4959}$ ] & 147734 & 10.00% + @xmath5\:\lambda 5007}$ ] & 339212 & 22.96% + @xmath30\:\lambda 5197}$ ] & 1318 & 0.09% + @xmath30\:\lambda 5200}$ ] & 758 & 0.05% + @xmath31\:\lambda 5518}$ ] & 160 & 0.01% + @xmath31\:\lambda 5538}$ ] & 190 & 0.01% + @xmath2\:\lambda 5755}$ ] & 4888 & 0.33% + @xmath32 & 75256 & 5.09% + @xmath33\:\lambda 6300}$ ] & 119328 & 8.08% + @xmath34\:\lambda 6312}$ ] & 5552 & 0.38% + @xmath33\:\lambda 6363}$ ] & 16100 & 1.09% + @xmath2\:\lambda 6548}$ ] & 361641 & 24.48% + @xmath35 & 613338 & 41.51% + @xmath2\:\lambda 6584}$ ] & 562659 & 38.08% + @xmath36 & 19084 & 1.29% + @xmath22\:\lambda 6717}$ ] & 410903 & 27.81% + @xmath22\:\lambda 6731}$ ] & 341386 & 23.11% + @xmath37 & 4232 & 0.29% + @xmath38\:\lambda 7135}$ ] & 24034 & 1.63% + @xmath18\:\lambda 7318,19,29,30}$ ] & 659 & 0.04% + our concern in this paper is not to analyse a volume- or magnitude - limited sample of galaxies , we therefore do not impose a redshift cut nor a magnitude limit . since the blue wavelength cut - off of the sdss spectrograph is @xmath39 , the @xmath18\:\lambda3727,3729}$ ] doublet falls outside the spectral range for @xmath40 . this is a concern , because as we will see later 55% of our final sample fall in this region of redshift space and their oxygen abundances are therefore somewhat uncertain . when possible we use the @xmath18\:\lambda7318 - 7330}$ ] quadruplet instead @xcite , but as this is a fairly weak line and falls in a region with significant sky emission , we can not always make use of this line . we select our sample requiring signal to noise ratio @xmath41 in @xmath0 , the resulting data set is given in table [ table5].shirazi / sb011/. ] when the width of the @xmath10ii line is consistent with that of the strong forbidden lines , we make the assumption that it has a nebular origin . given that @xmath10ii lines from individual wr stars typically are considerably broader than the forbidden lines in galaxies ( e.g. b08 ) , we feel this is a reasonable assumption . in addition we require a s / n@xmath42 in each of @xmath43 , @xmath5\:\lambda5007}$ ] , @xmath44 and @xmath2\:\lambda6584}$ ] emission lines to reliably classify our galaxies ( brinchmann et al . 2004 , hereafter b04 ) . the resulting sample contains 2865 spectra with strong nebular @xmath0 emission . in parallel , the spectra of the sample galaxies are examined for wr signatures using the approach discussed by b08 . this resulted in a total sample of 385 spectra with likely and secure wr features ( class 1 , 2 and 3 from b08 ) . while we do not discuss the sample of all wr galaxies in dr7 with class 13 in detail here , we note that it intersects that of the @xmath10ii sample but is not a strict subset of it ( see table [ table1 ] ) . figure [ fig1 ] shows the redshift distribution for the fraction of the @xmath10ii sample in the sdss as a shaded grey histogram . the cut - off at @xmath45 is due to @xmath44 falling outside the spectrograph range . the red histogram shows the redshift distribution for the fraction of just the _ star - forming _ ( sf ) galaxies in the sdss showing @xmath0 emission ( see below for a discussion of the classification ) . for each class we have divided the number of @xmath10ii emitting galaxies in that redshift bin by the number of similarly classified galaxies in the parent sample ( sdss galaxies that have s / n@xmath42 in each of @xmath43 , @xmath5\:\lambda5007}$ ] , @xmath44 and @xmath2\:\lambda6584}$ ] ) in that redshift bin . a constant value would therefore indicate a similar redshift distribution of the sample and the parent sample . it is clear from this that full sample closely follows the overall distribution of the sdss , but that the star - forming galaxies with @xmath0 emission are predominantly found at low redshift . we can also see that less than 2% of all galaxies , and less than 1% of the star - forming galaxies in the sdss dr7 show @xmath0 emission in their spectra . to classify the dominant ionization source in each galaxy we follow previous studies in using the ( * bpt ) line ratio diagnostic diagram of @xmath5\:\lambda5007}/{\ensuremath{\mathrm{h}\beta}}$ ] versus @xmath2\:\lambda6584}/{\ensuremath{\mathrm{h}\alpha}}$ ] as our starting point ( figure [ fig2 ] ) . as has been discussed extensively ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) this diagram allows a separation of agn and star - forming galaxies because of their significantly different ionizing spectra , typically leading to high @xmath5\:\lambda5007}/{\ensuremath{\mathrm{h}\beta}}$ ] and @xmath2\:\lambda6584}/{\ensuremath{\mathrm{h}\alpha}}$ ] when an agn is dominating the output of ionizing photons . ke01 used a combination of stellar population synthesis and photoionization models to compute a theoretical maximum starburst line that isolates objects whose emission line ratios can be accounted for by photoionization by massive stars ( below and to the left of the curve ) from those where some other source of ionization is required . ka03 defined an empirical upper limit to the @xmath11ii region sequence of sdss galaxies in the bpt diagram . the region lying between these two lines represents objects more naturally explained as having a composite spectrum combining @xmath11ii region emission with a harder ionizing source . as we will see later , this interpretation is further corroborated by the @xmath16 ratios we find for our sample . in the present study , we adopt a two - stage classification methodology . we start out by classifying all galaxies using the bpt diagram , and we will then refine our classifications for a subset of the galaxies using detailed inspection of the spectra and the @xmath15 line properties . for the initial classification we adopt a similar methodology to that of b04 and use the separation criteria defined by ke01 and ka03 to divide galaxies into different classes . galaxies which are distributed above the ke01 dividing line are considered agns , galaxies between the ke01 and ka03 limits have composite classification , which means their source of ionizing radiation could be a combination of star formation and agn activity . finally , galaxies below the ka03 line have a sf classification , which as we will see might be modified subsequently . figure [ fig2 ] shows the bpt diagram for our sample , the ke01 and ka03 classification lines are shown as solid and dotted lines , respectively . the distribution of all emission line galaxies in the sdss with @xmath46 in @xmath43 , @xmath5\:\lambda5007}$ ] , @xmath44 and @xmath2\:\lambda6584}$ ] is shown as a grey - scale 2d distribution , where the grey - scale shows the logarithm of the number of galaxies in each bin . blue circles show star - forming galaxies , triangles show composite galaxies and red squares mark agns . the grey dashed - dotted line shows the n2 ( * ? ? ? * hereafter pp04 ) metallicity calibration for @xmath47 ( @xmath2\:\lambda6584}/{\ensuremath{\mathrm{h}\alpha}}=-1.2 $ ] ) . these classifications are the final ones and incorporate further information as described in the following . while the bpt diagram is a useful classification diagram , it is not particularly sensitive to low levels of agn contamination and some progress can be made by including lines originating in the mostly neutral ism ( e.g. * ? ? ? * ) . for our purposes we however need to be very confident in the lack of agns in our sample and we will use the @xmath1 ratio for this purpose . as remarked earlier , only photons with energy in excess of 54.4 ev can ionize he@xmath48 and thence produce the @xmath15 recombination line . if we consider single stellar population models from starburst99 @xcite at an age of 2 myr ( so that all stars are on the main sequence ) , we find that the integrated spectrum of this population typically contain 2 - 3 orders of magnitude fewer photons at this energy than at the energy required to ionize o@xmath48 ( 35.5 ev ) , which is needed to produce the @xmath5\:\lambda5007}$ ] line from collisionally excited o@xmath49 . this line is therefore a very sensitive probe of agn activity , particularly if used in conjunction with other line ratios . we therefore make use of the @xmath1 vs @xmath2\:\lambda6584}/{\ensuremath{\mathrm{h}\alpha}}$ ] diagram to further refine the classification of our sources . we show this diagram in figure [ fig3 ] where the symbols and colours are the same as in figure [ fig2 ] . note that there is generally a very significant offset between the star - forming galaxies and agns in this diagram , in contrast to the gradual transition in much of the bpt diagram . to make a quantitative separation , we draw a random set of star - forming galaxies from the sdss and gradually add their emission line fluxes to that of a random set of agns . we quantify this by the variable @xmath50 which is defined to be the fraction of the total @xmath43 flux comes from the agn . the total flux is therefore @xmath51 . changing @xmath50 will trace out a path in the diagnostic diagram in figure [ fig3 ] as illustrated by the lines with colour gradients in the figure . the colouring of the lines corresponds to @xmath52 , as indicated by the colour bar on the side . we repeat this process for a thousand agn and sf objects located in different bins in the /@xmath43 diagram . based on this analysis we find that there is a well defined locus where 10% percent of the @xmath0 flux comes from an agn ( in this case @xmath53% of the @xmath43 flux comes from the agn , see section [ sec : agn ] for further details ) . a good fit to this relation is given by : @xmath54\:\lambda{6584}}}}{f_{{\ensuremath{\mathrm{h}\alpha } } } } ) + 1.32},\ ] ] which is shown as a dotted line in figure [ fig3 ] . the solid line shows a theoretical maximum starburst line , similar to the ke01 line in the bpt diagram we discuss this further below . finally , we look at the spectra of galaxies that we would classify as star - forming on the basis of their location in the bpt diagram , but that are offset from the rest of the star - forming galaxies in the /@xmath43 diagram and check whether they show agn features such as broad balmer lines , strong @xmath55 , feii emission or if they show a similar @xmath5\:\lambda4363}/{\ensuremath{\mathrm{h}\gamma}}$ ] ratio to that of agns . if some of these features are present , we change the classification from star - forming to agn ( these objects are plotted as red squares containing a blue cross in figures [ fig2 ] and [ fig3 ] ) . this happens for 127 ( 39% ) galaxies . this is a conservative approach as we would exclude star - forming galaxies with strong outflows and hence a broad base to the balmer lines for instance . to estimate the maximum starburst line in figure [ fig3 ] , we adopted the charlot & longhetti ( 2001 , hereafter cl01 ) models . these combine evolving stellar populations models from bruzual & charlot ( unpublished bc00 models using padova 1994 tracks , see bruzual & charlot 2003 ( bc03 ) for the current models ) with the photoionization code cloudy @xcite and adopt the simple dust attenuation prescription of @xcite . the main model parameters for our calculations are the metallicity @xmath56 , the ionization parameter , @xmath57 , the dust attenuation @xmath58 and the dust - to - metal ratio , @xmath59 , the model parameters used are given in table [ table2 ] , see b04 for a more detailed discussion . we use the cl01 single stellar population ( ssp ) models since these achieve the highest possible @xmath1 values . we then identify the maximum ratio reached by the different models and use this upper envelope to define the maximum starburst line shown as a solid line in figure [ fig3 ] . .an overall numerical summary of the @xmath10ii sample . see section [ sec : data ] for a summary of the selection and classification details . see section [ sec : origin ] for details on the wr classification . [ cols="<,<,^ " , ] + & & & & & & & & features & class & other names + 0752 - 52251 - 340 & 0.00291 & + 00 09 53.09 & + 15 44 04.80 & 51 & -1.35 & -1.67 & [email protected] & non - wr & 0 & n / a + 0390 - 51900 - 291 & 0.09437 & + 00 17 28.29 & -00 56 24.98 & 13 & -1.57 & -0.56 & [email protected] & wr & 2 & n / a + 0390 - 51900 - 445 & 0.09840 & + 00 21 01.03 & + 00 52 48.08 & 14 & -1.96 & -1.10 & [email protected] & wr & 2 & um 228 + 0753 - 52233 - 094 & 0.01424 & + 00 24 25.95 & + 14 04 10.65 & 52 & -1.98 & -1.36 & [email protected] & wr & 3 & n / a + 0418 - 51884 - 319 & 0.01791 & + 00 32 18.59 & + 15 00 14.17 & 19 & -1.97 & -1.52 & [email protected] & wr & 2 & shoc 022 + 0691 - 52199 - 389 & 0.19554 & + 00 42 52.31 & + 00 27 30.09 & 48 & -0.75 & -0.82 & [email protected] & wr & 2 & n / a + 1905 - 53706 - 628 & 0.15928 & + 00 49 13.88 & + 00 24 01.99 & 138 & -1.64 & -2.05 & [email protected] & non - wr & 0 & n / a + 0394 - 51913 - 075 & 0.16742 & + 00 55 27.46 & -00 21 48.77 & 15 & -1.76 & -0.89 & [email protected] & wr & 2 & n / a + 0695 - 52202 - 137 & 0.00556 & + 01 15 33.82 & -00 51 31.17 & 49 & -2.32 & -1.27 & [email protected] & wr & 3 & ngc 0450 + 2329 - 53725 - 536 & 0.00975 & + 01 25 34.19 & + 07 59 24.40 & 169 & -2.02 & -2.22 & [email protected] & non - wr & 0 & n / a + 0429 - 51820 - 495 & 0.05662 & + 01 47 07.04 & + 13 56 29.29 & 20 & -1.95 & -1.52 & [email protected] & non - wr & 0 & n / a + 0666 - 52149 - 331 & 0.15155 & + 01 59 53.07 & -08 13 48.99 & 44 & -1.73 & -1.43 & [email protected] & non - wr & 0 & shoc 099 + 1073 - 52649 - 409 & 0.03994 & + 02 13 06.62 & + 00 56 12.44 & 73 & -1.97 & -1.58 & [email protected] & non - wr & 0 & um 411 + 0667 - 52163 - 634 & 0.00492 & + 02 15 13.98 & -08 46 24.39 & 45 & -1.12 & -1.81 & [email protected] & non - wr & 0 & shoc 111 + 0456 - 51910 - 306 & 0.08222 & + 02 40 52.20 & -08 28 27.43 & 24 & -1.60 & -1.63 & [email protected] & non - wr & 0 & shoc 133 + 1070 - 52591 - 072 & 0.00421 & + 02 42 39.86 & -00 00 58.64 & 72 & -1.41 & -0.33 & [email protected] & non - wr & 0 & m077 + 0456 - 51910 - 076 & 0.00456 & + 02 48 15.94 & -08 17 16.51 & 23 & -2.11 & -1.90 & [email protected] & non - wr & 0 & n / a + 1666 - 52991 - 310 & 0.03992 & + 03 14 31.71 & + 41 05 25.95 & 117 & -1.95 & -1.40 & [email protected] & wr & 1 & n / a + 0413 - 51821 - 480 & 0.06868 & + 03 17 43.12 & + 00 19 36.84 & 16 & -0.77 & -0.54 & [email protected] & wr & 2 & n / a + 1733 - 53047 - 528 & 0.07881 & + 07 29 30.29 & + 39 49 41.62 & 121 & -1.97 & -0.93 & [email protected] & non - wr & 0 & n / a + 1922 - 53315 - 588 & 0.06978 & + 08 06 19.50 & + 19 49 27.31 & 139 & -2.05 & -1.19 & [email protected] & wr & 2 & n / a + 0761 - 52266 - 361 & 0.04592 & + 08 22 27.44 & + 42 23 31.14 & 53 & -1.09 & -0.37 & [email protected] & wr & 1 & n / a + 1267 - 52932 - 384 & 0.04722 & + 08 23 54.97 & + 28 06 21.75 & 78 & -2.16 & -0.95 & [email protected] & wr & 1 & n / a + 0548 - 51986 - 503 & 0.00706 & + 08 26 04.80 & + 45 58 07.36 & 30 & -2.07 & -0.76 & [email protected] & wr & 1 & ugc 04393 + 2425 - 54139 - 372 & 0.01974 & + 08 29 32.66 & + 14 27 06.92 & 173 & -2.00 & -1.35 & [email protected] & non - wr & 0 & n / a + 0445 - 51873 - 404 & 0.00245 & + 08 37 43.48 & + 51 38 30.26 & 21 & -2.38 & -1.71 & [email protected] & non - wr & 0 & mrk 0094 + 0828 - 52317 - 148 & 0.14746 & + 08 38 43.64 & + 38 53 50.50 & 56 & -1.88 & -1.32 & [email protected] & wr & 1 & n / a + 2278 - 53711 - 411 & 0.07219 & + 08 40 00.37 & + 18 05 31.01 & 168 & -2.02 & -1.53 & [email protected] & non - wr & 0 & n / a + 0564 - 52224 - 216 & 0.09109 & + 08 44 14.24 & + 02 26 21.10 & 34 & -2.13 & -1.20 & [email protected] & wr & 1 & n / a + 1875 - 54453 - 549 & 0.07498 & + 08 51 03.67 & + 62 13 26.93 & 136 & -2.06 & -1.39 & [email protected] & non - wr & 0 & n / a + 1785 - 54439 - 201 & 0.09186 & + 08 51 15.65 & + 58 40 55.02 & 130 & -2.01 & -1.83 & [email protected] & non - wr & 0 & n / a + 2430 - 53815 - 117 & 0.07592 & + 08 52 21.72 & + 12 16 51.76 & 174 & -2.06 & -1.45 & [email protected] & non - wr & 0 & n / a + 0551 - 51993 - 279 & 0.00961 & + 08 52 58.21 & + 49 27 33.91 & 31 & -1.83 & -0.62 & [email protected] & wr & 3 & sbs 0849 + 496 + 2086 - 53401 - 458 & 0.00981 & + 09 05 26.34 & + 25 33 02.57 & 154 & -1.93 & -1.12 & [email protected] & wr & 3 & ugc 04764 + 0566 - 52238 - 497 & 0.03909 & + 09 05 31.08 & + 03 35 30.38 & 35 & -1.93 & -1.78 & [email protected] & non - wr & 0 & n / a + 1194 - 52703 - 397 & 0.00513 & + 09 10 28.78 & + 07 11 17.97 & 76 & -2.02 & -1.77 & [email protected] & wr & 1 & n / a + 0899 - 52620 - 594 & 0.02727 & + 09 14 34.95 & + 47 02 07.24 & 60 & -2.13 & -1.55 & [email protected] & non - wr & 0 & sbs 0911 + 472 + 0554 - 52000 - 190 & 0.00767 & + 09 20 55.92 & + 52 34 07.34 & 33 & -2.10 & -1.83 & [email protected] & non - wr & 0 & n / a + 0553 - 51999 - 602 & 0.00772 & + 09 20 56.07 & + 52 34 04.32 & 32 & -1.71 & -1.63 & [email protected] & wr & 1 & n / a + 0485 - 51909 - 550 & 0.01366 & + 09 30 06.43 & + 60 26 53.40 & 25 & -2.17 & -1.59 & [email protected] & non - wr & 0 & sbs 0926 + 606a + 2580 - 54092 - 470 & 0.10060 & + 09 38 01.64 & + 13 53 17.07 & 186 & -1.18 & -0.43 & [email protected] & wr & 2 & n / a + 1594 - 52992 - 563 & 0.01486 & + 09 42 52.78 & + 35 47 25.98 & 110 & -1.92 & -1.44 & [email protected] & non - wr & 0 & n / a + 1305 - 52757 - 269 & 0.01085 & + 09 42 56.74 & + 09 28 16.26 & 80 & -2.38 & -1.51 & [email protected] & wr & 3 & ugc 05189 + 0266 - 51630 - 100 & 0.00478 & + 09 44 01.87 & -00 38 32.18 & 2 & -1.88 & -1.98 & [email protected] & non - wr & 0 & cgcg 007 - 025 + 1947 - 53431 - 448 & 0.01730 & + 09 50 00.77 & + 30 03 41.04 & 140 & -1.97 & -1.51 & [email protected] & non - wr & 0 & n / a + 0769 - 54530 - 086 & 0.04626 & + 09 51 31.77 & + 52 59 36.05 & 54 & -2.05 & -1.64 & [email protected] & wr & 2 & sbs 0948 + 532 + 1306 - 52996 - 005 & 0.00486 & + 09 54 49.56 & + 09 16 15.94 & 81 & -1.90 & -0.42 & [email protected] & wr & 3 & ngc 3049 + 2583 - 54095 - 062 & 0.01522 & + 09 56 42.49 & + 15 38 11.34 & 187 & -2.02 & -0.91 & [email protected] & wr & 3 & ugc 05342 + 2364 - 53737 - 618 & 0.00423 & + 10 10 32.81 & + 22 00 39.63 & 171 & -2.12 & -1.91 & [email protected] & wr & 1 & was 05 + 1745 - 53061 - 475 & 0.06131 & + 10 10 42.54 & + 12 55 16.81 & 123 & -1.92 & -1.40 & [email protected] & wr & 3 & n / a + 2588 - 54174 - 369 & 0.05563 & + 10 10 59.30 & + 15 42 23.53 & 188 & -1.92 & -1.70 & [email protected] & non - wr & 0 & n / a + 1745 - 53061 - 196 & 0.00957 & + 10 12 27.02 & + 12 20 37.50 & 122 & -2.24 & -1.77 & [email protected] & wr & 2 & n / a + 1955 - 53442 - 354 & 0.03762 & + 10 14 10.58 & + 34 20 34.79 & 141 & -1.66 & -0.54 & [email protected] & wr & 1 & kug 1011 + 345 + 1427 - 52996 - 221 & 0.00388 & + 10 16 24.52 & + 37 54 45.97 & 101 & -1.70 & -2.21 & [email protected] & non - wr & 0 & n / a + 2366 - 53741 - 124 & 0.01897 & + 10 24 02.75 & + 21 04 50.04 & 172 & -1.88 & -1.56 & [email protected] & wr & 2 & lsbc d568 - 03 + 0575 - 52319 - 521 & 0.03319 & + 10 24 29.25 & + 05 24 51.02 & 36 & -2.07 & -1.72 & [email protected] & non - wr & 0 & n / a + 2591 - 54140 - 222 & 0.04432 & + 10 26 23.65 & + 17 10 14.36 & 189 & -1.12 & -0.64 & [email protected] & non - wr & 0 & n / a + 0999 - 52636 - 517 & 0.04453 & + 10 33 28.53 & + 07 08 01.76 & 69 & -1.99 & -0.77 & [email protected] & wr & 2 & cgcg 037 - 076 + 0947 - 52411 - 569 & 0.00747 & + 10 34 10.15 & + 58 03 49.06 & 62 & -1.60 & -1.97 & [email protected] & non - wr & 0 & mrk 1434 + 0875 - 52354 - 226 & 0.02850 & + 10 35 08.88 & + 49 21 42.47 & 58 & -1.79 & -0.76 & [email protected] & wr & 2 & sbs 1032 + 496 + 1998 - 53433 - 304 & 0.00489 & + 10 36 13.22 & + 37 19 27.57 & 146 & -1.93 & -0.53 & [email protected] & wr & 3 & ngc 3294 + 0274 - 51913 - 187 & 0.01856 & + 10 39 24.38 & -00 23 21.44 & 3 & -2.16 & -0.85 & [email protected] & wr & 1 & ic 0633 + 2478 - 54097 - 370 & 0.00398 & + 10 41 09.60 & + 21 21 42.80 & 176 & -2.13 & -1.81 & [email protected] & wr & 3 & mrk 0724 + 0578 - 52339 - 060 & 0.01287 & + 10 44 57.79 & + 03 53 13.15 & 37 & -1.80 & -2.55 & [email protected] & non - wr & 0 & n / a + 2147 - 53491 - 514 & 0.05487 & + 10 45 20.42 & + 09 23 49.10 & 160 & -2.18 & -1.30 & [email protected] & wr & 2 & schg 1042 + 097 + 0275 - 51910 - 445 & 0.02620 & + 10 45 54.78 & + 01 04 05.84 & 4 & -2.28 & -1.34 & [email protected] & non - wr & 0 & shoc 308 + 1749 - 53357 - 499 & 0.01061 & + 10 46 53.99 & + 13 46 45.77 & 124 & -2.23 & -1.80 & [email protected] & non - wr & 0 & n / a + 1981 - 53463 - 438 & 0.02945 & + 10 47 23.61 & + 30 21 44.29 & 143 & -2.18 & -1.15 & [email protected] & wr & 3 & ton 0542 + 2483 - 53852 - 254 & 0.08443 & + 10 50 32.51 & + 15 38 06.31 & 177 & -1.79 & -1.76 & [email protected] & non - wr & 0 & n / a + 0275 - 51910 - 622 & 0.05061 & + 10 50 46.59 & + 00 36 40.11 & 5 & -1.44 & -0.46 & [email protected] & wr & 2 & n / a + 0876 - 52669 - 175 & 0.00435 & + 10 53 10.82 & + 50 16 53.21 & 59 & -1.70 & -1.75 & [email protected] & non - wr & 0 & n / a + 2359 - 53826 - 205 & 0.00452 & + 10 54 21.87 & + 27 14 22.16 & 170 & -1.93 & -0.74 & [email protected] & wr & 3 & ngc 3451 + 1362 - 53050 - 617 & 0.03745 & + 11 00 24.90 & + 43 01 11.93 & 91 & -2.11 & -1.29 & [email protected] & non - wr & 0 & n / a + 2213 - 53792 - 359 & 0.00214 & + 11 04 58.30 & + 29 08 16.55 & 164 & -1.73 & -1.84 & [email protected] & non - wr & 0 & n / a + 2211 - 53786 - 486 & 0.00214 & + 11 04 58.54 & + 29 08 15.72 & 163 & -2.18 & -1.83 & [email protected] & non - wr & 0 & n / a + 1363 - 53053 - 510 & 0.02154 & + 11 05 08.12 & + 44 44 47.24 & 92 & -2.05 & -1.28 & [email protected] & wr & 2 & mrk 0162 + 2494 - 54174 - 361 & 0.00492 & + 11 17 46.30 & + 17 44 24.69 & 178 & -2.04 & -1.69 & [email protected] & non - wr & 0 & n / a + 1223 - 52781 - 128 & 0.00345 & + 11 27 10.93 & + 08 43 51.70 & 77 & -1.71 & -1.27 & [email protected] & wr & 3 & ic 2828 + 1014 - 52707 - 254 & 0.00988 & + 11 27 32.67 & + 53 54 54.47 & 70 & -1.98 & -1.62 & [email protected] & non - wr & 0 & mrk 1446 + 2500 - 54178 - 084 & 0.00470 & + 11 29 14.15 & + 20 34 52.01 & 179 & -2.32 & -1.44 & [email protected] & wr & 2 & ic 0700 + 1754 - 53385 - 151 & 0.01764 & + 11 32 35.35 & + 14 11 29.83 & 125 & -2.19 & -1.48 & [email protected] & wr & 2 & n / a + 0967 - 52636 - 302 & 0.00084 & + 11 33 28.95 & + 49 14 13.01 & 64 & -1.22 & -1.76 & [email protected] & wr & 3 & mrk 0178 + 0967 - 52636 - 339 & 0.02601 & + 11 34 45.72 & + 50 06 03.33 & 65 & -2.25 & -1.41 & [email protected] & wr & 2 & mrk 1448 + 1442 - 53050 - 599 & 0.01018 & + 11 36 23.82 & + 47 09 29.08 & 102 & -2.07 & -1.92 & [email protected] & non - wr & 0 & n / a + 2012 - 53493 - 407 & 0.00483 & + 11 36 39.57 & + 36 23 42.89 & 150 & -1.82 & -1.52 & [email protected] & wr & 1 & ngc 3755 + 2506 - 54179 - 357 & 0.02082 & + 11 36 54.01 & + 19 55 34.80 & 180 & -1.77 & -0.77 & [email protected] & wr & 3 & mrk 0182 + 2008 - 53473 - 467 & 0.00601 & + 11 41 07.49 & + 32 25 37.22 & 149 & -1.82 & -1.82 & [email protected] & non - wr & 0 & kug 1138 + 327 + 0967 - 52636 - 540 & 0.00558 & + 11 45 06.26 & + 50 18 02.44 & 66 & -1.55 & -2.04 & [email protected] & non - wr & 0 & n / a + 2513 - 54141 - 309 & 0.02676 & + 11 48 05.45 & + 21 49 45.35 & 183 & -1.95 & -1.27 & [email protected] & wr & 2 & mrk 1459 + 2510 - 53877 - 560 & 0.04512 & + 11 48 27.34 & + 25 46 11.77 & 182 & -2.10 & -1.70 & [email protected] & non - wr & 0 & n / a + 2508 - 53875 - 615 & 0.07911 & + 11 48 40.87 & + 17 56 33.02 & 181 & -2.00 & -1.78 & [email protected] & non - wr & 0 & n / a + 1761 - 53376 - 636 & 0.00245 & + 11 50 02.73 & + 15 01 23.48 & 126 & -2.23 & -1.59 & [email protected] & wr & 3 & mrk 0750 + 0330 - 52370 - 471 & 0.00353 & + 11 52 37.68 & -02 28 06.39 & 8 & -2.13 & -1.75 & [email protected] & non - wr & 0 & n / a + 1446 - 53080 - 134 & 0.00337 & + 11 54 41.22 & + 46 36 36.35 & 103 & -1.81 & -1.68 & [email protected] & non - wr & 0 & n / a + 1313 - 52790 - 423 & 0.01726 & + 11 55 28.34 & + 57 39 51.97 & 82 & -2.20 & -1.77 & [email protected] & non - wr & 0 & mrk 0193 + 0516 - 52017 - 315 & 0.05811 & + 11 57 12.45 & + 02 28 27.88 & 27 & -1.99 & -1.15 & [email protected] & wr & 1 & um 469 + 1991 - 53446 - 584 & 0.01097 & + 11 57 31.73 & + 32 20 30.17 & 145 & -2.14 & -1.13 & [email protected] & wr & 3 & ngc 3991n + 2226 - 53819 - 157 & 0.08188 & + 12 00 16.49 & + 27 19 59.01 & 165 & -1.93 & -1.86 & [email protected] & non - wr & 0 & n / a + 1763 - 53463 - 094 & 0.06675 & + 12 00 33.42 & + 13 43 07.99 & 127 & -2.05 & -1.08 & [email protected] & wr & 2 & n / a + 2227 - 53820 - 389 & 0.05588 & + 12 01 49.90 & + 28 06 10.67 & 166 & -1.92 & -1.71 & [email protected] & non - wr & 0 & n / a + 0517 - 52024 - 504 & 0.00429 & + 12 08 11.11 & + 02 52 41.82 & 28 & -1.84 & -0.27 & [email protected] & wr & 2 & ngc 4123 + 2004 - 53737 - 439 & 0.04894 & + 12 09 24.64 & + 32 44 02.05 & 148 & -1.44 & -0.43 & [email protected] & wr & 1 & n / a + 2644 - 54210 - 188 & 0.02435 & + 12 09 27.95 & + 22 06 16.69 & 193 & -1.91 & -0.64 & [email protected] & wr & 2 & ugc 07137 + 2610 - 54476 - 421 & 0.00277 & + 12 15 18.60 & + 20 38 26.72 & 191 & -2.59 & -1.80 & [email protected] & wr & 2 & mrk 1315 + 1625 - 53140 - 386 & 0.00864 & + 12 16 47.89 & + 08 02 56.28 & 112 & -1.81 & -1.44 & [email protected] & non - wr & 0 & vcc 0207 + 2001 - 53493 - 146 & 0.00061 & + 12 17 49.31 & + 37 51 55.50 & 147 & -2.28 & -1.72 & [email protected] & non - wr & 0 & n / a + 2880 - 54509 - 277 & 0.00428 & + 12 22 25.79 & + 04 34 04.77 & 198 & -2.28 & -1.33 & [email protected] & non - wr & 0 & n / a + 0955 - 52409 - 608 & 0.00234 & + 12 25 05.41 & + 61 09 11.29 & 63 & -1.97 & -1.96 & [email protected] & wr & 3 & sbs 1222 + 614 + 2880 - 54509 - 095 & 0.09423 & + 12 26 11.90 & + 04 15 36.06 & 197 & -2.00 & -1.66 & [email protected] & non - wr & 0 & n / a + 1453 - 53084 - 322 & 0.00100 & + 12 26 15.69 & + 48 29 38.43 & 106 & -2.27 & -2.11 & [email protected] & non - wr & 0 & ugca 281 + 1371 - 52821 - 053 & 0.00071 & + 12 28 09.26 & + 44 05 08.02 & 93 & -1.67 & -1.09 & [email protected] & wr & 1 & ngc 4449 + 1371 - 52821 - 059 & 0.00072 & + 12 28 13.86 & + 44 07 10.43 & 94 & -2.50 & -1.95 & [email protected] & wr & 3 & ngc 4449 + 1452 - 53112 - 011 & 0.00157 & + 12 30 28.33 & + 41 41 22.07 & 104 & -2.33 & -1.23 & [email protected] & wr & 3 & ngc 4485 + 1452 - 53112 - 016 & 0.00192 & + 12 30 38.45 & + 41 39 11.31 & 105 & -2.14 & -1.31 & [email protected] & wr & 3 & ngc 4490 + 1615 - 53166 - 120 & 0.00418 & + 12 30 48.60 & + 12 02 42.82 & 111 & -1.44 & -2.04 & [email protected] & wr & 1 & n / a + 1768 - 53442 - 476 & 0.00697 & + 12 31 54.67 & + 15 07 36.48 & 128 & -1.82 & -0.68 & [email protected] & wr & 3 & ic 0797 + 2613 - 54481 - 507 & 0.04855 & + 12 38 29.93 & + 19 59 21.36 & 192 & -1.90 & -0.99 & [email protected] & wr & 2 & n / a + 0494 - 51915 - 007 & 0.08778 & + 12 40 49.89 & + 66 24 20.17 & 26 & -1.88 & -1.49 & [email protected] & non - wr & 0 & shoc 379 + 1372 - 53062 - 072 & 0.04190 & + 12 41 34.25 & + 44 26 39.24 & 95 & -0.19 & -0.93 & [email protected] & non - wr & 0 & n / a + 1975 - 53734 - 498 & 0.00198 & + 12 43 56.70 & + 32 10 14.67 & 142 & -1.77 & -1.88 & [email protected] & wr & 3 & ngc 4656 + 2236 - 53729 - 038 & 0.00350 & + 12 45 16.87 & + 27 07 30.78 & 167 & -1.83 & -1.15 & [email protected] & wr & 3 & ngc 4670 + 1455 - 53089 - 556 & 0.02377 & + 12 48 46.36 & + 47 42 53.45 & 107 & -2.14 & -1.48 & [email protected] & wr & 2 & n / a + 1989 - 53772 - 089 & 0.02782 & + 12 53 06.56 & + 36 49 11.41 & 144 & -1.85 & -0.52 & [email protected] & wr & 3 & ngc 4774 + 2461 - 54570 - 089 & 0.00797 & + 12 54 23.74 & + 58 53 40.67 & 175 & -2.03 & -1.15 & [email protected] & wr & 3 & n / a + 0602 - 52072 - 369 & 0.02766 & + 13 02 49.20 & + 65 34 49.27 & 39 & -1.92 & -1.57 & [email protected] & wr & 1 & n / a + 2018 - 53800 - 096 & 0.03595 & + 13 03 54.44 & + 37 14 01.88 & 152 & -2.11 & -1.39 & [email protected] & non - wr & 0 & n / a + 0339 - 51692 - 083 & 0.00450 & + 13 04 32.27 & -03 33 22.12 & 9 & -2.26 & -1.36 & [email protected] & wr & 3 & ugca 322 + 0782 - 52320 - 022 & 0.11183 & + 13 04 45.63 & + 62 24 20.88 & 55 & -1.84 & -0.95 & [email protected] & wr & 2 & n / a + 2016 - 53799 - 185 & 0.01620 & + 13 06 24.19 & + 35 13 43.04 & 151 & -2.08 & -1.71 & [email protected] & non - wr & 0 & n / a + 0602 - 52072 - 019 & 0.03678 & + 13 14 26.56 & + 63 33 11.37 & 38 & -1.62 & -1.15 & [email protected] & wr & 3 & n / a + 2023 - 53851 - 263 & 0.00288 & + 13 14 47.37 & + 34 52 59.81 & 153 & -2.37 & -1.45 & [email protected] & wr & 3 & ugc 08323 + 0526 - 52312 - 097 & 0.18400 & + 13 22 11.96 & + 01 30 34.39 & 29 & -2.04 & -1.00 & [email protected] & wr & 2 & f13196 + 0146 + 0341 - 51690 - 606 & 0.02246 & + 13 23 47.46 & -01 32 51.95 & 10 & -1.88 & -2.48 & [email protected] & non - wr & 0 & um 570 + 1282 - 52759 - 057 & 0.01637 & + 13 25 19.89 & + 48 02 26.15 & 79 & -1.93 & -1.64 & [email protected] & non - wr & 0 & sbs 1323 + 483 + 2112 - 53534 - 557 & 0.01462 & + 13 25 49.42 & + 33 03 54.38 & 157 & -2.05 & -1.57 & [email protected] & wr & 2 & was 69 + 1376 - 53089 - 637 & 0.02797 & + 13 28 44.05 & + 43 55 50.51 & 96 & -2.16 & -1.28 & [email protected] & wr & 2 & mrk 0259 + 2606 - 54154 - 474 & 0.09425 & + 13 29 16.56 & + 17 00 21.00 & 190 & -1.92 & -1.13 & [email protected] & wr & 3 & n / a + 2110 - 53467 - 499 & 0.01613 & + 13 30 17.38 & + 31 19 58.02 & 156 & -2.14 & -1.41 & [email protected] & non - wr & 0 & ugc 08496 + 1464 - 53091 - 370 & 0.01170 & + 13 31 26.91 & + 41 51 48.29 & 108 & -2.13 & -2.12 & [email protected] & non - wr & 0 & n / a + 1801 - 54156 - 583 & 0.14730 & + 13 39 24.24 & + 07 39 27.61 & 131 & -1.80 & -0.82 & [email protected] & wr & 1 & n / a + 2094 - 53851 - 487 & 0.00320 & + 13 41 56.48 & + 30 31 09.62 & 155 & -2.19 & -1.73 & [email protected] & wr & 2 & mrk 0067c3 + 1043 - 52465 - 308 & 0.00594 & + 13 42 51.85 & + 52 42 30.57 & 71 & -1.79 & -1.53 & [email protected] & wr & 1 & mrk 1480 + 1803 - 54152 - 448 & 0.05450 & + 13 44 24.06 & + 07 45 00.08 & 132 & -1.78 & -1.00 & [email protected] & wr & 2 & n / a + 0854 - 52373 - 514 & 0.03041 & + 13 45 31.50 & + 04 42 32.71 & 57 & -1.97 & -1.55 & [email protected] & wr & 1 & tololo 1343 + 049 + 1776 - 53858 - 632 & 0.02156 & + 13 46 49.45 & + 14 24 01.68 & 129 & -1.75 & -0.42 & [email protected] & wr & 2 & mrk 0796 + 1158 - 52668 - 062 & 0.03383 & + 13 59 50.92 & + 57 26 22.98 & 75 & -1.89 & -1.42 & [email protected] & non - wr & 0 & mrk 1486 + 2770 - 54510 - 583 & 0.02775 & + 14 01 07.12 & + 21 14 34.60 & 195 & -2.14 & -1.20 & [email protected] & wr & 1 & ugc 08929 + 1324 - 53088 - 271 & 0.00058 & + 14 02 28.23 & + 54 16 33.08 & 86 & -2.12 & -0.89 & [email protected] & wr & 3 & ngc 5447 + 1378 - 53061 - 023 & 0.00442 & + 14 02 36.07 & + 39 13 13.28 & 97 & -1.92 & -1.46 & [email protected] & wr & 2 & n / a + 1323 - 52797 - 002 & 0.00056 & + 14 03 01.17 & + 54 14 29.40 & 83 & -2.59 & -1.75 & [email protected] & wr & 3 & n / a + 1323 - 52797 - 014 & 0.00087 & + 14 03 34.06 & + 54 18 36.91 & 84 & -2.25 & -0.86 & [email protected] & wr & 3 & ngc 5461 + 1325 - 52762 - 356 & 0.00085 & + 14 03 39.84 & + 54 18 56.87 & 88 & -2.29 & -0.67 & [email protected] & wr & 3 & ngc 5461 + 1325 - 52762 - 350 & 0.00093 & + 14 04 11.24 & + 54 25 18.67 & 87 & -2.40 & -1.48 & [email protected] & non - wr & 0 & n / a + 1642 - 53115 - 155 & 0.01194 & + 14 04 14.87 & + 36 43 32.67 & 113 & -2.14 & -0.88 & [email protected] & wr & 3 & mrk 1369 + 1324 - 53088 - 234 & 0.00097 & + 14 04 28.63 & + 54 23 52.80 & 85 & -2.41 & -1.30 & [email protected] & non - wr & 0 & n / a + 1325 - 52762 - 412 & 0.07731 & + 14 09 56.76 & + 54 56 48.89 & 89 & -2.00 & -1.32 & [email protected] & wr & 3 & sbs 1408 + 551a + 2786 - 54540 - 084 & 0.00855 & + 14 18 51.13 & + 21 02 39.74 & 196 & -1.97 & -2.24 & [email protected] & non - wr & 0 & n / a + 2746 - 54232 - 104 & 0.00771 & + 14 23 48.53 & + 14 38 16.54 & 194 & -2.04 & -0.48 & [email protected] & wr & 3 & n / a + 1381 - 53089 - 470 & 0.02230 & + 14 26 28.17 & + 38 22 58.67 & 98 & -1.87 & -1.96 & [email protected] & non - wr & 0 & n / a + 1644 - 53144 - 564 & 0.08624 & + 14 28 05.51 & + 36 27 10.40 & 114 & -1.93 & -1.60 & [email protected] & non - wr & 0 & n / a + 1827 - 53531 - 503 & 0.17350 & + 14 29 47.01 & + 06 43 34.97 & 133 & -1.80 & -1.17 & [email protected] & non - wr & 0 & n / a + 0305 - 51613 - 604 & 0.01341 & + 14 30 53.51 & + 00 27 46.35 & 6 & -2.28 & -1.67 & [email protected] & wr & 1 & n / a + 2137 - 54206 - 310 & 0.01502 & + 14 31 08.88 & + 27 14 12.29 & 158 & -2.15 & -1.20 & [email protected] & non - wr & 0 & mrk 0685 + 1709 - 53533 - 215 & 0.00466 & + 14 32 48.36 & + 09 52 57.15 & 119 & -2.31 & -1.42 & [email protected] & wr & 2 & ngc 5669 + 0920 - 52411 - 575 & 0.02740 & + 14 48 05.38 & -01 10 57.72 & 61 & -2.02 & -1.73 & [email protected] & wr & 3 & shoc 486 + 1646 - 53498 - 616 & 0.00267 & + 14 48 52.02 & + 34 42 42.99 & 115 & -1.48 & -1.66 & [email protected] & non - wr & 0 & ugc 09540 + 1383 - 53116 - 110 & 0.00396 & + 14 50 56.56 & + 35 34 19.59 & 99 & -2.24 & -1.41 & [email protected] & wr & 1 & n / a + 2145 - 54212 - 388 & 0.08403 & + 14 51 33.55 & + 26 46 03.56 & 159 & -1.51 & -0.67 & [email protected] & wr & 1 & n / a + 1843 - 53816 - 087 & 0.00942 & + 14 54 12.15 & + 30 12 36.25 & 134 & -2.38 & -1.52 & [email protected] & non - wr & 0 & n / a + 1844 - 54138 - 311 & 0.00610 & + 14 56 36.63 & + 30 13 52.36 & 135 & -2.31 & -1.50 & [email protected] & wr & 2 & n / a + 1399 - 53172 - 299 & 0.03255 & + 15 09 34.18 & + 37 31 46.11 & 100 & -2.01 & -2.09 & [email protected] & non - wr & 0 & n / a + 2911 - 54631 - 344 & 0.04684 & + 15 19 47.15 & + 39 45 37.85 & 199 & -1.98 & -0.63 & [email protected] & wr & 2 & n / a + 1651 - 53442 - 255 & 0.06816 & + 15 23 32.19 & + 29 31 12.08 & 116 & -1.96 & -1.79 & [email protected] & non - wr & 0 & n / a + 1679 - 53149 - 384 & 0.00826 & + 15 26 30.31 & + 41 17 22.34 & 118 & -1.83 & -1.04 & [email protected] & wr & 3 & ugc 09856 + 2163 - 53823 - 546 & 0.03395 & + 15 34 56.40 & + 24 51 39.24 & 161 & -1.68 & -0.64 & [email protected] & wr & 2 & n / a + 0616 - 52442 - 364 & 0.00225 & + 15 37 04.18 & + 55 15 50.62 & 40 & -2.46 & -1.33 & [email protected] & non - wr & 0 & n / a + 1725 - 54266 - 068 & 0.03772 & + 15 45 43.55 & + 08 58 01.35 & 120 & -1.69 & -2.01 & [email protected] & non - wr & 0 & n / a + 2167 - 53889 - 071 & 0.01098 & + 15 46 58.88 & + 17 53 03.07 & 162 & -1.72 & -0.40 & [email protected] & wr & 3 & ngc 5996 + 2524 - 54568 - 146 & 0.10704 & + 16 06 27.54 & + 13 55 47.88 & 184 & -1.99 & -1.28 & [email protected] & wr & 1 & n / a + 2527 - 54569 - 147 & 0.01217 & + 16 15 17.02 & + 13 01 33.08 & 185 & -1.96 & -1.57 & [email protected] & non - wr & 0 & n / a + 0624 - 52377 - 361 & 0.00237 & + 16 16 23.54 & + 47 02 02.32 & 42 & -1.56 & -1.91 & [email protected] & wr & 3 & arp 2 + 0364 - 52000 - 187 & 0.03133 & + 16 24 10.11 & -00 22 02.58 & 12 & -2.11 & -1.58 & [email protected] & wr & 3 & shoc 536 + 0624 - 52377 - 092 & 0.02993 & + 16 26 04.26 & + 46 22 05.79 & 41 & -1.36 & -0.39 & [email protected] & wr & 2 & n / a + 1570 - 53149 - 021 & 0.00910 & + 16 47 10.66 & + 21 05 14.51 & 109 & -1.83 & -2.06 & [email protected] & non - wr & 0 & n / a + 1342 - 52793 - 112 & 0.03205 & + 16 49 05.27 & + 29 45 31.61 & 90 & -1.52 & -0.39 & [email protected] & wr & 1 & kug 1647 + 298 + 0976 - 52413 - 600 & 0.01195 & + 17 12 36.63 & + 32 16 33.42 & 67 & -1.99 & -1.96 & [email protected] & wr & 1 & n / a + 0978 - 52441 - 118 & 0.01483 & + 17 18 53.45 & + 30 11 36.20 & 68 & -1.92 & -0.84 & [email protected] & wr & 2 & iras 17169 + 3014 + 0358 - 51818 - 504 & 0.04723 & + 17 35 01.25 & + 57 03 08.55 & 11 & -2.17 & -1.31 & [email protected] & wr & 2 & shoc 579 + 1115 - 52914 - 309 & 0.01381 & + 20 47 59.21 & -00 10 53.98 & 74 & -2.18 & -1.17 & [email protected] & wr & 1 & n / a + 0673 - 52162 - 312 & 0.06669 & + 22 25 10.13 & -00 11 52.84 & 46 & -1.85 & -1.81 & [email protected] & non - wr & 0 & n / a + 1893 - 53239 - 476 & 0.02061 & + 22 38 31.12 & + 14 00 29.78 & 137 & -2.16 & -2.22 & [email protected] & non - wr & 0 & n / a + 0742 - 52263 - 179 & 0.03029 & + 23 01 23.59 & + 13 33 14.79 & 50 & -2.06 & -1.68 & [email protected] & non - wr & 0 & n / a + 0677 - 52606 - 533 & 0.03312 & + 23 02 10.00 & + 00 49 38.84 & 47 & -1.83 & -2.21 & [email protected] & non - wr & 0 & n / a + 0650 - 52143 - 330 & 0.03592 & + 23 56 21.96 & -09 04 07.42 & 43 & -1.51 & -1.43 & [email protected] & wr & 1 & n / a +

random interlacements ( at level ) is a one parameter family of random subsets of introduced by sznitman in . the vacant set at level is the complement of the random interlacement at level . while the random interlacement induces a connected subgraph of for all levels , the vacant set has a non - trivial phase transition in , as shown in and . in this paper , we study the effect of small quenched noise on connectivity properties of the random interlacement and the vacant set . for a positive , we allow each vertex of the random interlacement ( referred to as occupied ) to become vacant , and each vertex of the vacant set to become occupied with probability , independently of the randomness of the interlacement , and independently for different vertices . we prove that for any and , almost surely , the perturbed random interlacement percolates for small enough noise parameter . in fact , we prove the stronger statement that bernoulli percolation on the random interlacement graph has a non - trivial phase transition in wide enough slabs . as a byproduct , we show that any electric network with i.i.d . positive resistances on the interlacement graph is transient , which strengthens our result in . as for the vacant set , we show that for any , there is still a non - trivial phase transition in when the noise parameter is small enough , and we give explicit upper and lower bounds on the value of the critical threshold , when .

1109.5086

the model of random interlacements was recently introduced by sznitman in @xcite in order to describe the local picture left by the trajectory of a random walk on the discrete torus @xmath6 , @xmath7 when it runs up to times of order @xmath8 , or on the discrete cylinder @xmath9 , @xmath10 , when it runs up to times of order @xmath11 , see @xcite , @xcite . informally , the random interlacement poisson point process consists of a countable collection of doubly infinite trajectories on @xmath1 , and the trace left by these trajectories on a finite subset of @xmath1 `` looks like '' the trace of the above mentioned random walks . the set of vertices visited by at least one of these trajectories is the random interlacement at level @xmath0 of sznitman @xcite , and the complement of this set is the vacant set at level @xmath0 . these are one parameter families of translation invariant , ergodic , long - range correlated random subsets of @xmath1 , see @xcite . we call the vertices of the random interlacement occupied , and the vertices of the vacant set vacant . while the set of occupied vertices induces a connected subgraph of @xmath1 for all levels @xmath0 , the graph induced by the set of vacant vertices has a non - trivial phase transition in @xmath0 , as shown in @xcite and @xcite . the effect of introducing a small amount of quenched disorder into a system with long - range correlations on the phase transition has got a lot of attention ( see , e.g. , @xcite , @xcite , @xcite , @xcite ) . in this paper we consider how small quenched disorder affects the connectivity properties of the random interlacement and the vacant set . for @xmath12 , given a realization of the random interlacement , we allow each vertex independently to switch from occupied to vacant and from vacant to occupied with probability @xmath2 , and we study the effect it has on the existence of an infinite connected component in the graphs of occupied or vacant vertices . we prove that for any @xmath3 and @xmath4 , almost surely , the set of occupied vertices percolates for small enough noise parameter @xmath2 . in fact , we prove the stronger statement that bernoulli percolation on the random interlacement graph has a non - trivial phase transition in wide enough slabs . the two main ingredients of our proof are a strong connectivity lemma for the interlacement graph proved in @xcite and sznitman s decoupling inequalities from @xcite . as a byproduct , we show that any electric network with i.i.d . positive resistances on the interlacement graph is transient , which strengthens our result in @xcite . we also prove that for any @xmath3 , the set of vacant vertices still undergoes a non - trivial phase transition in @xmath0 when the noise parameter @xmath2 is small enough , and give explicit upper and lower bounds on the value of the threshold , when @xmath5 . the bounds that we derive suggest that the vacant set phase transition is robust with respect to noise , which we state as a conjecture . for @xmath13 , @xmath3 , let @xmath14 be the law of a simple random walk @xmath15 on @xmath1 with @xmath16 . let @xmath17 be a finite subset of @xmath1 . the equilibrium measure of @xmath17 is defined by @xmath18 , \quad \mbox{for } x\in k , \ ] ] and @xmath19 for @xmath20 . the capacity of @xmath17 is the total mass of the equilibrium measure of @xmath17 : @xmath21 since @xmath3 , for any finite set @xmath22 , the capacity of @xmath17 is positive . therefore , we can define the normalized equilibrium measure by @xmath23 let @xmath24 be the space of doubly - infinite nearest - neighbor trajectories in @xmath1 ( @xmath3 ) which tend to infinity at positive and negative infinite times , and let @xmath25 be the space of equivalence classes of trajectories in @xmath24 modulo time - shift . we write @xmath26 for the canonical @xmath27-algebra on @xmath24 generated by the coordinate maps , and @xmath28 for the largest @xmath27-algebra on @xmath25 for which the canonical map @xmath29 from @xmath30 to @xmath31 is measurable . let @xmath32 be a poisson point measure on @xmath25 . for a finite subset @xmath17 of @xmath1 , denote by @xmath33 the restriction of @xmath32 to the set of trajectories from @xmath25 that intersect @xmath17 , and by @xmath34 be the number of trajectories in @xmath35 . the point measure @xmath33 can be written as @xmath36 , where @xmath37 are doubly - infinite trajectories from @xmath24 parametrized in such a way that @xmath38 and @xmath39 for all @xmath40 and for all @xmath41 . for @xmath4 , we say that a poisson point measure @xmath32 on @xmath25 has distribution @xmath42 if the following properties hold : * [ distr:1 ] the random variable @xmath34 has poisson distribution with parameter @xmath43 . * [ distr:2 ] given @xmath34 , the points @xmath44 , @xmath41 , are independent and distributed according to the normalized equilibrium measure on @xmath17 . * [ distr:3 ] given @xmath34 and @xmath45 , the corresponding forward and backward paths are conditionally independent , @xmath46 are distributed as independent simple random walks , and @xmath47 are distributed as independent random walks conditioned on not hitting @xmath17 . properties ( 1)-(3 ) uniquely define @xmath42 , as proved in theorem 1.1 in @xcite . in fact , theorem 1.1 in @xcite gives a coupling of the poisson point measures @xmath48 with distribution @xmath42 for all @xmath4 . we refer the reader to @xcite for more details . let @xmath49 be the set of edges of @xmath1 , i.e. , @xmath50 . we will use the following convention throughout the paper . for a subset @xmath51 of @xmath49 , the subgraph of the lattice @xmath52 with the vertex set @xmath1 and the edge set @xmath51 will be also denoted by @xmath51 . for a poisson point measure @xmath32 with distribution @xmath42 , the _ random interlacement _ @xmath53 ( at level @xmath0 ) is defined in @xcite as the set of vertices of @xmath1 visited by at least one of the trajectories from @xmath54 . this is a translation invariant and ergodic random subset of @xmath1 , as shown in ( * ? ? ? * theorem 2.1 ) . the law of @xmath55 is characterized by the identity ( see ( 0.10 ) and remark 2.2 ( 2 ) in @xcite ) : @xmath56 = e^{-u \mathrm{cap}(k ) } , \quad \text{for all finite $ k\subseteq \z^d$ } .\ ] ] we denote by @xmath57 the set of edges of @xmath49 traversed by at least one of the trajectories from @xmath54 . the corresponding random subgraph @xmath58 of @xmath52 ( with the vertex set @xmath1 and the edge set @xmath58 ) is called the _ random interlacement graph _ ( at level @xmath0 ) . it follows from theorem 2.1 and remark 2.2(4 ) of @xcite that @xmath58 is a translation invariant ergodic random subgraph of @xmath52 . let @xmath59 be the _ vacant set _ at level @xmath0 . given a parameter @xmath60 , we consider the family @xmath61 , @xmath13 , of independent bernoulli random variables ( an independent noise ) with parameter @xmath2 , and define @xmath2-disordered analogues of the random interlacement @xmath62 and the vacant set @xmath63 as follows . we say that @xmath64 if @xmath65 and @xmath66 or @xmath67 and @xmath68 . in other words , the vertices of the random interlacement get an @xmath2-chance to become vacant , and the vertices of the vacant set get an @xmath2-chance to become occupied . let @xmath69 . we are interested in percolative properties of @xmath62 and @xmath63 . it follows from remark 1.6(4 ) in @xcite that for any @xmath3 and @xmath4 , @xmath70 \asymp \left(1+|x - y|_\infty\right)^{2-d},\quad \mbox{for } x , y\in\z^d , \ ] ] where @xmath71 denotes the covariance under @xmath42 . this displays the presence of long - range correlations in @xmath72 . non - rigorous study of the effect of small quenched noise on the critical behavior of a system with long - range correlations was initiated in @xcite . it was shown , among other results , in @xcite that the random interlacement graph @xmath58 consists of a unique infinite connected component and isolated vertices . ( refinements of this result were obtained in @xcite . ) in @xcite , we showed that the random interlacement graph is almost surely transient for any @xmath4 in dimensions @xmath3 . in theorem [ thm : slabs ] of the present paper , we prove that for any @xmath4 and small enough @xmath12 , the set @xmath62 still contains an infinite connected component . in fact , theorem [ thm : slabs ] implies that @xmath55 and @xmath58 still have an infinite connected component in wide enough slabs , even after a small positive density of vertices of @xmath55 , respectively edges of @xmath58 , is removed . one might interpret all these results as an evidence of the heuristic statement that the geometry of the interlacement graph is similar to that of the underlying lattice @xmath1 . recently , this question has been settled in @xcite by a clever refinement of the techniques in @xcite . it was proved in @xcite ( and later in @xcite with a different , model independent proof ) that the graph distance in @xmath55 is comparable to the graph distance in @xmath1 , and a shape theorem holds for balls with respect to graph distance on @xmath55 . first results about heat - kernel bounds for the random walk on @xmath55 have been recently obtained in ( * ? ? ? * theorem 2.3 ) . an important role in understanding the local picture left by the trajectory of a random walk on the discrete torus @xmath6 , @xmath7 or the discrete cylinder @xmath9 , @xmath10 is played by @xmath73=0 \}\ ] ] ( see , e.g. , @xcite ) . it follows from ( * ? ? ? * ( 1.53 ) and ( 1.55 ) ) that for @xmath74 , the set @xmath75 is stochastically dominated by @xmath72 . therefore , for all @xmath76 , @xmath77=0 $ ] . moreover , by @xcite , @xmath78 , i.e. , there is a non - trivial phase transition for @xmath72 in @xmath0 at @xmath79 . in theorem [ thm : vuvarepsilon ] of this paper , we prove that for small enough @xmath2 , the @xmath2-disordered vacant set @xmath63 still undergoes a non - trivial phase transition in @xmath0 . in theorem [ thm : u*varepsilon ] we give explicit upper and lower bounds on the phase transition threshold for @xmath63 , as @xmath5 . these bounds suggest that the phase transition is actually robust with respect to noise . we state it as a conjecture in remark [ rem : conjecture : u*varepsilon ] . for @xmath80 , we define the random subset @xmath81 of @xmath82 by deleting each edge with probability @xmath83 and retaining it with probability @xmath84 , independently for all edges , and , similarly , the random subset @xmath85 of @xmath1 by deleting every vertex of @xmath1 with probability @xmath83 and retaining it with probability @xmath84 , independently for all vertices . we look at the random subgraphs of @xmath86 with vertex set @xmath1 and edge set @xmath87 , and the one induced by the set of vertices @xmath88 . our first theorem states that the graphs @xmath55 and @xmath58 have infinite connected subgraphs in a wide enough slab , moreover , bernoulli bond percolation on @xmath58 and bernoulli site percolation on @xmath55 restricted to this slab have a non - trivial phase transition . [ thm : slabs ] let @xmath3 and @xmath4 . there exist @xmath89 and @xmath90 such that , almost surely , the random graphs @xmath91 and @xmath87 contain infinite connected components in the slab @xmath92 . as a byproduct of the proof of theorem [ thm : slabs ] , we obtain the following generalization of the main result in @xcite . [ thm : transience ] let @xmath3 and @xmath4 . let @xmath93 , @xmath94 be independent identically distributed positive random variables . the electric network @xmath95 with resistances @xmath93 is almost surely transient , i.e. , the effective resistance between any vertex in @xmath58 and infinity is finite . theorem [ thm : transience ] is a generalization of the main result of @xcite , since the transience of the unique infinite connected component of the random interlacement graph @xmath58 follows from the case when @xmath93 are almost surely equal to @xmath96 ( see , e.g. , @xcite ) . the result of theorem [ thm : transience ] is equivalent ( see the main result of @xcite ) to the following statement : for any @xmath4 , there exists @xmath89 such that the graph @xmath87 contains a transient component , i.e. , the simple random walk on it is transient . the proof of this fact will come as a byproduct of the proof of theorem [ thm : slabs ] . the main idea of the proofs of theorems [ thm : slabs ] and [ thm : transience ] is renormalization . we partition the graph @xmath1 into disjoint blocks of equal size . a block is called good if the graph @xmath58 contains a unique large connected component in this block and all the edges of the block are in @xmath81 , otherwise it is called bad . a more precise definition will be given in section [ sec : connectivity ] . it will be shown that paths of good blocks contain paths of @xmath97 . in particular , percolation of good blocks implies percolation of @xmath97 . using the strong connectivity result of @xcite , stated as lemma [ l : connectivity ] below , we show that a block is good with probability tending to @xmath96 , as the size of the block increases . we then use the decoupling inequalities of @xcite , stated as theorem [ thm : decoupling ] below , to show in lemma [ l : badpaths ] that @xmath98-connected components of bad blocks are small . with the result of lemma [ l : badpaths ] , the existence statement of theorem [ thm : slabs ] follows using a standard duality argument , and the proof of theorem [ thm : transience ] is reminiscent of the proof of theorem 1 in section 3 of @xcite . in our next theorem , we show that for small enough @xmath12 , the @xmath2-disordered vacant set @xmath63 undergoes a non - trivial phase transition in @xmath0 . let @xmath99=0 \ } .\ ] ] [ thm : vuvarepsilon ] let @xmath3 . for any @xmath100 and @xmath101 , @xmath102=0 .\ ] ] in other words , for @xmath100 , the @xmath2-disordered vacant set @xmath63 undergoes a phase transition in @xmath0 at @xmath103 . moreover , there exists @xmath104 such that for all @xmath105 , @xmath106 the first statement of theorem [ thm : vuvarepsilon ] is proved in lemma [ l : coupling ] . it follows from a standard coupling argument and the fact that the set @xmath75 is stochastically dominated by @xmath107 for @xmath74 ( see ( * ? ? ? * ( 1.53 ) and ( 1.55 ) ) ) . the second statement of theorem [ thm : vuvarepsilon ] follows from the more general statement of theorem [ thm : u*varepsilon ] , in which we give explicit upper and lower bounds on @xmath103 , as @xmath5 . the proof of theorem [ thm : u*varepsilon ] uses renormalization , and is very similar in spirit to the proof of theorem [ thm : slabs ] . the bounds on @xmath103 that we obtain in theorem [ thm : u*varepsilon ] are in terms of certain thresholds describing local behavior of @xmath72 in sub- and supercritical regimes ( see and definition [ def : tildeu ] , respectively ) . in particular , they are purely in terms of @xmath72 and not @xmath63 . as we discuss in remark [ rem : conjecture : u*varepsilon ] , these thresholds are conjectured to coincide with @xmath79 , therefore it is reasonable to believe that the phase transition of @xmath72 is stable with respect to small random noise . in other words , the following conjecture holds : @xmath108 finally , note that it is essential for @xmath109 that the parameter @xmath2 is small . for example , since @xmath110 has the same law as the bernoulli site percolation with parameter @xmath111 , which is supercritical in dimensions @xmath3 ( see @xcite ) , we obtain that @xmath112 . we now describe the structure of the remaining sections of the paper . we recall the strong connectivity lemma of @xcite and the decoupling inequalities of @xcite in section [ section : notationandproperties ] . in section [ subsection : seed ] we construct and study seed events which are used in section [ sec : connectivity ] to define good blocks . lemma [ l : badpaths ] , the main ingredient of the proofs of theorems [ thm : slabs ] and [ thm : transience ] , is proved in section [ sec : connectivity ] . the proofs of theorems [ thm : slabs ] and [ thm : transience ] are given in section [ sec : proofs ] , and the proof of theorem [ thm : vuvarepsilon ] is given in section [ sec : vacantset ] , where we also give explicit bounds on @xmath103 , as @xmath113 . in this section we introduce basic notation and collect some properties of the random interlacements , which are recurrently used in our proofs . for @xmath114 , we write @xmath115 for the absolute value of @xmath116 , and @xmath117 for the integer part of @xmath116 . for @xmath118 , we write @xmath119 for the @xmath120-norm of @xmath121 , i.e. , @xmath122 , and @xmath123 for the @xmath124-norm of @xmath121 , i.e. , @xmath125 . for @xmath126 and @xmath13 , let @xmath127 be the @xmath120-ball of radius @xmath128 centered at @xmath121 , and @xmath129 . for @xmath13 and integers @xmath130 , we write @xmath131 for the set of vertices @xmath132 with @xmath133 for all @xmath134 . for @xmath135 , we write @xmath136 if both of its endvertices are in @xmath131 . if @xmath137 , we denote by @xmath138 the set of edges of @xmath139 with both endvertices in @xmath131 . for @xmath140 , we write @xmath141 in @xmath139 , if @xmath121 and @xmath142 are in the same connected component of the graph @xmath139 . let @xmath143 , with @xmath144 and the canonical @xmath27-algebra @xmath145 , be the probability space on which @xmath58 is defined . for @xmath146 , we say that @xmath147 is in @xmath58 when @xmath148 . let @xmath149 , with @xmath150 and the canonical @xmath27-algebra @xmath151 , be the probability space on which @xmath81 is defined . for @xmath152 , we say that @xmath147 is in @xmath81 when @xmath148 . finally , let @xmath153 denote the probability space on which the random interlacement graph @xmath58 and bernoulli bond percolation configuration @xmath81 are jointly defined . throughout the paper , we use the following notational agreement . for events @xmath154 and @xmath155 , we denote the corresponding events @xmath156 and @xmath157 in @xmath158 also by @xmath159 and @xmath160 , respectively . we denote by @xmath161 the indicator of event @xmath162 and by @xmath163 the complement of @xmath162 . for @xmath164 , given a random subset @xmath165 of @xmath49 , with @xmath166 , and an event @xmath167 , we define @xmath168 where for @xmath135 , @xmath169 equals @xmath96 if @xmath170 , and @xmath171 otherwise . conversely , for an element @xmath172 , let @xmath173 ( by our convention , we also denote by @xmath174 the graph with the vertex set @xmath1 and the edge set @xmath175 . ) an event @xmath176 is called increasing , if for any @xmath177 , all the elements @xmath178 with @xmath179 are in @xmath162 . the event @xmath162 is called decreasing , if @xmath163 is increasing . throughout the text , we write @xmath180 and @xmath181 for small positive and large finite constants , respectively , that may depend on @xmath182 and @xmath0 . their values may change from place to place . the following strong connectivity lemma follows from proposition 1 in @xcite . [ l : connectivity ] let @xmath3 , @xmath4 , and @xmath12 . there exist constants @xmath183 and @xmath184 such that for all @xmath185 , @xmath186 \geq 1 - c\exp(-cr^{1/6 } ) .\ ] ] lemma [ l : connectivity ] may seem more general than proposition 1 in @xcite , but , in fact , the two results are equivalent . in order to see this , the reader may check how proposition 1 is derived from lemma 13 in @xcite . let @xmath187 ( the choice of @xmath188 will be justified in the proof of lemma [ l : badpaths ] . ) let @xmath189 and @xmath190 be positive integers . we introduce the geometrically increasing sequence of length scales @xmath191 for @xmath192 , we introduce the renormalized lattice graph @xmath193 by @xmath194 for @xmath195 and @xmath192 , let @xmath196 let @xmath197 denote the canonical coordinates on @xmath198 . for @xmath199 , let @xmath200 be a @xmath201-measurable event . we call events of the form @xmath202 _ seed events_. we denote the family of events @xmath203 by @xmath204 . examples of seed events important for this paper will be considered in section [ subsection : seed ] . the reader should think about the events @xmath202 as `` bad '' events . now we recursively define bad events on higher length scales using seed events . for @xmath205 and @xmath206 , denote by @xmath207 the event that there exist @xmath208 with @xmath209 such that the events @xmath210 and @xmath211 occur : @xmath212 ( for simplicity , we omit the dependence of @xmath213 on @xmath189 from the notation . ) note that @xmath213 is @xmath214-measurable . ( this can be shown by induction on @xmath215 . ) recall the definition . the following theorem is a special case of theorem 3.4 in @xcite ( modulo some minor changes that we explain in the proof ) . [ thm : decoupling ] for all @xmath7 , @xmath216 and @xmath217 , there exists @xmath218 such that for all @xmath192 , @xmath219 , and @xmath220 a multiple of @xmath188 , we have 1 . if @xmath221 are decreasing events , then for all @xmath222 , @xmath223 \leq \left(l_0^{2d } \sup_{x\in { \mathbb{g}}_0\cap[0,l_n)^d } \mathbb p\left[\badseed_x(\intgraph^u)\right ] + \frac14\right)^{2^n } , \ ] ] 2 . if @xmath221 are increasing events , then for all @xmath224 , @xmath225 \leq \left(l_0^{2d } \sup_{x\in { \mathbb{g}}_0\cap[0,l_n)^d } \mathbb p\left[\badseed_x(\intgraph^u)\right ] + \frac14\right)^{2^n } .\ ] ] we refer the reader to section 3 of @xcite for the notation . our events @xmath213 correspond to the events @xmath226 of @xcite , @xmath227 plays the role of @xmath228 , thus @xmath229 and @xmath230 in definition 3.1 of @xcite . there are a number of comments we would like to make before applying results derived in section 3 of @xcite : \(1 ) even though the events @xmath226 in @xcite pertain to the occupancy of vertices ( i.e. , they are subsets of @xmath231 ) , theorem 3.4 in @xcite also applies in the setting when the events @xmath226 pertain to the occupancy of edges ( i.e. , they are subsets of @xmath198 ) , see theorem 2.1 , remark 2.5(3 ) and corollary 2.1 of @xcite . \(2 ) the constant @xmath188 is taken to be @xmath232 in definition 3.1 in @xcite , but theorem 3.4 in @xcite works for any large enough constant @xmath188 , with @xmath233 also large enough . \(3 ) the events @xmath213 defined by are not cascading in the sense of definition 3.1 in @xcite , because @xmath234 of @xcite only holds for @xmath235 rather than for all @xmath236 which is a multiple of @xmath232 . nevertheless , the statement and the proof of theorem 3.4 in @xcite only involve events @xmath226 , with @xmath237 for some previously fixed @xmath219 and @xmath238 ( where @xmath238 is large enough ) . taking the above remarks into account , we can apply theorem 3.4 of @xcite to the events @xmath213 . in order to derive and from theorem 3.4 of @xcite , we choose @xmath238 large enough , so that @xmath239 , @xmath240 , and @xmath241 . ( see , e.g. , the calculations in ( 3.37 ) of @xcite . ) currently , theorem 3.4 in @xcite ( and , as a result , theorem [ thm : decoupling ] of this paper ) is proved only for increasing and decreasing events . it would be interesting to show that the result of theorem 3.4 in @xcite holds for a more general class of events . [ cor : decoupling ] let @xmath7 , @xmath4 and @xmath217 . let @xmath221 be all increasing events and @xmath242 , or all decreasing events and @xmath243 . if @xmath244 = 0 , \ ] ] then there exist @xmath245 such that for all @xmath192 , @xmath246 \leq 2^{-2^n } .\ ] ] moreover , if the limit in ( as @xmath247 ) exists and equals to @xmath171 , then there exists @xmath218 such that the inequality holds for all @xmath220 a multiple of @xmath248 , @xmath249 ( for some constant @xmath250 ) , and @xmath192 . in this section we apply corollary [ cor : decoupling ] to two families of ( decreasing and increasing ) bad events defined in terms of @xmath251 . we also recursively define a similar ( but simpler ) family of bad events in terms of @xmath81 and derive results analogous to corollary [ cor : decoupling ] for this family given that @xmath84 is close enough to @xmath96 . the corresponding seed events will be used in section [ sec : connectivity ] to define good vertices in @xmath252 . the good vertices will have the property that the existence of an infinite path of good vertices in @xmath252 implies the existence of an infinite path in the graph @xmath97 , as stated formally in lemma [ l : fromg0tozd ] . we define the density of the interlacement at level @xmath0 ( see , e.g. , ( 1.58 ) in @xcite ) by @xmath253 where @xmath254 is the green function of the simple random walk on @xmath1 started at @xmath171 . the function @xmath255 is continuous . note that @xmath256 if and only if @xmath257 for some @xmath258 , thus @xmath55 is a measurable function of @xmath58 . it follows from theorem 2.1 and remark 2.2(4 ) of @xcite that @xmath58 is a translation invariant ergodic random subset of @xmath49 . by an appropriate ergodic theorem ( see , e.g. , theorem viii.6.9 in @xcite ) , we get @xmath259 in this subsection we define and study a family of bad decreasing @xmath260-measurable events @xmath261 with ( see ) @xmath262 for @xmath205 , and @xmath263 \leq 2^{-2^n}$ ] . in order to define the bad decreasing seed event @xmath264 , we define its complement , the `` good '' increasing event @xmath265 . [ def : exu ] fix @xmath4 . recall the definition of the graph @xmath174 in . let @xmath266 be the measurable subset of @xmath198 such that @xmath267 iff a. [ def_exu_a ] for all @xmath268 , the graph @xmath269 contains a connected component with at least @xmath270 vertices , b. [ def_exu_b ] all of these @xmath271 components are connected in the graph @xmath272 . note that @xmath266 is an increasing @xmath273-measurable event . moreover , if @xmath165 is a random translation invariant subset of @xmath49 , then @xmath274 = \mathbb p[e_0^u(\gengraphrand)]$ ] for all @xmath275 . [ l : ex ] for any @xmath4 there exists @xmath276 such that @xmath277 \to 1 , \quad \mbox{as } \quad l_0\to\infty .\ ] ] let @xmath4 . by the continuity of @xmath278 , we can choose @xmath12 and @xmath276 so that @xmath279 with such a choice of @xmath2 and @xmath280 , for @xmath219 , we obtain @xmath281 let @xmath282 . we consider the boxes @xmath283 the volume of @xmath284 is @xmath285 . using and , we get that with probability tending to @xmath96 as @xmath247 , each of the boxes @xmath284 , @xmath268 contains at least @xmath286 vertices of @xmath287 . now by lemma [ l : connectivity ] , all the vertices of @xmath288 are connected in @xmath289 for all @xmath268 with probability tending to @xmath96 as @xmath247 . this shows that the event in definition [ def : exu ] holds with probability tending to @xmath96 as @xmath247 . again by lemma [ l : connectivity ] , the vertices of @xmath290 , @xmath268 are all connected in @xmath291 . this , together with the previous conclusion , implies that the event in definition [ def : exu ] holds with probability tending to @xmath96 as @xmath247 . hence we have established . [ cor : ex ] for each @xmath4 , there exists @xmath292 such that for all integers @xmath220 a multiple of @xmath248 ( see ) , @xmath293 ( for some constant @xmath294 ) , and @xmath192 , @xmath295 \leq 2^{-2^n } .\ ] ] indeed , it immediately follows from corollary [ cor : decoupling ] and lemma [ l : ex ] . in this subsection we define and study a family of bad increasing @xmath260-measurable events @xmath296 with ( see ) @xmath297 for @xmath205 , and @xmath298 \leq 2^{-2^n}$ ] . in order to define the bad increasing seed event @xmath299 , we define its complement , the `` good '' decreasing event @xmath300 . [ def : fxu ] let @xmath4 . let @xmath301 be the measurable subset of @xmath198 such that @xmath302 iff for all @xmath268 , the graph @xmath303 contains at most @xmath304 vertices in connected components of size at least @xmath305 , i.e. , @xmath306 note that @xmath301 is a decreasing @xmath273-measurable event . moreover , if @xmath307 is a random translation invariant subset of @xmath49 , then @xmath308 = \mathbb p[f_0^u(\gengraphrand)]$ ] . [ l : fx ] for any @xmath4 there exists @xmath217 such that @xmath309 \to 1 , \quad \mbox{as } \quad l_0\to\infty .\ ] ] let @xmath4 . by the continuity of @xmath278 , we can choose @xmath276 so that @xmath310 therefore , implies that , with probability tending to @xmath96 as @xmath247 , the inequality with @xmath174 replaced by @xmath311 is satisfied for all @xmath268 . this implies . [ cor : fx ] for each @xmath4 , there exists @xmath292 such that for all integers @xmath220 a multiple of @xmath248 ( see ) , @xmath293 ( for some constant @xmath294 ) , and @xmath192 , @xmath312 \leq 2^{-2^n } .\ ] ] indeed , it immediately follows from corollary [ cor : decoupling ] and lemma [ l : fx ] . in this subsection we define and study a family of bad decreasing @xmath260-measurable events @xmath313 in the spirit of the definition : @xmath314 for @xmath205 , and @xmath315 \leq 2^{-2^n}$ ] when @xmath89 is close enough to @xmath96 . we define the bad decreasing seed event @xmath316 as the measurable subset of @xmath198 such that @xmath317 iff there is an edge in the box @xmath318 which is not in @xmath174 ( remember that an edge @xmath319 is in @xmath131 if both its endvertices are in @xmath131 ) , i.e. , @xmath320 note that @xmath321 is a decreasing @xmath273-measurable event . moreover , if @xmath307 is a random translation invariant subset of @xmath49 , then @xmath322 = \mathbb p[\overline d_0(\gengraphrand)]$ ] . [ l : hx ] for any integers @xmath219 and @xmath323 there exists @xmath89 such that for all @xmath192 , @xmath324 \leq 2^{-2^n } .\ ] ] since the probability of @xmath325 is at most @xmath326 , we can choose @xmath327 so that @xmath328 < 1/2 .\ ] ] note that for @xmath329 , @xmath330 , the events @xmath331 and @xmath332 are independent and have the same probability . therefore , since @xmath333 , we get @xmath324 \leq l_0^{2d } \mathbb p \left[\overline d_{0,n-1}(\berbond^p)\right]^2 \leq \dots \leq \left(l_0^{2d}\right)^{1 + 2+\ldots+2^{n-1}}\left(\mathbb p \left[\overline d_0(\berbond^p)\right]\right)^{2^n } \leq \left(l_0^{2d } \mathbb p \left[\overline d_0(\berbond^p)\right]\right)^{2^n } .\ ] ] the result follows from the choice of @xmath84 . for @xmath334 , we say that @xmath121 and @xmath142 are nearest - neighbors in @xmath252 if @xmath335 , and @xmath98-neighbors in @xmath252 if @xmath336 . we say that @xmath337 is a nearest - neighbor path in @xmath252 , if for all @xmath338 , @xmath339 and @xmath340 are nearest - neighbors in @xmath252 , and a @xmath98-path in @xmath252 , if for all @xmath338 , @xmath339 and @xmath340 are @xmath98-neighbors in @xmath252 . let @xmath4 and @xmath341 . recall the definitions of the bad seed events @xmath342 , @xmath343 and @xmath321 from definition [ def : exu ] , definition [ def : fxu ] and , respectively . we say that @xmath344 is a _ bad _ vertex if the event @xmath345 occurs . otherwise , we say that @xmath121 is _ good_. the following lemma will be useful in the proofs of theorems [ thm : slabs ] and [ thm : transience ] . [ l : fromg0tozd ] let @xmath121 and @xmath142 be nearest - neighbors in @xmath252 , and assume that they are both good . \(a ) each of the graphs @xmath346 , with @xmath347 , contains the unique connected component @xmath348 with at least @xmath270 vertices , and \(b ) @xmath349 and @xmath350 are connected in the graph @xmath351 . + in particular , this implies that if there is an infinite nearest - neighbor path @xmath352 of good vertices in @xmath252 , then the set @xmath353 contains an infinite nearest - neighbor path of @xmath97 . let @xmath121 and @xmath142 be nearest - neighbors in @xmath252 , and assume that they are both good . by definition [ def : exu ] , the graphs @xmath354 and @xmath355 contain connected components of size at least @xmath356 , which are connected in the graph @xmath357 . by definition [ def : fxu ] , each of the graphs @xmath354 and @xmath355 contains at most @xmath304 vertices in connected components of size at least @xmath305 . since @xmath358 , there can be at most one connected component of size @xmath359 in each of the graphs @xmath354 and @xmath355 . this impies that each of the graphs @xmath360 , with @xmath361 , contains the unique connected component @xmath348 with at least @xmath362 vertices , and @xmath349 and @xmath350 are connected in the graph @xmath357 . finally , by , @xmath363 . therefore , all the edges of the graph @xmath357 are present in @xmath81 . for @xmath344 , and @xmath364 which are divisible by @xmath189 , let @xmath365 be the event that @xmath366 is connected to the boundary of @xmath367 by a @xmath98-path of bad vertices in @xmath252 . let @xmath368 be the event that @xmath121 is connected to the boundary of @xmath367 by a @xmath98-path of bad vertices in @xmath252 . [ l : badpaths ] for any @xmath4 , there exist @xmath219 , @xmath89 , @xmath369 and @xmath370 ( all depending on @xmath0 ) such that for all @xmath371 divisible by @xmath189 , we have @xmath372 \leq c e^{- n^c } .\ ] ] we may assume that @xmath373 . it suffices to show that for @xmath192 , @xmath374 \leq c e^{- l_n^c } .\ ] ] indeed , choose @xmath215 so that @xmath375 . then @xmath376\leq \mathbb p[\overline{h}^*(0,l_n,2l_n ) ] \leq c e^{-l_n^c } \leq c'e^{-n^{c ' } } .\ ] ] let @xmath4 . choose @xmath377 , @xmath219 , and @xmath89 such that corollaries [ cor : ex ] and [ cor : fx ] and lemma [ l : hx ] hold . for @xmath192 and @xmath378 , we say that @xmath121 is @xmath215-_bad _ if the event @xmath379 occurs . otherwise , we say that @xmath121 is @xmath215-_good_. ( in particular , @xmath121 is @xmath171-bad if and only if @xmath121 is bad . ) by the definition of @xmath380 , @xmath381 and @xmath382 , @xmath383 in order to prove , it suffices to show that for all @xmath192 and @xmath195 , @xmath384 indeed , since the number of vertices in @xmath385 equals @xmath386 , we obtain by translation invariance that @xmath387 \leq 4^d \left(\mathbb p[\overline d_{0,n}(\berbond^p ) ] + \mathbb p[\overline e_{0,n}^u(\intgraph^u ) ] + \mathbb p[\overline f_{0,n}^u(\intgraph^u)]\right ) \leq 4^d\cdot 3\cdot 2^{-2^n } \leq c e^{-l_n^c } .\ ] ] we prove by induction on @xmath215 . the statement is obvious for @xmath388 . we assume that holds for all integers smaller than @xmath205 , and will show that it also holds for @xmath215 . it suffices to prove the induction step for @xmath389 . the proof goes by contradiction . assume that @xmath390 occurs and all the vertices in @xmath391 are @xmath215-good . let @xmath392 be a @xmath98-path of bad vertices in @xmath252 from @xmath393 to the boundary of @xmath394 . let @xmath395 . note that the path @xmath392 intersects the boundary of each of the boxes @xmath396 , for @xmath397 . therefore , there exist @xmath398 such that for all @xmath397 , ( a ) @xmath399 and ( b ) @xmath400 ( see figure [ fig : yi ] ) . -1 cm is as the closest vertex in @xmath401 to the point of the first intersection of @xmath392 with @xmath402 . concentric boxes are not drawn to scale here : the innermost box is @xmath393 , the outermost box is @xmath394 , and the intermediate boxes are @xmath396 , for @xmath403 . the smallest and second smallest boxes along the path @xmath392 are @xmath404 and @xmath405 , respectively , for @xmath397 . ] by the definition of @xmath406 and @xmath407 s , all the boxes @xmath405 are disjoint and contained in @xmath408 , and the path @xmath392 connects @xmath404 to the boundary of @xmath405 , i.e. , the event @xmath409 occurs for all @xmath397 . we will show that @xmath410 which will contradict our assumption that holds for @xmath411 . since all the vertices in @xmath412 are @xmath215-good by assumption , it follows from that @xmath413 note that each of the balls @xmath414 contains at most @xmath415 different @xmath407 s . therefore , the union of the balls @xmath416 ( with @xmath417 s defined in ) contains at most @xmath418 different @xmath407 s , which is strictly smaller than @xmath406 by the choice of @xmath188 in . we conclude that there exists @xmath419 such that @xmath420 we assume that @xmath238 is chosen large enough so that @xmath421 , i.e. , @xmath422 . with this choice of @xmath238 , @xmath423 therefore , follows from and , which is in contradiction with the assumption that holds for @xmath411 . this implies that holds for all @xmath192 . the proof of lemma [ l : badpaths ] is completed . in this section , we derive theorems [ thm : slabs ] and [ thm : transience ] from lemmas [ l : fromg0tozd ] and [ l : badpaths ] . the two results of theorem [ thm : slabs ] can be proved similarly ( note that the results of sections [ section : notationandproperties]-[sec : connectivity ] can be trivially adapted to site percolation on @xmath55 ) , therefore we only provide a proof for the case of bond percolation on @xmath58 . choose @xmath189 and @xmath89 such that lemma [ l : badpaths ] holds . remember the definitions of a bad vertex and the event @xmath424 from section [ sec : connectivity ] . let @xmath425 be a positive integer . note that the probability that there exists a @xmath98-circuit of bad vertices in @xmath426 around @xmath427 is at most @xmath428 \leq c\sum_{n = m}^\infty e^{-n^c } \leq 1/2 , \ ] ] for large enough @xmath425 . if there is no such circuit , then , by planar duality ( see , e.g. , ( * ? ? * chapter 3.1 ) ) , there is a nearest - neighbor path @xmath429 of good vertices in @xmath426 that connects @xmath427 to infinity . namely , for all @xmath430 , @xmath431 , @xmath432 , @xmath433 is good , @xmath434 , and @xmath435 as @xmath436 . it follows from lemma [ l : fromg0tozd ] that the graph @xmath437 contains an infinite connected component of @xmath438 . therefore , the probability that an infinite nearest - neighbor path in @xmath438 visits @xmath439 is at least @xmath111 . by the ergodicity of @xmath438 , an infinite nearest - neighbor path in @xmath440 exists with probability @xmath96 . we will use the main result of @xcite that for an infinite graph @xmath441 and i.i.d . positive random variables @xmath442 , @xmath443 , the following statements are equivalent : ( a ) almost surely , the electric network @xmath444 is transient , and ( b ) for some @xmath89 , independent bond percolation on @xmath445 with parameter @xmath84 contains with positive probability a cluster on which simple random walk is transient . ( in the proof , we will only use the easy implication , namely , that ( b ) implies ( a ) . ) therefore , in order to prove theorem [ thm : transience ] , it suffices to show that for some @xmath89 , with positive probability , the graph @xmath87 contains a transient subgraph . the proof of this fact is similar to the proof of theorem 1 in @xcite , so we only give a sketch here . let @xmath3 and @xmath4 . denote by @xmath446 the @xmath182-dimensional euclidean unit sphere . we will show that , for any @xmath447 , there exists an event @xmath448 of probability @xmath96 such that if @xmath448 occurs , then @xmath449 ( the set @xmath450 is roughly shaped like a paraboloid with an axis parallel to @xmath451 . ) after that , one can proceed , as in section 3 of @xcite , to show that this infinite connected subgraph of @xmath87 is transient . remember the definitions of the bad vertex and the event @xmath452 from section [ sec : connectivity ] . let @xmath189 and @xmath89 satisfy lemma [ l : badpaths ] . by and the borel - cantelli lemma , for any @xmath453 , the following event @xmath448 has probability @xmath96 : there exists a ( random ) @xmath255 such that for all @xmath344 with @xmath454 , the event @xmath455 does not occur . it remains to show that if the event @xmath448 occurs , then holds . we will first prove that the event @xmath448 implies that \(a ) for each @xmath456 , there is a nearest - neighbor path @xmath457 of good vertices in @xmath458 that connects @xmath459 to infinity , and \(b ) all the paths @xmath457 are connected by nearest - neighbor paths of good vertices in @xmath460 . + indeed , assume first that ( a ) fails , i.e. , there exists @xmath456 such that the set of vertices @xmath461 connected to @xmath459 by a nearest - neighbor path of good vertices in @xmath462 is finite . by ( * ? ? ? * lemma 2.1 ) or ( * ? ? ? * theorem 3 ) , the boundary of this set contains a @xmath98-connected subset @xmath463 of bad vertices in @xmath462 such that any nearest - neighbor path from @xmath459 to infinity in @xmath462 intersects @xmath463 . in particular , there exists @xmath344 with @xmath454 , such that the event @xmath455 occurs ; and , therefore , the event @xmath448 does not occur . similarly , if ( a ) holds and ( b ) fails , then there exist at least two disjoint connected components of good vertices of diameter @xmath464 in @xmath465 that intersect @xmath466 . therefore , by ( * ? ? ? * lemma 2.1 ) or ( * ? ? ? * theorem 3 ) , there exists @xmath344 with @xmath467 such that the event @xmath468 occurs . this again implies that the event @xmath448 does not occur . it remains to notice that by ( a ) , ( b ) and lemma [ l : fromg0tozd ] , the occurence of @xmath448 implies . indeed , lemma [ l : fromg0tozd ] and ( a ) imply that there is an infinite path of @xmath87 in every set @xmath450 , @xmath456 , and lemma [ l : fromg0tozd ] and ( b ) imply that all these infinite paths are in the same connected subgraph of @xmath87 . therefore , we have constructed the event @xmath448 of probability @xmath96 which implies . in order to show that the infinite cluster in is transient , we proceed identically to the proof of theorem 1 in section 3 of @xcite . we omit further details . in this section we prove theorem [ thm : vuvarepsilon ] . the first statement of theorem [ thm : vuvarepsilon ] is proved in section [ sec : thm : vuvarepsilon : existence ] . in section [ sec : def : overlineul ] we state theorem [ thm : u*varepsilon ] , which implies the second statement of theorem [ thm : vuvarepsilon ] . the result of theorem [ thm : u*varepsilon ] is more general than the one of theorem [ thm : vuvarepsilon ] , since it also provides explicit upper and lower bounds on @xmath103 , as @xmath5 . we prove theorem [ thm : u*varepsilon ] in section [ sec : proof : u*varepsilon ] . [ l : coupling ] for any @xmath469 and @xmath100 , the set @xmath470 is stochastically dominated by @xmath63 . in particular , for any @xmath471 , almost surely , the set @xmath63 does not contain an infinite connected component . let @xmath100 . let @xmath474 , @xmath13 , be independent bernoulli random variables with parameter @xmath2 , and @xmath475 , @xmath13 , independent bernoulli random variables with parameter @xmath476 , the two families are mutually independent , and also independent from the random interlacement @xmath55 . let @xmath477 . it is easy to see that given @xmath55 , the @xmath478 are independent , and the probability that @xmath479 equals @xmath2 for @xmath67 , and @xmath480 for @xmath65 . therefore , the set of vertices @xmath481 has the same distribution as @xmath62 . since , for @xmath74 , @xmath55 is stochastically dominated by @xmath287 , we deduce that @xmath62 is stochastically dominated by @xmath482 , and , therefore , @xmath470 is stochastically dominated by @xmath63 . in this section we state that for small enough @xmath12 , @xmath483 and give explicit upper and lower bounds on @xmath103 , as @xmath113 . the main result of this section is theorem [ thm : u*varepsilon ] , which will be proved in section [ sec : proof : u*varepsilon ] . in order to state the theorem , we need to define the critical thresholds @xmath484 and @xmath485 . the earlier version of this paper contained a different proof of the fact that @xmath483 . it was based on a new notion of the so - called strong supercriticality in slabs . that proof is available in the first version of this paper on the arxiv @xcite . the proof we present here is significantly simpler and relies on recent local uniqueness results of @xcite . [ def : tildeu ] let @xmath3 . let @xmath486 be the supremum over all @xmath487 such that for each @xmath0 smaller than @xmath487 , there exist constants @xmath488 and @xmath292 such that for all @xmath205 , we have @xmath489 \geq 1 - c e^{-n^c } , \ ] ] and @xmath490 \geq 1 - c e^{-n^c } .\ ] ] note that definition [ def : tildeu ] implicitly implies that the right hand side of must be positive for all @xmath491 and large enough @xmath215 . in particular , we conclude that @xmath492 . it was recently proved in ( * ? ? ? * theorem 1.1 ) ( and , for @xmath493 , earlier in ( * ? ? ? * ( 1.2 ) and ( 1.3 ) ) ) that @xmath494 let us also recall the definition of @xmath485 from ( * ? ? ? * ( 0.6 ) ) and ( * ? ? ? * ( 0.10 ) ) : @xmath495 = 0 \right\ } .\ ] ] it follows from @xcite that @xmath496 we prove the following theorem . [ rem : conjecture : u*varepsilon ] it would be interesting to understand whether the phase transition of @xmath72 is actually stable with respect to small random noise . in other words , is it true that @xmath498 based on , an affirmative answer to will be obtained as soon as one proves that @xmath499 note that the thresholds @xmath484 and @xmath485 are defined purely in terms of @xmath72 , and not @xmath63 . the statement is about local connectivity properties of sub- and supercritical phases of @xmath72 . in the context of bernoulli percolation , similar thresholds can be defined , and it is known that they coincide with the threshold for the existence of an infinite component ( see , e.g. , ( * ? ? ? * ( 5.4 ) and ( 7.89 ) ) ) , i.e. , the analogue of holds . the main challenge in proving comes from the long - range dependence in @xmath72 and the lack of the so - called bk - inequality ( see , e.g. , ( * ? ? ? * ( 2.12 ) ) ) , and hence it is interesting in its own . recall the definition of @xmath500 from the beginning of section [ sec : mainresults ] . in order to prove , it suffices to show that @xmath501>0,\quad \mbox{and } \\ \label{stable_subcritical } \forall\ , u > u _ { * * } \quad \exists \ , \varepsilon_0(u)>0 \quad \forall \ , \varepsilon < \varepsilon_0(u ) \ , : \quad & \mathbb{p } [ 0 \stackrel{\mathcal{v}^u \cup \mathcal{b}^{\varepsilon}}{\longleftrightarrow } \infty ] = 0 .\ \end{aligned}\ ] ] the proofs of these statements are very similar to the proof of theorem [ thm : slabs ] . therefore , we only sketch the main ideas here . by and definition [ def : vacantk ] , @xmath508\geq \eta(u)$ ] and @xmath509 \to \eta(u)$ ] as @xmath510 . therefore , by an appropriate ergodic theorem ( see , e.g. , ( * ? ? * theorem viii.6.9 ) and ( * ? ? ? * theorem 2.1 ) ) , we get @xmath511 * for all @xmath268 , the graph @xmath512 contains a connected component with at least @xmath513 vertices , and all these @xmath271 components are connected in @xmath514 , * for all @xmath268 , @xmath515 , * @xmath516 . definition [ def : good ] is similar to the definition of a good vertex in section [ sec : connectivity ] , except that now we are dealing with @xmath520 , rather than with @xmath58 . in particular , the event @xmath517 pertains to the occupancy of the vertices of @xmath1 rather than the edges . the event in ( i ) corresponds to the event @xmath521 , the event in ( ii ) corresponds to the event @xmath522 , and the event in ( iii ) corresponds to the complement of the event @xmath523 . the role of the continuous function @xmath278 in definitions [ def : exu ] and [ def : fxu ] is played by @xmath524 ( see and compare to ) . the role of lemma [ l : connectivity ] is played by the following lemma . * @xmath531 is connected to the boundary of @xmath532 in @xmath72 , and * every two nearest - neighbor paths from @xmath533 to the boundary of @xmath534 in @xmath72 are in the same connected component of @xmath535 . let @xmath536 . by and , there exist constants @xmath537 and @xmath538 , such that for all @xmath128 , we have @xmath539\geq 1 - \widetilde c e^{-r^{\widetilde c } } .\ ] ] therefore , it suffices to show that @xmath540 let @xmath541 . let @xmath349 and @xmath350 be the connected components of @xmath121 and @xmath142 in @xmath542 . we will show that if @xmath543 occurs then @xmath544 . note that by the choice of @xmath545 and @xmath546 , @xmath349 contains a path from @xmath121 to the boundary of @xmath547 , and @xmath350 contains a path from @xmath142 to the boundary of @xmath548 . assume that @xmath543 occurs . take a nearest - neighbor path @xmath549 in @xmath550 from @xmath121 to @xmath142 . for each @xmath551 , the occurrence of the events @xmath552 and @xmath553 implies that ( a ) there exist nearest - neighbor paths @xmath554 and @xmath555 in @xmath72 , @xmath554 from @xmath556 to the boundary of @xmath557 , and @xmath555 from @xmath558 to the boundary of @xmath559 , and ( since both paths connect @xmath560 to the boundary of @xmath561 ) ( b ) any two such paths are connected in @xmath562 . this implies that @xmath349 and @xmath350 must be connected in @xmath563 . this finishes the proof of and of the lemma . using , , and lemma [ l : connectivity : bigdiam ] , we can proceed similarly to the proof of ( see also the proofs of corollaries [ cor : ex ] and [ cor : fx ] and lemma [ l : hx ] ) to show that for any @xmath525 , there exist @xmath219 , @xmath369 and @xmath370 such that for all @xmath371 divisible by @xmath189 , we have @xmath564 \leq c e^{-n^c } .\ \end{split}\ ] ] we now use planar duality , similarly to the proof of theorem [ thm : slabs ] , to show that implies that for large enough @xmath189 , @xmath565>0 .\ ] ] similarly to lemma [ l : fromg0tozd ] , we observe that if there exists an infinite nearest - neighbor path @xmath566 of good vertices in @xmath252 , then the set @xmath567 contains an infinite nearest - neighbor path of @xmath568 . this , together with , implies . it follows from the definition of @xmath485 and the choice of @xmath2 ( similarly to the proof of ) that for any @xmath569 , there exist @xmath219 , @xmath369 and @xmath370 such that for all @xmath371 divisible by @xmath189 , we have @xmath575 \leq c e^{-n^c } .\ \end{split}\ ] ] in particular , for any @xmath569 and large enough @xmath189 , almost surely , there is no infinite nearest - neighbor cluster of bad vertices in @xmath252 . finally , note that if @xmath392 is an infinite path in @xmath576 from the origin , then the origin is in an infinite nearest - neighbor path of bad vertices in @xmath252 . this implies . we thank a .- s . sznitman for pointing out a connection between our results and questions of robustness to noise , which led to results discussed in section [ sec : vacantset ] , and for a careful reading of the manuscript .

we explore how the seesaw sector in neutrino mass models may be constrained through symmetries to be completely determined in terms of low - energy mass , mixing angle and-violating phase observables . the key ingredients are intra - family symmetries to determine the neutrino dirac mass matrix in terms of the charged - lepton or quark mass matrices , together with inter - family or flavor symmetries to determine diagonalization matrices . implications for leptogenesis and collider detection of heavy neutral leptons are discussed . we show that leptogenesis can succeed in small regions of parameter space for the case where the neutrino dirac mass matrix equals the up - quark mass matrix . the model where the neutrino dirac mass matrix equals the charged - lepton mass matrix can yield a heavy neutral lepton as light as about 1 tev , but detecting such a particle will be difficult .

0810.1104

neutrino oscillation experiments involving neutrinos and antineutrinos coming from astrophysical and terrestrial sources @xcite have found compelling evidence that neutrinos have mass . to accommodate this observation , the minimal standard model ( sm ) must be extended . some sensible ways to do this include : ( a ) type i seesaw with three heavy right - handed ( rh ) majorana neutrinos @xcite , ( b ) the use of an electroweak higgs triplet to directly provide the left - handed ( lh ) neutrinos with small majorana masses ( type ii seesaw @xcite ) , ( c ) introducing a fermion triplet ( type iii seesaw @xcite ) , ( d ) the generation of three dirac neutrinos through an exact parallel of the sm method of giving mass to charged fermions , and ( e ) the radiative generation of neutrino masses as per the zee or babu models @xcite . but in the absence of more experimental data , it is impossible to tell which , if any , of these is actually correct . the focus of this paper is on method ( a ) , the seesaw framework with three heavy rh majorana neutrinos ( denoted @xmath1 throughout ) . it is an attractive possibility because it simply posits the existence of these @xmath1 s to parallel the multiplet structure of the other fermions while providing a simple explanation for why the light neutrinos are so much less massive than the charged leptons . since the setup uses the most general renormalisable lagrangian consistent with the sm gauge symmetry , both the yukawa couplings of the lh leptons to the rh neutrinos and bare majorana masses are permitted for the rh neutrinos . consequently , the additional assumption that the rh majorana mass scale is much higher than that of the charged fermions leads to a tiny mass for ordinary neutrinos through the famous seesaw relation : @xmath2 , where @xmath3 is the majorana mass for a light neutrino and @xmath4 is a large rh majorana mass ( @xmath5 ) with @xmath6 being most naturally of the order of a charged fermion mass . the three light neutrino mass eigenstates are accompanied by three heavy neutral lepton mass eigenstates . depending on the parameter space for the rh neutrino bare masses and yukawa couplings , additional benefits may flow : thermal leptogenesis @xcite if there are appropriate @xmath0-violating decays and if the lightest of the heavy @xmath1 s has mass greater than about @xmath7 gev @xcite ; leptogenesis through @xmath0-violating oscillations of the @xmath1 s as in the akhmedov - rubakov - smirnov mechanism @xcite ; or @xmath1 s as a warm dark matter candidate @xcite . since the mass eigenstate heavy neutral leptons are to a good approximation sterile with respect to gauge interactions , they are difficult to detect experimentally . this is especially true if they are also extremely massive , as in the thermal leptogenesis alternative . on the other hand , if they are not as massive and are in the tev scale , then they can be looked for in colliders through their yukawa interactions , and through their suppressed but nonzero weak interactions ( induced through the mass mixing with regular active neutrinos ) . to experimentally test the seesaw scenario , it would be helpful if one knew the parameters governing the @xmath1-sector including their interactions with other sm particles . in the minimal seesaw model , these parameters are arbitrary , so one has to go beyond the minimal model to achieve this goal . the purpose of this paper is to illustrate how symmetries may be used to determine the rh majorana mass matrix as a function of low - energy mass , mixing angle and @xmath0-violating phase observables by constructing several representative models . we then examine these models to see if thermal leptogenesis can succeed or if experimentally accessible heavy @xmath1 s are predicted . in the next section we discuss the general model building symmetry requirements for relating the @xmath1-sector parameters to low - energy observables . section [ sec : symmetry - basics ] then revises the basic properties of the identified symmetries , followed by sec . [ sec : models ] which details specific models . section [ sec : phenomenology ] presents a phenomenological study of those models , and we conclude in sec . [ sec : conclusion ] the effective light majorana neutrino mass matrix @xmath3 , defined through @xmath8 is given by @xmath9 where @xmath10 is the neutrino dirac mass matrix , defined through @xmath11 whilst the rh majorana mass matrix @xmath4 is defined through @xmath12 let @xmath13 be the mass - eigenstates for the light majorana neutrinos , where @xmath14 is the unitary diagonalization matrix for @xmath3 . the diagonalized @xmath3 is therefore @xmath15 eq.([eq : ml ] ) then implies that @xmath16 the matrix @xmath17 has been experimentally determined up to an absolute light neutrino mass scale , which we shall conveniently parameterize through the lightest @xmath3 eigenvalue . for the normal hierarchy ( nh ) with @xmath18 , we have ( for @xmath19 ) @xmath20 where @xmath21 ev@xmath22 and @xmath23 ev@xmath22 are the `` solar '' and `` atmospheric '' squared - mass difference respectively @xcite . for the inverted hierarchy ( ih ) with @xmath24 , we obtain @xmath25 in order to connect the high- and low - energy sectors , one must have @xmath4 completely determined by known quantities . hence , our goal is to have @xmath4 constructed from some combination of @xmath17 , the charged - fermion mass matrices , @xmath26 with @xmath27 , and the lepton and quark mixing matrices ( @xmath28 and @xmath29 ) respectively . as a consequence , the first necessary condition , according to eq.([eq : ml2 ] ) , is : @xmath30 the simplest possibility is that @xmath31 there are custodial @xmath32 , unification and quark - lepton symmetries that can enforce each of these conditions at tree - level , as we shall review in the next section . for the moment , let us just accept that they are all possible . equation ( [ eq : ml2 ] ) now becomes @xmath33 introducing the diagonalized fermion mass matrix @xmath34 where the @xmath35 and @xmath36 are the left- and right - diagonalization matrices for @xmath6 respectively , eq.([eq : ml3 ] ) can be rewritten as @xmath37 which in turn reveals the second necessary condition : @xmath38 because the known weak interactions are left - handed , the right - diagonalization matrix can not be measured . therefore , to satisfy condition ( [ eq : cond2 ] ) , @xmath36 needs to be determined by the theory , and this usually means a _ flavor symmetry _ is required . in the next section we shall review how flavor symmetries can give rise to fully determined diagonalization matrices , where their entries are just numbers , usually related to the clebsch - gordan coefficients of the flavor symmetry group . the product @xmath39 is similar in form to both the pmns and ckm matrices , which are , respectively , @xmath40 the simplest ansatze are that @xmath41 equals either @xmath28 or @xmath29 . the next simplest class would see @xmath39 equal to a product of either the pmns or ckm matrix and a known matrix predicted by the flavor symmetry selected . let us consider some special cases . the simplest possibility suggested by the above equations would be that @xmath42 these relations may be achieved by imposing a @xmath43 and flavor symmetry respectively . the rh majorana mass matrix is then completely determined through @xmath44 two other possibilities , arising from the enforcement of ( @xmath45 ) and the appropriate flavor symmetries , are that @xmath46 leading to @xmath47 because of the automatic presence of @xmath14 in the formula for @xmath4 , it is relatively straightforward to find symmetries leading to eqs . ( [ eq : mr_e_pmns ] ) and ( [ eq : mr_ud_pmns ] ) where the leptonic pmns mixing matrix is a key feature . but it may also be of interest to consider symmetry structures that can lead to the pmns matrix being replaced by the ckm matrix ( or a product of the two ) . one way to try this would be to arrange symmetries such that @xmath3 would be necessarily diagonal , giving @xmath48 . then , the condition @xmath49 , together with @xmath50 , will lead to @xmath51 , and hence the relation @xmath52 would be obtained . a similar relation with @xmath53 and @xmath54 interchanging roles could equally well be contemplated . the delicate part would be obtaining a diagonal @xmath3 without forcing a diagonal @xmath10 . if the latter were diagonal , then the relations @xmath55 or @xmath56 would also imply that @xmath57 or respectively @xmath58 , and hence leading to @xmath59 at tree - level . finally , there is of course the relatively mundane case where all of the diagonalization matrices in the formula for @xmath4 are equal to the identity , so that one simply gets @xmath60 interestingly , this is not possible for the @xmath61 choice , because the pmns matrix is known to be very dissimilar to the identity . however , flavor symmetries allowing , eq.([eq : mundane ] ) can in principle be achieved for @xmath62 or @xmath54 . in these situations , one would then get @xmath63 and @xmath64 and @xmath65 . although the analysis above was framed in terms of the leading seesaw expression @xmath66 , it generalizes to cases where additional terms on the right - hand side are kept , because the higher - order terms contain _ a priori _ the same unknowns as does the leading term . in summary , the general properties of enforcing a ( @xmath67 ) symmetry in parallel with some flavor symmetries motivate relations of the form @xmath68 of which eqs.([eq : mr_e_pmns ] ) , ( [ eq : mr_ud_pmns ] ) , ( [ eq : mr_d_ckm ] ) and ( [ eq : mundane ] ) are important examples . the aim of this section is to briefly illustrate how mass relations of the type @xmath69 may be enforced , as well as the role of flavor symmetry in determining the diagonalization matrices of interest . we will present some concrete examples that utilize these ideas to good effect in the next section . it is well known that in a minimal @xmath70 framework one obtains the mass relations @xmath71 , because all fermions are in the same multiplet and the electroweak higgs lies in a real fundamental of @xmath70 . these relations are too powerful from a phenomenological perspective : while the neutrino dirac mass matrix is related to that of another fermion as desired , the other mass relations @xmath72 are _ not _ wanted . however , this observation motivates the search for gauge groups that contain the sm as a subgroup and have enough power to establish the mass relation we seek without violating any known experimental constraints . indeed , subgroups of @xmath70 are good starting points for such a search . outside of @xmath70 , the use of discrete rather than continuous symmetries to relate different multiplets constitutes another sensible strategy . let us consider the following groups , motivated by being subgroups of @xmath70 , but not necessarily to be thought of as arising from an underlying @xmath70 theory : the standard @xmath73 unification group @xcite , its flipped extension @xmath74 @xcite and the left - right group @xmath75 @xcite . standard @xmath73 has the lh charged leptons and lh down antiquarks in the @xmath76 representation , while their mass partners are in the @xmath77 . in the minimal model a single yukawa term couples those two multiplets to a higgs in the @xmath76 , leading to the relation @xmath78 . the up - quark and neutrino dirac masses are governed by independent yukawa couplings , so they are unrelated to each other and unrelated to @xmath79 and @xmath80 . in flipped @xmath73 , the down antiquarks and the up antiquarks flip roles , as also do the charged antileptons and antineutrinos . the minimal model thus supplies @xmath56 with unrelated @xmath80 and @xmath79 entries . for our purposes , standard @xmath73 is not useful , but flipped @xmath73 is interesting can also be used to enforce @xmath56 ] . the third subgroup , the left - right group , has the power to enforce mass degeneracy between weak isospin partners : @xmath82 and @xmath83 @xcite . such a degeneracy follows from requiring a bidoublet higgs to be _ real _ , which at the @xmath70 level follows from having the higgs 10-plet being real . this basically causes @xmath84 to become custodial @xmath32 . so we conclude that flipped @xmath73 which can give @xmath56 and the left - right group which can give @xmath85 are relevant @xmath70 subgroups for our purposes . the other obvious mass relation @xmath55 will be obtained in the next section not from @xmath70 or any of its subgroups , but rather by using the idea of discrete quark - lepton symmetry @xcite . the idea here is to extend the gauge group by including an @xmath86 color group for leptons , with standard leptons identified as one of the colours after spontaneous symmetry breaking . the gauge structure now permits a discrete interchange symmetry between quarks and ( generalized ) leptons to be imposed , from which @xmath55 can follow . though we shall not pursue this line of thought further in this paper , we should also remark that the relation between @xmath10 and @xmath87 need not be a direct equality . at the @xmath70 level one can consider embedding the electroweak higgs doublet not in the @xmath77 but in a higher - dimensional representation . in that case a matrix of clebsch - gordan coefficients relates @xmath10 with the other fermion mass matrix , as a generalization of the well - known georgi - jarlskog @xcite modification of the @xmath79 to @xmath80 relation in @xmath73 unification . once the appropriate fermion - mass - constraining group is selected , the remaining challenges are twofold . the first , as well - illustrated by minimal @xmath70 , is the removal of byproducts such as unwanted mass relations or interactions . the second is the need to have predictable diagonalization matrices . quite frequently , it is possible to meet both of these challenges by introducing a _ flavor symmetry _ and a non - minimal higgs sector . in cases where this is not sufficient , unbroken global non - flavor symmetries may be imposed to eliminate all undesirable terms . the key concept is that of a `` form - diagonalizable matrix '' @xcite . this is a matrix containing relations amongst its elements and perhaps also texture zeros so as to make the diagonalization matrices fully determined while leaving the eigenvalues arbitrary . special flavor symmetries exist to enforce form - diagonalizability , and they have in recent years been widely used to try to understand the `` tribimaximal '' form @xcite that is consistent with the experimentally measured pmns matrix . in the models presented below , combined effect of the mass - relating symmetry and the flavor symmetry will be to produce a relation of the form @xmath88 , where @xmath89 is given by a known diagonalization matrix . in this section , we construct three realistic models that can enforce @xmath88 , and subsequently lead to relations ( [ eq : mr_e_pmns ] ) and ( [ eq : mr_ud_pmns ] ) respectively . we consider a flipped @xmath73 group @xcite augmented by @xmath91 flavor symmetry @xcite : @xmath92 with hypercharge @xmath93 given by a linear combination of @xmath94 and @xmath95 . the choice of this gauge group is for the reason discussed in the previous section : one naturally obtains the useful mass relation @xmath56 while avoiding @xmath78 . the role of the flavor symmetry is then purely to ensure that all diagonalization matrices are completely determined . for this model , the particle contents and their transformation properties under @xmath96 are given by : @xmath97 where the superscripts 1,2 and 3 and greek letters are the color and @xmath73 indices respectively . in matrix form , the @xmath96 invariant interaction lagrangian then contains the following terms : @xmath98 note that @xmath99 , which provides the heavy majorana mass , breaks @xmath96 down to the sm , and is expected to be at a much higher energy scale than @xmath100 which breaks electroweak symmetry . writing out the @xmath91 structure of the @xmath101- and @xmath102-terms in eq . ( [ eq : su5_lag2 ] ) with the vacuum @xmath103 where @xmath104 , one gets @xmath105 expanding out the @xmath91 invariants using the results in the appendix , one obtains @xmath106 where @xmath107 . from ( [ eq : su5_me_mu_mnud ] ) , we deduce that @xmath108 and hence @xmath109 putting this into ( [ eq : ml2 ] ) gives @xmath110 returning to eq . ( [ eq : su5_lag2 ] ) , if we expand the @xmath111- and @xmath112-term in flavor space , it becomes apparent that the @xmath53-quark mass matrix , @xmath80 , and the rh majorana mass matrix , @xmath4 , are both arbitrary complex symmetric matrices . consequently , the diagonalization matrices @xmath113 and @xmath114 ( since @xmath3 is a function of @xmath4 ) are both arbitrary unitary matrices in this model . this implies that the model places no restrictions on the neutrino mixing matrix , @xmath115 , and the quark mixing matrix , @xmath116 , and so one simply sets them to match the experimental values . next , we construct a slightly more complicated model within the framework of a discrete quark - lepton symmetry @xcite . as well as the usual @xmath91 flavor symmetry , we also introduce an additional unbroken @xmath118 global symmetry to forbid certain interaction terms in the lagrangian . the symmetry group is @xmath119 where @xmath120 is the discrete quark - lepton symmetry that relates @xmath121 while hypercharge @xmath93 is given by a linear function of @xmath95 and @xmath94 . the field contents are @xmath122 where @xmath123 @xmath124 are exotic leptonic - color partners of the usual leptons . the discrete @xmath120 symmetry is broken and these exotic leptons gain mass when @xmath125 picks up a nonzero vev : @xmath126 we arrange @xmath127 to give a large majorana mass while keeping @xmath128 . the @xmath129 s will break electroweak symmetry as usual . in order to avoid domain walls and allow the implementation of the seesaw mechanism , we demand the following hierarchy for the energy scales : @xmath130 overall , the @xmath131 invariant interaction lagrangian takes the form : @xmath132\epsilon^{\alpha\beta\gamma } \nonumber\\ % % % % % % % % % % % % % % % % % % % & \qquad + \lambda_{g1}\left(\overline{q}_l u_r \phi_1 + \overline{f}_l e_r \phi_2\right ) + \lambda_{g2}\left(\overline{q}_l u_r \phi_2^c + \overline{f}_l e_r \phi_1^c\right ) + \lambda_{g3}\left(\overline{q}_l d_r \phi_d + \overline{f}_l n_r \phi_d^c\right)\nonumber\\ % % % % % % % % % % % % % % % % % % % % & \qquad + \lambda_{h1}\left(\overline{n^c_r}_{\alpha}\ , n_{r\beta}\ , \delta_1^{\alpha\beta } + \overline{d^c_r}_{\alpha}\ , d_{r\beta}\ , \delta_2^{\alpha\beta}\right ) + \text{h.c.}\ ; , \label{ql : lag}\end{aligned}\ ] ] where @xmath133 are @xmath134 indices and the terms proportional to @xmath135 are the mass terms for the exotic fermions . from ( [ ql : lag ] ) and taking @xmath136 and @xmath137 , we expect the following mass relations : @xmath138 so , in general , @xmath139 but @xmath140 . writing out the @xmath91 structure for the above matrices , we have : @xmath141 choosing the vacuum patterns : @xmath142 and following the @xmath91 rules in the appendix , we get @xmath143 where @xmath144 , @xmath145 and @xmath146 . in addition , it can be shown that when the @xmath91 singlets , and acquire nonzero vevs , the resulting neutrino majorana mass matrix , @xmath4 , is an arbitrary complex symmetric matrix . using the results ( [ eq : ql_me_mu_md_mnud ] ) and ( [ eq : ql_vs ] ) , we can conclude that in this model @xmath147 where @xmath148 which is arbitrary , whilst we have @xmath149 . so , at tree - level , this model predicts no quark mixing . however , since the symmetry enforcing this result is now broken , radiative corrections will generate nonzero quark mixing . we have not attempted to prove that realistic mixing angles can be obtained , since our focus in this paper is on the lepton sector . it is interesting that the form of the mixing matrices predicted by this model is consistent with small quark mixing ( @xmath150 ) , whereas neutrino mixing ( @xmath151 ) is large @xcite . this is because @xmath152 is a trimaximal mixing matrix , and so , unless @xmath153 , one expects the product of the two would be very dissimilar to the identity . finally , we consider a left - right model @xcite with @xmath91 flavor symmetry . the symmetry group is @xmath155 here , the imposition of the discrete @xmath156 parity symmetry is not necessary , and hence will be omitted for simplicity . the complete list of relevant particle contents for this setup is : @xmath157 where we have deliberately embedded the same higgs doublet into @xmath158 to form a _ real _ bidoublet . in matrix form , the @xmath159 invariant lagrangian has the following terms : @xmath160 when the symmetry is broken spontaneously by the nonzero vevs , @xmath161 where @xmath162 and @xmath163 , we obtain mass relations of the form : @xmath164 in flavor space , the charged - lepton and neutrino dirac - mass terms become @xmath165\nonumber\\ % % % % % % & \qquad + \widetilde{y}_1 \,(\overline{e}_l\,\vev{\phi^{0*}})_{\underline{1}}\,e_r + \widetilde{y}_1 ' \,(\overline{e}_l\,\vev{\phi^{0*}})_{\underline{1}'}\,e_r '' + \widetilde{y}_1 '' \,(\overline{e}_l\,\vev{\phi^{0*}})_{\underline{1}''}\,e_r ' + \text{h.c.}\;,\label{eq : lr_me}\\ % % % % % % % % % % % % % % % % % % % % % % % \mnud\;:\qquad & y_1 \,(\overline{\nu}_l\,\vev{\phi^{0}})_{\underline{1}}\,\nu_r + y_1 ' \,(\overline{\nu}_l\,\vev{\phi^{0}})_{\underline{1}'}\,\nu_r '' + y_1 '' \,(\overline{\nu}_l\,\vev{\phi^{0}})_{\underline{1}''}\,\nu_r ' \nonumber\\ & \qquad -\left [ \widetilde{y}_1 \,(\overline{\nu}_l\,\vev{\phi^{0}})_{\underline{1}}\,\nu_r + \widetilde{y}_1 ' \,(\overline{\nu}_l\,\vev{\phi^{0}})_{\underline{1}'}\,\nu_r '' + \widetilde{y}_1 '' \,(\overline{\nu}_l\,\vev{\phi^{0}})_{\underline{1}''}\,\nu_r ' \right ] + \text{h.c.}\;.\label{eq : lr_mnud}\end{aligned}\ ] ] taking @xmath166 and then comparing eqs . ( [ eq : lr_me ] ) and ( [ eq : lr_mnud ] ) , one gets @xmath167 where @xmath168 . whereas the neutrino majorana mass matrix is a general complex symmetric just like in our other examples , the quark mass matrices have a special form . for @xmath169 , the expanded lagrangian , @xmath170 gives rise to a mass matrix of the form @xmath171 while it can be shown that mass matrix @xmath80 also has a similar structure : @xmath172 where @xmath173 , @xmath174 and @xmath175 are complicated functions of the vevs and yukawa couplings . equations ( [ eq : lr_mu ] ) and ( [ eq : lr_md ] ) imply that the diagonalization matrices @xmath176 and @xmath177 are not completely arbitrary . however , it is easy to see that there are enough degrees of freedom in the resulting @xmath178 such that experimental data can be fitted . returning to ( [ eq : lr_relation ] ) , it is clear that the main prediction of this model is @xmath179 where @xmath148 is _ a priori _ arbitrary and to be fitted to the experimental observations . the general conclusion from the previous section is that it is possible to use symmetries to construct the relation @xmath180 that links the high - energy seesaw sector to low - energy observables . using the current experimental data on quarks and leptons , the properties of the heavy rh majorana neutrinos in these models can therefore be inferred directly , and interesting consequences may arise . whilst the mixing matrix @xmath28 can be in general written as @xmath181 where @xmath182 , @xmath183 is the @xmath0-violating dirac phase , and @xmath184 and @xmath185 denote the two majorana phases , it is often more convenient to absorb the majorana phases into @xmath17 in ( [ eq : ph_mr ] ) and allow the @xmath186 s to be complex masses instead . when numerical analysis is required , we use the best fit values with @xmath187 errors for the mixing angles @xcite : @xmath188 but for our analytical work , we assume that @xmath28 has an exact tribimaximal form @xcite , with @xmath189 the inputs to the light neutrino mass matrix @xmath17 are restricted by the squared - mass differences : @xmath190 obtained from neutrino oscillation experiments @xcite and the cosmological bound on the sum of all neutrino masses : @xmath191 ev ( 95% c.l . ) @xcite which implies an absolute upper limit of @xmath192 in the following , we study ( [ eq : ph_mr ] ) by taking a generic form for @xmath193 where @xmath194 is assumed . it is obvious that once @xmath26 has been chosen ( i.e. @xmath195 s are known ) , only @xmath196 and @xmath197 ( or @xmath198 for the inverted hierarchy case ) can potentially change the form of @xmath4 and its eigenvalue spectrum . moreover , if @xmath199 , it is expected that the dirac phase , @xmath183 , would not play a significant role . , the choice of dirac phase can influence the mass eigenvalues by almost two orders of magnitude for certain sets of majorana phases and @xmath200 values , as our parameter space scans have indicated . ] so , to understand the leading behaviors of the mass spectrum for @xmath4 , we approximate @xmath28 with the tribimaximal form ( see ( [ eq : tb_values ] ) ) and absorb @xmath201 into @xmath202 respectively . after expanding out the rhs of ( [ eq : ph_mr ] ) , we have @xmath203 there are two limiting cases of eq . ( [ eq : mexact ] ) which can provide important insights into the dependence of the heavy rh majorana masses @xmath204 on the mass scale of the lightest lh neutrino . for the normal hierarchy scheme , we have @xmath205 with @xmath206 related to @xmath197 via ( [ eq : nh ] ) . therefore , in this limit , we can write eq . ( [ eq : mexact ] ) as @xmath207 is the dominant part of the matrix as @xmath205 , while @xmath208 is considered to be a small perturbation . suppose that the true eigenvalues and eigenvectors for @xmath4 can be expressed as @xmath209 and @xmath210 respectively for all @xmath211 , where @xmath212 is assumed . then , perturbation theory implies that the variation in the eigenvalues is given by @xmath213 where @xmath214 s are chosen to be orthonormal to each other . solving @xmath212 for @xmath215 , one immediately gets @xmath216 and subsequently @xmath217 in the limit of @xmath218 and @xmath219 . combining eqs . ( [ eq : mr_e0 ] ) and ( [ eq : mr_de ] ) , the heavy rh neutrino masses are @xmath220 hence , we can see that due to the large neutrino mixing , the expected correspondence between @xmath186 and the dirac masses , @xmath221 , no longer holds and that only the largest rh neutrino mass is a function of @xmath197 . substituting in the running fermion masses @xmath222 at @xmath223 @xcite as typical values for @xmath195 s , we have the following predictions for rh neutrino masses in the normal hierarchy case : @xmath224 the plots of @xmath225 as a function of @xmath197 for the case @xmath226 and for many different values of @xmath227 are shown in fig . [ fig : mu_m123 ] . these numerical results validate the trend predicted by the theoretical analysis . the tallest spikes in the diagrams of fig . [ fig : mu_m123 ] are locations where level crossing occurs ( @xmath228 or @xmath229 are quasi - degenerate ) for certain special values of dirac and majorana phases , an effect that has been previously studied in @xcite . plots of @xmath228 for the case @xmath230 are shown in fig . [ fig : md_me_m12 ] . for the inverted hierarchy scheme , @xmath231 , and hence , we take @xmath232 which then leads to the following expressions for the @xmath4 masses : @xmath233 the resulting numerical values for this case are similar to those shown in eqs . ( [ eq : mr_mass_u_case ] ) to ( [ eq : mr_mass_e_case ] ) although they are in general slightly smaller . when the lightest neutrino mass approaches the upper bound of ( [ eq : nu_mass_limit ] ) , we get @xmath234 . assuming that the majorana phases @xmath201 are negligible , then eq . ( [ eq : mexact ] ) becomes @xmath235 from this , we can immediately deduce the approximate scale for the @xmath204 s : @xmath236 these estimates agree well with the numerical results shown in fig . [ fig : mu_m123 ] and [ fig : md_me_m12 ] . vs. @xmath197 in the @xmath237 case ( left column ) and @xmath238 case ( right column ) with normal hierarchy for light neutrino masses assumed . input running masses used : ( left ) @xmath239 mev , @xmath240 mev , @xmath241 gev , and ( right ) @xmath242 mev , @xmath243 mev , @xmath244 gev , where @xmath245 gev . each plot contains approximately @xmath246 data points produced by systematically sweeping @xmath197 and the @xmath247 parameter space.,title="fig : " ] vs. @xmath197 in the @xmath237 case ( left column ) and @xmath238 case ( right column ) with normal hierarchy for light neutrino masses assumed . input running masses used : ( left ) @xmath239 mev , @xmath240 mev , @xmath241 gev , and ( right ) @xmath242 mev , @xmath243 mev , @xmath244 gev , where @xmath245 gev . each plot contains approximately @xmath246 data points produced by systematically sweeping @xmath197 and the @xmath247 parameter space.,title="fig : " ] vs. @xmath197 in the @xmath237 case ( left column ) and @xmath238 case ( right column ) with normal hierarchy for light neutrino masses assumed . input running masses used : ( left ) @xmath239 mev , @xmath240 mev , @xmath241 gev , and ( right ) @xmath242 mev , @xmath243 mev , @xmath244 gev , where @xmath245 gev . each plot contains approximately @xmath246 data points produced by systematically sweeping @xmath197 and the @xmath247 parameter space.,title="fig : " ] vs. @xmath197 in the @xmath237 case ( left column ) and @xmath238 case ( right column ) with normal hierarchy for light neutrino masses assumed . input running masses used : ( left ) @xmath239 mev , @xmath240 mev , @xmath241 gev , and ( right ) @xmath242 mev , @xmath243 mev , @xmath244 gev , where @xmath245 gev . each plot contains approximately @xmath246 data points produced by systematically sweeping @xmath197 and the @xmath247 parameter space.,title="fig : " ] using the @xmath4 mass spectrum information presented above , several general comments on the possibility of baryon asymmetry generation via thermal leptogenesis for the models discussed in section [ sec : models ] can be made . first of all , from the fact that @xmath248 is typically in the range of @xmath249 gev for all setups , it is clear that conventional leptogenesis where the asymmetry is generated predominantly by the decays of @xmath250 s would not be successful @xcite . however , there exist other special solutions to the leptogenesis scenario . as was pointed out earlier , the tall spikes in the plots of figs . [ fig : mu_m123 ] and [ fig : md_me_m12 ] indicate that there are regions in the parameter space for these models where @xmath251 and @xmath252 become almost degenerate . consequently , it has been shown in @xcite that a sufficient baryon asymmetry can be generated from resonant enhancement @xcite to the raw @xmath0 asymmetry in the decays of @xmath250 s . furthermore , a similar enhancement to the decay of the next - to - the - lightest rh neutrino @xmath253 , when @xmath252 and @xmath254 become degenerate , can also produce the desired asymmetry in principle , as long as washout effects mediated by the lighter @xmath250 s are insufficient @xcite . another interesting observation is that , recently , ref . @xcite discussed the possibility of successful leptogenesis ( without the need for resonant enhancement ) in models with @xmath70-inspired mass relations which have properties similar to those presented here ( see also @xcite ) . in the analysis of @xcite , they explored the situation where the asymmetry is first generated by @xmath253 decays at a temperature where flavor effects @xcite are important . specifically , the relevant range of @xmath255 gev leads to a two - flavor regime where the lepton asymmetry will be stored in the @xmath256-component , as well as a coherent superposition of @xmath257-components . subsequently , flavor dependent washout effects coming from interactions with @xmath250 s may not completely erase all components of the asymmetry generated by the @xmath253 s . one central conclusion in @xcite is that , for this mechanism to generate enough asymmetry , the mass of the next - to - the - lightest rh neutrino must be about @xmath258 gev . inspecting the @xmath252-plot of fig . [ fig : mu_m123 ] ( corresponding to the @xmath259 case ) , we can see that the condition of @xmath258 gev can be marginally met by a small region of the parameter space ( near the various spikes in the region where @xmath197 is between @xmath260 to @xmath261 ev ) , whereas the @xmath262 cases are definitely ruled out for this scenario due to the smallness of @xmath252 . therefore , it appears that for some special values of @xmath197 with certain sets of phases @xmath263 , leptogenesis via @xmath253 decays taking into account the effects of flavor is also possible ( for the @xmath259 model ) in addition to resonant leptogenesis . moreover , if this picture of flavored @xmath253-leptogenesis is indeed the mechansim responsible for generating the baryon asymmetry of the universe , then the corresponding sets of low energy phases in our model @xmath263 which make this possible will generally lead to modifications of the neutrinoless double beta decay rate through the quantity @xcite @xmath264 for example , taking @xmath265 ev and assuming normal hierarchy , the phases implied by @xmath253-leptogenesis will lead to @xmath266 ev , which is a noticeable reduction from 0.070 ev in cases where both majorana phases are turned off . ev in this discussion is because so far we have not found any set of phases for @xmath267 in which @xmath253-leptogenesis is actually viable . ] however , present experimental upper limits on @xmath268 lie somewhere between 0.16 and 0.68 ev @xcite , and so it is difficult to distinguish such differences . the detection of this may only be possible in future experiments such as cuore @xcite and gerda @xcite which have a projected sensitivity down to about 0.05 ev . in summary , while the models presented in sec . [ sec : models ] do not generically lead to successful baryon asymmetry generation via thermal leptogenesis , some fine - tuned special cases do exist . it is possible that the enlargement of the workable parameter space for leptogenesis can result from modifications to the higgs sector of these models , but such analyses are beyond the scope of this work . it is interesting to note that in the model with @xmath269 , the lightest heavy majorana neutrino mass @xmath248 can be about 1 tev making one wonder if it is possible to see signals of such a particle at the large hadron collider ( lhc ) and or a future international linear collider ( ilc ) . however , since the heavy majorana neutrinos are dominantly right - handed singlets which do not have gauge interactions , the interactions of the heavy neutral leptons with sm gauge bosons arise through their mixing between light neutrinos . the interaction lagrangians are parameterized through mixing angles @xmath270 ( @xmath271 ) of order @xmath272 as per @xmath273 where @xmath274 . with these interactions , it is possible to produce signals for heavy neutral leptons through @xmath275 followed by @xmath276 or @xmath277 . the production of @xmath1 by @xmath278 is much harder to study due to large backgrounds . however , in a model - independent study in ref . @xcite , such a mechanism was found to lead to a detectable heavy neutral lepton signal only if the mass is of order 100 gev or less , for the initial stage of lhc running with luminosity of order 10 @xmath279 . besides , the amplitudes of @xmath270 in our models are far too small . even assuming @xmath280 ev which will saturate the bound of ( [ eq : nu_mass_limit ] ) and in the best case scenario with inverted hierarchy and special choice of phases , one obtains @xmath281 ( with @xmath282 gev ) which is much less than the minimum required to produce an observable signal in any of the channels @xcite . the suppression is even greater for the @xmath283 or @xmath256 flavor . as a result , it is very difficult to detect the heavy neutral leptons through this mechanism even with an integrated luminosity up to 300 @xmath279 . if there is only one higgs doublet , there is also a light neutrino and heavy neutral lepton interaction with the higgs particle given by @xmath284 this interaction , although not of much help in the production of heavy neutral leptons through @xmath285 , does provide another channel for @xmath1 decay . if the higgs mass is not too much larger than the w boson mass , the decay rate is similar to that for @xmath286 or @xmath277 . in the models we are considering , there are several higgs doublets . the neutral higgs couplings to light neutrinos and heavy neutral leptons are then not necessarily proportional to @xmath287 and can increase the decay rate . also , in our models there are charged higgs bosons interacting with light neutrinos and heavy neutral leptons which also provide additional channels for detection of the @xmath1 s . but given the smallness of the mixing @xmath270 mentioned above , it is still very difficult to detect a heavy neutral lepton with mass of order 1 tev at the lhc even with 300 @xmath279 of luminosity . charged - higgs couplings to charged - leptons and heavy neutral leptons may have interesting signals at an ilc through @xmath288 with t - channel heavy higgs exchange , and @xmath289 with u - channel @xmath1 exchange @xcite . in particular the processes @xmath289 , are very sensitive the heavy neutral lepton mass . it has been shown in ref . @xcite that if @xmath290 is in the range of @xmath291 to @xmath292 , the ilc with an energy of 500 gev can probe heavy neutral lepton masses up to @xmath293 tev . in our case , the charged - higgs coupling to charged - leptons and heavy neutral leptons can be larger than @xmath294 , but still too small to be probed using the processes mentioned above . the main point of this paper was to demonstrate through general arguments backed up by explicit models that symmetries can be used to connect the rh majorana neutrino mass matrix to low - energy observables such as charged - fermion masses , mixing angles and @xmath0-violating phases . if a model of this type were to actually describe nature , then the benefit would be that the high - mass seesaw sector would be completely determined from low - energy observations , improving the predictability and testability of the seesaw neutrino mass generation mechanism . since this mechanism can also be used to understand the cosmological matter - antimatter asymmetry through leptogenesis , such constrained models are also important for cosmology . we focused on the simplest models of this type , which yielded @xmath295 where @xmath27 . our phenomenological analysis showed that successful leptogenesis is possible for the @xmath296 case in certain fine - tuned corners of parameter space . we also noted that the @xmath297 case can also supply a heavy neutral lepton with a mass of about 1 tev , opening the prospect for collider detection , though detailed analysis showed that this mass is still too large to plausibly expect detection at either the lhc or a future ilc . future work in this are could explore a possible role for the ckm matrix rather than the more obvious pmns matrix in the formula for @xmath4 . also , the use of clebsch - gordan coefficients to generalize the relationship between the neutrino dirac mass matrix and @xmath26 away from being a strict equality is another obvious line of investigation . finally , our explicit models used flavor symmetry to render the right - handed diagonalization matrices to be simply identity matrices . it could also be of interest to loosen this constraint . sscl and rrv thank nicole bell for discussions . xgh would like to thank kitpc for hospitality where part of this work was done . this work was supported in part by the australian research council , the commonwealth of australia , the nsc and ncts . @xmath91 is the alternating group of order 4 . it is isomorphic to the group representing the proper rotational symmetries of a regular tetrahedron . it has 12 elements and 4 conjugacy classes : one set containing the identity , two sets containing four 3-fold rotations each and one set of three 2-fold rotations . by the dimensionality theorem , we know that @xmath91 must have four irreducible representations : @xmath298 and @xmath299 , where @xmath300 is the trivial representation , @xmath301 and @xmath302 are non - trivial one - dimensional representations that are complex conjugate of each other , while @xmath299 is a real three - dimensional representation . some basic tensor product rules : @xmath303 where subscripts @xmath304 and @xmath305 denote `` asymmetric '' and `` symmetric '' respectively . suppose @xmath306 and @xmath307 are triplets in @xmath91 . then eq . ( [ app:3x3 ] ) means @xmath308 where @xmath309 and we have abbreviated @xmath310 with @xmath311 . b. t. cleveland _ et al . _ , nucl . phys . proc . * 38 * , 47 ( 1995 ) ; y. fukuda _ et al . _ [ super - kamiokande collaboration ] , phys . lett . * 82 * , 2430 ( 1999 ) [ arxiv : hep - ex/9812011 ] ; y. fukuda _ et al . _ [ super - kamiokande collaboration ] , phys . rev . lett . * 81 * , 1562 ( 1998 ) [ arxiv : hep - ex/9807003 ] ; k. lande _ et al . _ , nucl . suppl . * 77 * , 13 ( 1999 ) ; d. n. abdurashitov _ et al . _ [ sage collaboration ] , nucl . suppl . * 77 * , 20 ( 1999 ) ; t. a. kirsten [ gallex and gno collaborations ] , nucl . suppl . * 77 * , 26 ( 1999 ) ; y. fukuda _ et al . _ [ super - kamiokande collaboration ] , phys . b * 436 * , 33 ( 1998 ) [ arxiv : hep - ex/9805006 ] ; c. athanassopoulos _ et al . _ [ lsnd collaboration ] , phys . rev . c * 54 * , 2685 ( 1996 ) [ arxiv : nucl - ex/9605001 ] ; c. athanassopoulos _ et al . _ [ lsnd collaboration ] , phys . c * 58 * , 2489 ( 1998 ) [ arxiv : nucl - ex/9706006 ] ; q. r. ahmad _ et al . _ [ sno collaboration ] , phys . rev . lett . * 89 * , 011301 ( 2002 ) [ arxiv : nucl - ex/0204008 ] ; s. n. ahmed _ et al . _ [ sno collaboration ] , phys . lett . * 92 * , 181301 ( 2004 ) [ arxiv : nucl - ex/0309004 ] ; k. eguchi _ et al . _ [ kamland collaboration ] , phys . * 90 * , 021802 ( 2003 ) [ arxiv : hep - ex/0212021 ] . p. minkowski , phys . b * 67 * , 421 ( 1977 ) ; t. yanagida , in _ workshop on unified theories _ , kek report 79 - 18 p.95 ( 1979 ) ; m. gell - mann , p. ramond , r. slansky , in _ supergravity _ ( north holland , amsterdam , 1979 ) eds . van nieuwenhuizen , d. freedman , p.315 ; s. l. glashow , in _ 1979 cargese summer institute on quarks and leptons _ ( plenum press , new york , 1980 ) eds . m. levy , j .- l . basdevant , d. speiser , j. weyers , r. gastmans and m. jacobs , p.687 ; r. barbieri , d. v. nanopoulos , g. morchio and f. strocchi , phys . b * 90 * , 91 ( 1980 ) ; r. n. mohapatra and g. senjanovic , phys . lett . * 44 * , 912 ( 1980 ) ; g. lazarides , q. shafi and c. wetterich , nucl . b * 181 * , 287 ( 1981 ) . w. konetschny and w. kummer , phys . b * 70 * , 433 ( 1977 ) ; t. p. cheng and l. f. li , phys . d * 22 * , 2860 ( 1980 ) ; g. lazarides , q. shafi and c. wetterich , nucl . b * 181 * , 287 ( 1981 ) ; j. schechter and j. w. f. valle , phys . rev . d * 22 * , 2227 ( 1980 ) ; r. n. mohapatra and g. senjanovic , phys . rev . d * 23 * , 165 ( 1981 ) . r. foot , h. lew , x. g. he and g. c. joshi , z. phys . c * 44 * , 441 ( 1989 ) . m. fukugita and t. yanagida , phys . b * 174 * , 45 ( 1986 ) . see for example : w. buchmller , p. di bari and m. plmacher , annals phys . * 315 * , 305 ( 2005 ) [ arxiv : hep - ph/0401240 ] , w. buchmuller , r. d. peccei and t. yanagida , ann . nucl . part . * 55 * , 311 ( 2005 ) [ arxiv : hep - ph/0502169 ] . e. k. akhmedov , v. a. rubakov and a. y. smirnov , phys . * 81 * , 1359 ( 1998 ) [ arxiv : hep - ph/9803255 ] . see for example : k. abazajian , g. m. fuller and m. patel , phys . d * 64 * , 023501 ( 2001 ) [ arxiv : astro - ph/0101524 ] . t. asaka , m. shaposhnikov and a. kusenko , phys . b * 638 * , 401 ( 2006 ) [ arxiv : hep - ph/0602150 ] , t. asaka , s. blanchet and m. shaposhnikov , phys . b * 631 * , 151 ( 2005 ) [ arxiv : hep - ph/0503065 ] , and references within . s. n. ahmed _ et al . _ [ sno collaboration ] , phys . lett . * 92 * , 181301 ( 2004 ) [ arxiv : nucl - ex/0309004 ] ; m. h. ahn _ [ k2k collaboration ] , phys . lett . * 90 * , 041801 ( 2003 ) [ arxiv : hep - ex/0212007 ] ; g. l. fogli , e. lisi , a. marrone , a. palazzo and d. montanino , nucl . suppl . * 118 * , 177 ( 2003 ) [ arxiv : hep - ph/0310012 ] ; g. l. fogli , e. lisi , a. marrone , a. melchiorri , a. palazzo , p. serra and j. silk , phys . d * 70 * , 113003 ( 2004 ) [ arxiv : hep - ph/0408045 ] ( and references therein ) ; w. m. yao _ et al . _ [ particle data group ] , j. phys . g * 33 * , 1 ( 2006 ) . t. schwetz , m. tortola and j. w. f. valle , arxiv:0808.2016 [ hep - ph ] . z. g. berezhiani and m. y. khlopov , sov . j. nucl . * 51 * , 935 ( 1990 ) [ yad . fiz . * 51 * , 1479 ( 1990 ) ] . h. georgi and s. l. glashow , phys . lett . * 32 * , 438 ( 1974 ) . s. m. barr , phys . b * 112 * , 219 ( 1982 ) . r. n. mohapatra and j. c. pati , phys . d * 11 * , 2558 ( 1975 ) , r. n. mohapatra and j. c. pati , phys . d * 11 * , 566 ( 1975 ) . j. c. pati and a. salam , phys . d * 10 * , 275 ( 1974 ) [ erratum - ibid . d * 11 * , 703 ( 1975 ) ] . r. foot and h. lew , phys . d * 41 * , 3502 ( 1990 ) , r. foot , h. lew and r. r. volkas , phys . d * 44 * , 1531 ( 1991 ) . h. georgi and c. jarlskog , phys . b * 86 * , 297 ( 1979 ) . c. i. low and r. r. volkas , phys . d * 68 * , 033007 ( 2003 ) [ arxiv : hep - ph/0305243 ] . p. f. harrison , d. h. perkins and w. g. scott , phys . b * 530 * , 167 ( 2002 ) [ arxiv : hep - ph/0202074 ] , p. f. harrison and w. g. scott , phys . b * 535 * , 163 ( 2002 ) [ arxiv : hep - ph/0203209 ] ; phys . b * 557 * , 76 ( 2003 ) [ arxiv : hep - ph/0302025 ] , z. z. xing , phys . b * 533 * , 85 ( 2002 ) [ arxiv : hep - ph/0204049 ] , x. g. he and a. zee , phys . b * 560 * , 87 ( 2003 ) [ arxiv : hep - ph/0301092 ] . e. ma and g. rajasekaran , phys . d * 64 * , 113012 ( 2001 ) [ arxiv : hep - ph/0106291 ] , k. s. babu , e. ma and j. w. f. valle , phys . b * 552 * , 207 ( 2003 ) [ arxiv : hep - ph/0206292 ] , e. ma , phys . d * 70 * , 031901 ( 2004 ) [ arxiv : hep - ph/0404199 ] , e. ma , arxiv : hep - ph/0409075 , g. altarelli and f. feruglio , nucl . b * 720 * , 64 ( 2005 ) [ arxiv : hep - ph/0504165 ] , g. altarelli and f. feruglio , nucl . b * 741 * , 215 ( 2006 ) [ arxiv : hep - ph/0512103 ] , k. s. babu and x. g. he , arxiv : hep - ph/0507217 , a. zee , phys . b * 630 * , 58 ( 2005 ) [ arxiv : hep - ph/0508278 ] , e. ma , phys . d * 72 * , 037301 ( 2005 ) [ arxiv : hep - ph/0505209 ] , e. ma , mod . phys . a * 20 * , 2601 ( 2005 ) [ arxiv : hep - ph/0508099 ] . h. lew and r. r. volkas , phys . d * 47 * , 1356 ( 1993 ) [ arxiv : hep - ph/9209288 ] . x. g. he , y. y. keum and r. r. volkas , jhep * 0604 * , 039 ( 2006 ) [ arxiv : hep - ph/0601001 ] . e. komatsu _ et al . _ [ wmap collaboration ] , arxiv:0803.0547 [ astro - ph ] . e. k. akhmedov , m. frigerio and a. y. smirnov , jhep * 0309 * , 021 ( 2003 ) [ arxiv : hep - ph/0305322 ] . g. c. branco , r. gonzalez felipe , f. r. joaquim and m. n. rebelo , nucl . b * 640 * , 202 ( 2002 ) [ arxiv : hep - ph/0202030 ] . h. fusaoka and y. koide , phys . d * 57 * , 3986 ( 1998 ) [ arxiv : hep - ph/9712201 ] . m. flanz , e. a. paschos and u. sarkar , phys . b * 345 * , 248 ( 1995 ) [ erratum - ibid . b * 382 * , 447 ( 1996 ) ] [ arxiv : hep - ph/9411366 ] ; m. flanz , e. a. paschos , u. sarkar and j. weiss , phys . lett . b * 389 * , 693 ( 1996 ) [ arxiv : hep - ph/9607310 ] ; a. pilaftsis , phys . d * 56 * , 5431 ( 1997 ) [ arxiv : hep - ph/9707235 ] . see also a. pilaftsis and t. e. j. underwood , nucl . b * 692 * , 303 ( 2004 ) [ arxiv : hep - ph/0309342 ] ; a. pilaftsis and t. e. j. underwood , phys . d * 72 * , 113001 ( 2005 ) [ arxiv : hep - ph/0506107 ] ; a. de simone and a. riotto , jcap * 0708 * , 013 ( 2007 ) [ arxiv:0705.2183 [ hep - ph ] ] and references therein . p. di bari , nucl . b * 727 * , 318 ( 2005 ) [ arxiv : hep - ph/0502082 ] . bari and a. riotto , arxiv:0809.2285 [ hep - ph ] . a. abada , p. hosteins , f. x. josse - michaux and s. lavignac , arxiv:0808.2058 [ hep - ph ] . r. barbieri , p. creminelli , a. strumia and n. tetradis , nucl . b * 575 * , 61 ( 2000 ) [ arxiv : hep - ph/9911315 ] , o. vives , phys . d * 73 * , 073006 ( 2006 ) [ arxiv : hep - ph/0512160 ] , e. nardi , y. nir , e. roulet and j. racker , jhep * 0601 * , 164 ( 2006 ) [ arxiv : hep - ph/0601084 ] , a. abada , s. davidson , f. x. josse - michaux , m. losada and a. riotto , jcap * 0604 * , 004 ( 2006 ) [ arxiv : hep - ph/0601083 ] , a. abada , s. davidson , a. ibarra , f. x. josse - michaux , m. losada and a. riotto , jhep * 0609 * , 010 ( 2006 ) [ arxiv : hep - ph/0605281 ] , s. blanchet and p. di bari , jcap * 0703 * , 018 ( 2007 ) [ arxiv : hep - ph/0607330 ] . s. m. bilenky and s. t. petcov , rev . phys . * 59 * , 671 ( 1987 ) [ erratum - ibid . * 61 * , 169.1989 errat,60,575 ( 1989 errat,60,575 - 575.1988 ) ] ; s. r. elliott and p. vogel , ann . nucl . part . sci . * 52 * , 115 ( 2002 ) [ arxiv : hep - ph/0202264 ] . c. arnaboldi _ et al . _ [ cuoricino collaboration ] , phys . c * 78 * , 035502 ( 2008 ) [ arxiv:0802.3439 [ hep - ex ] ] . c. arnaboldi _ et al . _ [ cuore collaboration ] , nucl . instrum . a * 518 * , 775 ( 2004 ) [ arxiv : hep - ex/0212053 ] ; e. fiorini , phys . rept . * 307 * , 309 ( 1998 ) ; r. ardito _ et al . _ , arxiv : hep - ex/0501010 .

we propose the use of the kinetic sunyaev - zeldovich ( ksz ) effect to probe the circumgalactic medium ( cgm ) , with the aid of a spectroscopic survey covering the same area of a sz survey . one can design an optimal estimator of the ksz effect of the cgm with a matched filter , and construct the cross correlation between the estimator and the peculiar velocity recovered from the galaxy survey , which can be measured by stacking a number of galaxies . we investigate two compelling profiles for the cgm , the mb profile and the profile , and estimate the detectability against the synergy of a fiducial galaxy survey with number density mpc and an act - like sz survey . we show that the shape of the filter does not change much with redshift for the profile , while there are significant side lobes at for the mb profile . by stacking milky way - size halos around z , one can get 1 signal to noise ( s / n ) for the both profiles . the s / n increases with decreasing redshift before it reaches a maximum ( 7.5 at z 0.15 for the mb profile , at for the profile ) . due to the large beam size , a planck - like cmb survey can marginally detect the ksz signal by stacking the same number of galaxies at . the search for the cgm in realistic surveys will involve dividing the galaxies into subsamples with similar redshift and mass of host halos , and scaling the results presented here to obtain the s / n . cosmology : theory cosmic background background methods : statistical galaxies : intergalactic medium

1602.08932

cosmologists have noticed the @xmath8 50% shortfall of the cosmic baryons at low redshift compared to the baryon content synthesized in the big bang for a long time @xcite . the cosmic baryon budget has been predicted from a series of cosmic microwave background ( cmb ) surveys @xcite , e.g. @xmath12 from the most recent measurement @xcite . at high redshifts , ly@xmath13 forest can account for all baryons expected @xcite . however , careful census at low redshift revealed that baryons in collapsed objects such as galaxies , galaxy groups and clusters only account for about @xmath8 10% of the cosmic mean of the baryonic content , and @xmath830% of the baryons lie in the gas phase revealed by the ly@xmath13 absorption line . the rest of cosmic baryons are missing . in light of cosmological hydrodynamic simulations , majority of the missing baryons are believed to reside in a phase called warm hot intergalactic medium ( whim ) with temperature between @xmath14 k @xcite . the whim in the absorbers of ultraviolet ( uv ) and x - ray absorption lines signatured by ly@xmath13 , ovi , ovii and other highly ionized metals has been extensively investigated ( e.g. * ? ? ? * ; * ? ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and the evidence of the whim is accumulating . a related problem is the galactic missing baryon problem @xcite . observationally , the baryonic fraction increases monotonically with the mass of the gravitational system , and only galaxy clusters and massive galaxies contain the cosmic mean fraction of baryons @xcite . for low mass galaxies , stars and interstellar medium ( ism ) only account for a small fraction of the baryons , leaving @xmath6 80% of the baryons undetected in observations . the theoretical concept of a hot ( @xmath8 10@xmath15k ) diffuse corona has been discussed and modeled in numerous literatures @xcite . these hot coronae could have been shock heated by accretion , or ejected from galaxies into the intergalactic medium ( igm ) or still detained in the galactic dark halo as cgm , depending on the feedback energy in the structure formation process . there have been a number of tentative detections of the cgm for both early type galaxies and late type galaxies in uv and x - ray band , via both emission and absorption ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? on the other hand , for our own galaxy , the presence of hot halo can explain the lack of hi signature in local dwarf spheroidal galaxies @xcite and the head tail structure of the high velocity clouds ( hvc)@xcite . the x - ray emission measure also puts a constraint on the baryonic content in the cgm of the milky way ( mw ) . @xcite discussed that an extended hot gaseous halo profile @xcite can agree well with the existent constraints . although no broad conclusion of the extent and the profile of the cgm has been drawn , the search for the cgm has been a critical issue , and it s possible to detect the cgm through a careful design of observation considering the ubiquity of the cgm in galaxies . the sunyaev - zeldovich ( sz ) effect @xcite is one of the most promising methods of detecting the missing baryons . after the reionization epoch , our universe is almost completely ionized after z=6 , and free electrons are prevailing in galaxy clusters as intracluster medium ( icm ) and in between as igm . they will scatter off cmb photons via inverse compton scattering , and generate secondary cmb anisotropies known as the sz effect . the sz effect is therefore contributed by all electrons , and it can serve as a tool to probe the electrons in ionized gas and hence the missing baryons in the icm / igm / cgm . due to relatively low density and temperature of the cgm , it s difficult to detect them via the thermal sz ( tsz ) effect ( @xmath16 ) and x - ray emission ( @xmath17 ) . the kinetic sz ( ksz ) effect , however , directly measures the electron momentum @xmath18 along the line of sight , and is thus less weighted toward hot gas ( icm ) in galaxy groups and clusters . for less massive halos with virial temperature around @xmath19 k , the ksz effect turns out to be larger than the tsz effect @xcite . considering the @xmath20% fraction of missing baryons , the prevailing igm / cgm around galaxies can become a detectable ksz effect source . blind detection of the ksz signal from the auto angular correlation alone seems impossible because of the overwhelming contamination of cosmic infrared background ( cib ) and the degeneracy between the ksz effect and the primary cmb . the state - of - art observations like the planck satellite , south pole telescope ( spt ) , and atacama cosmology telescope ( act ) alone are not able to measure the auto - correlation signal of the kinetic sz effect . most of these surveys alone can only put an upper limit of the ksz effect @xcite . taking into account the additional information of the tsz bispectrum can help improve the constraint on the ksz effect @xcite . however , the results inevitably depend on both the thermal and kinetic sz templates . since the ksz effect traces the peculiar momenta of the free electrons and thus leaves imprints in the cmb sky , it can be used to constrain cosmological models or cosmological parameters . recent works have devoted to measuring the ksz effect on the cmb sky at the positions of galaxy clusters or galaxies . works have focused on measuring the bulk flow of the local universe using galaxy clusters @xcite , however , the existence of the bulk flow remained controversial due to low significance @xcite . the first solid detection of the ksz effect @xcite reported a 3.8 @xmath7 significance by measuring the pairwise ksz effect of galaxy clusters , taking advantage of the synergy between the boss spectroscopic galaxy group catalogue and the act survey . alternatively , the ksz effect was proposed to probe the missing baryons @xcite . progress has been made towards the measurement of the ksz effect using galaxies . @xcite claimed the detection of the local bulk flow by measuring the ksz effect toward the direction of nearby galaxies , which indicates the existence of hot plasma around galaxies . moreover , recent works @xcite found the evidence of ksz signature out to about three times the mean virial radius of the central galaxy catalogue ( cgc ) sample , in which they employed both the pairwise momentum estimator and the ksz temperature - velocity field correlation estimator . these results suggest that the hot plasma around galaxies may be responsible for the missing baryons of galaxies , and produce detectable ksz effect . considering the cgm may account for a significant fraction of the missing baryons of galaxies , we here present a work to probe the prevailing cgm through the measurement of the kinetic sz effect at the galaxy positions in the cmb map . previous work by @xcite has theoretically studied the tsz / ksz effect of the cgm , and discussed the detectability of the sz effect of the cgm as well as the constraints on the gas fraction . we here alternatively propose to measure the ksz effect of the cgm and the detectability of the signal with different profiles . we try to design an unbiased minimum variance estimator of the ksz effect of the cgm in order to minimize the noise , and then stack the velocity weighted ksz signal of galaxies to further improve the measurement . the minimum variance estimator involves finding out an optimal filter to filter out the large scale contamination of the primary cmb while keep the ksz signal of cgm conserved . the design is unprejudicedly constructed for both the icm in galaxy groups and the cgm around galaxies , provided we have the knowledge of the profile of the optical depth profile of the icm / cgm ( e.g. see * ? ? ? * for the application to mock sample of clusters ) . _ in particular , we here focus on the ksz effect of the cgm of milky way(mw)-size halos . _ since the baryonic budget and the radial extent of the cgm are still under debate @xcite , we here confine the cgm to two compelling profiles : the hot extended halo profile ( * ? ? ? * hereafter mb profile ) and the empirical @xmath0 profile @xcite . in this work , we are mostly interested in the relatively flatter mb profile , as extensively discussed in @xcite concerning the milky way , and it proves a promising profile of the missing baryons of the mw - size halos . we treat the @xmath0 profile as a comparison model , which can give some insight to the upper limit of detectability of the cgm . the fiducial cosmology we adopt in this work is the standard flat @xmath21cdm cosmology @xcite : @xmath22 , the cosmic baryon fraction @xmath23 . we focus on the detectability of ska - like galaxy surveys and act - like sz surveys , and discuss the possible application to other surveys in [ sec : discussion ] . ska will survey @xmath24 hi galaxies in a wide sky area and a deep volume , and will make a fair sample of the reservoir of the cgm of late type galaxies . for mw- size galaxies to be resolved , we need sz surveys with the resolution like the act / spt survey . the organization of the paper is as follows . we first introduce the models and show the ksz effect of the cgm in [ sec : modeling_cgm_ksz ] , and then raise the formalism and methodology to measure the ksz effect from the cgm in section [ sec : methodology ] . we then detail the cross correlating ksz effect with recovered velocity in [ sec : isolating_ksz ] and further apply the matched filter to improve the detection and predict the detectability in [ sec : results ] . we present the conclusion and discussion in [ sec : discussion ] . when cmb photons travel along the line of sight , the high energy electrons in the intervening medium ( e.g. icm , igm or cgm ) will scatter off the cmb photons and distort the energy spectrum of the cmb via the inverse compton scattering . this process gives rise to the tsz effect due to the thermal motion of electrons , and to the ksz effect if the scattering medium has a bulk motion with respect to the rest frame of the cmb . for the cgm of a milky way - size galaxy , the ksz effect is larger than the tsz effect @xcite . suppose we have a galaxy with its centre at the position @xmath25 in the 2d sky , where we have adopted the flat sky approximation as the galaxy s scale is very small @xcite . the peculiar momenta of the electrons in the surrounding cgm will induce the kinetic sz effect @xmath26 . in this work , we assume the ksz signal of the cgm is spherical symmetry around the galaxy centre , and we can always put the galaxy centre @xmath27 at the origin , therefore from herein we write any function of @xmath28 as a function of the radial separation @xmath29 . the induced ksz signal is proportional to the electron momentum integrated along the line of sight @xmath30 @xmath31 is the optical depth profile of the cgm , where @xmath32 is the scale factor , @xmath33 is the comoving radial distance , @xmath34 is the electron number density profile , and @xmath35 is the thompson cross section . since for a single galaxy , the peculiar velocity along the line of sight @xmath36 is nearly constant , the ksz induced temperature distortion is proportional to the product of the optical depth and the line of sight velocity @xmath37 here , we have defined @xmath38 as @xmath39 . the observed optical depth profile is the convolution @xmath40 between the underlying profile @xmath41 and the beam window @xmath42 . we assume the beam of the sz survey is a gaussian profile @xmath43 , with @xmath44 , and @xmath45 the full width half maximum ( fwhm ) of the beam profile . accordingly , the beam window in fourier space can be written as @xmath46 . in practice , we calculate the smoothed optical depth profile in fourier space as @xmath47 . the mass fraction and the underlying spatial distribution of the cgm are still under debate . self - regulated galaxy simulations @xcite indicate that the baryonic fraction of galaxies can vary substantially . a reasonable range covers 30 - 100% of the cosmic mean @xmath48 . on the other hand , observations have made little advance in constraining the mass fraction of the cgm , which covers an even broader range of the mass fraction @xmath8 10 - 100% @xcite . the profile of the cgm is also an open question , and the empirical @xmath0 profile is often adopted in the analysis of the observational data , especially for elliptical and massive spiral galaxies @xcite . however , @xcite found that the mb profile @xcite can account for the missing baryons of the milky way , and is favoured by current observational constraints . in this work , we take the mb profile as a fiducial profile of the cgm for mw - size spiral galaxies , and the @xmath0 model as a comparison profile . we calculate the ksz effect in the context of future galaxy surveys such as ska . such survey will detected a huge number of hi field galaxies and is less subject to the contamination of the hot icm . we expect that most galaxies will be located in halos with mass range @xmath49@xmath50 , and assume that gas within the virial radius is in the hydrostatic equilibrium . given the redshift distribution and mass distribution of the surveyed galaxies , we can in principle calculate the differential contribution of host halos in certain mass and redshift bin , and obtain the total stacked signal . though , we will take the example of the typical milky way - size halo of @xmath51 m@xmath52 in this work . the radial density profile of the mb model ( also known as extended hot halo model ) follows @xmath53 with the profile @xmath54^{3/2}\ , \ ] ] where @xmath55 is the normalized value of gas density at virial radius ( please find details in @xcite ) . @xmath56 , and @xmath57 is the characteristic radius defined by @xmath58 , where the virial radius @xmath59 kpc and the concentration @xmath60=12 have been adopted for milky way - size host halo . assuming a baryonically closed system , the baryonic mass of the cgm within the virial radius is @xmath61 . in the following , we assume the baryon fraction @xmath62 , and the gas fraction @xmath63 within the virial radius . we assume the missing mass of hot gas is @xmath64 as in their work , which implies @xmath65 of the cosmic mean . for comparison , besides the fiducial profile , we also test the detectability of a steeper density profile , the @xmath0 profile , which follows @xmath66 the @xmath0 profile usually fits well to the hot gas around elliptical galaxies @xcite and galaxy groups and clusters@xcite . a realistic @xmath0 profile would require fitting two parameters , the core radius @xmath67 and the profile index @xmath0 , to the x - ray emission data . we here assume the index @xmath68 , and fix @xmath69 for a direct comparison between the mb profile and the @xmath0 profile . we also require a density normalization that integrated gas mass within the virial radius to be @xmath70 . though the @xmath0 profile is often adopted for the hot gas around elliptical galaxies and icm in galaxy clusters , it has been applied to giant spiral galaxies lately @xcite . whether @xmath0 profile is appropriate for cgm in late type galaxies is still under debate @xcite . we here choose the @xmath0 profile to demonstrate the detectability of the cgm , which may put constraints on its upper limit . the profile of the optical depth can be obtained by integrating the gas density alone the sightline : @xmath71 where @xmath72 is the ionization fraction , and @xmath73 is the projection length . at low redshift , we assume the universe is fully ionized such that @xmath74 . @xmath75 is the projected distance of the line of sight to the centre of the galaxy , and @xmath76 is the upper range of the integration which means we cut the integration at the virial radius . @xmath77 is the electron molecular weight , and @xmath78 is the proton mass . the observed optical depth @xmath79 is the convolution of the underlying optical depth @xmath80 and the beam profile @xmath42 , i.e. @xmath81 . for a mw - size halo with a virial mass @xmath82 , the typical virial radius is @xmath5 arcmin at @xmath83 . it can be marginally resolved by the act - like sz survey . at higher redshift , the galaxies are not resolved or just marginally resolved so that most of the signal is within the beam and concentrated in the central pixel regardless of spatial profile . thus a similar optical depth profile for different cgm profiles is expected after the beam convolution at higher redshifts . below redshift @xmath83 , the galaxy can be well resolved into several concentric rings of pixels . the beam convolved ksz profiles of the cgm are shown in fig . [ fig : tksz ] at several redshifts ( @xmath84 ) . at redshifts @xmath3 , the ksz profile of the mb model is much flatter and the signal amplitude is much smaller than those of the @xmath0 profile in the central region . we expect less benefit from filtering the mb s profile , because the filter may not well separate the spatial pattern between the ksz effect of extended hot halo and the primary cmb . profile ( lower panel ) . the ksz signal of mb model is much flatter in the entire region , and smaller than @xmath0 profile in the central region . [ fig : tksz],width=307 ] the ksz effect is immersed in the overwhelming contamination constituted by the thermal sz effect @xmath85 , the primary cmb fluctuation @xmath86 , the foreground cosmic infrared background @xmath87 @xcite , and the instrument noise @xmath88 imposed by the detector . the noisy temperature distortion profile @xmath89 of the cgm at galaxy position is given by @xmath90 where we have contracted @xmath91 as a term @xmath92 , as they are all noise to the ksz effect . in realistic surveys , the sky map is pixelated on the receiver . for each pixel in the observed cmb map , both sky and the instrument noise contribute to the measured temperature . all the components except the instrument noise @xmath88 should be smoothed by the telescope beam . to model the instrument noise , we assume that errors are uncorrelated between pixels and have uniform variance @xmath93 , i.e. @xmath94 @xcite . the power spectrum of the instrument noise thus can be written as @xmath95 for act - like survey , we adopt @xmath96 arcmin for the beam size , pixel size @xmath97 arcmin , and @xmath98k noise per pixel . these fiducial parameters are adopted through the paper unless specified otherwise . the problem of probing the ksz effect of the cgm is now equivalent to the problem of optimally solving for @xmath99 in eq . [ eq : noisy_map ] . the basic idea is outlined as follows . the problem requires separating the ksz signal from the noisy cmb map by constructing a correlation , and then beating the noise from all the other components , mainly from the primary cmb . as shown in previous works , the ksz effect of the cgm is correlated with an effective velocity field @xmath100 , which we can reconstruct from spectroscopic galaxy surveys . the correlation is only contributed by ksz effect , but the noise term is contributed by all the components in the cmb sky plus the instrument noise . we can then optimally remove the noise with the matched filter . this would require the knowledge of the optical profile of the underlying ksz signal of the cgm . finally we stack galaxies of the same host halo mass to boost the signal to noise @xmath101 . the presence of the overwhelming noise , e.g. the primary cmb , the cib from dusty star forming galaxies , the tsz from the very source of ksz effect , and the potential correlation between the cib and the tsz , makes the direct detection of the ksz effect nearly impossible . in principle , one can separate the signal if there are different spectral dependences between different components of the noise term , by taking advantage of modern cmb surveys with multiple bands . however , the ksz effect itself is not only much smaller than other components , but is also degenerate with the primary cmb , as it lacks spectral dependence . the local velocity or momentum is a good tracer of the underlying ksz effect . we can therefore weight the temperature distortion profile @xmath102 around galaxies with an effective velocity @xmath103 at the galaxy position . galaxy density @xmath104 traces the underlying dark matter distribution @xmath105 within an accuracy of a bias factor , i.e. @xmath106 . with the aid of spectroscopic galaxy survey , we can reconstruct an effective velocity field @xmath103 that traces the underlying peculiar velocity @xmath107 , by assuming that the matter density @xmath105 follows the linear growth theory . such approximation was widely adopted , and the reconstructed velocity traces the underlying velocity well ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * for more discussion ) . given the galaxy distribution , we can construct the underlying dark matter density field @xmath105 , which should follow the continuity equation of dark matter in the linear approximation @xmath108 where @xmath107 is the underlying peculiar velocity . in the regime of the linear perturbation theory , one can reconstruct the effective velocity field @xmath103 with @xmath109 where @xmath110 is the hubble parameter at redshift @xmath111 , @xmath112 and @xmath113 is the linear density growth rate . we have employed that the density follows the linear perturbation theory , and the galaxy is linearly biased against the dark matter with @xmath106 , where @xmath114 is the linear bias of galaxies . from galaxy surveys , we can measure the galaxy bias @xmath114 beforehand . though there are stochasticities in galaxy bias , studies have shown that the uncertainty in galaxy bias is likely to be at the level of 10% @xcite , and can be negligible at the scales of interest @xcite . this construction cancels out the cross correlation between the velocity and all the other components except the ksz effect , since they do nt have the characteristic directional dependences and should be uncorrelated with the reconstructed velocity . details can be found in @xcite . therefore , we are able to isolate the ksz signal of the cgm . statistically , the line of sight component of the recovered velocity @xmath115 is correlated with underlying velocity field @xmath36 , and is thus correlated with the ksz effect @xmath116 ( momentums ) . so that we obtain the correlation of the recovered velocity and the cmb map @xmath117 \right > \\ & = \langle v_{\parallel } t_{\rm ksz}(\theta ) \rangle \\ & = \langle v_{\parallel } \tilde{u } \rangle \tau_b ( \theta ) . \end{split}\ ] ] in the second line of the equation , we have dropped out @xmath118 . when we stack the correlation , we assume that all the galaxies share the same optical depth profile @xmath119 ( see the discussion below for the possible uncertainty of @xmath80 profile ) . now that one separates the ksz signal from the noisy map by constructing the correlation estimator between the reconstructed velocity and the underlying ksz signal of the cgm , one need to assess the performance of the estimator . one of the most important statistics to evaluate the performance is the measurement error , which is known as the variance of the correlation @xmath120 , @xmath121 ^ 2 \rangle \\ & = \langle [ v_\parallel \delta t(\theta)]^2 \rangle- \langle v_\parallel \delta t(\theta ) \rangle^2 \\ & \simeq \langle v^2_\parallel \rangle \langle \delta t^2(\theta ) \rangle \ . \end{split}\ ] ] in the last line of eq . [ eq : stacking_var_nofilter ] , we omit @xmath122 since it s much smaller than the variance @xmath123 . we also made an assumption that @xmath115 and @xmath102 follow the distribution of gaussian fields . we then expand the equation using wick theorem , and omit any terms concerning 3-point correlation . the measurement error is constituted by the product of the velocity dispersion @xmath124 and the temperature variance @xmath125 of each pixel in the cmb survey . it s clear from eq . [ eq : stacking_var_nofilter ] that the temperature variance of each pixel is contributed by all the components in the pixel , including the predominant cmb variance . therefore , cross correlating the cmb map with the reconstructed velocity @xmath115 merely separates the ksz effect of cgm from other components of the noisy cmb map , while the detection is still subject to large measurement error . as shown later in [ subsec : signal_to_noise_nofilter ] , stacking @xmath8 1 million mw - size halos is necessary to achieve 1 @xmath7 detection . thus a direct cross correlation between the ksz effect of the cgm with the reconstructed velocity picks out only the ksz effect . to improve the detection , we need to further reduce the measurement error by removing the contamination of all the other components . this involves utilizing the knowledge of the covariance matrix of the noise , and the spatial distribution of the ksz signal of the cgm , which in turn lies in the optical depth profile @xmath119 as in eq . [ eq : stacking_corr ] . given the profile of @xmath80 , we can design an unbiased estimator @xmath126 of the underlying velocity field @xmath39 with a matched filter @xmath127 to optimally remove the noise mainly from the primary cmb . then by cross correlating @xmath128 one can achieve a much higher s / n . we now turn to the problem of finding out an optimal estimator of @xmath129 from eq . [ eq : noisy_map ] . the primary cmb dominates the variance in each pixel . as shown in [ subsec : signal_to_noise_nofilter ] , the primary cmb gets the most of the contribution from large scale multipoles , while the ksz signal from the cgm is mainly from the scale of galaxy size . thus we can take advantage of the different scale dependences and optimally subtract the primary cmb from large scales with an unbiased minimum variance estimator @xmath126 , by employing matched filter ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? an optimal matched filter @xmath130 should satisfy the following requirements : ( 1 ) @xmath126 is an unbiased estimator of the amplitude @xmath39 of the underlying ksz signal ; ( 2 ) the variance between the estimator @xmath126 and underlying signal @xmath39 is minimized . the nominal temperature distortion profile around one galaxy is @xmath131 . we can construct an effective quantity @xmath126 as an estimate of @xmath39 , @xmath132 where @xmath29 the radial separation from the galaxy centre and @xmath130 is the matched filter . the first condition requires an unbiased estimator , thus the expectation of @xmath126 should be @xmath39 @xmath133 which means @xmath134 in the above equation , we have used the property that the ensemble average of the noise term @xmath92 vanishes . the second condition of the estimator requires minimizing of the variance @xmath135 between the estimator and the underlying amplitude . we can write the variance in fourier space using parseval theorem @xmath136 ^ 2 \rangle - \langle \tilde{u } \rangle ^2 \\ & = \frac{1}{(2\pi)^4 } \langle \left [ \int\!\mathrm{d}^2 l \phi(l ) n(l)^{\ast } \right ] \left[\int \!\mathrm{d}^2 l^{\prime } \phi^{\ast}(l^{\prime } ) n(l^{\prime } ) \right ] \rangle \\ & = \frac{1}{(2\pi)^2 } \int\!\mathrm{d}^2 l \int\!\mathrm{d}^2 l^{\prime } \delta^d(l - l^{\prime } ) ( c_l b_l^2+\sigma_p^2 \theta_p^2 ) \phi(l ) \phi^{\ast } ( l^{\prime } ) \\ & = \frac{1}{(2\pi)^2}\int\!\mathrm{d}^2 l ( c_l b_l^2 + \sigma_p^2 \theta_p^2)\ , \left|\phi(l)\right|^2 \ , \end{split}\ ] ] in the following transform convention @xmath137 which has been reduced to the zero - order hankel transform due to circular symmetry . @xmath138 is the angular power spectrum of the primary cmb , for which we have used the camb code @xcite . we have assumed that there s no correlation between the primary cmb and the instrument noise . therefore , solving the matched filter can be reduced to finding the local extremum of a function that is subject to an equality constraint . specifically , here we need to solve for @xmath127 and a lagrangian multiplier @xmath139 that minimize @xmath140 , \ ] ] with the constraint @xmath141 hence the following equation is demanded @xmath142 = 0 \ .\ ] ] we thus have the solution by solving eq . [ eq : unbiased_fourier ] and [ eq : lambda_fourier ] @xmath143 where the normalization @xmath139 is given by @xmath144^{-1 } \ .\ ] ] we then apply this filter to the noisy cmb map , and the cross correlation signal would be @xmath145 ^ 2 \rangle \\ & = \langle v_\parallel \tilde{u } \rangle \left ( \int d^2\theta \phi(\theta ) \tau_b(\theta)\right ) \\ & = \langle v_\parallel u_\parallel \rangle t_{\rm cmb}/c . \end{split}\ ] ] by constructing the correlation of the optimal estimator and the recovered velocity , we expect not only to get the ksz effect isolated from the noisy map , but also to have a greatly reduced measurement error . by stacking galaxies with the same host halo mass , we will further beat the statistical noise and get a higher s / n . if no filter is applied to the noisy cmb map , emission from all scales of every component in the cmb sky will be retained in the measurement error . from eq . [ eq : stacking_corr ] and [ eq : stacking_var_nofilter ] , we can now estimate the signal to noise ratio of this correlation , @xmath146 ^ 2 } { \langle \delta t^2(\theta)\rangle } \\ & = r_{uv}^2 \frac{\langle \delta t^2_{\rm ksz}(\theta ) \rangle}{\langle \delta t^2 \rangle } , \end{split}\ ] ] where @xmath147 the correlation coefficient defined as @xmath148 . thus the signal to noise ratio of each pixel is essentially proportional to the correlation coefficient @xmath147 and the ratio of the variance of the ksz effect to the overall variance of the cmb sky . all the components in the cmb map constitute the pixel noise to the ksz signal , and it can be obtained from the following equation @xmath149 @xmath138 is the measured power spectrum of the cmb sky , which is the sum of the primary cmb , the sz effect , the cib , etc . . however , the primary cmb dominates on scales @xmath150 and gets most of its variance around the first several acoustic peaks . the power spectra of other components will only exceed cmb on smaller scales , and get suppressed by the beam window of the telescope , and thus contribute negligible power to the pixel variance . hence we calculate the pixel variance in eq . [ eq : pixel_variance ] with only the primary cmb , which we get from the camb code . the resulting pixel variance of the temperature fluctuation is @xmath151 k@xmath152 , which means @xmath153 k of the r.m.s . noise in a pixel . in the context of the standard cosmology and the linear perturbation theory , both baryons and dark halos emmersed in the same gravitational potential exerted by the large scale density perturbation and shall follow the same velocity field on large scales . in linear regime and even quasilinear regime , one can reconstruct an effective velocity field using the continuity equation , given a galaxy density field from observations . on small scales , shell crossing will complicate the dynamics and fail the linear prediction . nevertheless , the velocity field reconstructed from spectroscopic galaxy surveys can trace the underlying velocity flow on large scales . hydrodynamic plus n - body simulations have examined the relation between the underlying velocity and the reconstructed velocity . it has been shown @xcite that the recovered velocity field @xmath103 follows the underlying baryonic velocity @xmath107 fairly good . for simplification we here denote @xmath154 and @xmath100 as the projected velocity component along any given direction of the 3d velocity @xmath107 and @xmath103 . to quantify the performance of the velocity reconstruction , one usually introduces in the fourier space the cross correlation coefficient @xmath155 , and the velocity bias @xmath156 ( see e.g. * ? ? ? * for a different but related definition of the coefficient @xmath157 ) , where the power spectrum @xmath158 , and @xmath159 is defined as @xmath160 . for the underlying baryon velocity , @xmath161 for the recovered velocity from real space matter density , @xmath162 for the cross power spectrum between the two . the recovered velocity power spectrum and the cross power spectrum , given matter density in redshift space , are shown as @xmath163 and @xmath164 . the shot noise power spectrum of velocity , which has been scaled to the fiducial galaxy number density , is shown in triangles . _ the lower two panels _ : the cross correlation coefficient @xmath157 between the underlying baryon velocity @xmath154 and the recovered velocity given galaxy density in real space and in redshift space.,width=307 ] as a function of @xmath165 at z=0 ( left panel ) and z=0.5 ( right panel ) , defined as @xmath166 . the cross correlation and auto correlation are integrated up to @xmath165 in the convention of eq . [ eq : cross_correlation ] . thick and thin lines denote the nr run and ww run respectively . here @xmath147 encodes the uncertainties induced by the redshift space distortion , the gas physics , and the shot noise . , width=307 ] a perfect reconstruction of the underlying velocity means @xmath167 and @xmath168 at all scales . in realistic galaxy surveys , there are a number of issues that will complicate the velocity reconstruction @xcite , so that @xmath157 and @xmath169 both deviate from unity . however , deterministic bias does not affect the estimation of s / n as clearly shown in eq . [ eq : signal_to_noise_nofilter ] , leaving the dependence of s / n only on @xmath147 . hence we focus on @xmath147 to evaluate the reconstruction . several important issues which may degrade the velocity reconstruction are listed here : ( 1 ) eq . [ eq : vk_reconstruction ] only holds when @xmath170 , and we have also simplified the equation by assuming the linear density evolution , and a deterministic bias between the galaxy density @xmath104 and the matter density @xmath105 ; ( 2 ) the galaxy number density in the redshift space is what is measured in galaxy survey , while the equation holds in the real space ; ( 3 ) in the reconstruction we lose the information of velocity vorticity which contributes significantly to the ksz signal on small scales ; ( 4 ) the lack of knowledge of the underlying gas dynamics will induce uncertainties in the recipes of gas dynamics employed in simulation and thus in the reconstructed velocity field ; and finally ( 5 ) the mask and geometry of galaxy surveys will reduce the effectiveness of the reconstruction . these factors will of course degrade the reconstruction . fortunately , the velocity correlation length is @xmath171 @xmath172mpc , and the reconstruction is fairly good on linear scale and even on quasilinear scales . we test several issues mentioned above against a set of hydrodynamic simulations . the simulations are run with the gadget2 code @xcite in a lcdm cosmology with parameters : @xmath173 . the box size of the simulation is l = 300 @xmath172mpc on each side , in which @xmath174 dark matter particles and @xmath174 gas particles are initially seeded . two runs are implemented with the same initial condition , one of which is non - radiative ( nr run ) , and the other radiative ( ww run ) with star formation and weak wind feedback ( wind speed is 480 km / s ) . we measure the power spectra of the underlying velocity of baryons @xmath175 , the recovered velocity @xmath176 given dark matter distribution , the cross power spectrum @xmath177 , and the corresponding power spectra in the redshift space , and show them in the upper panels of fig . [ fig : cor_uv ] . it s shown that the recovered power spectrum of the velocity in the redshift space is larger than that in the real space on large scales , and on scales @xmath178 @xmath179mpc@xmath180the power spectrum in redshift space drops rapidly due to the finger of god effect . also shown is the shot noise power spectrum , which has been scaled to the fiducial galaxy number density @xmath181@xmath182mpc@xmath2 , and it will dominate the velocity power spectrum at @xmath183@xmath179mpc@xmath180 . comparing the nr run ( thin lines ) and ww run ( thick lines ) , we find gas physics only modify small scale power spectrum . we also show the correlation coefficient in fourier space @xmath157 in the lower two panels of fig . [ fig : cor_uv ] . the cross correlation coefficient is large than 0.9 on scale @xmath184 @xmath179mpc@xmath180 , and decreases rapidly on smaller scales when nonlinear evolution dominates , and redshift space distortion further reduces the coefficient . comparing nr run and ww run , we find gas physics only affects the reconstruction on scales @xmath185 @xmath179mpc@xmath180 . all of above factors play a role in shaping the recovered velocity . though there are indeed significant deviations between the power spectra of recovered velocity and the underlying velocity , what we really measure is the cross correlation of the recovered velocity and the cmb distortion at galaxy positions . thus to quantity the performance of the reconstruction , we measure the cross correlation coefficient @xmath186 . the cross correlation between the reconstructed velocity and the underlying velocity along the line of sight ( l.o.s ) can be calculated by integrating the cross power spectrum as @xmath187 the cross correlation converges rapidly to the auto correlation of the underlying velocity . on one hand , linear theory predicts that peak contribution comes from @xmath188 around @xmath189@xmath179mpc@xmath180(see fig . [ fig : cor_uv ] ) , and 70% ( 90% ) of the velocity dispersion is contributed on scales @xmath184@xmath179mpc@xmath180(@xmath190@xmath179mpc@xmath180 ) . on the other hand , observations and simulations have also shown that only 15% of the velocity dispersion comes from @xmath191@xmath179mpc@xmath180 , where the linear approximation holds very well @xcite . we measure @xmath192 , @xmath193 and @xmath194 in the simulation with the convention as shown in eq . [ eq : cross_correlation ] . as shown in fig . [ fig : cor_uv ] , most of the correlation signal comes from large scale modes , which makes sure that the correlation @xmath194 converges on large scales . we then calculate the cross correlation coefficient in the redshift space @xmath147 as a function of @xmath165 , and show them in fig . [ fig : coef ] . here @xmath147 encodes errors induced by the redshift space distortion , the gas physics , and the shot noise . we find that the correlation coefficient @xmath147 is larger than 0.8 for galaxy surveys sampling a galaxy density corresponding to @xmath195@xmath179mpc@xmath180 , and then decrease to below 0.6 at smaller scales when the shot noise becomes overwhelming . therefore in this work we choose @xmath196 as the fiducial value , and the signal to noise ratio can be scaled accordingly for other values of @xmath147 . we have tested the correlation coefficient @xmath147 against various uncertainties in the velocity reconstruction in hydrodynamic simulations , and we assume @xmath147 applies to our fiducial cosmology . for our fiducial cosmology , the underlying zero - lag auto - correlation along the line of sight @xmath197 is calculated using the linear perturbation theory in this work @xmath198 where @xmath199 is the power spectrum of the dark matter obtained from camb . we have adopted the assumption that the velocity is isotropic . the linear prediction of the velocity variance is @xmath200 km / s in our fiducial cosmology . as shown in the eq . [ eq : signal_to_noise_nofilter ] , the signal to noise ratio is proportional to the r.m.s of the underlying velocity dispersion and the correlation coefficient @xmath147 . simple calculation shows that the central ksz signal at @xmath201 for a mw - size halo is @xmath202k ( eq . [ fig : tksz ] ) . therefore , a direct stacking without optimal filtering requires more than one million galaxies to achieve 1 @xmath7 detection . even with more pixels of resolved galaxies at lower redshifts , the required galaxy number scales inversely to the number of pixels per galaxy , and is still beyond current galaxy surveys for late type galaxies . compared to the variance of the background sky , the instrument noise term is very small . therefore , the measurement error of the cgm s ksz signal is almost entirely contributed by the primary cmb . thanks to the different scale dependences between the ksz effect and the primary cmb , we can reduce the variance by filtering out the contribution of the large scale primary cmb . for extended hot halo model ( mb ) profile and @xmath0 profile in fourier space . the first peaks of the filters are almost at @xmath203 5000- 6000 . the filters are smooth and they look similar for both models at high redshift , while at low redshift the filter for the flatter mb model fluctuates much more severely than that for the @xmath0 profile . [ fig : phil],width=307 ] for act - like cmb survey for both mb profile ( upper panel ) and empirical @xmath0 density profile ( lower panel ) for cgm . [ fig : phit],width=307 ] as shown in eq . [ eq : phi_fourier ] , the filter profile is the most sensitive to the scales where the signal becomes relatively dominant with respect to the primary cmb plus the instrument noise . on large scales ( low @xmath204 ) , @xmath205 almost vanishes since @xmath206 is much smaller than the primary cmb spectrum . however , the primary cmb is damped exponentially , while @xmath206 decreases slowly for @xmath207 . in the case of act - like cmb survey , the filter gets its maximum contribution from those scales where the instrument noise is comparable to the primary cmb , i.e. @xmath208 at almost all redshifts . at lower redshifts z@xmath209 0.06 , there are several comparable troughs and peaks for the mb model , as shown in the fourier space in fig . [ fig : phil ] . because of the acoustic fluctuation in the damping tail of the cmb , there s small dithering of the filter . we can also see the behaviour of the filter @xmath130 in fig . [ fig : phit ] in real space , in which we have adopted the normalization so that @xmath210 . for all the selected redshifts shown here , the first trough is located at @xmath211 arcmin away from the galaxy center , which is just around the beam size ( the first trough of 0-th bessel function ) . however , at z@xmath209 0.1 , the mb profile is very shallow and the ksz signal from scales larger than the beam size is non - negligible , and thus part of the ksz signal is compromised by the filter . that s why one sees that at low redshifts , the filter fluctuates heavily even in the outer region of the halo . we also show in fig . [ fig : phit ] the filter for the @xmath0 profile which turns out to be similar for all redshifts . as shown in the figure , although the spatial extent of the @xmath0 profile ( see fig . [ fig : tksz ] ) increases by an order of magnitude from z=0.3 to the local universe , the shape of the filter at different redshifts changes only slightly . this feature will simplify the stacking procedure , as the filter follows almost the same shape , and thus apply to all galaxies at different redshifts . also as shown in fig . [ fig : phit ] , when galaxies became well resolved by the sz survey , the filter profiles diverge significantly between the two cgm profiles . the different behaviours will in turn shed light on how to discriminate between the two cgm profiles , as the ksz effect of a given profile only responds to the corresponding matched filter , and gives the optimal measurement accordingly . the matched filter gives an optimal estimator @xmath126 of the underlying ksz signal which minimizes the variance . with this estimator in hand , we can cross correlate the estimator @xmath126 and the constructed velocity field @xmath100 , and evaluate the signal to noise ratio for a single galaxy as follows @xmath212^{-1 } \\ & = \left [ 1+\frac{\langle v_\parallel^2 \rangle \langle \tilde{u}^2 \rangle } { \langle v_\parallel \tilde{u } \rangle^2 } + \frac{\langle v_\parallel^2\rangle}{\langle v_\parallel \tilde{u } \rangle^2 } \frac { ( 2\pi)^2\int\mathrm{d}^2 l \phi^2(l)\left [ c_l b^2(l)\!+\!\sigma^2_p \theta^2_p\right ] } { \left [ \int\!\mathrm{d}^2 l \phi(l ) \tau(l ) b(l ) \right]^2 } \right]^{-1 } . \end{split}\ ] ] we have transformed the estimation into fourier space in the last line . as addressed in [ subsec : uncertainties_in_r ] , the signal to noise ratio is strongly correlated with the cross correlation between the reconstructed velocity and the underlying velocity . even if there s a deterministic bias of velocity between @xmath100 and @xmath154 , the bias will cancel out in the denominator and the numerator , leaving the dependence only on the underlying velocity dispersion and the correlation coefficient @xmath147 . we can rewrite the signal to noise estimation in the case of matched filtering as @xmath213 } { t^2_{\rm cmb } \left [ \int\mathrm{d}^2 l \phi(l ) \tau(l ) b(l ) \right]^2 } \right]^{-1 } . \end{split}\ ] ] in real survey , even if the contamination is filtered , the residual noise is still much larger than the ksz signal of mw - size halos . thus the first two terms can be neglected , if the correlation coefficient is good enough . therefore , the signal to noise ratio is proportional to the correlation coefficient @xmath147 , and that s why a good velocity tracer is the key quantity in this method . it s clear from fig . [ fig : phil ] that @xmath205 will remove the contribution from large scale modes of the primary cmb @xmath138 . though the large scale modes of the ksz effect @xmath206 are also removed , the loss of signal is negligible and the total signal is conserved by design . after applying this filter , the contribution of the primary cmb to the noise is greatly reduced , and the r.m.s . noise is dramatically reduced from @xmath214 k to a level of @xmath215 k at @xmath201 . assuming we stack galaxies of the same host halo mass , for example 10000 mw - size halos , in a narrow redshift bin around @xmath216 , we will reduce the statistical noise and achieve @xmath6 1 @xmath7 detection . at lower redshifts , the cgm will have higher ksz effect amplitude , and cover more extended pixels , which make the filter much more effective . therefore , the s / n increases until @xmath217 , and with 10000 mw - size halos one can achieve @xmath218 detection . below this redshift , the signal to noise decreases , as the ksz effect get non - negligible contribution from large scales for spatially extended halos , and thus the filter can not effectively suppress the primary cmb on these scales . the signal to noise ratio as a function of the redshift for 10000 galaxies in a redshift bin ( @xmath219 for @xmath220 , @xmath221 for @xmath3 ) is plotted in fig . [ fig : snz ] . in fact , we have simplified the estimation here , and the total signal to noise ratio will inevitably depend on the number distribution , the redshift distribution , and the mass distribution of the host halos . careful calculation is needed when applied to the real galaxy catalogue . also shown in the figure is the s / n for the @xmath0 profile , which is much steeper in the outer region than the fiducial mb profile . for the @xmath0 profile , the mass is more concentrating to the centre , leading to a higher ksz effect in the central region , with little ksz signal coming from large scale wave mode . the ksz amplitude of the @xmath0 profile is much higher than the mb profile , and thus we can expect a higher signal to noise ratio at the same redshift . as shown in the lower panel of fig . [ fig : phil ] , the filter for the @xmath0 profile effectively removes the large scale contribution from the primary cmb , and retains the ksz signals very well . thus the s / n increases with decreasing redshift till the turnover redshift , z=0.03 for a @xmath0 profile , where we get a maximum s / n=19 and a dramatic drop - off below this redshift . besides the intrinsic shape of the ksz profile of the cgm , the beam size of the telescope also determine the shape of the filter . for comparison , we also predict the detectability for the two profiles in the case of a planck - like cmb survey , i.e. , @xmath222 arcmin , @xmath223k per pixel . the results are also shown in the fig . [ fig : snz ] . there s similar trends except much lower signal to noise ratio . the signal to noise ratio reaches the maximum at z@xmath9 1 ( z@xmath9 0.03 ) for the mb profile ( the @xmath0 ) profile when the size of the halo is about 5 arcmin , just around the beam size of planck . the signal to noise ratio will be @xmath8 1 and 2 for the mb profile and @xmath0 profile , respectively . for 10000 mw - size halos as a function of redshift for an act - like sz survey , for both the mb profile and the @xmath0 profile with @xmath68 . also shown are the predictions for sz surveys like planck . [ fig : snz],width=307 ] in this work , we investigate the ksz effect of the cgm for mw - size halos . by weighting the cmb sky with the reconstructed velocity at the galaxy position , one can pick out the ksz effect of the cgm . the cross correlation is however subject to heavy contamination mainly from the primary cmb . to remove the large scale contaminations from the primary cmb , one can take advantage of different scale dependences between the primary cmb and the ksz signal , and choose a minimum variance estimator by designing a matched filter . one can then stack the filtered map to further suppress the statistical error . with this method , we alternatively provide a statistical means to detect the cgm via the ksz effect . furthermore , in its own right , ksz effect of the missing galactic baryons can be cleanly measured , given field galaxies with similar host halo mass . since current detection of the ksz effect from the missing galactic baryons is mixed with the ksz effect from icm / igm , a clean sample of galaxies , such as field galaxies from ska , should be essential . for comparison , we take the compelling mb profile as well as the empirical @xmath0 model to test the detectability . the matched filter is designed to remove the primary cmb on large scales , and keep the small scale ksz signal conserved . for the mb profile , given an act - like sz survey and 10000 milky way - size halos in a narrow redshift bin , we can achieve more than @xmath224 for the cgm of mw - size halos at @xmath216 . for galaxies occupying a single pixel , the matched filter degenerates to the aperture photometry ( ap ) filter , which can be easily implemented in the cmb map by subtracting the average of an outer ring of equal area from the central part ( see * ? ? ? * for an example ) . at lower redshifts , more resolved pixels of each galaxy are involved to regulate the filter , the s / n increase greatly with the same mount of galaxies , and reaches a maximum s / n about 7.5 at @xmath225 . the cgm with the @xmath0 model is more concentrating and the s / n turns out to be much higher . though at high redshift the two profiles show similar s / n , the ksz signal of @xmath0 profile has much higher s / n at @xmath226 . it can reach up to a maximum 19 @xmath7 detection at a lower redshift ( z@xmath8 0.03 ) . although it is unclear a steeper profile like the @xmath0 model is favoured by observational constraints for the cgm in milky way - size halos , ksz effect of such profile can give insight to its upper limit . due to the different responses of the minimum variance estimator to the filter , mainly at low redshift ( @xmath3 ) , it is possible to discriminate mb profile from the @xmath0 profile , as the underlying cgm will optimally respond to the corresponding filter profile and show the highest significance of detection among a set of filters . we also test a planck - like survey , and with the same number of galaxies , it can only marginally detect the ksz signal of the cgm at very low redshift . though this work targets the cgm of late type galaxies , it can also serve to probe the ksz effect for early type galaxies given the knowledge of the cgm s profile . in this work , we have made a set of assumptions , and some of them are still under debate . we discuss several major concerns here . 1 . _ it is still under debate the total mass fraction and the spatial distribution of the cgm . _ such information depends on the halo mass , the accretion history , the star formation , the feedback energy deposited and the merging process . on one hand , simulations of galaxy formation use a broad baryon fraction ranging from 30 - 100% . on the other hand , there are controversial constraints on baryon fraction @xmath48 within the virial radius . for example , @xcite found a consistent @xmath48 with the cosmic mean , while @xcite draw conclusions that only 10 - 20% missing baryons can be accounted for within the virial radius of the host halo . the radial extent of x - ray observations , the spectral modeling , and the metallicity distribution suggests different conclusions . therefore , in this work , we take @xmath22762.5% of the cosmic mean , which is the missing baryon fraction of the milky way , as a fiducial fraction of the hot coronae constituting the missing part of baryons . we can easily scale the signal to noise ratio with @xmath228 with other values using eq . [ eq : optical_depth ] and eq . [ eq : signal_to_noise_filtered ] . the uncertainty of the optical depth profile . _ we can see that for a given sz survey with fixed resolution , the two profiles of cgm lead to similar filter profile and thus similar s / n at high redshifts ( @xmath229 for act , @xmath230 for planck ) when galaxies are unresolved . when galaxies become well resolved at lower redshifts , the more concentrating @xmath0 profile has steeper ksz effect , and the filter profiles diverge significantly . since either the mb profile or the @xmath0 model may not account for the true profile of the underlying cgm , or the profiles of halos with the same mass can vary due to different galaxy formation histories , the uncertainty in the profile can induce variance in the estimator and degrade the detectability . future application of this method requires more observational constraints on the average profile and the scatter of the profile for the cgm . _ mass range of host halos . _ in this work , we have shown that for mw - size halos , the method works well . in real survey , host halos of galaxies span a much broader mass range . this technique can be carried out with careful design . on one hand , as shown in eq . [ eq : optical_depth ] that there s degeneracy between @xmath228 and m@xmath231 , high mass with low @xmath228 ( e.g. beta profile for massive ellipticals ) can lead to similar s / n for halos with low mass and high @xmath228 . on the other hand , we can always bin host halos of a broad mass range into subsamples of narrow mass bin . we can then apply the filter to galaxies to obtain a measure of @xmath232 for each subsample , and stack them to reduce the statistical noise further . the projection effect . _ as a temperature distortion upon the cmb , the ksz signal is an effect that involves projecting all the free electrons thus the ionized gas along the line of sight . therefore , the ksz effect of the cgm would suffer from the contamination of the chance alignment of background galaxies , clusters or diffuse baryons around galaxies , which are within the velocity correlation length ( see * ? ? ? * for the observational evidence of baryons beyond the virial radius of halos ) . one can in principle model the contribution these sources and filter them out , by accounting for the two - halo term of the ksz effect , given the cosmology and the halo occupation distribution for late type galaxies or the distribution of the icm for clusters . for example , @xcite calculated the power spectrum of the ksz from the cgm and clusters using halo model , and they found the ksz signal from cgm would only dominate the clusters on scales @xmath233 . we can alternatively compute the cross correlation between the icm and the cgm using hydrodynamic simulations , and quantify the influence of the projection of icm along the line of sight of the cgm . another possible contamination is the potentially prevailing igm or filamentary structures bridging galaxies beyond the virial radius . for field galaxies from ska hi survey , this problem is less significant . the question is still open whether the missing baryons in galaxies are detained in the cgm or have been expelled to the igm , or have never been accreted into galaxies in the galaxy formation process in the first place . the projection effect is thus to some extent degenerate to the gas fraction @xmath228 within the virial radius . taking @xmath228 as a free parameter and assigning @xmath228 with some empirical distribution would be a reasonable choice . the stochasticity in the velocity correlation . _ a number of issues can induce uncertainties in the velocity correlation . we test the velocity reconstruction and the cross correlation under the influence of the redshift space distortion , the shot noise , and the gas physics using a set of hydrodynamic simulations . we have found these effect will not fail the reconstruction . we have also assumed a deterministic galaxy bias @xmath114 in the velocity reconstruction . though there are stochasticities in @xmath114 , the uncertainties are under control @xcite on scales of interest . as discussed in [ subsec : velocity_dispersion ] , the peak contribution to the velocity correlation is from around @xmath234@xmath179mpc@xmath180 , and 70% contribution is from @xmath235@xmath179mpc@xmath180 . given a galaxy survey with number density larger than @xmath236mpc@xmath2 , the recovered velocity traces the underlying velocity very well on scales of interest , as shown in fig . [ fig : coef ] , and we take @xmath196 as a fiducial value in the s / n estimation . the application of the velocity reconstruction to the real survey has also proved effective @xcite . 6 . _ the number density and the redshift distribution of galaxy surveys_. the realistic measurement of the ksz signal of cgm of galaxies will depend on the number density and the redshift distribution of the galaxies of the real survey . on one hand for example , ska @xcite will provide a wide and deep survey of hi galaxies covering the act / spt sky partly , and map @xmath24 field galaxies . it will be sufficient to recover the velocity field from the density field , and thus recover the ksz effect of the cgm . on the other hand , 2mass redshift survey ( 2mrs ) @xcite have independently completed the peculiar velocity survey , and there are roughly 6000 edge - on spiral galaxies in local universe . this catalogue alternatively provide a measure of the velocity field . however , the spiral galaxy number ( @xmath8 150 ) covering the act sky ( 780 deg@xmath152 released in the southern sky ) is so small that the cgm of these galaxies may not be detectable . with even smaller beam size and data release of full 2500 @xmath237 sky in the future , spt can serve to probe the cgm of spiral galaxies in 2mrs . the cgm of early type galaxies . _ the method presented in this paper can in principle apply to the cgm of all galaxies with given profiles . for example , the cgm in massive early type galaxies , even with low baryon fraction , can contain more hot gas than in late type galaxies @xcite . if this is the prevailing case , early type galaxies should contribute a huge reservoir of cgm for the ksz effect . 6df galaxy redshift survey ( 6dfgrs ) @xcite has measured the peculiar velocity of @xmath8 9000 early type galaxies at @xmath238 in the southern sky . a very rough estimate shows that there are @xmath8 300 galaxies in the area covering the act sky . given the @xmath0 profile , as a usually adopted profile for the cgm in elliptical galaxies , the cgm in the halo with mass of @xmath51 m@xmath52 ( as shown in fig . [ fig : snz ] ) can be detected with @xmath239 sigma level with a rough estimate via this stacking scheme . massive elliptical galaxies with higher mass can show even higher s / n . we thank pengjie zhang for helpful discussion , and weipeng lin for kindly providing the simulation data . js is supported by the national natural science foundation of china under grant no . tf was partially supported by the national natural science foundation of china grant no . 11273021 , by the strategic priority research program `` the emergence of cosmological structures '' of the chinese academy of sciences , grant no . xdb09000000 , and by the national natural science funds for distinguished young scholar grant no .

we present high resolution ( r = 75,000100,000 ) mid - infrared spectra of the high - mass embedded young star irs 1 in the ngc 7538 star - forming region . absorption lines from many rotational states of ch ,cch , ch , ch , nh , hcn , hnco , and cs are seen . the gas temperature , column density , covering factor , line width , and doppler shift for each molecule are derived . all molecules were fit with two velocity components between 54 and 63 . we find high column densities ( ) for all the observed molecules compared to values previously reported and present new results for ch and hnco . several physical and chemical models are considered . the favored model involves a nearly edge - on disk around a massive star . radiation from dust in the inner disk passes through the disk atmosphere , where large molecular column densities can produce the observed absorption line spectrum .

0902.0610

the ngc 7538 ( s158 ) molecular cloud , located in the perseus arm at a distance of 2.8 kpc @xcite , contains many protostellar objects , making it a good candidate to study different stages of star formation . furthermore , there seems to be a progression in evolution from northwest ( the visible h ii region , ngc 7538 , and irs 2 , an only moderately obscured compact hii region ) to southeast ( irs 9 , a deeply embedded core ) @xcite . the near- and mid - infrared sources irs 1 , 2 , and 3 all lie near the northwest end of the star - forming region . of these , irs 1 is the brightest mid - infrared and radio continuum source @xcite . its exciting star has a luminosity of @xmath6 , which corresponds to a zams spectral type earlier than o7.5 , and an ionizing photon flux of @xmath7 @xcite . irs 1 is embedded in a dense molecular cloud . millimeter continuum emission is seen from the surrounding dust @xcite . @xcite mapped the hco@xmath8 and hcn emission from within @xmath91 of irs 1 with 3 resolution . they found evidence for a shell - like structure on scales of 1020 around irs 1 . @xcite observed nh@xmath3 absorption toward irs 1 , and @xcite mapped the absorption with the vla . the absorption traces warm gas with a high column density of nh@xmath3 and is centered near @xmath10 . the emission , on the other hand , is centered near @xmath11 . the systemic velocity for this source is @xmath12 @xcite . more recently , @xcite mapped nh@xmath3 emission from a more extended region in ngc 7538 . this emission probably probes the outer envelope as indicated by the colder temperature and the smaller column density . using both single - dish and interferometer observations of various molecular tracers , @xcite studied the density and temperature structure of both the cold outer envelope and the warm inner material ( 24072000 au ) . the inner warm region is characterized by temperatures of a few hundred k. outflows are often present toward protostars , and ngc 7538 irs 1 is no exception . @xcite observed irs 1 with the vla at 5 and 15 ghz . she found a pair of very compact lobes of continuum radiation , separated in declination by 0.2 , with emission extending out to @xmath132 . her preferred model involves a bipolar ionized outflow from a late o star , collimated by a core of dense gas extending from @xmath1465 au to @xmath1525,000 au . further evidence of outflows was found when @xcite observed irs 1 with the vla in the h66@xmath16 line and the 22 ghz continuum . their continuum image is similar to that of campbell , but shows additional structure . they found broad recombination line emission in the two lobes , with a minimum line - to - continuum ratio , which they attribute to high electron density , between the lobes . they propose a model involving a high velocity stellar wind interacting with photo - evaporating knots of neutral gas . more recently , @xcite have modeled the radio continuum observations of irs 1 as due to a wind produced by photo - evaporation of a circumstellar disk with a radius of 500 au exposed to uv radiation from the central o star . maser line emission is seen from h@xmath0o @xcite , oh @xcite , h@xmath0co @xcite , ch@xmath3oh @xcite , and nh@xmath3@xcite . @xcite used vlbi observations of the methanol masers to infer the presence of a nearly edge - on rotating disk traced by one of the maser clusters toward irs 1 . the central velocity of this cluster , which is the brightest cluster near irs 1 in both ch@xmath3oh and nh@xmath3 masers , is 56.2 . a second cluster of masers , 0.25 to the south , is at 61.0 . the disk masers lie along a line centered near the gap between the lobes of the free - free continuum radiation , although they are aligned along a position angle @xmath930 from the symmetry plane . @xcite made a detailed model of maser emission from a disk around irs 1 , supporting the suggestion of @xcite they concluded that the maser emission comes from disk radii of @xmath17 au . however , more recently , @xcite show that the ch@xmath3oh masers may trace knots in the outflow instead . also , @xcite identify a disk - like structure perpendicular to the outflow direction in mid - infrared continuum . the extent of the infrared emission is @xmath9450 au . infrared absorption by dust grains and their icy mantles has also been observed toward irs 1 . @xcite and @xcite observed 9.7 @xmath18 m silicate absorption and absorption by various ices , including h@xmath0o . the ice bands toward irs 1 are relatively weak , especially in comparison with ngc 7538 irs 9 , which must lie in a colder region of the ngc 7538 cloud . @xcite observed the complete 2.5 - 20 @xmath18 m region with the short - wavelength spectrometer ( sws ) on the _ infrared space observatory _ , . h@xmath0o , co , and co@xmath0 ices are clearly present , as well as other less securely identified features . toward ngc 7538 irs 9 , @xmath91 southeast of irs 1 , a much richer ice spectrum is seen , with ch@xmath3oh , xcn ( probably ocn@xmath19 ) , and ch@xmath4 also confidently identified , and ocs , h@xmath0co , hcooh , and nh@xmath3 likely present . gas - phase absorption in the infrared has been observed toward irs 1by several groups . @xcite observed absorption by @xmath2co and @xmath1co in their 5 @xmath18 m @xmath20 bands . they did not spectrally resolve the lines , and they used a curve of growth analysis assuming pure absorption by gaussian lines with fwhm of 8 to derive the gas temperature and column density from the less - saturated @xmath1co lines . they found two absorbing components : cold gas at 25 k , and warm gas at 176 k. a gas density of @xmath21 was required to maintain the population of the high-@xmath22 levels in the warm gas . using the -sws , @xcite detected warmer gas than the co observed by @xcite . they observed the @xmath23 band of c@xmath0h@xmath0 at 13.7 @xmath18 m and the @xmath24 band of hcn at 14.0 @xmath18 m . with the sws resolution of @xmath91800 , they were not able to resolve individual line shapes , and were most sensitive to the blended lines of the q branches , but they were able to derive temperatures and column densities from the shapes of the q branches and the depths of the @xmath20 fundamental and @xmath25 hot band q branches . they derived @xmath26 = 800 k and @xmath27 for c@xmath0h@xmath0 , and @xmath26 = 600 k and @xmath28 for hcn . @xcite derived @xmath26 = 500 k for c@xmath0h@xmath0 from an updated reduction of the spectra . @xcite and @xcite studied gas absorption from co@xmath0 and h@xmath0o using data . they found enhanced abundances toward the inner warm material compared to the cold envelope indicating that grain mantles are sublimating and enriching the gas - phase chemistry close to the protostar . gas - phase ch@xmath4 was also detected toward irs 9 @xcite but not irs 1 . the wealth of observational data available suggest the following physical scenario for ngc 7538 irs 1 . the massive young star is surrounded by a massive cold envelope on scales @xmath9 72000 au @xcite . at scales below 1000 au , the temperature increases to @xmath29 100 k. inside this radius , there is evidence for a disk @xcite and small scale knotty outflows @xcite , both of which can be affected by the uv and x - ray radiation from the protostar ( e.g. , * ? ? ? * ; * ? ? ? * ) . in this paper , we present high resolution mid - infrared spectroscopy of ngc 7538 irs 1 showing a rich absorption spectrum containing lines from _ seven _ molecules : c@xmath0h@xmath0 , hcn , ch@xmath4 , ch@xmath3 , nh@xmath3 , hnco , and cs . mid - infrared absorption is much more sensitive to the inner warm gas than radio observations . previous infrared studies suffered from limited spectral resolution . section [ sec : obs ] describes the observations . we then present the model used to derive the column densities and temperatures for the various species in section [ sec : fit ] and describe the results for each of the molecules in section [ sec : results ] . subsequently , in section [ sec : mod ] , we discuss the possible scenario in which we are probing chemistry in a circumstellar disk . in section [ sec : conc ] , we provide some concluding remarks . the observations were made with texes , the texas echelon cross echelle spectrograph @xcite , on the nasa infrared telescope facility ( irtf ) in 2001 june and november , 2002 september and december , and 2005 december . texes is a high - resolution cross - dispersed spectrograph operating at mid - infrared wavelengths between 5 and 25 @xmath18 m . it achieves a spectral resolution of @xmath30 = 75,000 - 100,000 ( @xmath31 ) shortward of 14 @xmath18 m with a slit width of @xmath91.5 . ( of the observations presented here , only the spectra near 11 @xmath18 m , which used a 1 slit , achieved the higher resolution . ) the spatial resolution along the north - south oriented slit is @xmath91 , slightly larger than the diffraction limit of the irtf . the spectral and spatial sampling by the @xmath32 pixel si : as detector array are 1.0 and 0.35 , respectively . at each spectral setting , 5 - 10 orders of the high resolution echelon grating are recorded , giving a spectral coverage of @xmath330.5% . at wavelengths shortward of 11 @xmath18 m full order widths are observed , giving continuous spectral coverage , but longward of 11 @xmath18 m the order width exceeds the array width , leaving gaps in the observed spectra . the location of the gaps can be shifted in order to optimize line coverage . however , since the gaps are small and , at 13 , there are many telluric lines , it would be impractical to repeat observations with small shifts in order to get complete spectral coverage for each setting . much of the data were taken in high water vapor conditions , but the observing conditions did not seriously affect the spectra except for increased noise , especially on atmospheric lines . observations of ngc 7538 irs 1 were interspersed with observations of a bright asteroid ( usually ceres ( @xmath34 jy ) , but also hygeia ( @xmath35 jy ) , which served as a comparison source for removal of telluric absorption features . the telescope was nodded by 35 , moving the source along the spectrograph entrance slit to allow subtraction of background emission . at the beginning of each set of nodded observations an ambient temperature blackbody was observed for flat - fielding . the first nods were used to peak up on the source by maximizing the throughput of the slit . the nod pairs taken during peak up were not included in the final sum . data reduction followed the standard texes procedure @xcite of first subtracting the two nod position spectra , flat - fielding with the blackbody - sky difference echellogram @xcite , and interpolating over spikes and bad pixels . optimally - weighted point - source spectra were then extracted from the echellograms . the spectra were linearized in ln@xmath36 based on the known optical distortions , and absolute wavenumber calibration was obtained from telluric emission features and corrected for the earth s motion relative to lsr . the resulting wavenumber scale is correct to within 1 . the same procedure was used for the asteroid spectra , and then the irs 1 spectra were divided by the asteroid spectra with correction for differences in airmass . with this procedure , telluric absorption lines as deep as 90% can be removed , although of course the noise increases on lines . however , broad response variations often remained , which were removed by fitting the continuum in each echelon order to a quartic polynomial and dividing . this procedure also sets the continuum in each order to one . ten spectral settings were observed in the 728820 @xmath37(13.712.2 @xmath18 m ) region , two in the 860930 @xmath37 ( 11.510.7 @xmath18 m ) region ( see figure [ fig : twelveum ] ) , and three in the 12401312 @xmath37 ( 8.07.6 @xmath18 m ) region ( see figure [ fig : eightum ] ) . a total of 110 echelon orders , or @xmath970 @xmath37 , were observed . the original observations were meant to study the line profiles of c@xmath0h@xmath0 and hcn , which had previously been detected . in addition to the prominent fundamental band lines of c@xmath0h@xmath0 and hcn , many c@xmath0h@xmath0 hot band lines were detected in our spectra including lines from the @xmath38 + @xmath23 @xmath38 and the 2@xmath23 @xmath23bands . after the hot bands were identified , unidentified lines still remained , some of which were double dipped . the double lines turned out to be ch@xmath3 lines which are split due to spin - rotation interactions ( see appendix ) . this is the first detection of ch@xmath3toward dense gas and the first ground - based detection of ch@xmath3 . previously , ch@xmath3 had only been observed with toward the galactic center @xcite . in efforts to detect ethane , c@xmath0h@xmath39 , at 810 - 820 @xmath37 , many lines were detected , though none corresponded to c@xmath0h@xmath39 . some were due to nh@xmath3 , but many lines remained unidentified ( see figures [ fig : hncoall ] and [ fig : hncoblowup ] ) . based on their separation we were able to identify the lines as due to hnco . interstellar hnco is a well known hot core molecule at radio wavelengths @xcite , but it had never been seen at infrared wavelengths . in total , five molecules were observed in the 728820 @xmath37 region : c@xmath0h@xmath0 ( including @xmath1c@xmath2ch@xmath0 ) , hcn , hnco , ch@xmath3 , and nh@xmath3 . at higher frequencies , ch@xmath4 , c@xmath0h@xmath0 and cs were observed in the 12401320 @xmath37 region . ch@xmath4 observations from the ground are possible for this source because of the large doppler shift with respect to the telluric lines . table [ sumlines ] gives a summary of the detected lines . at least 5 lines were observed for each molecule , with some molecules like c@xmath0h@xmath0 ( including isotopic lines and hot bands ) , and nh@xmath3 having more than 40 lines detected . however , the number of hnco lines dominates over lines from the other molecules , with over 100 lines detected . table [ linelist ] ( full version is available online only ) lists all the individual lines detected in our spectra . a selection of spectra including all observed molecules are shown in figures [ fig : twelveum ] and [ fig : eightum ] , along with fits described in [ sec : fit ] . for most molecules the line strengths were taken from the geisa03 database . the ch@xmath3 band strength was obtained from @xcite and the hnco oscillator strength was obtained from @xcite . the cs oscillator strength was taken from @xcite . lcc c@xmath0h@xmath0 & 25 & @xmath23 + c@xmath0h@xmath0@xmath40 & 2 & @xmath38 + @xmath23 + c@xmath0h@xmath0@xmath41 & 20 & @xmath38 + @xmath23@xmath38 + @xmath1c@xmath2ch@xmath0@xmath42 & 7 & @xmath23 + hcn & 10 & @xmath24 + nh@xmath3 & 64 & @xmath24 + hnco & 125 & @xmath38 + ch@xmath3 & 11 & @xmath24 + ch@xmath4 & 12 & @xmath24/@xmath38 dyad + cs & 6 & @xmath24 [ sumlines ] lcccc c@xmath0h@xmath0 & q(5 ) & 729.289 & 35.30 & @xmath23 + c@xmath0h@xmath0 & q(6 ) & 729.343 & 49.42 & @xmath23 + c@xmath0h@xmath0 & q(7 ) & 729.406 & 65.89 & @xmath23 + c@xmath0h@xmath0 & q(8 ) & 729.477 & 84.71 & @xmath23 + c@xmath0h@xmath0 & q(9 ) & 729.558 & 105.89 & @xmath23 + c@xmath0h@xmath0 & q(10 ) & 729.648 & 129.41 & @xmath23 + c@xmath0h@xmath0 & q(11 ) & 729.747 & 155.29 & @xmath23 + c@xmath0h@xmath0 & q(14 ) & 730.096 & 247.02 & @xmath23 + c@xmath0h@xmath0 & q(15 ) & 730.230 & 282.30 & @xmath23 + c@xmath0h@xmath0 & q(16 ) & 730.373 & 319.93 & @xmath23 + c@xmath0h@xmath0 & q(18 ) & 730.686 & 402.22 & @xmath23 + c@xmath0h@xmath0 & q(19 ) & 730.855 & 446.89 & @xmath23 + c@xmath0h@xmath0 & q(20 ) & 731.034 & 493.90 & @xmath23 + c@xmath0h@xmath0 & q(26 ) & 732.287 & 825.20 & @xmath23 + c@xmath0h@xmath0 & q(27 ) & 732.526 & 888.62 & @xmath23 + [ linelist ] to derive physical parameters of the absorbing gas , model spectra were fitted to the observations . for this purpose , the data were divided into two groups . all data at wavenumbers between 700 and 1000 @xmath37 ( 10.713.7 ) were fitted simultaneously , and data at 12401280 @xmath37 ( near 8 ) were fitted separately . the marquardt fitting procedure @xcite was used , which minimizes @xmath43 , the summed squared deviation normalized by the squared noise , between the data and the model . fitting was first attempted with a single component for each molecule , but the residuals suggested that a better fit could be obtained with multiple velocity components . the model that we used assumed that each molecule is found in one or two absorbing components . each component is described by its doppler shift , @xmath44 , its gaussian line width ( @xmath45 half width = ( 2kt / m)@xmath46 ) , @xmath47 , the molecule s column density , @xmath48 , the rotational temperature of the gas , @xmath26 , and the covering factor , @xmath49 , ( the fraction of the background continuum source covered by the component ) . the fitted column density is the average over the partially covered source , so the column density in the covered portion would be @xmath50 . a covering factor of , e.g. , 10% means that lines saturate at 90% of the continuum flux and the column density in the absorbing component is ten times the average over the continuum source . the effect modeled in this way could also result from veiling ( continuum emission from foreground or surrounding material ) or from re - emission in the lines by the absorbing molecules . the components were assumed to overlap by the products of their covering factors ( see fig . [ fig : comp ] ) . the following equation was used to calculate the observed transmission : @xmath51 where @xmath52 and @xmath53 are the covering factors for the two components . this results in a greater absorption in saturated lines than would result if the optical depths were added first and then the transmission spectrum was calculated from the optical depth spectrum . this approach was chosen because it gave a better fit to the data than adding the optical depths first and assuming the same covering factor , but it assumes that different components absorb along different lines of sight to a partially covered continuum source , which may not be the case . a frequency correction was allowed for each spectral setting observed , to correct for errors in wavelength calibration . with only one exception , the correction was less than a 1 doppler shift . in addition , a broadening of the instrumental resolution was permitted in the fitting to allow for imperfect internal instrument focusing , or possibly a ` macroturbulent ' broadening in the absorbing gas . the fitting program chose a resolution a factor of 1.2 greater than that derived from gas cell data . a constant frequency resolution ( as opposed to a constant doppler resolution ) was used in each of the two fitted regions . this represents well the improving resolving power toward shorter wavelengths in the 10.7 - 13.7 @xmath18mregion . in addition to these parameters , the continuum , slope , and curvature of each echelon order were varied , to allow correction of the baseline fitting done during data reduction . the continuum fitting and frequency correction required about 300 free , but rather easily determined , parameters . fewer parameters are needed to determine the physical conditions such as temperature , column density , line width and covering factor : 70 parameters for the 10.7 - 13.7 @xmath18 m spectra and 21 for the 8 @xmath18 m region . for the 10 - 13 region , we have over 3000 points to constrain the 70 parameters of interest if we characterize the constraining points by the number of lines times the number of pixels for each line . for c@xmath0h@xmath0 and hnco , lines of several bands were observed . for c@xmath0h@xmath0 , in addition to lines of the q and r branches of the @xmath23 fundamental , lines of several @xmath38 + @xmath23 @xmath23 and 2@xmath23 @xmath23 bands ( involving absorption from the excited @xmath38 and @xmath23 vibration states ) were observed . the vibrational temperatures describing the populations of the @xmath38 and @xmath23 states were included as free parameters in the fit . r - branch lines of the @xmath23 band of @xmath1c@xmath2ch@xmath0 were also observed , and the @xmath1c@xmath2ch@xmath0/c@xmath0h@xmath0 abundance ratio was allowed to vary , although it was kept the same in all the c@xmath0h@xmath0components . two c@xmath0h@xmath0 lines of the @xmath38 + @xmath23 band at 7.6 were observed as well . for hnco , lines from the @xmath38 band were observed . hnco is an only slightly non - linear molecule , and its spectrum consists of a series of sub - bands resembling those of a linear molecule , like c@xmath0h@xmath0 . its dipole moment oscillates diagonally , giving it a - type sub - bands , with no change in the angular momentum about its long axis and only p and r branches , and b - type sub - bands , in which the angular momentum about its long axis changes , and p , q , and r branches are seen . see appendix for further discussion of the hnco spectrum . since we did not have laboratory data regarding the relative strengths of the different sub - bands , we considered them to be caused by different species and summed the abundances to obtain the hnco abundance . the hcn lines observed were from the @xmath24 bending mode . we included hcn 2@xmath24 @xmath24 lines in the fit . although no hot hcn lines were obvious in the spectra , their inclusion lowered @xmath54 significantly ( @xmath55 ) and changed the best - fitting @xmath56 noticeably . although it might not be apparent from the spectra , given the shallowness of the lines , many of the detected lines are at least moderately saturated , and some are very saturated . the lines only appear weak because of the small covering factors ( or emission filling in the absorption lines ) and to a lesser extent the small line widths . this conclusion is very robust ; the relative depths of lines of different opacity requires substantial saturation . this is most apparent in the near equality of the depths of the ortho and para lines in the c@xmath0h@xmath0 q branch , near 730 @xmath37 ( see top panel of figure [ fig : twelveum ] ) . hcn in the 60 component and nh@xmath3 in the 56.5 component are so saturated that only lower limits could be placed on their abundances until their isotopomers were included in the fit to provide upper limit constraints . by including isotopomers and intrinsically weak lines , a wide enough range of line optical depths was observed to allow meaningful constraints to be placed on the molecular abundances in spite of the saturation . the observed lines of ch@xmath3 were in the r branch of the @xmath24 out - of - plane bending mode of this planar molecule . each of these lines is doubled by the interaction between the electron spin and the molecular rotation ( see appendix ) . the ch@xmath3 @xmath24 band center is at 16.5 @xmath18 m , whereas all observed lines were shortward of 13.5 @xmath18 m , with @xmath57 7 , making us rather insensitive to cold ch@xmath3 . lines of the p and q branches of the nh@xmath3 @xmath24 band were observed . there was no evidence of @xmath58nh@xmath3 lines , but they were included in the fit to constrain the nh@xmath3 abundance , as was done with h@xmath1cn . in the 8 @xmath18 m region , which was fitted separately , lines of the @xmath59 dyad of ch@xmath4 were observed . because of the strong absorption by low-@xmath22 lines of ch@xmath4 in the earth s atmosphere , most of the observed lines originate at @xmath60 . we were able to observe the r(0 ) line due to the favorable doppler shift of the source ( @xmath61 ) . the r(0 ) line helps constrain the temperature derived from the high @xmath22 lines . we also detected and fitted absorption by cs . the derived component parameters are given in table [ tabparms ] . uncertainties for the parameters , given in parentheses , are three times the square roots of the diagonal elements of the error matrix , which are uncertainties allowing all other parameters to vary . the noise in the fitted spectra is assumed to be purely photon statistical noise . with this noise estimate , the reduced @xmath62 , indicating that non - statistical noise sources contribute moderately , or perhaps that a different model for the line profiles is justified . including non - statistical noise , the uncertainties given might reasonably be taken to be 95% confidence intervals . the larger fractional errors are asymmetric , and would be more symmetric in the log of the parameters . for column densities , this is a result of the non - linearity of the curve of growth for moderately saturated lines . column densities with errors greater than their values are not consistent with zero . for unsaturated lines , the derived covering factors are very uncertain because equivalent widths depend only on the product of the covering factor and the column density . for example , the column densities of hnco and ch@xmath3 have smaller fractional uncertainties than those of c@xmath0h@xmath0 and hcn , whereas the covering factors have smaller uncertainties for c@xmath0h@xmath0 and hcn . all of the fitted molecules are detected with very high confidence . lccccccccccc c@xmath0h@xmath0 & 55.7 ( 0.3 ) & 0.6 ( 0.1 ) & 225 ( 20 ) & 3.0 ( 0.6 ) & 5.9 ( 0.6 ) & & 59.4 ( 0.3 ) & 0.6 ( 0.1 ) & 191 ( 10 ) & 2.8 ( 0.2 ) & 23.7 ( 1.0 ) + hcn & 56.3 ( 0.3 ) & 0.8 ( 0.1 ) & 256 ( 30 ) & 5.6 ( 2.8 ) & 8.4 ( 0.9 ) & & 60.0 ( 0.3 ) & 1.0 ( 0.4 ) & 456 ( 127 ) & 1.3 ( 1.2 ) & 68.4 ( 49.7 ) + hnco & 57.2 ( 0.3 ) & 1.8 ( 0.2 ) & 319 ( 27 ) & 0.4 ( 0.3 ) & 3.8 ( 0.3 ) & & 60.2 ( 0.2 ) & 0.7 ( 0.4 ) & 171 ( 79 ) & 0.1 ( 0.3 ) & 97.0 ( 11 ) + ch@xmath3 & 54.2 ( 0.4 ) & 1.8 ( 0.6 ) & 258 ( 17 ) & 1.7 ( 0.2 ) & 97.0 ( 15 ) & & 62.8 ( 0.5 ) & 1.5 ( 0.9 ) & 927 ( 161 ) & 0.6 ( 0.1 ) & 97.3 ( 18 ) + nh@xmath3 & 57.3 ( 0.3 ) & 0.9 ( 1.0 ) & 278 ( 31 ) & 5.2 ( 1.5 ) & 8.9 ( 1.2 ) & & 60.1 ( 0.2 ) & 0.4 ( 0.1 ) & 248 ( 32 ) & 2.8 ( 0.8 ) & 15.8 ( 3.6 ) + ch@xmath4 & 56.3 ( 0.3 ) & 0.2 ( 0.04 ) & 674 ( 360 ) & 20 ( 8) & 5.8 ( 1.9 ) & & 60.1 ( 0.3 ) & 0.6 ( 0.3 ) & 668 ( 250 ) & 16 ( 13 ) & 99 ( 13 ) + cs & 55.2 ( 0.4 ) & 1.0 ( 0.3 ) & 224 ( 173 ) & 0.02 ( 0.03 ) & 100 ( 8) & & 59.4 ( 0.5 ) & 0.7 ( 0.3 ) & 249 ( 100 ) & 7.0 ( 3.0 ) & 5.5 ( 0.3 ) + [ tabparms ] for most molecules , we found two components with doppler shifts of @xmath63 and @xmath64 , with widths of @xmath65 ( see table [ tabparms ] ) . most of the component temperatures were in the range 200 - 400 k and covering factors were typically 0.04 - 0.08 , with c@xmath0h@xmath0 requiring somewhat larger values of 0.06 and 0.18 . hnco has somewhat broader component line widths but other parameters are similar . ch@xmath3 has the broadest component line widths , with a greater velocity spacing between the two components and the hottest component . both cs and ch@xmath3 are fitted with much larger covering factors ( consistent with 100@xmath66 ) than were the other molecules . however , these covering factors are quite uncertain due to the weakness of the observed lines , and the large values may not be significant . the model spectra resulting from the best fit parameters can be seen in figures [ fig : twelveum ] and [ fig : eightum ] . in addition to agreeing among the different molecules ( except ch@xmath3 ) , the doppler shifts agree with the two most commonly observed values for molecules seen at radio wavelengths . this gives us some confidence that the two velocity components , at 56.5 and 60 , are real , although whether they are two separate components , or just a way of describing a non - gaussian velocity distribution is difficult to determine from the data . generally , the temperatures of the two components for a given molecule are not very different , although a fit constraining them to be equal was significantly worse than the fit allowing all temperatures to vary . line widths , column densities , and covering factors vary more . the fit was very poor when they were constrained to be equal for all molecules . note that our quoted column densities are averages over the lines of sight to the continuum emitting material , or column densities in the absorbing gas multiplied by the covering factors . there are several possible interpretations for the covering factor parameters : the continuum source could in fact be partially covered , foreground or surrounding emission could veil the spectrum , or re - emission in the lines could fill in the absorption . in all of these cases , the column densities through the absorbing material are given by n / c , rather than n , the average column density over the line of sight in the partial covering interpretation . consequently , if we could determine the column density along the lines of sight on which we see absorption we should compare that to our n / c values . before discussing the results of our measurements of individual molecules , we attempt to derive and co column densities from previous measurements . the column toward ngc 7538 irs 1 has not been directly measured , but it may be estimated from the extinction measurements of @xcite with the assumption of a normal interstellar gas - to - dust ratio . this method gives = @xmath67 . if the continuum source is non - uniformly covered , this column density represents an average over the source . hence , it should probably be compared to the values we derive for average column densities , n , of the molecules we observe . however , the silicate optical depth includes extinction from all material along the line of sight to ngc 7538 , most of which is unlikely to contain our molecules . this suggests that @xmath67 is an overestimate of in the relevant gas . on the other hand , we suggest below that we may be observing absorption by gas in the atmosphere of a protoplanetary disk , and dust in such a region could be depleted by settling toward the disk midplane . this effect would result in an underestimate of from the extinction . we give the ratios of our derived column densities to an column density of 7.5 @xmath68 in table [ tab : abun ] . lcc c@xmath0h@xmath0 & 5.7 & 7.7 + hcn & 6.9 & 9.2 + hnco & 0.5 & 0.6 + ch@xmath3 & 2.3 & 3.1 + nh@xmath3 & 8.1 & 10.8 + ch@xmath4 & 36 & 48 + cs & 7.0 & 9.3 + [ tab : abun ] it may be preferable to compare our column densities to the column density of co , which has been observed in absorption toward ngc 7538 irs 1 by @xcite . they obtain a column density of @xmath69 @xmath70 for @xmath1co in cold gas at 25 k and @xmath71 @xmath70 in warm gas at 176 k. we would not be sensitive to their cold gas . presumably their warm gas includes both of our components , although our temperatures are generally higher than theirs . if we assume @xmath2co/@xmath1co = 72 , using the @xmath2c/@xmath1c ratio we derived from our c@xmath0h@xmath0 observations , we derive the abundances of our molecules relative to co listed in table [ tab : abun ] . ( note that @xcite use @xmath2c/@xmath1c = 89 . ) however , we should note that we have measured one spectral setting including co lines , while searching for ocs absorption near 5 @xmath18 m . the noise is rather high , but it is apparent that the @xmath1co lines are not well described by moderately saturated gaussians , as assumed by @xcite . the @xmath1co lines are deeper and broader than the lines of any of our other molecules , and have prominent blue - shifted shoulders probably tracing an outflow . we suspect that the lines are more saturated , and the co column density larger than @xcite concluded . the different shapes and greater depths of the co lines than our c@xmath0h@xmath0 lines also indicates that co probes different gas . likely , c@xmath0h@xmath0 and the other molecules we observe in the mid - infrared are found only in unusual regions , whereas co is distributed along the entire line of sight . in general , we conclude that the ratios of column densities to those of either or co given in table [ tab : abun ] are quite uncertain because the different column density tracers may be sensitive to different components along the line of sight . the column densities of the different molecules we observe should be much more comparable . since our initial search for molecules began with c@xmath0h@xmath0 and hcn , the first models only included those two molecules . in these models , we derived similar temperatures and column densities as derived from observations @xcite . once we include absorption from other molecules , hnco in particular , the temperatures for hcn and especially for c@xmath0h@xmath0 are lowered significantly . some hnco lines overlap high-@xmath22 lines of c@xmath0h@xmath0 , which means that more hot c@xmath0h@xmath0 is necessary to produce the observed absorption when hnco is not included in the model . in our final model including all of the observed molecules , our derived column densities of c@xmath0h@xmath0 , @xmath72 , and hcn , @xmath73 , ( see table [ tabparms ] ) are larger than those derived in @xcite , @xmath74 and @xmath75 . part of the difference is due to the different doppler parameter used : @xmath76 is used in @xcite , whereas , we derive @xmath77 . we derive lower temperatures , @xmath78 k and @xmath79 k than derived by @xcite , @xmath80 k and @xmath81 k , and @xcite , @xmath82 k. we also find that we need a small covering factor to account for the saturated yet shallow lines . the level of saturation is difficult to determine from the data . in order to verify that our fit to the texes data is also consistent with the observations , we take our model at the -sws resolution and compare to the data ( figure [ fig : isofit ] ) . the model matches the fundamental q - branches for the two molecules as well as a hot band q - branch of c@xmath0h@xmath0 . c@xmath0h@xmath0 is expected to be in rotational local thermal equilibrium ( lte ) since it has no permanent dipole moment . the fact that the two excited vibrational states of c@xmath0h@xmath0 from which absorption was observed , @xmath38 and @xmath23 , are found to be populated near lte at the c@xmath0h@xmath0 rotational temperature , or even slightly higher , provides information about the physical conditions of the gas . for both bands , the hot bands indicate they are slightly out of lte since the vibrational temperatures ( @xmath83 ) are higher than the rotational temperatures ( @xmath84 ) . for the @xmath38 + @xmath23 @xmath38 bands , the @xmath83 temperatures for the two components were 346 k and 271 k while for the 2@xmath23 @xmath23 bands , they were 277 k and 227 k. the rotational temperatures for the @xmath23 fundamental are 225 k and 191 k for the two components . in lte , the vibrational temperatures for the hot bands would equal the rotational temperature measured by the fundamental band lines . the @xmath38 level has no allowed radiative transitions to the ground vibrational level , and none of the rotational levels of the ground vibrational level are coupled radiatively , in both cases due to the symmetry of the c@xmath0h@xmath0 molecule . this suggests that the populations of the ground and @xmath38 states should be set by collisions , and so should be in lte at the same temperature . however , the @xmath38 level can be radiatively pumped by absorption in the 7.6 @xmath18 m @xmath38 + @xmath23 band followed by emission of photons in the @xmath38 + @xmath23 @xmath38 band . the @xmath23 state must be populated predominantly radiatively for any plausible density , since the critical density is @xmath85 . from these considerations , we conclude that the brightness temperature of the radiation field seen by the observed gas must be close to the kinetic temperature of the gas . in addition , the fact that the measured rotational temperature for hcn is similar to that of c@xmath0h@xmath0 indicates that the rotational levels of hcn are kept populated either by collisions , which would require a gas density @xmath86 , or by infrared pumping through the @xmath24 transitions . @xcite conclude that either excitation mode is possible . the presence of c@xmath0h@xmath0 2@xmath23 @xmath23 absorption requires infrared pumping , although @xmath38 + @xmath23 @xmath38 absorption does not . we can attempt to estimate the extent to which emission in our lines fills in the absorption , at least for c@xmath0h@xmath0 , by using the 2@xmath23 @xmath23 lines as probes of the @xmath23 population . our spectra require a @xmath23 vibrational temperature @xmath9300 k. we need to compare this number to the blackbody brightness temperature of the background continuum radiation . from the shape of the 813 @xmath18 m spectrum , @xcite derived a dust temperature of 330 k assuming the emitting dust is optically thick , or 370 k assuming optically thin silicate emission . he preferred the latter model , which gives a brightness temperature only when combined with an assumption about the solid angle subtended by the source . in either case , our vibrational temperature is only @xmath910% less than the background temperature , so that the source function is @xmath930% below the background intensity , and lines would be expected to saturate at @xmath970% of the continuum . however , the continuum spectrum from which @xcite derived a dust temperature is the average over the continuum source . if the gas we observe covers only a small fraction of the source , the brightness temperature of the radiation passing through that gas could be much larger than the average . given the many assumptions and uncertainties in this calculation , it is difficult to make a strong statement . the level at which our c@xmath0h@xmath0 lines saturate , @xmath985% of the continuum , which we model with a small covering factor , could instead be due partly or entirely to emission by the absorbing gas . it is possible for the lines to saturate at 85% of the continuum even with small covering factors ( e.g. , 24% and 6% for c@xmath0h@xmath0 ) since the line widths are narrower than our spectral resolution . presumably the same statement applies to the other observed molecules . from our infrared observations , we find that nh@xmath3 has two components : one at 57.3 and the other at 60.1 . most of the nh@xmath3 is found in the first component , for which the temperature is @xmath9 278 k and the column density is 5.2 @xmath87 . the second component has a lower temperature ( @xmath9248 k ) and the column density is 2.8 @xmath88 ( see table [ tabparms ] ) . nh@xmath3 absorption toward irs 1 has also been studied in the radio by @xcite and @xcite . they found @xmath89 in gas with @xmath26 = 170220 k and @xmath90 . although their temperature and doppler shift is in reasonable agreement with our bluer absorption component , their column density is nearly 20 times ours . without constraining the @xmath58n/@xmath91n isotopic ratio , our data allow an nh@xmath3 column as large as that from radio observations , but with this nh@xmath3 column @xmath58nh@xmath3 lines should have been apparent in our spectra . it is possible that we sample only a fraction of the column observed at cm wavelengths as a result of larger dust opacity at mid - infrared wavelengths preventing us from observing lines of sight through the densest regions of a knotty gas distribution . it may be notable that @xcite found , as did we , that although relative depths of lines ( in their case hyperfine components ) require large optical depths , line shapes do not appear flat - bottomed as would be expected if they are optically thick . they suggest that the absorbing gas may be in small spectrally and spatially unresolved knots , each of which is optically thick , and which combine to make a line that does not have the shape of a thick line . our use of two components in our fit may be a way of approximating this situation . the velocities for the two components are 57.2 and 60.2 , with corresponding temperatures of 319 and 171 k and column densities of 4@xmath92 and 1@xmath92 . since the hnco temperatures and velocities derived from our observations agree with the values for the other molecules , we can assume that hnco is in the same gas . it appears the region probed by hnco is very small in angular size . @xcite detect hnco emission in the radio and derive a beam - averaged column density of 1.1@xmath9310@xmath91 . this value is an order of magnitude lower than our observed column density . however , beam dilution in the radio observations can explain the difference in the column densities . ch@xmath3 is detected for the first time toward warm , dense gas . the parameters derived for this molecule do not agree as well with the other molecules , although the weakness of the observed lines makes the derived parameters rather uncertain . the separation between the centroid velocities for the two components , @xmath94 = -54.2 and 62.8 , is nearly twice the separation found for the other molecules . also , the covering factor for both of the components is consistent with 100% , whereas as it is closer to 10% for the other molecules . in addition , the temperature of the second component is much hotter ( @xmath15900 k ) than the temperatures derived for the other molecules . we note though that the fact that the observed lines are of high excitation makes them particularly sensitive to a hot gas component , which may not have been noticed for other molecules . and the weakness of the lines makes the derived covering factor ( which is inferred from relative line depths indicating saturation ) very uncertain . ch@xmath4 gas - phase absorption had been observed toward the neighboring protostar irs 9 @xcite but had not previously been seen toward irs 1 . we find centroid velocities of 56.3 and 60.1 , which are similar to the other molecules . the temperatures are higher than for most molecules ( @xmath9 670 k for both components ) but have large uncertainties ( see table [ tabparms ] ) . the best fit seems to indicate that one of the components covers the entire source whereas the other covers about 6% of the source . of all of the observed molecules , ch@xmath4 has the highest column density . the implications of the high ch@xmath4 column density the chemistry are discussed in [ sec : chemmod ] . we detect only 6 lines of cs ( see table [ linelist ] full table available online only ) . the lines are rather weak and parameters are poorly constrained . despite uncertainties , the parameters agree with those for other molecules . the centroid velocities are at 55.2 and 59.4 . the temperatures , @xmath95 and @xmath96 , are similar to those found other molecules ( except for ch@xmath4 and ch@xmath3 ) . from radio studies of the envelope around irs 1 , a cs abundance of @xmath97 is found @xcite . however , also using radio observations , @xcite derive a cs abundance of @xmath98 . our abundance is about 2 orders of magnitude larger than the largest estimate of the abundance in the envelope indicating that our observations are probing material closer to the protostar . we now consider the implications of the observations for the chemistry of the observed material . hot core models can be used to represent two of the scenarios discussed in [ sec : intro ] : 1 ) material in a circumstellar disk and 2 ) photo - evaporating knots of neutral molecular material . hot cores are small ( @xmath99 pc ) , dense ( @xmath100 ) and hot ( @xmath101 k ) regions associated with high - mass young stellar objects @xcite . chemically , hot cores are identified by an enhanced abundance of fully hydrogenated molecules such as h@xmath0o and nh@xmath3 , which are usually observed through rotational transitions at sub - millimeter and millimeter wavelengths . these abundances are enhanced with respect to the cold molecular envelope . a hot core is thought to form as a result of the protostar heating nearby material , which triggers ice sublimation from the grain mantles @xcite . hnco is another hot core molecule which has been found in the orion hot core with an abundance @xmath98 relative to @xcite . gas - phase hnco is possibly coming from evaporation of grain mantles . the 4.62 feature observed toward some protostars has been attributed to ocn@xmath102 frozen in grain mantles ( e.g. , * ? ? ? * ; * ? ? ? * ) . recent results of solid state features toward protostars show an ocn@xmath102 upper limit of 10@xmath103 toward irs 1 @xcite . the total hnco column derived in this work is 5.4@xmath92 , a factor of 2 lower than the solid ocn@xmath102 limit . in general , irs 1 does not show many ice features compared to the colder nearby source irs 9 ( see * ? ? ? * ) , which has a column density @xmath48(ocn@xmath102 ) = 1.2@xmath104 . if solid ocn@xmath102 sublimates and captures a proton to form hnco , the expected column density for hnco would be comparable to the observed ocn@xmath102 ice toward cold lines of sight ( like irs 9 ) . however , the gas - phase column observed is at least an order of magnitude smaller than that of the solid state ion toward sources like irs 9 . this indicates that when ocn@xmath102 sublimates , a fraction of the ices goes into forming hnco , while most of hnco is destroyed in the warm gas - phase chemistry on short time scales . further studies of how ocn@xmath102 reacts in the gas - phase are needed in order to understand the difference between ice and gas phase column densities . similarly , ch@xmath4 is probably also evaporating from the grain mantles , and thus the gas - phase abundance is highly enhanced compared with the cold molecular envelope . studies of the solid ch@xmath4 content toward irs 1 show that the column density is 14 times less than our observed gas - phase column @xcite , suggesting that the solid ch@xmath4 on grains has sublimated . in comparison , irs 9 has a higher content of solid ch@xmath4 than gas - phase ch@xmath4 @xcite . if we consider that irs 1 is likely a more evolved object than irs 9 @xcite , the higher gas - phase ch@xmath4 toward irs 1 also points to grain mantle evaporation . the high abundance of ch@xmath4 means that daughter products such as ch@xmath3 and ch@xmath0 should be abundant . while the photo - dissociation of ch@xmath4 preferentially forms ch@xmath0 rather than ch@xmath3 , our observations of ch@xmath3 agree with this branching ratio . we predict that ch@xmath0 is also present toward irs 1 with a higher abundance than ch@xmath3 . it is uncertain where c@xmath0h@xmath0 is formed in the gas - phase chemistry or is evaporating from grain mantles . however , most models do not predict the observed column densities from formation in gas - phase chemistry only . from our observations we see a high abundance of the parent molecules nh@xmath3 and ch@xmath4 . high abundances of the other molecules are also observed . @xcite present a time dependent chemical evolution of molecules in a hot core . the models begin at the time when the protostar turns on ( @xmath105 yrs ) . table [ tab : mods ] shows the abundances with respect to co at an age of 10@xmath106 years . the hot core model seems to underpredict abundances with respect to co of all our observed molecules . however , as mentioned in [ sec : c2h2 ] , the co column density found by @xcite may not be the appropriate number with which to compare our observations . if the co column along the lines of sight containing our molecules is larger than the column measured by @xcite , then perhaps the abundances from the model will more closely resemble the observations . we can also compare the relative abundances between our observed molecules , avoiding the uncertainties in whether the co and dust absorption trace the lines of sight that our molecules are found in . the observed and predicted values for @xmath48(ch@xmath3)/@xmath48(ch@xmath4 ) are similar , @xmath90.01 . for the other molecules the ratios from the models at an age of 10@xmath106 years are very different from the observed values . the values for the models assume that we can see all the material and comparing the model results directly with the observations without accounting for line - of - sight does not make for a good comparison . figure [ fig : chemmod ] shows the column density evolution for 6 of the detected molecules as determined by @xcite corrected for the observed beam . for each molecule , the observed range of column densities is shown in the horizontal shaded area . this hot core model can predict the observed column densities for 4 of the 6 molecules plotted . however , the ages indicated by each molecule do not give a consistent overall age . for ch@xmath4 and hcn the model overpredicts the observed column densities . the other molecules suggest ages between 2@xmath107 and 2@xmath108 years . while ages vary by three orders of magnitude , it should be noted that the model has not been optimized to be an accurate representation of this source . with chemical models that more closely represent the source , it may be possible to determine a more consistent age . lccc c@xmath0h@xmath0 & 5.7 & 0.88 & 0.06 0.1 + hcn & 6.9 & 0.02 & 16 24 + ch@xmath3 & 2.3 & 0.016 & 0.02 1.6 + nh@xmath3 & 8.1 & 0.76 & 1 6 + ch@xmath4 & 36 & 1.6 & 7 35 + cs@xmath109 & 7.0 & 0.0031 & 0.03 0.07 + [ tab : mods ] if we consider the disk scenario , we can also compare our observations to a chemical model of a photo - evaporating disk . @xcite present a chemical model for a disk around a 10 m@xmath110 star . the model consists of a flared disk with a photo - evaporating layer on the surface . table [ tab : mods ] compares our derived abundances with respect to co for the various molecules to the predicted values from disk models . the summarized results of models taking the midplane temperature for the temperature of the layers below the photo - dissociation region ( pdr ) are discussed here . the difference among the three models is the variation in the x - ray ionization rate , @xmath111 ( 1@xmath111 = 1.3@xmath9310@xmath112s@xmath113 ) , from 1 to 10@xmath114 . table [ tab : mods ] shows that the abundance for c@xmath0h@xmath0 is underpredicted . in the chemical model , c@xmath0h@xmath0 is produced only by gas phase reactions . one possible explanation is that c@xmath0h@xmath0 is formed as ice on grain mantles . the abundance of nh@xmath3 is also underpredicted . the model with the highest x - ray ionization rate produces ch@xmath3 abundances close to the observed value . for that model , @xmath48(ch@xmath3)/@xmath48(ch@xmath4 ) is 4 times larger than the observed ratio . these models do not fit our observations very well but can serve as a guide to the expected abundances in such a scenario . the model by @xcite has not been modified to try to fit data . so , adapting the model for this object may give predictions closer to the observations . the star in ngc 7538 irs 1 is believed to be a 30 m@xmath110 star . the different radiation field may affect the resulting chemistry . @xcite present a model of chemical evolution of the envelopes of massive protostars using afgl 2591 as an example . from that model , they conclude that enhancements of c@xmath0h@xmath0 , hcn and ch@xmath4 are possible at late times in the evolution ( @xmath115 years ) and at high temperature ( @xmath116 800 k ) . our derived temperatures indicate cooler material , yet the molecules seem to be enhanced in the line of sight . alternatively , these species could be produced at high temperature in the inner disk , and subsequently brought outward by the disk wind . including uv and x - ray radiation in the model affects the chemistry especially relating to hcn . @xcite find that increasing the uv and x - ray flux can lead to enhanced hcn starting by the ionization of n@xmath0 . the column densities for hcn and cs from the @xcite model of protostar afgl 2591 agree with our observed abundances while their column densities for c@xmath0h@xmath0 , ch@xmath4 , and nh@xmath3 are lower than our observed values . hot core and disk chemistry models predict the enhancement of molecules such as c@xmath0h@xmath0 , ch@xmath4 , and nh@xmath3 . however , the observed abundance of c@xmath0h@xmath0 is higher than what models predict from warm gas - phase chemistry ( see table [ tab : mods ] ) . this indicates that c@xmath0h@xmath0 is probably frozen on dust grains and sublimates along with molecules like ch@xmath4 . @xcite studied the solid features of c@xmath0h@xmath0 , especially when mixed with h@xmath0o or co. they found that the c@xmath0h@xmath0 features broaden substantially when mixed with h@xmath0o and to a lesser extent when mixed with co. they compare the laboratory data to data for the colder neighbor , irs 9 , which resulted in an upper limit of @xmath117 for the column density of solid c@xmath0h@xmath0 . this upper limit is consistent with the observed gas - phase column density seen toward irs 1 . so , it is possible that solid c@xmath0h@xmath0 is sublimating from grain mantles as protostars heat the environment . the observed ch@xmath3/ch@xmath4 ratio agrees with the branching ratio for the destruction of ch@xmath4 . based on these results ch@xmath0 should also be very abundant . we now attempt to construct a physical and geometrical model of ngc 7538 irs 1 . within the possible scenarios presented from various radio and infrared observations ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , we propose a scenario in which the molecular absorption presented here comes from a circumstellar disk . we will examine other possibilities first . figure [ fig : cartoon ] depicts the possible scenarios allowed by the available observations . we can quickly rule out a simple model in which the absorbing molecular gas is in the foreground molecular cloud and not closely associated with irs 1 . our observations require the gas to be much hotter ( @xmath116 300 k ) and denser ( @xmath118 to maintain rotational lte of hcn out to @xmath119 ) than is found away from luminous sources in molecular clouds . according to @xcite , temperatures do not reach the observed values until you get to within 400 au of the central star . in addition , the infraredradiation field must have a brightness temperature @xmath9300 k to populate the c@xmath0h@xmath0 @xmath23 level sufficiently to account for the observed 2@xmath23@xmath23 absorption . undoubtedly , the absorbing molecular gas is in close proximity to the infraredcontinuum source irs 1 . it is not quite so easy to rule out a model in which the absorbing molecules are in boundary region between the envelope around the irs 1hypercompact h ii region and the outflow . this gas could be compressed by the ionized wind , and if it is as close as 0.1 , or 280 au , from a @xmath120 source it would have a temperature near 300 k. the temperature structure determined by @xcite indicates that indeed temperatures range from 200 400 k at radii of 220380 au . however , interaction with the ionized wind , which has a velocity @xmath9100 , would be expected to accelerate the gas , causing broad , blue - shifted absorption ( see * ? ? ? in contrast , the observed lines have widths of @xmath121 and have centroids within a few of those seen in surrounding molecular gas . another argument against the presence of the observed gas in an envelope around irs 1 is the fact that net absorption is seen . since the vibrational temperature is comparable to the brightness temperature of the continuum radiation , emission lines would be seen if the molecular gas had a larger extent than the continuum source . if the lines arose in molecular shell surrounding the hypercompact h ii region , this probably would be the case . on the other hand , @xcite proposes that the centimeter continuum from irs 1 results from partially ionized material in an outflow . the centimeter continuum emission has been spatially resolved into knots by @xcite , who suggest that the emission comes from photo - evaporation of knots of neutral molecular material ( see fig . [ fig : cartoon ] ) . the ch@xmath3oh masers seen by @xcite also seem to trace the knotty structure ( in addition to the disk described in [ sec : intro ] ) probed by the cm continuum emission . the knots may be material stripped from the disk . the stellar wind would then blow on these knots . however , if our molecules are in these knots we would expect large velocity difference between the knots resulting in broad lines , which are not observed . if we compare the velocities we derive for the two components of our molecules to those observed for the ch@xmath3oh masers , we find that they match the two velocities found for the a cluster of masers . this indicates that the infrared observations may be probing the same gas as the ch@xmath3oh masers . this scenario is further supported by @xcite . however , @xcite also find that the interpretation of the masers tracing a disk also agrees with their observations . we propose that the absorbing material is in a circumstellar disk . there is evidence for a disk close to edge - on from other observations and models ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? in this situation , hot dust in the surface layers or the inner rim of a flared disk could provide the continuum source . with a nearly edge - on orientation , radiation from the inner region could pass through the outer disk atmosphere , where absorption lines could be formed . emission from some of the dust on the surface of a slightly inclined disk or in an outflow could be observed directly , accounting for the dilution of the spectrum that we modeled with a small covering factor . this scenario is similar to the one described by @xcite to describe c@xmath0h@xmath0 and hcn absorption toward the low - mass protostar irs 46 in ophiuchus ( ylw 16b ) . detailed modeling of the disk structure of irs 46 showed that it was possible to have the absorbing material in the disk illuminated by the continuum from the surface . figure [ fig : cartoon ] is a cartoon of our proposed picture , where the orange rays represent the continuum . some of the continuum radiation passes through the absorbing outer disk . in this scenario , dust settling is likely to occur , allowing for observations of large columns of gas without too much dust . here we present a summary of our conclusions : \1 . ngc 7538 irs 1 has a rich spectrum in the mid - infrared . at high spectral resolution , lines of previously observed molecules such as c@xmath0h@xmath0and hcn are seen as well as weak lines of molecules such as hnco , ch@xmath3 , nh@xmath3 , ch@xmath4 , and cs . we have presented the first infrared detection of interstellar hnco . this is also the first detection of ch@xmath3 toward dense gas . the data show shallow yet saturated lines indicating that either the absorbing gas does not fully cover the continuum source or there is emission filling in the absorption . the observed ortho : para equivalent width ratio for c@xmath0h@xmath0 is @xmath9 1.5 instead of the expected value of 3 . the rotational temperatures for the various molecules range between 200 and 400 k with the exception of ch@xmath3 , which has one hot component ( @xmath122 k ) . from the derived temperatures , we find that the gas is within @xmath9 400 au of the central star @xcite . we suggest that the material traces a close to edge - on circumstellar disk as determined from other observations ( e.g. , * ? ? ? however , it can also be tracing hot , knotty material close in to the star ( see figure [ fig : cartoon ] ) . \5 . because it is possible to use hot core chemistry to describe chemistry in a disk , the different chemical models do not help identify the physical location of the absorbing gas . however , the models do seem to indicate that ngc 7538 irs 1 is an evolved protostar with an age @xmath123 years . our observations help constrain the chemistry in massive protostars . the abundances for all the molecules are enhanced , except for possibly cs ( see discussion in [ sec : ch4cs ] ) . species known to be present in icy mantles such as ch@xmath4 and nh@xmath3 have high column densities suggesting that they have recently sublimated from grain mantles . the high abundance of the daughter molecules such ch@xmath3 gives constraints on the gas - phase chemistry happening after sublimation of ices . hnco also provides some constraints about the presence of ocn@xmath102 on dust grains . it is unclear whether c@xmath0h@xmath0 is a parent molecule or a product of warm gas - phase chemistry . for molecules like hcn , uv / x - ray radiation is important in producing the observed abundances . ch@xmath3 is a nearly planar molecule and thus do not expect to see inversion transitions as observed in nh@xmath3 . we observe the out - of - plane vibrational mode ( similar to the umbrella mode in nh@xmath3 ) . we observe splitting due to spin - rotation interactions . for symmetric - top molecules , the splitting of a rotational level is given by where @xmath125 and @xmath126 are the spin - rotation coupling constants for the b and c axes , respectively , at a given vibrational state , @xmath127 . we use equation ( [ eq : ch3 ] ) in deriving the splitting for the lines we observed . the values for @xmath125 and @xmath128 were taken from @xcite . hnco is a quasilinear , nearly symmetric - top molecule . the moment of inertia about the figure axis is small and the moments of inertia about the other two axes are much larger ( @xmath9 100 times larger ) and about equal to each other . because it is a slightly asymmetric molecule , the levels with k @xmath129 0 are split into two components . @xcite refer to these levels as upper ( u ) and lower ( l ) . table [ rules ] shows the selection rules for hnco . in figure [ fig : hncoblowup ] , the hnco lines labeled r(j)1u correspond to lines where @xmath136 and @xmath137 and k = 1 upper . likewise , the lines labeled r(j)1l are lines where @xmath136 and @xmath137 and k = 1 lower . because the splitting between the two levels is small the lines overlap in the spectral region shown in figures [ fig : hncoall ] and [ fig : hncoblowup ] . some lines from the upper and lower k states are blended together and give the appearance of double peaked lines . , f. , van dishoeck , e. f. , boogert , a. c. a. , pontoppidan , k. m. , blake , g. a. , dullemond , c. p. , evans , n. j. , hogerheijde , m. r. , jrgensen , j. k. , kessler - silacci , j. e. , & knez , c. 2006 , , 636 , l145

computation of the marginal likelihood from a simulated posterior distribution is central to bayesian model selection but is computationally difficult . the often - used harmonic mean approximation uses the posterior directly but is unstably sensitive to samples with anomalously small values of the likelihood . the laplace approximation is stable but makes strong , and often inappropriate , assumptions about the shape of the posterior distribution . it is useful , but not general . we need algorithms that apply to general distributions , like the harmonic mean approximation , but do not suffer from convergence and instability issues . here , i argue that the marginal likelihood can be reliably computed from a posterior sample by careful attention to the numerics of the probability integral . posing the expression for the marginal likelihood as a lebesgue integral , we may convert the harmonic mean approximation from a sample statistic to a quadrature rule . as a quadrature , the harmonic mean approximation suffers from enormous truncation error as consequence . this error is a direct consequence of poor coverage of the sample space ; the posterior sample required for accurate computation of the marginal likelihood is much larger than that required to characterize the posterior distribution when using the harmonic mean approximation . in addition , i demonstrate that the integral expression for the harmonic - mean approximation converges slowly at best for high - dimensional problems with uninformative prior distributions . these observations lead to two computationally - modest families of quadrature algorithms that use the full generality sample posterior but without the instability . the first algorithm automatically eliminates the part of the sample that contributes large truncation error . the second algorithm uses the posterior sample to assign probability to a partition of the sample space and performs the marginal likelihood integral directly . this eliminates convergence issues . the first algorithm is analogous to standard quadrature but can only be applied for convergent problems . the second is a hybrid of cubature : it uses the posterior to discover and tessellate the subset of that sample space was explored and uses quantiles to compute a representative field value . qualitatively , the first algorithm improves the harmonic mean approximation using numerical analysis , and the second algorithm is an adaptive version of the laplace approximation . neither algorithm makes strong assumptions about the shape of the posterior distribution and neither is sensitive to outliers . based on numerical tests , we recommend a combined application of both algorithms as consistency check to achieve a reliable estimate of the marginal likelihood from a simulated posterior distribution . * keywords : * bayesian computation , marginal likelihood , algorithm , bayes factors , model selection

0911.1777

a bayesian data analysis specifies joint probability distributions to describe the relationship between the prior information , the model or hypotheses , and the data . using bayes theorem , the posterior distribution is uniquely determined from the conditional probability distribution of the unknowns given the observed data . the posterior probability is usually stated as follows : @xmath0 where @xmath1 is the marginal likelihood . the symbol @xmath2 denotes the assumption of a particular model and the parameter vector @xmath3 . for physical models , the sample space @xmath4 is most often a continuous space . in words , equation ( [ eq : bayes ] ) says : the probability of the model parameters given the data and the model is proportional to the prior probability of the model parameters and the probability of the data given the model parameters . the posterior may be used , for example , to infer the distribution of model parameters or to discriminate between competing hypotheses or models . the latter is particularly valuable given the wide variety of astronomical problems where diverse hypotheses describing heterogeneous physical systems is the norm ( see * ? ? ? * for a thorough discussion of bayesian data analysis ) . for parameter estimation , one often considers @xmath5 to be an uninteresting normalization constant . however , equation ( [ eq : evidence ] ) clearly admits a meaningful interpretation : it is the support or _ evidence _ for a model given the data . this see this , assume that the prior probability of some model , @xmath6 say , is @xmath7 . then by bayes theorem , the probability of the model given the data is @xmath8 . the posterior odds of model @xmath9 relative to model @xmath10 is then : @xmath11 if we have information about the ratio of prior odds , @xmath12 , we should use it , but more often than not our lack of knowledge forces a choice of @xmath13 . then , we estimate the relative probability of the models given @xmath14 over their prior odds by the bayes factor @xmath15 ( see * ? ? ? * for a discussion of additional concerns ) . when there is no ambiguity , we will omit the explicit dependence on @xmath16 of the prior distribution , likelihood function , and marginal likelihood for notational convenience . the bayes factor has a number of attractive advantages for model selection @xcite : ( 1 ) it is a consistent selector ; that is , the ratio will increasingly favor the true model in the limit of large data ; ( 2 ) bayes factors act as an occam s razor , preferring the simpler model if the `` fits '' are similar ; and ( 3 ) bayes factors do not require the models to be nested in any way ; that is , the models and their parameters need not be equivalent in any limit . there is a catch : direct computation of the marginal likelihood ( eq . [ eq : evidence ] ) is intractable for most problems of interest . however , recent advances in computing technology together with developments in markov chain monte carlo ( mcmc ) algorithms have the promise to compute the posterior distribution for problems that have been previously infeasible owing to dimensionality or complexity . the posterior distribution is central to bayesian inference : it summarizes all of our knowledge about the parameters of our model and is the basis for all subsequent inference and prediction for a given problem . for example , current astronomical datasets are very large , the proposed models may be high - dimensional , and therefore , the posterior sample is expensive to compute . however , once obtained , the posterior sample may be exploited for a wide variety of tasks . although dimension - switching algorithms , such as reversible - jump mcmc @xcite incorporate model selection automatically without need for bayes factors , these simulations appear slow to converge for some of our real - world applications . moreover , the marginal likelihood may be used for an endless variety of tests , ex post facto . @xcite presented a formula for estimating @xmath17 from a posterior distribution of parameters . they noted that a mcmc simulation of the posterior selects values of @xmath3 distributed as @xmath18 and , therefore , @xmath19 or @xmath20_{p({\boldsymbol{\theta}}|{{\bf d } } ) } , \label{eq : hmean}\ ] ] having suppressed the explicit dependence on @xmath16 for notational clarity . this latter equation says that the marginal likelihood is the harmonic mean of the likelihood with respect to the posterior distribution . it follows that the harmonic mean computed from a sampled posterior distribution is an estimator for the marginal likelihood , e.g. : @xmath21^{-1}. \label{eq : hsamp}\ ] ] unfortunately , this estimator is prone to domination by a few outlying terms with abnormally small values of @xmath22 ( e.g. see * ? ? ? * and references therein ) . @xcite describes convergence criteria for equation ( [ eq : hsamp ] ) and @xcite present augmented approaches with error estimates . alternative approaches to computing the marginal likelihood from the posterior distribution have been described at length by @xcite . of these , the laplace approximation , which approximates the posterior distribution by a multidimensional gaussian distribution and uses this approximation to compute equation ( [ eq : evidence ] ) directly , is the most widely used . this seems to be favored over equation ( [ eq : harmonic ] ) because of the problem with outliers and hence because of convergence and stability . in many cases , however , the laplace approximation is far from adequate in two ways . first , one must identify all the dominant modes , and second , modes may not be well - represented by a multidimensional gaussian distribution for problems of practical interest , although many promising improvements have been suggested ( e.g. * ? ? ? * ) . @xcite explored the use of the savage - dickey density ratio for cosmological model selection ( see also * ? ? ? * for a full review of the model selection problem for cosmology ) . finally , we may consider evaluation of equation ( [ eq : evidence ] ) directly . the mcmc simulation samples the posterior distribution by design , and therefore , can be used to construct volume elements in @xmath23-dimensional parameter space , @xmath24 , e.g. when @xmath25 . although the volume will be sparsely sampled in regions of relatively low likelihood , these same volumes will make little contribution to equation ( [ eq : evidence ] ) . the often - used approach from computational geometry , delaney triangulation , maximizes the minimum angle of the facets and thereby yields the `` roundest '' volumes . unfortunately , the standard procedure scales as @xmath26 for a sample of @xmath27 points . this can be reduced to @xmath28 using the flip algorithm with iterative construction @xcite but this scaling is prohibitive for large @xmath27 and @xmath23 typical of many problems . rather , in this paper , we consider the less optimal but tractable kd - tree for space partitioning . in part , the difficulty in computing the marginal likelihood from the sampled posterior has recently led ( * ? ? ? * nesting sampling ) to suggest an algorithm to simulate the marginal likelihood rather than the posterior distribution . this idea has been adopted and extended by cosmological modelers @xcite . the core idea of nesting sampling follows by rewriting equation ( [ eq : zdef0 ] ) as a double integral and swapping the order of integration , e.g. @xmath29 the nested sampler is a monte carlo sampler for the likelihood function @xmath30 with respect to the prior distrbution @xmath31 so that @xmath32 . the generalization of the construction in equation ( [ eq : nest ] ) for general distributions and multiple dimensions is the lebesgue integral ( see [ sec : intgr ] ) . clearly , this procedure has no problems with outliers with small values of @xmath33 . of course , any algorithm implementing nested sampling must still thoroughly sample the multidimensional posterior distribution and so retains all of the intendant difficulties that mcmc has been designed to solve . in many ways , the derivation of the nested sampler bears a strong resemblance to the derivation of the harmonic mean but without any obvious numerical difficulty . this led me to a careful study of equations ( [ eq : evidence ] ) and ( [ eq : zdef0 ] ) to see if the divergence for small value of likelihood could be addressed . indeed they can , and the following sections describe two algorithms based on each of these equations . these new algorithms retain the main advantage of the harmonic mean approximation ( hma ) : direct incorporation of the sampled posterior without any assumption of a functional form . in this sense , they are fully and automatically adaptive to any degree multimodality given a sufficiently large sample . we begin in [ sec : intgr ] with a background discussion of lebesgue integration applied to probability distributions and monte carlo ( mc ) estimates . we apply this in [ sec : evid ] to the marginal likelihood computation . this development both illuminates the arbitrariness in the hma from the numerical standpoint and leads to an improved approach outlined in [ sec : algo ] . in short , the proposed approach is motivated by methods of numerical quadrature rather than sample statistics . examples in [ sec : examples ] compare the application of the new algorithms to the hma and the laplace approximation . the overall results are discussed and summarized in [ sec : summary ] . using riemann and lebesgue integration . for riemann integration , we sum thin vertical rectangles of height @xmath34 about the abscissa point @xmath35 for some width @xmath36 . for lebesgue integration , we sum thin horizontal rectangles of width @xmath37 about the ordinate point @xmath38 of height @xmath39 . in both cases , we sum the area under the curve @xmath40 . in the lebesgue case , we must add in the rectangle of width @xmath41 and height @xmath42.,scaledwidth=50.0% ] assume that we have a mc generated sample of random variates with density @xmath40 . recall that any moment or expectation with respect to this density may be computed as a single sum , independent of the parameter - space dimension ! this powerful and seemingly innocuous property follows from the power of lebesgue integration ( e.g. * ? ? ? * ) . to see this , let us begin by considering a one - dimensional integral @xmath43 where @xmath40 is non - negative , finite and bounded , that is : @xmath44 for @xmath45 $ ] . more formally , we may define the lebesgue integral @xmath46 over the measure of sets of points @xmath47 $ ] with measure @xmath48 as follows . we assume that @xmath40 is measurable over @xmath4 and , following @xcite , define the measure function @xmath49 . clearly , @xmath50 is monotonic with @xmath51 and @xmath52 . let @xmath53 be a partition of @xmath54 $ ] with @xmath55 and @xmath56 . in our case , the partition @xmath57 is our mc sample . the choice of @xmath57 induces a partition of @xmath58 through @xmath59 although the sequence in @xmath58 will no longer be monotonic . for convenience , reorder the @xmath58 such that @xmath60 . now consider the lebesgue integral of @xmath40 over @xmath4 , @xmath61 we interpret this geometrically as the area under the curve @xmath62 ; in other words , we have swapped the abscissa and ordinate . to see this , define @xmath63 for the partition @xmath57 . define the interval @xmath64 . clearly @xmath65\sup(\lambda_j ) = \left[m(b ) - m(a)\right]\sup(\lambda_j)\ ] ] and , therefore , @xmath66 since @xmath50 is monotonic and @xmath67 . using this , we may evaluate the integral in equation ( [ eq : int2 ] ) as follows : @xmath68 the sums in equation ( [ eq : lun ] ) have the form of a rectangular quadrature of the measure over the _ range _ of @xmath40 . this geometry is illustrated in figure [ fig : geom ] . although the lebesgue integration theory is general , the equivalence of the one - dimensional integral and equation ( [ eq : lebesgue2 ] ) is easily seen by rewriting equation ( [ eq : int1 ] ) as a two dimensional integral and changing the order of integration using elementary techniques as follows : @xmath69 where @xmath70 and @xmath71 are the minimum and maximum values of @xmath40 in @xmath54 $ ] . an average of the sums in equation ( [ eq : lun ] ) gives us a trapezoidal rule analog : @xmath72(y_{i } - y_{i-1 } ) . \label{eq : tn}\ ] ] further generalization is supported by the lebesgue theory of differentiation . a central result of the theory is that a continuous , monotonic function in some interval is differentiable almost everywhere in the interval @xcite . this applies to the measure function @xmath37 . this result may be intuitively motivated in the context of our marginal likelihood calculation as follows . our measure function describes the amount of density with likelihood smaller than some particular value . a typical likelihood surface for a physical model is smooth , continuous , and typically consists of several discrete modes . consider constructing @xmath37 by increasing the value of @xmath38 beginning at the point of maximum likelihood peak . since @xmath73 , this is equivalent to beginning at @xmath74 and decreasing @xmath30 . recall that @xmath37 decreases from 1 to 0 as @xmath38 increases from @xmath75 to @xmath76 . therefore , we construct @xmath37 from @xmath77 by finding the level set corresponding to some value of @xmath38 and subtracting off the area of the likelihood surface constructed from the perimeter of the set times @xmath78 . the function @xmath37 will decrease smoothly from unity at @xmath75 until @xmath38 reaches the peak of the second mode . at this point , there may be a discontinuity in the derivative of @xmath37 as another piece of perimeter joins the level set , but it will be smooth thereafter until the peak of the third mode is reached , and so on . since we expect the contribution to @xmath17 to be dominated by a few primary modes , this suggests that we can evaluate the integral @xmath46 numerically using the quadrature implied by equation ( [ eq : lebesgue2 ] ) and possibly even higher - order quadrature rules . these arguments further suggest that partitioning @xmath4 into separate domains supporting individual modes would improve the numerics by removing discontinuities in the derivative of @xmath37 and explicitly permitting the use of higher - order quadrature rules . this partition may be difficult to construct automatically , however . to better control the truncation error for this quadrature , we might like to choose a uniform partition in @xmath58 , @xmath79 , to evaluate the sums @xmath80 and @xmath81 . for mc integration , this is not possible . rather , mc selects @xmath57 with irregular spacings and this induces a partition of @xmath58 . motivated by kernel density estimation , we may then approximate @xmath50 by @xmath82 where @xmath83 monotonically increases from 0 to 1 in the vicinity of @xmath84 . for example , we may choose @xmath85 to be the heaviside function @xmath86 which assigns a `` step '' to the upper value of the range in @xmath58 for each @xmath87 . alternatively , we may consider smoothing functions such as @xmath88 \label{eq : nummeas3}\ ] ] where @xmath89 denotes the error function . then , upon substituting equation ( [ eq : nummeas ] ) into equation ( [ eq : lebesgue2 ] ) for @xmath90 , we get : @xmath91 where @xmath92 and @xmath93 by construction and the final equality follows by gathering common factors of @xmath94 in the earlier summation . for integration over probability distributions , we desire @xmath95 distributed according some probability density @xmath96 , @xmath97 = \int_{\mathbb{r } } f(x ) g(x)\,dx , \label{eq : probint}\ ] ] which yields @xmath98 with the normalization @xmath99 , and therefore , @xmath100 $ ] . despite the appearance of the density @xmath101 in equation ( [ eq : probint ] ) , only the value of measure has changed , not the formalism , i.e. the points @xmath87 are now sampled from @xmath96 rather than uniformly . let us now consider the connection between equation ( [ eq : itilden ] ) and a classic mc integration , although we do not need this for later sections of this paper . for a classic mc integration , we choose @xmath102 by construction , and with this substitution @xmath103 becomes the mc approximation : @xmath104 for integration over probability distributions , we assign @xmath105 to its expectation value @xmath106 and the mc integral becomes @xmath107 although intuition suggests that the explicit form from equation ( [ eq : itilden ] ) will yield a better result . the development leading to equation ( [ eq : itilden ] ) remains nearly unchanged for a multidimensional integral : @xmath108 as in the one - dimensional case , the lebesgue integral becomes @xmath109 where the only difference is that @xmath50 is now the measure of the set of points with @xmath110 with @xmath111 . perhaps more remarkable than the similarity of equation ( [ eq : lebesgue2 ] ) with equation ( [ eq : lebesgue3 ] ) is that the numerical lebesgue integral is one - dimensional independent of the dimensionality @xmath23 . however , this does not simplify the computational work ; one still needs to sample a subset of @xmath112 to evaluate equation ( [ eq : lebesgue3 ] ) . the mc version proceeds similarly as well : replace @xmath87 by @xmath113 in equation ( [ eq : mci ] ) . in summary , monte carlo integration is most often posed as the expectation over a distribution , which , more generally , is a lebesgue integral . lebesgue integration and differentiation theory suggests alternative computational approaches akin to traditional riemann - based numerical analysis , if the underlying likelihood function and prior probability density are well - behaved functions . we will see in the next section that a truncation - error criterion applied to the marginal likelihood integral in the form of equation ( [ eq : lebesgue3 ] ) can improve the hma . now , given a mc - computed sample from the posterior distribution @xmath114 with prior distribution @xmath115 and likelihood function @xmath33 , how does one compute the marginal likelihood ? the integral in equation ( [ eq : evidence ] ) states that marginal likelihood is the expectation of the likelihood with respect to the prior distribution . this is the same as equation ( [ eq : lebesgue3 ] ) with @xmath116 replacing @xmath117 , @xmath33 replacing @xmath118 , @xmath115 replacing @xmath119 . alternatively , returning to equation ( [ eq : zdef0 ] ) , the integral @xmath120 is implicitly defined by @xmath121 the value @xmath122 is the probability of @xmath31 over @xmath123 . we will assume that @xmath124 ; this implies that @xmath125 since @xmath126 . in addition , the existence of equation ( [ eq : zdef ] ) implies that @xmath127 almost everywhere in @xmath4 . defining @xmath128 , it follows that the lebesgue integral of the integral on the left - hand - side of equation ( [ eq : zdef ] ) is @xmath129 with measure @xmath130 intuitively , one may interpret this construction as follows : divide up the parameter space @xmath116 into volume elements sufficiently small that @xmath114 is approximately constant . then , sort these volume elements by their value of @xmath131 . the probability element @xmath132 is the prior probability of the volume between @xmath38 and @xmath133 . clearly @xmath134 $ ] and may be trivially computed from a mcmc - generated posterior distribution . using our finite mc - sampled distributed as the posterior probability , @xmath135 , and converting the integral to a sum , we have the following simple estimate for @xmath136 : @xmath137 } \equiv \frac1n \sum_{j=1}^n \mathbf{1}_{\{y_j > y_i\ } } , \qquad m_i^{[u ] } \equiv \frac1n \sum_{j=1}^n \mathbf{1}_{\{y_j\ge y_i\ } } , \qquad m_i \equiv \frac{m_i^{[l ] } + m_i^{[u]}}{2 } , \label{eq : measurex3}\ ] ] where we have defined the left and right end points from equation ( [ eq : nummeas2 ] ) and the mean separately so that @xmath138 } \le m_i \le m_i^{[u]}$ ] . the indicator function @xmath139 enforces the inclusion of a contribution @xmath140 for index @xmath141 only if @xmath142 or @xmath143 for the lower and upper form , respectively . alternatively , these sums may be expressed using equations ( [ eq : nummeas])([eq : itilden ] ) . we may now estimate the marginal likelihood from equation ( [ eq : zdef ] ) using the second part of equation ( [ eq : lebesgue3 ] ) for finite @xmath27 by gathering terms in @xmath144 to get @xmath145 in deriving equation ( [ eq : harmonic0 ] ) , the leading term @xmath146 from equation ( [ eq : lebesgue3 ] ) is absorbed into the sum and @xmath147 . assuming that @xmath148 in equation ( [ eq : zdef ] ) and using equation ( [ eq : harmonic0 ] ) yields @xmath149 this is an alternative derivation for the `` harmonic mean '' approximation ( hma ) to the marginal likelihood . the evaluation of the integral @xmath150 ( eq . [ eq : intk ] ) may fail both due to insufficient sampling and intrinsic divergence . as an example of the former , a sparse sampling may lead to large intervals @xmath151 and inflated truncation error ( eq . [ eq : numerick ] ) . we will consider this further in the next section . as an example of the latter , consider the textbook inference of an unknown mean @xmath152 from a sample of @xmath27 normally distributed points @xmath153 . the likelihood function is @xmath154 where @xmath155 and @xmath156 is the sample mean . let the prior distribution for @xmath152 be @xmath157 . we use an mcmc algorithm to sample the posterior distribution @xmath152 . now , let us evaluate @xmath150 using lebesgue integration for this example . to begin , we need the measure function @xmath158 with @xmath73 we may use equation ( [ eq : likeex ] ) to solve for @xmath159 , noting that the solution has two branches . after some algebra , we have : @xmath160 \label{eq : myg}\ ] ] where @xmath161 @xmath162 the value @xmath163 is the variance weighted mean of the prior mean and the sample mean , and @xmath164 is the harmonic mean of the variance of the prior distribution and the variance of the sample mean . the value @xmath75 is the minimum value for @xmath38 and @xmath165 describes the offset of @xmath152 with increasing @xmath38 . note that @xmath166 . since the values of @xmath38 are obtained by a sampling , the sample will not cover @xmath167 but will be limited from above by the smallest sampled value of the likelihood @xmath168 . we define this limited value of the integral @xmath150 as @xmath169 where the last equality uses @xmath170 . clearly @xmath171 . the magnitude of the truncation due to finite sampling , @xmath172 , depends critically on the width of the likelihood distribution relative to the prior distribution . we describe this ratio of widths by @xmath173 . the convergence condition @xmath174\rightarrow0 $ ] requires that @xmath37 decreases faster than @xmath175 for some @xmath176 . for @xmath177 and large @xmath38 , @xmath178 increases as @xmath179 . for @xmath180 , @xmath178 decreases at least as fast @xmath181 . figure [ fig : gauss_ex1 ] shows @xmath182 as a function of @xmath183 and suggests that @xmath184 is sufficient to obtain convergence for practical values of @xmath27 . qualitatively , a prior distribution that limits @xmath38 from above ( or , equivalently , @xmath30 from below ) will prevent the divergence . the integral @xmath185 is shown as a function of @xmath186 for various values of the ratio @xmath173 . for an uninformative prior distribution @xmath187 and @xmath188 and @xmath189 diverges with increasing @xmath186 . for an informative prior distribution @xmath190 and @xmath189 convergences quickly with @xmath186.,scaledwidth=60.0% ] similar asymptotic expressions for @xmath189 may be derived for multivariate normal distributions . for simplicity , assume that data is distributed identically in each of @xmath23 dimensions and , for ease of evaluation , take @xmath191 . then @xmath192 where @xmath193 is the upper incomplete gamma function and @xmath194 is the complete gamma function . using the standard asymptotic formula for @xmath193 in the limit of large @xmath38 , one finds that @xmath195^{k/2 - 1 } \left(\frac{y}{y_0}\right)^{-1-b } \qquad \mbox{for\ } y\gg k/2 . \label{eq : mulimygasymp}\ ] ] this expression reduces to equation ( [ eq : myg ] ) when @xmath196 , but more importantly , this shows that the tail magnitude of @xmath37 increases with dimension @xmath23 . figure [ fig : gauss_ex2 ] illustrates this for various values of @xmath23 and @xmath183 . the divergence of @xmath150 in the limit @xmath197 shows that normally distributed likelihood function with an uninformative prior is divergent . moreover , figures [ fig : gauss_ex1 ] and [ fig : gauss_ex2 ] further demonstrate that weakly informative prior with very small @xmath183 is likely to be numerically divergent even if @xmath150 formally converges . intuitively , the cause is clear : if the markov chain never samples the wings of the prior distribution that still make significant contribution to @xmath150 , then @xmath150 will increase with sample size . analytically , the failure is caused by the measure decreasing too slowly as @xmath38 increases ( as described at the beginning of [ sec : evid ] ) . empirically , the convergence of the integral @xmath150 may be tested by examining the run of @xmath189 for increasing @xmath38 . overall , [ sec : evid ] highlights the marginal likelihood as a numerical quadrature . we have considered the path of approximations leading from the quadrature to standard expression for the hma . we have also considered the intrinsic requirements on the prior distribution so that the meausure @xmath37 is convergent . this development suggests that there are two sources of error in evaluating equation ( [ eq : zdef ] ) using equations ( [ eq : evidencex ] ) and ( [ eq : numerick ] ) . the first is a truncation error of order @xmath198 . the second is a bias error resulting from specifying the computational grid by a monte carlo procedure . we will consider each in turn . thinking again in terms of numerical quadrature , we can trust the sum in equation ( [ eq : numerick ] ) when adjacent values of @xmath22 are close . the error in @xmath199 will be dominated by the extremely small values of @xmath200 that lead to @xmath201 . specifically , the error for such a term will be @xmath202 , and a small number of such terms can dominate the error for @xmath199 . numerical analysis is based on functional approximation of smooth , continuous , and differentiable functions . this truncation error estimate assumes that @xmath37 is such a function . although not true everywhere , we have argued in [ sec : intgr ] ) that this assumption should be valid over a countable number of intervals in practice . in addition , we expect the sample to be strongly clustered about the value of the posterior mode . by the central limit theorem for a likelihood dominated posterior , the distribution of @xmath152 tends toward a normal distribution . therefore , larger @xmath27 will yield more extremal values of @xmath30 and _ increase _ the divergence of @xmath203 . eventually , for proper prior distributions and sufficiently large @xmath27 , the smallest possible value of @xmath30 will be realized as long as @xmath204 ( see eq . [ eq : zdef ] and following discussion ) . further sampling will reduce the largest intervals , and this will lead to the decrease of large @xmath205 , and finally , convergence . the second source of error is closely related to the first . after eliminating the divergent samples with @xmath206 , the sampled domain @xmath123 will be a subset of the originally defined domain @xmath207 . that is , the mcmc sample will not cover all possible values of the parameter vector @xmath208 . this implies that the numerical quadrature of equation ( [ eq : zdef ] ) will yield @xmath209 . identification of these error sources immediately suggests solutions . note that this observation does not change the problem definition in some new way , but rather , allows us to exploit the mcmc - chosen domain @xmath123 to eliminate the divergence for small @xmath183 described in [ sec : gauss_ex ] . first , we may decrease the truncation error in @xmath199 by ordering the posterior sample by increasing values of @xmath38 and truncating the sequence at the point where @xmath210 for some choice @xmath211 . next , we need a consistent evaluation for @xmath212 . we may use the sampled posterior distribution itself to estimate the sampled volume in @xmath207 . this may be done straightforwardly using a space partitioning structure . a computationally efficient structure is a binary space partition ( bsp ) tree , which divides a region of parameter space into two subregions at each node . the most easily implemented tree of this type for arbitrary dimension is the kd - tree ( short for k - dimensional tree ) . the computational complexity for building the tree from the @xmath27 sampled points in parameter space scales as @xmath213 using the quicksort algorithm at each successive level ( this may be improved , see * ? ? ? each leaf has zero volume . each non - leaf node has the minimum volume enclosing the points in the node by coordinate planes . assigning the volume containing a fixed number of leaves @xmath214 ( e.g. @xmath215 or @xmath216 ) , and some representative value of the prior probability in each node ( such as a @xmath217-quantile or mean value ) , one may immediately sum product of each volume and value to yield an estimate of @xmath122 . for modest values @xmath27 , we will almost certainly find that @xmath209 . since the mcmc chain provides the values of @xmath115 and @xmath218 , we may use the same tree to evaluate both @xmath219 and @xmath212 over the sampled volume @xmath123 . the example in [ sec : gauss_ex ] suggests that evaluation of @xmath150 may stymied by poor convergence unless the prior distribution is restrictive . therefore , if the value of @xmath150 is divergent or very slowly convergent , the evaluation of @xmath17 using @xmath220 will fail whether or not we use the improved truncation criteria . direct evaluation of the @xmath17 is free from this divergence and remains an practical option in this case . the advantage of a direct evaluation is clear : the converged markov chain samples the domain @xmath4 proportional to the integrand of equation ( [ eq : evidence ] ) , and therefore , we expect @xmath221 for large sample size by construction . we propose a hybrid of cubature and monte carlo integration . the bsp tree provides a _ tiling _ of multidimensional volume by using the posterior distribution to define volume elements , @xmath222 . we use a @xmath217-quantile ( such as the @xmath223 median ) or mean value of the posterior probability or the prior probability to assign a probability value to each volume element . an approximation to the integrals @xmath17 and @xmath122 follow from summing the field values over the volume elements , analogous to a multidimensional riemann rule . although @xmath224 for a _ infinite _ posterior sample , there are several sources of error in practice . first , the variance in the tessellated parameter - space volume will increase with increasing volume and decreasing posterior probability . this variance may be estimated by bootstrap . secondly , the truncation error of the cubature increases with the number of points per element . as usual , there is a variance bias trade off choosing the resolution of the tiling : the bias of the probability value estimate increases and the variance decreases as the number of sample points per volume element increases . the prior probability value will be slowing varying over the posterior sample for a typical likelihood - dominated posterior distribution , so the bias will be small . this suggests that larger numbers of points per cell will be better for the evaluation of @xmath122 and a smaller number will be better for @xmath17 . some practical examples suggest that the resulting estimates are not strongly sensitive to the number of points per cell ( @xmath225 or 32 appears to be a good compromise ) . almost certainly , there will be a bias toward larger volume and therefore larger values of @xmath219 and this bias will increase with dimension most likely . to summarize , we have described two approaches for numerically computing @xmath17 from a mcmc posterior simulation . the first evaluates of the integral @xmath150 by numerical lebesgue integration , and the second evaluates @xmath17 directly by a parameter space partition obtained from the sampled posterior distribution . the first is closely related to the hma . it applies ideas of numerical analysis the integral that defines the hma . the second is more closely related to the laplace approximation . in some sense , laplace approximation is an integral of a parametric fit to the posterior distribution . the tree integration described above is , in essence , an integral of a non - parametric fit to the posterior distribution . the advantage of the first method its amenability to analysis . the disadvantage is the limitation on convergence as illustrated in [ sec : gauss_ex ] . the advantage of the second method is its guaranteed convergence . the disadvantage is its clear , intrinsic bias and variance . the variance could be decreased , presumably , using high - dimensional voronoi triangulation but not without dramatic computational cost . the hma is treated as an expectation value in the literature . one of the failures pointed out by @xcite and others is that the hma is particularly awful when the sample is a single datum . in the context of the numerical arguments here , this is no surprise : one can not accurately evaluate a quadrature with a single point ! even for larger samples , the hma is formally unstable in the limit of a thin - tailed likelihood function owing to the divergence of the variance of the hma . @xcite address this failure of the statistic directly , proposing methods for stabilizing the harmonic mean estimator by reducing the parameter space to yield heavier - tailed densities . this is a good alternative to the analysis presented here when such stabilization is feasible . as previously mentioned , @xcite presents conditions on the posterior distribution for the consistency of the hma . intuitively , there is a duality with the current approach . trimming the lebesgue quadrature sum so that the interval @xmath226 is equivalent to lopping off the poorly sampled tail of the posterior distribution . this truncation will be one - sided in the estimate of the marginal likelihood @xmath219 since it removes some of the sample space . however , this may be compensated by an appropriate estimation of @xmath212 . the exposition and development in [ sec : evid ] identifies the culprits in the failure of the hma : ( 1 ) truncation error in the evaluation of the measure @xmath37 ; and ( 2 ) the erroneous assumption that @xmath148 when @xmath207 in practice . we now present two new algorithms , the _ numerical lebesgue algorithm _ ( nla ) and the _ volume tessellation algorithm _ ( vta ) , that implement the strategies described in [ sec : newevid ] to diagnose and mitigate this error . nla computes @xmath199 and vta computes @xmath212 and , optionally , @xmath219 directly from equation ( [ eq : evidence ] ) . in the following sections , we assume that @xmath25 . we begin with a converged mcmc sample from the posterior distribution . after the initial sort in the values of @xmath22 , the nla computes the difference @xmath227 with @xmath228 to find the first value of @xmath229 satisfying @xmath230 . the algorithm then computes the @xmath231 for @xmath232 using equation ( [ eq : measurex3 ] ) . for completeness , we compute the @xmath231 using both the restriction @xmath233 and @xmath234 to obtain lower and upper estimate for @xmath37 . then , these may be combined with @xmath235 and @xmath236 from [ sec : intgr ] to riemann - like upper and lower bounds on @xmath199 . see the listing below for details . excepting the sort , the work required to implement this algorithm is only slightly harder than the hma . the vta uses a kd - tree to partition the @xmath27 samples from posterior distribution into a spatial regions . these tree algorithms split @xmath112 on planes perpendicular to one of the coordinate system axes . the implementation described here uses the median value along one of axes ( a _ balanced _ kd - tree ) . this differs from general bsp trees , in which arbitrary splitting planes can be used . there are , no doubt , better choices for space partitioning such as voronoi tessellation as previously discussed , but the kd - tree is fast , easy to implement , and published libraries for arbitrary dimensionality are available . traditionally , every node of a kd - tree , from the root to the leaves , stores a point . in the implementation used here , the points are stored in leaf nodes only , although each splitting plane still goes through one of the points . this choice facilitates the computation of the volume spanned by the points for each node as follows . let @xmath237 be the number of parameter - space points @xmath238 } , n=1,\ldots,{\bar m}_j$ ] in the @xmath239 node . let @xmath240}$ ] denote the field quantities at each point @xmath238}$ ] . some relevant field quantities include the values of the unnormalized posterior probability and the prior probability . the volume for node @xmath141 is @xmath241}_i,\ldots,\theta^{[{{\bar m}}_j]}_i ) - \min(\theta^{[1]}_i,\ldots,\theta^{[{{\bar m}}_j]}_i ) \right ] . \label{eq : volj}\ ] ] the set of nodes with @xmath242 for some fixed integer @xmath243 , determines an exclusive volume partition of the parameter space spanned by the point set , the _ frontier_. the value of @xmath243 is chosen large enough to limit the sampling bias of field quantities in the volume but small enough resolve the posterior modes of interest . the values @xmath244 $ ] seem to be good choices for many applications . each node in the frontier is assigned a representative value @xmath245 . i use @xmath217-quantiles with @xmath246 for tests here . the resulting estimate of the integrals @xmath212 and/or @xmath219 follow from summing the product of the frontier volumes with their values @xmath245 . both the nla and the vta begin with a sort of the likelihood sequence @xmath30 and this scales as @xmath247 . in the nla , the computation of the @xmath248 followed by the computation of @xmath219 is @xmath249 . the sequence @xmath250 is useful also for diagnosis as we will explore in the next section . however , in many cases , we do not need the individual @xmath231 but only need the differential value @xmath251 to compute @xmath219 , which contains a single term . the values of likelihood may range over many orders of magnitude . owing to the finite mantissa , the differential value be necessary to achieve adequate precision for large @xmath27 , and the nla may be modified accordingly . the algorithm computes the lower , upper , and trapezoid - rule sums ( eqns . [ eq : lun][eq : tn ] ) for the final integral @xmath219 . for large posterior samples , e.g. @xmath252 , the differences between @xmath253 and @xmath254 are small . indeed , a more useful error estimate may be obtained by a random partitioning and subsampling of the original sequence @xmath255 to estimate the distribution of @xmath219 ( see the examples in [ sec : examples ] ) . in practice , computing the marginal likelihood from a posterior sample with @xmath256 takes 0.2 cpu seconds on a single 2ghz opteron processor . although nla could be easily parallelized over @xmath257 processors to reduce the total runtime by @xmath258 this seems unnecessary . the kd - tree construction in vta scales as @xmath259 followed by a tree walk to sum over differential node volumes to obtain the final integral estimates that scales as @xmath260 . this scaling was confirmed empirically using the multidimensional example described in [ sec : highd ] with dimension @xmath261 $ ] and sample size @xmath262 $ ] . computing the marginal likelihood from a posterior sample with @xmath256 and @xmath263 takes 4.4 cpu seconds on a single 2ghz opteron processor , and , therefore , the computation is unlikely to be an analysis bottleneck , even when resampling to produce a variance estimate . the leading coefficient appears to vary quite weakly the distribution , although there may be undiscovered pathological cases that significantly degrade performance . the required value of @xmath27 increases with parameter dimension @xmath23 ; @xmath256 is barely sufficient for @xmath264 in tests below . subsampling recommends the use of even larger chains to mitigate dependence of the samples . therefore , the first practical hardware limitation is likely to be sufficient ram to keep the data in core . likelihood values @xmath265 from the simulated posterior distribution sort the sequence so that @xmath266 @xmath267 @xmath268 @xmath269 @xmath270 and/or @xmath271 save the values @xmath231 @xmath272 @xmath273 @xmath274 save the estimated marginal likelihood , @xmath219 and algorithmic error likelihood values @xmath265 from the simulated posterior distribution change variables @xmath275 sort the sequence so that @xmath276 create an empty point set @xmath277 @xmath278 @xmath279 add @xmath280 } , { \bf f}^{[n]})$ ] to @xmath277 @xmath281 find the set of frontier nodes @xmath282 with the desired number of points @xmath214 @xmath283 compute the median value of @xmath240}$ ] among the @xmath214 points @xmath284 save the estimated value of the integrals @xmath285 buildkd(@xmath277 ) a set posterior points and values @xmath280 } , { \bf f}^{[n]})\in{\cal p}$ ] compute and store the range for each coordinate and the volume for this node locate the coordinate dimension @xmath286 with maximum variance determine the median value of @xmath287}$ ] ( e.g. by quicksort ) split @xmath277 into two subsets by the hyperplane defined by @xmath288 : @xmath289 . to estimate the marginal likelihood using the methods of the previous section , we may use either the nla to estimate @xmath150 and the vta to estimate @xmath122 or use the vta alone to estimate @xmath17 . examples below explore the performance of these strategies . the mcmc posterior simulations are all computed using the umass bayesian inference engine ( bie , * ? ? ? * ) , a general - purpose parallel software platform for bayesian computation . all examples except that in [ sec : highd ] simulate the posterior distribution using the parallel tempering scheme @xcite with @xmath292 and 20 temperature levels . convergence was assessed using the subsampling algorithm described in @xcite , a generalization of the @xcite test . for a simple initial example , let us compute the marginal likelihood for a data sample @xmath14 of 100 points @xmath293 modelled by @xmath294 with prior distribution for @xmath295 . the marginal likelihood @xmath17 can be computed analytically from @xmath14 for this simple example . the final 200,000 converged states of the mcmc - generated chain were retained . application of the nla for @xmath199 and the vta for @xmath212 gives a value of @xmath296 ( 95% confidence interval ) , close to but systematically smaller than the analytic result : @xmath297 . a value of @xmath298 seems appropriate from numerical considerations , although experiments suggest that the algorithm is not sensitive to this choice as long as @xmath299 is not so small to decimate the sample or so large that error - prone outliers are included . it is prudent to check a range of @xmath299 to determine the appropriate value of each problem . the vta yields @xmath300 , consistent with the analytic result . the bias in the first estimate appears to be caused by an overestimate of @xmath301 produced by the vta . this might be improved by a space partition whose cells have smaller surface to volumes ratios ( [ sec : highd ] for a graphical example ) . the bias is much less pronounced in the direct estimate of @xmath219 by the vta owing to smallness of the posterior probability in the extrema of the sample . these extrema result in anomalously small value of @xmath302 for the hma . figure [ fig : mk ] illustrates the details of the nla applied to this computation . panel ( a ) plots @xmath303 from equation ( [ eq : measurex3 ] ) . the run of @xmath303 with @xmath38 rises rapidly near the posterior mode and drops rapidly to zero for small likelihood values . the inset in this figure shows @xmath303 in linear scale . the measure function @xmath303 , and hence the integral @xmath304 , is dominated by large values of @xmath30 as long as @xmath305 decreases sufficiently fast ( see [ sec : gauss_ex ] ) . panel ( b ) plots the accumulating sum defining the quadrature of @xmath219 in equations ( [ eq : lun])([eq : tn ] ) , beginning with the largest values of likelihood first . the contribution to @xmath219 is dominated at the likelihood peak , corresponding to the steeply rising region of @xmath303 in panel ( a ) . in other words , the samples with small values of @xmath30 that degrade the hma make a negligible contribution to the marginal likelihood computation as long as @xmath306 . in addition , nla provides upper and lower bounds , and thereby some warning when the estimate is poorly conditioned , e.g. owing to an inappropriate choice for @xmath299 . the plot in figure [ fig : mk]b will readily reveal such failures . a more realistic error assessment can be obtained by subsampling the sequence @xmath255 . the cpu time for these algorithms is sufficiently small that this procedure should be practical in general . consider the following experiment : ( 1 ) the posterior is simulated by mcmc ( as described above ) to obtain a chain of 400,000 states ; ( 2 ) the first half of the chain is discarded ; ( 3 ) the second - half is randomly subsampled with replacement to obtain 128 samples of 10,000 states ; ( 4 ) the marginal likelihood for each is computed using the nla , vta , the laplace approximation and the hma ( approximately 2 cpu minute in total ) . for all but the hma , increasing the number of states decreases the variance for each distribution ; samples with 10,000 states best revealed the differences between the algorithms with a single scale . figure [ fig : subsample ] illustrates the relative performance with different prior distributions . figure [ fig : subsample]a is the model described at the beginning of this section ; the range of the prior distribution is much larger than the values sampled from the posterior distribution . the prior distributions for each successive panel have smaller ranges as indicated . the colors are composited over @xmath307 yields the new value @xmath308 . ] with @xmath309 ( e.g. hma over vta is brown , hma over nla is purple , laplace over hma is blue - grey , laplace over vta is blue - green ) . in panel ( d ) , the range is within the range of values sampled by the posterior in panel ( a ) . the overall trends are as follows : 1 ) the hma has unacceptably large variance unless the domain of the prior roughly coincides with the domain sampled by the mcmc algorithm ; 2 ) the vta and laplace approximation have the smallest variances , followed by hma ; 3 ) the nla is consistently biased below the analytic value ; and 4 ) the vta and laplace approximation are closed to the expected analytic value . indeed , the laplace approximation is an ideal match to and should do well for this simple unimodal model . in the final panel , there are no outlier values of @xmath30 and the harmonic mean approximation is comparable to the others . these tests also demonstrate that the same outliers that wreck the hma have much less affect on nla and vta . further experimentation reveals that the results are very insensitive to the threshold value @xmath299 . in fact , one needs an absurdly large value of @xmath299 , @xmath310 , to produce failure . here , we test these algorithms on the radiata pine compressive strength data analyzed by han and carlin ( 2001 ) and a number of previous authors . we use the data tabled by han and carlin from williams ( 1959 ) . these data describe the maximum compressive strength parallel to the grain @xmath311 , the density @xmath312 , and the resin - adjusted density @xmath313 for @xmath314 specimens . carlin and chib ( 1995 ) use these data to compare the two linear regression models : @xmath315 with @xmath316 , and @xmath317 . we follow han and carlin ( 2001 ) and carlin and chib ( 1995 ) , adopting @xmath318 priors on @xmath319 and @xmath320 , and @xmath321^{-1}\right)$ ] priors on @xmath322 and @xmath323 , where @xmath324 is the inverse gamma distribution with density function @xmath325 where @xmath326 and @xmath327 . han and carlin point out these priors are approximately centered on the least - squares solution but are otherwise rather vague . using direct integration , green and ohagan ( 1998 ) find a bayes factor of about 4862 in favor of model 2 . .marginal likelihood for non - nested linear regression models [ cols="<,<,<,<,<",options="header " , ] for each model of dimension @xmath23 , we compute a markov chain using the differential evolution algorithm ( de , * ? ? ? this algorithm evolves an ensemble of chains with initial conditions sampled from the prior distribution . a proposal is computing by randomly selecting pairs of states from the ensemble and using a multiple of their difference ; this automatically ` tunes ' the proposal width . we have further augmented this algorithm by including a tempered simulation step @xcite after every 20 de steps ( see * ? ? ? * for more details ) . each row describes of table [ tab : highd ] describes the application of the nla , vta , and laplace approximation to a model of dimension @xmath23 . the mcmc simulations produce approximately 1.4 million converged states . convergence is testing using the gelman - rubin statistic ( op . each converged chain is resampled with replacement to provide 1024 subsamples of @xmath257 states . the value @xmath328 $ ] is chosen to achieve 95% confidence intervals approximately 1% of @xmath219 or smaller . the 95% confidence intervals on @xmath219 are indicated as sub- and super - scripts . recall that the standard vta determines volume spanned @xmath214 samples and approximates the integral by multiplying the volume by the median value of the sample . to assess the variance inherent in this choice , i quote the results for two other p - quantiles , @xmath329 and @xmath330 . finally , for each algorithm , the table presents the relative error : @xmath331 . both the nla and vta results are very encouraging : the relative error is within a few percent for @xmath332 . for @xmath264 , i computed @xmath219 with samples sizes of 400,000 states . both the nla and vta tend to slightly overestimate @xmath17 for large @xmath23 . the laplace approximation results are disappointing for small @xmath23 and improve for large @xmath23 , but still are less precise than either the nla or vta . . , scaledwidth=70.0% ] figure [ fig : k2d ] illustrates the kd - tree construction for a single @xmath333 sample . each two - dimensional cell is colored by the median value of the posterior probability for the @xmath334 points in each cell and scaled to the peak value of posterior probability @xmath335 for the entire sample . a careful by - eye examination of the cell shape reveals a preponderance of large axis - ratio rectangles ; this is a well - known artifact of the kd - tree algorithm . for large values of @xmath335 , the volume elements are small , and with a sufficiently large sample , the gradient in @xmath335 across the volume are small . for small values of @xmath335 , the volume elements are large , the gradients are large , and the large - axis ratio rectangles distort the reconstructed shape of the true posterior . however , as described in [ sec : newevid ] , the values of @xmath224 for an infinite sample , so a small number of distorted rectangles will not compromise the end result . moreover , the values of @xmath336 at large volumes are smaller than those at small volume for these tests , and this further decreases the importance of the kd - tree cell - shape artifact . but for a sample from the cauchy distribution from fig . . the box shows the quantiles and median , the whisker shows the ( 10% , 90% ) intervals , followed by outlying points . the three distributions are ( 1 ) the hma ; ( 2 ) nla for @xmath199 and vta for @xmath212 ; and ( 3 ) vta for both @xmath337 and @xmath219.,scaledwidth=50.0% ] as an example of model selection , we first compute the marginal likelihood for the same data @xmath338 as in the first example of [ sec : test ] but assuming a cauchy - lorentz distribution , @xmath339^{-1},\ ] ] as the model with unknown location @xmath340 and scale @xmath183 parameters . for prior distributions , we take @xmath341 and @xmath342 where @xmath343 is the weibull distribution with scale parameter @xmath344 and shape parameter @xmath23 . nla yields @xmath345 , vta yields @xmath346 and the hma yields @xmath347 . the data and fits are shown in figure [ fig : cauchy]a . there should be no surprise that the true model ( with @xmath348)is strongly preferred . let us now repeat the experiment using 100 data points selected from the cauchy - lorentz distribution ( @xmath349 ) and compare the marginal likelihood values for a cauchy - lorentz distribution and a mixture of two normal distributions ( @xmath350 ) . nla and vta , respectively , yield @xmath351 and @xmath352 . the hma yields @xmath353 and @xmath354 . regardless of the algorithm performing the test , the bayes factor reveals strong evidence in favor of the true model . note from figure [ fig : cauchy]b that both models are reasonable fits `` by eye '' . however , the bayes factor overwhelmingly _ prefers _ the simpler ( in this case , true ) model . as expected , the distribution of @xmath219 for the heavy - tailed cauchy distribution is much better behaved ( see fig . [ fig : subsamplec ] ) . the results for nla and vta are consistent and the hma is systematically larger , but non enough to misguide a decision . in summary , much of the general measure - theoretic underpinning of probability and statistics naturally leads naturally to the evaluation of expectation values . for example , the harmonic mean approximation ( hma , * ? ? ? * ) for the marginal likelihood has large variance and is slow to converge ( e.g. * ? ? ? on the other hand , the use of analytic density functions for the likelihood and prior permits us to take advantage of less general but possibly more powerful computational techniques . in [ sec : intro][sec : algo ] we diagnose the numerical origin of the insufficiencies of the hma using lebesgue integrals . there are two culprits : 1 ) the integral on the left - hand side of equation ( [ eq : zdef0 ] ) may diverge if the measure function @xmath355 from equation ( [ eq : measurex ] ) decreases too slowly ; and 2 ) truncation error may dominate the quadrature of the left - hand side of equation ( [ eq : zdef0 ] ) unless the sample is appropriately truncated . using numerical quadrature for the marginal likelihood integral ( eqns . [ eq : evidence ] and [ eq : measurex ] ) leads to improved algorithms : the _ numerical lebesgue algorithm _ ( nla ) and the _ volume tessellation algorithm _ ( vta ) . our proposed algorithms are a bit more difficult to implement and have higher computational complexity than the simple hma , but the overall cpu time is rather modest compared to the computational investment required to produce the mcmc - sampled posterior distribution itself . for a sample of size @xmath27 , the sorting required by nla and vta has computational complexity of @xmath247 and @xmath356 , respectively , rather than @xmath249 for the harmonic mean . nonetheless , the computational time is a fraction of second to minutes for typical values of @xmath357 ( see [ sec : algo ] ) . the geometric picture behind nla is exactly that for lebesgue integration . consider integrating a function over a two - dimensional domain . in standard riemann quadrature , one chops the domain into rectangles and adds up their area . the sum can be refined by subdividing the rectangles ; in the limit of infinitesimal area , the resulting sum is the desired integral . in the lebesgue approach , one draws horizontal slices through the surface and adds up the area of the horizontal rectangles formed from the width of the slice and the vertical distance between slices . the sum can be refined by making the slices thinner when needed ; in the limit of slices of infinitesimal height , the resulting sum is the desired integral . in the riemann case , we multiply the box area in the domain , @xmath358 , by the function height , @xmath359 . in the lebesgue , we multiply the slice height in the range , @xmath360 , by the domain area , @xmath361 ( see fig . [ fig : geom ] ) . both algorithms easily generalize to higher dimension . for the lebesgue integral , the slices become level sets on the hypersurface implied by the integrand . therefore the lebesgue approach always looks one - dimensional in the level - set value ; the dimensionality @xmath23 is ` hidden ' in the area of domain ( _ hypervolume _ @xmath361 for @xmath362 ) computed by the measure function @xmath37 . the level - set value for the nla is @xmath363 . once determined , nla applies the trapezoidal rule to the sum over slices and compute the upper and lower rectangle sums as bounds . clearly , the error control on this algorithm might be improved by using more of the information about the run of @xmath361 with @xmath359 . having realized that the practical failure of the harmonic mean approximation is a consequence of the sparsely sampled parameter - space domain , nla addresses the problem by determining a well - sampled subset @xmath207 from the mcmc sample , ex post facto . restricted to this subset , @xmath123 , the value of the integral @xmath122 on the right - hand side of equation ( [ eq : zdef0 ] ) is less than unity . we determine @xmath123 by a binary space partitioning ( bsp ) tree and compute @xmath122 from this partition . a bsp tree recursively partitions a the k - dimensional parameter space into convex subspaces . the vta is implemented with a kd - tree @xcite for simplicity . in addition , one may use vta by itself to compute equation ( [ eq : evidence ] ) directly . judged by bias and variance , the test examples do not suggest a strong preference for either the nla or the vta . however , both are clearly better than the hma or the laplace approximation . conversely , because these algorithms exploit the additional structure implied by smooth , well - behaved likelihood and prior distribution functions , the algorithms developed here will be inaccurately and possibly fail miserably for _ wild _ density functions . the nla and the vta are not completely independent since the nla uses the tessellation from the vta to estimate the integral @xmath122 . however , the value of the integral @xmath150 tends to dominate @xmath17 , that is @xmath364 , and the contributions are easily checked . based on current results , i tentatively recommend relying preferentially on vta for the following reasons : 1 ) there is no intrinsic divergence ; 2 ) it appears to do as well as vta even in a high - dimensional space ; and 3 ) there is no truncation threshold @xmath299 . figure [ fig : k2d ] illustrates the potential for volume artifacts that could lead to both bias and variance . this error source affects both the vta and nla ( through the computation of @xmath212 ) but the affect on the nla may be larger ( [ sec : test ] ) . additional real - world testing , especially on high - dimensional multimodal posteriors , will provide more insight . in test problems described in this paper , i explored the effects of varying the threshold @xmath299 and the kd - tree bucket size @xmath214 . these parameters interact the sample distribution , and therefore , are likely to vary for each problem . i also recommend implementing both the nla , vta , htm , laplace approximation and comparing the four for each problem . we are currently testing these algorithms for astronomical inference problems too complex for a simple example ; the results will be reported in future papers . an implementation of these algorithms will be provided in the next release of the umass bayesian inference engine ( bie , * ? ? ? there are several natural algorithmic extensions and improvements not explored here . [ sec : intgr ] describes a smoothed approximation to the computation of @xmath37 ( eqns . [ eq : nummeas][eq : itilden ] ) rather than the step function used in [ sec : algo ] . the direct integration of equation ( [ eq : evidence ] ) currently ignores the location of field values in each cell volume . at the expense of cpu time , the accuracy might be improved by fitting the sampled points with low - order multinomials and using the fits to derive a cubature algorithm for each cell . in addition , a more sophisticated tree structure may decrease the potential for bias by providing a tessellation with `` rounder '' cells . in conclusion , the marginal likelihood @xmath365 may be reliably computed from a monte carlo posterior sample though careful attention to the numerics . we have demonstrated that the error in the hma is due to samples with very low likelihood values but significant prior probability . it follows that their posterior probability also very low , and these states tend to be outliers . on the other hand , the converged posterior sample is a good representation of the posterior probability density by construction . the proposed algorithms define the subdomain @xmath207 dominated by and well - sampled by the posterior distribution and perform the integrals in equation ( [ eq : zdef0 ] ) over @xmath123 rather than @xmath4 . although more testing is needed , these new algorithms promise more reliable estimates for @xmath17 from an mcmc simulated posterior distribution with more general models than previous algorithms can deliver . i thank michael lavine for thoughtful discussion and both neal katz and michael lavine for comments on the original manuscript . it is also a pleasure to acknowledge the thoughtful and helpful comments of two anonymous referees and the associate editor of the journal . this work was supported in part by nsf iis program through award 0611948 and by nasa aisr program through award nng06gf25 g .

we examine the notion and properties of the non - hermitian effective hamiltonian of an unstable system using as an example potential resonance scattering with a fixed angular momentum . we present a consistent self - adjoint formulation of the problem of scattering on a finite - range potential , which is based on separation of the configuration space into two segments , internal and external . the scattering amplitude is expressed in terms of the resolvent of a non - hermitian operator . the explicit form of this operator depends on both the radius of separation and the boundary conditions at this place , which can be chosen in many different ways . we discuss this freedom and show explicitly that the physical scattering amplitude is , nevertheless , unique although not all choices are equally adequate from the physical point of view . the _ energy - dependent _ operator should not be confused with the non - hermitian effective hamiltonian which is usually exploited to describe interference of overlapping resonances . we note that the simple breit - wigner approximation is as a rule valid for any individual resonance in the case of few - channel scattering on a flat billiard - like cavity , leaving no room for non - trivial to appear . the physics is appreciably richer in the case of an open chain of connected similar cavities whose spectrum has a band structure . for a fixed band of overlapping resonances , the smooth energy dependence of can be ignored so that the constant submatrix approximately describes the time evolution of the chain in the energy domain of the band and the complex eigenvalues of define the energies and widths of the resonances . we apply the developed formalism to the problem of a chain of barriers whose solution is also found independently in a closed form . we construct for the two commonly considered types of the boundary conditions ( neumann and dirichlet ) for the internal motion . although the final results are in perfect coincidence , somewhat different physical patterns arise of the trend of the system with growing openness . formation in the outer well of a short - lived doorway state shifted in the energy is explicitly demonstrated together with the appearance of long - lived states trapped in the inner part of the chain .

cond-mat0206176

to the best of our knowledge , the concept of the non - hermitian effective hamiltonian appeared first in feshbach s papers @xcite in connection with the general theory of resonance nuclear reactions @xcite and , independently , in livic s study of open systems @xcite . a typical atomic nucleus forms near an excitation energy @xmath6 above the threshold of nucleon emission a rich set of long - lived compound states with a very dense energy spectrum . for relatively small excitation energies , these states manifest themselves as narrow isolated resonances in collisions of nucleons and target nuclei . the sharp energy dependence of the corresponding cross sections near a given resonance is described by the universal breit - wigner formula . any smoother variations can be ignored in the domain of an isolated resonance . at higher @xmath6 , the resonance states begin to overlap and strongly interfere . nevertheless , one can still neglect the smooth energy dependence within a group of close interfering resonances . in this approximation , the propagation of the unstable system created on the intermediate stage of the collision is characterized by the resolvent of an energy - independent non - hermitian matrix @xcite . the latter describes in the time picture the irreversible evolution of the excited intermediate state and can therefore be interpreted as the effective hamiltonian @xmath1 whose anti - hermitian part , originating from on - shell self - energy contributions , is responsible for decays into open channels . the complex eigenvalues of the effective hamiltonian determine the energies and widths of the resonances . the method of the effective hamiltonian proved to be a success for describing many important resonance phenomena , especially in the context of chaotic scattering @xcite . it must be stressed that , similar to the breit - wigner formula , the notion of the effective hamiltonian is _ local _ in energy although the general approaches @xcite work in much wider energy intervals where smooth variations become important and should already be taken into account . for the very concept of the effective hamiltonian to be consistent , the scales of resonant and smooth variations must be appreciably different . otherwise the resonance behavior of scattering amplitudes will be distorted or even completely destroyed . the smooth dependence influences the background phases as well as the parameters of the resonances situated in different energy domains . such effects can not be described by simply enlarging the dimension of the matrix of the effective hamiltonian . instead , a large energy - dependent matrix @xmath7 emerges whose simple interpretation as a time - shift operator is no longer valid . there is , generally , no one - to - one correspondence between the complex energies of resonance states on the one hand and the @xmath6-dependent eigenvalues of this large matrix on the other . the current intensive studies , theoretical as well as experimental , of chaotic scattering of a particle by an open two - dimensional cavity ( see , e.g. , @xcite and references therein ) renewed interest in the hamiltonian approach to the resonance scattering theory and , in particular , to the concept and properties of the effective non - hermitian hamiltonian @xcite . although the scattering processes considered are purely potential , the theory is formulated in close analogy with the formalism developed in @xcite for nuclear collisions . the configuration space is divided into two parts , internal and external . the feshbach projection technique is employed to express the scattering matrix in the specific form that explicitly reveals the sharp resonant energy dependence while all smoother variations show up only indirectly via changes of the matrix elements . related is the problem of electromagnetic field quantization in open optical cavities @xcite . while the probability amplitudes of the physical processes are fixed unambiguously , the projection procedure exploited is not unique . there exists a rather wide freedom in choosing the surface of separation as well as the boundary conditions ( bcs ) on this surface . therefore , the amplitudes of interest are expressed in terms of quantities that depend on the details of the formalism . the independence of the final results of the calculations is not , as a rule , directly seen . therefore , a certain caution is necessary to avoid incorrect assertions . an unexpected dependence on the ( auxiliary ) boundary condition at the cavity - lead interface was found , for example , in @xcite . moreover , poor agreement with numerically calculated exact poles of the @xmath8 matrix was revealed for poles distant from the real axis . these points , partly attributed by the authors to numerical limitations , require further clarification . additional physical arguments should sometimes be involved to reasonably restrict the freedom . it is important to recognize that the density of levels in billiard - like cavities is much lower than that of the many - body nuclear systems . actually , strong overlap and interference of different resonance states are not , as a rule , possible in the single two - dimensional cavities ordinarily considered @xcite . the simple breit - wigner approximation is usually sufficient in this case for any individual resonance ( see , e.g. , @xcite ) . noticeable interference can occur only in the rare event of accidental near degeneracy of resonances . experimental observation of such an interference of few resonances in an open microwave cavity has recently been reported in @xcite . stronger overlap and a non trivial effective hamiltonian matrix can , however , appear when open chains of a number of similar potential wells connected to each other are considered . the energy spectrum has a band structure in this case with much denser spectrum within a given band . a schematic model of such a kind was considered in @xcite . a similar example was also investigated in @xcite in the framework of graph theory @xcite . some general aspects of scattering in periodic structures have also been considered in @xcite . in this paper we examine the notion of the effective hamiltonian in potential scattering . the questions we are concerned with do not depend on the regular or irregular character of the motion . thus , we restrict ourselves to the simplest case of the single - channel @xmath9-wave scattering . the extension to higher partial waves is straightforward . a consistent self - adjoint formulation of the problem of scattering on a finite - size potential in terms of the internal and external subproblems is presented in sec . ii . the ( one - dimensional ) @xmath8 matrix is expressed in terms of a non - hermitian energy - dependent operator @xmath10 whose form , together with the form of the associated @xmath11 function ( sec . iii ) , _ depends _ on the bc and the radius @xmath12 of separation in the configuration space . different bcs define different representations of the internal and external parts of the ( unique ) scattering wave function . the role of the fictitious direct reflection at the separation point @xmath12 is also discussed . in sec . iv the exactly solvable open kronig - penney model is considered as an example of potential scattering with interfering resonances . closed expressions are found for both scattering @xmath13 and staying - wave @xmath14 functions as well as wigner s function @xmath15 . their analytical properties in the complex @xmath16 and energy planes are analyzed in detail . hereafter we compare these exact results with those obtained in the framework of the projection formalism of the previous sections . we analyze two typical choices of the bcs for the internal motion : the cases of neumann and dirichlet bcs . in the second case , the internal problem corresponds to a closed counterpart of the system under consideration that allows one to follow the changes of the motion due to the interference of resonances when the openness grows . the latter is in line with numerous applications considered in the literature . non - hermitian effective hamiltonians for a fixed band of resonance states are built up in both cases . we summarize our main findings in the concluding sec . the radial motion in the @xmath9-wave scattering is described by the schrdinger equation ( we use the units @xmath17 throughout the paper ) @xmath18 with the boundary condition @xmath19 at the origin . we use below the specific method which , basically , goes back to bloch s paper @xcite ( see also @xcite ) . however , we present a derivation which leads to interrelated boundary problems for internal and external regions . making use of heaviside s step function one can decompose the wave function into internal and external parts as follows @xmath20 where the functions @xmath21 and @xmath22 are supposed to be continuous at the point @xmath12 together with their two first derivatives . the partition radius @xmath12 can be chosen arbitrarily ; we suggest only that @xmath23 . the second derivative of the function ( [ psi1 ] ) reads @xmath24\ , \delta(r - a ) - [ u(a)-\phi(a)]\ , \delta'(r - a)\,.\end{aligned}\ ] ] when substituting this expression into eq . ( [ schr ] ) , there exists a freedom in attributing the local terms in ( [ psi ] ) to the internal or external regions . one possible choice is to define the singular functions as @xmath25 and @xmath26 , where @xmath27 , thus considering the first as belonging to the internal subspace and the second to the external one , respectively . with such definitions , we can represent ( [ schr ] ) in the form @xmath28 with the following entries : [ n ] @xmath29 where @xmath30 stands for the kinetic energy operator . in particular , in the coordinate representation the matrix elements of the entries take the form [ cn ] @xmath31\delta(r - r ' ) \nonumber\\ & & -\delta(r - a_{-})\,\delta'(r'-a_{-})\ , , \\ \langle r| { { h_{\textrm{ext } } } ^{\textsc{(d ) } } } |r'\rangle & = & \textstyle -\frac{d^2}{dr^2}\,\delta(r - r ' ) \nonumber\\ & & -\delta'(r - a_{+})\,\delta(r'-a_{+ } ) \ , , \\ \langle r|v|r'\rangle & = & \langle r'|{v}^{\dagger}|r\rangle = \delta(r - a_{-})\,\delta'(r'-a_{+})\ , . \ ] ] the presence of the singular terms assures hermiticity of the operators @xmath32 and @xmath33 in contrast to @xmath30 alone @xcite . the relations @xmath34 , etc . , can be easily checked by means of partial integrations . the singular operators on the boundary provide the boundary conditions , the first one neumann for the internal region , the second one dirichlet for the external region . equation ( [ psi ] ) implies that both bcs are interrelated . this also leads to the adjointness of the coupling operators @xmath35 and @xmath36 . the dirichlet boundary operator is essentially different from that of neumann type and can not be produced by the boundary operators as used in @xcite . the range of operators in eq . ( [ cn ] ) within the hilbert space is defined by the functions on which the singular terms vanish . this requirement fixes the bcs at the point of separation @xmath12 . the full hilbert space is a direct sum of the space spanned by the eigenvectors of the internal problem @xmath37 with neumann bc , @xmath38 and that spanned by the eigenvectors of the external problem @xmath39 with dirichlet bc , @xmath40 the type of bc is explicitly indicated by the corresponding superscripts in eqs . ( [ htot])([cn ] ) . as usual , we assume normalization conditions [ norm ] @xmath41 in the discrete and continuous spectra , respectively . the external function @xmath42\propto e^{-ikr}-e^{-2ika}e^{ikr}\ ] ] describes a wave fully reflected at the point @xmath12 , the corresponding @xmath8 function being equal to @xmath43 . one can also proceed in the opposite way and ascribe the terms with @xmath44 and with @xmath4 to the internal and external regions , respectively . in this case , the entries in eq . ( [ htot ] ) read [ d ] @xmath45 that corresponds to the nonperturbed internal problem with dirichlet bc @xmath46 , when the external one has neumann bc @xmath47 . the latter results in an additional shift by @xmath48 of the reflection phase at the separation point @xmath12 : @xmath49 . the interplay of the internal and external motions due to the off - diagonal elements @xmath35 , ( [ n]c ) , distorts the outer waves @xmath22 . from the upper row in eq . ( [ htot ] ) we find for all @xmath50 , @xmath51=1,2 , ... @xmath52 here @xmath53 is the resolvent operator for the internal problem . then , the lower row transforms into @xmath54 or , in the position representation , @xmath55= 0\ , , \nonumber \\ \textstyle ( k^2 + \frac{d^2}{dr^2})\phi(r ) - \delta(r - a_{+})[\phi'(a ) - { { { g_{\textrm{int } } } ^{\textsc{(d ) } } } } ' ' ( a , a)\phi(a)]= 0 \nonumber\end{aligned}\ ] ] for the two cases considered . this yields change of the bcs of the external part of the exact wave function to [ bcext ] @xmath56 where the shorthand @xmath57 has been used . let us note that eq . ( [ bcextn ] ) is the conventional boundary condition in the wigner - eisenbud @xmath11-matrix theory of resonance nuclear reactions @xcite . in the external region @xmath39 , where the potential vanishes identically , the wave function has the form @xmath60 $ ] . therefore , the conditions ( [ bcext ] ) obtained above allow us to express the function @xmath61 in terms of the internal green s functions , [ sr ] @xmath62 both expressions are exact and equivalent to each other . the only difference is that they are written in different complete bases in the full hilbert space . the merits as well as limitations of each of these representations will be discussed below . we define further the phase shifts @xmath63 due to the influence of the internal region and the functions @xmath64 by [ r ] @xmath65 the spectral representations of the green s function are used here in the last steps . this shows that the energy levels of the corresponding internal problems are real poles of the @xmath11 functions in the complex energy plane . the formulas obtained are in close analogy with the representations found in @xcite in a different way . they are quite similar to those appearing in the @xmath11-matrix theory @xcite . we should , however , stress that functions @xmath11 defined in our manner differ from the standard wigner s @xmath11 function by trivial factors like @xmath66 or @xmath67 which we for the sake of convenience include in the definition . a certain caution is needed while taking the limit @xmath68 in eq . ( [ rd ] ) since the derivative of the green s function @xmath69 has a discontinuity when @xmath70 . the symbols @xmath71 and @xmath72 stand for solutions of the internal problem with dirichlet bcs only at the points @xmath73 or @xmath74 , respectively . it is readily seen that the mixed second partial derivative contains a singular contribution @xmath75\delta ( r - r ' ) = w\delta(r - r')\,,\ ] ] @xmath76 being the wronskian . this singularity at @xmath77 must be excluded and the second derivative must be understood as @xmath78 it immediately follows from this remark that the spectral sum @xmath79 diverges when @xmath70 . indeed , convergence of this sum depends on the contributions of the very high levels , @xmath80 . for such an excitation , we can neglect the potential @xmath81 whereupon the solution of the internal problem reduces simply to @xmath82 . the contribution @xmath83 to be dropped appears due to the factor @xmath84 which arises in the numerator after the double differentiation has been done . it is easy to see , that just the contribution of this kind mainly caused numerical problems in @xcite . for @xmath85 , the sum converges , although slowly , to the finite limit ( [ lim ] ) because of oscillations @xcite . it would be a mistake to rely upon the spectral representations in ( [ r ] ) , and interpret the eigenvalues @xmath86 as energies of resonance states and the phases @xmath87 or @xmath88 as smooth phases of the background scattering . indeed , the levels @xmath86 depend on what sort of boundary conditions have been used . more than that , many other forms of bc are equally permissible ( see , for example , the fundamental review @xcite ) . it can be shown that the most general possible bc , @xmath89 involve two arbitrary parameters @xmath90 ( here we include formally @xmath91 for the dirichlet bcs ) . returning to eq . ( [ sr ] ) , we see that , due to the relation @xmath92 , the levels @xmath93 are _ zeros _ of the function @xmath94 and , similarly , the levels @xmath95 are zeros of @xmath96 . in fact , the two phases are connected as @xmath97\,.\ ] ] the shift @xmath48 just compensates the similar shift of @xmath98 and the total scattering phase @xmath99 does not depend on the type of bc used . in particular , the positions of the poles of the function @xmath13 in the complex @xmath16 plane , which are found from the equation @xmath100 are bc independent because of the relation ( [ tancot ] ) . this is in agreement with the fact that both factors @xmath101 are entire functions in @xmath16 plane . the factorized form of the residues of poles in ( [ r ] ) allows one to represent the function @xmath13 in a different but fully equivalent form as @xmath102 where the non - hermitian symmetric operator @xmath0 is defined by @xmath103 here @xmath104 is a diagonal matrix of the eigenvalues of the corresponding internal problems and the column vector @xmath105 of coupling amplitudes , which originate from the off - diagonal element @xmath35 in eq . ( [ htot ] ) , has the components @xmath106 in the cases of neumann or dirichlet bc , respectively . a proof of equivalence to the expressions ( [ sr ] ) immediately comes from the following relation between resolvents ( see , for example , @xcite ) : @xmath107 it follows from eq . ( [ st ] ) that the poles of the @xmath8 function can also be found from the secular equation @xmath108=0\ ] ] where @xmath109 is a point in the complex @xmath16 plane . in the position representation this is equivalent to the spectral problem with a complex ( outgoing - wave ) boundary condition [ kp ] @xmath110 this clearly demonstrates again the independence of the choice of bcs for disconnected internal and external motions . actually , the form of the operator ( [ calh ] ) is provocative . one is tempted to interpret this operator as an effective hamiltonian whose _ @xmath16-dependent _ eigenvalues define complex energies of metastable resonance states formed by the potential @xmath81 . however , this interpretation is , generally , wrong . to make our points clearer , let us start with the following simple remark . let the potential vanish identically everywhere , @xmath111 , so that the @xmath8 function @xmath112 . nevertheless , neither the operator @xmath0 nor the function @xmath113 are trivial in themselves . for example , in the case of neumann bcs matrix elements of @xmath0 are @xmath114\,.\ ] ] any truncated finite - size @xmath115 matrix obtained from eq . ( [ calhn0 ] ) gives , when substituted in eq . ( [ seq ] ) , @xmath116 pairs of complex roots . however , the complex poles of the resolvent @xmath117^{-1}$ ] of the truncated matrix have nothing to do with the poles of the genuine @xmath8 function . the truncation procedure is not stable when @xmath118 . the difficulties become even worse in the case of dirichlet bcs when the imaginary part also grows , @xmath119\,.\ ] ] in fact , all poles found go to infinity in the limit @xmath118 . indeed , the true poles , because of the identity @xmath120=\det(z^2- { h_{\textrm{int } } } ) [ 1+\frac{i}{2}r(z)]$ ] , must satisfy the equation @xmath121 which is equivalent to ( [ s - pls ] ) . for example , in the case of neumann bc this equation runs as @xmath122 . in particular , when the potential @xmath81 vanishes identically , the green s function in the @xmath109 plane is equal to @xmath123 , eq . ( [ seceq ] ) looks as @xmath124 and , therefore , has no roots in any finite domain of this plane . this implies , in turn , that eq . ( [ kp ] ) has no non - trivial solutions . in fact , the function ( [ tils ] ) is simply equal to @xmath125 in this case and compensates exactly the phase shift due to the fictitious reflection at the separation point @xmath12 . the remark above is of quite a general nature . the radius of separation can be chosen arbitrarily . for the sake of simplicity , we suggest only that @xmath12 is larger than the finite radius of the potential . the actual wave function satisfies at this point the conditions @xmath126 of course , none of these quantities is known before the problem has been solved . any boundary condition used above generates a complete basis in the hilbert space , in which the actual wave functions can be expanded . all such bases are formally equivalent . but this does not mean that all of them are equally adequate from the physical point of view . in particular , the more the basis eigenvectors and their derivatives differ at the point @xmath12 from the real values ( [ actbc ] ) the more slowly the corresponding expansion converges near this point . a forced bc creates a false reflection at this point , which is described by the factor @xmath127 , whose phase should be fully compensated by the similar part of the total phase of the function @xmath128 . both factors separately depend on @xmath12 although the complete function @xmath13 does not . the main role in this compensation is played by the matrix elements @xmath129 with large @xmath130 when the influence of the potential @xmath131 becomes negligible , and we return to the situation described in the previous paragraph . however , the choice of the separation radius and bc influences the positions of the poles ( as well as the residues ) of the function @xmath11 and , consequently , the explicit form of the matrix @xmath132 . this influence is stronger the less adequate the choice made of bc and @xmath12 . in general , the parameters of the @xmath11 function can carry rather poor information about the actual complex poles . for this reason , diagonalization of a truncated matrix @xmath132 ( which is necessary in any numerical computation ) can lead in the case of a poor choice to strong disagreement with the characteristics of the actual poles of the function @xmath8 . the explicit dependence of the matrix elements @xmath132 on the wave number @xmath16 causes additional problems ( see next section ) . to extract the physically relevant effective hamiltonian from the formal operator @xmath0 , additional physical considerations must be engaged . for example , one may expect from the physical point of view that the most relevant choice of the separation point @xmath12 would be a distance matching an outer potential barrier which is strong enough to make immediate reflection at this point quite probable . to explore in more detail the questions briefly discussed above , we will apply in the next section the formal technique sketched here to the problem of scattering by a finite periodic set of @xmath4 barriers , which can be solved exactly . we will consider below @xmath9-wave scattering by a periodically disposed chain of @xmath2 @xmath4 barriers , @xmath133 to ensure formation of long - lived resonance states , at least some of the strength constants @xmath134 should be positive . the distance @xmath135 is measured in units of the size of the well formed by two neighboring barriers . below , we derive the effective hamiltonian starting directly from the schrdinger equation . because of the local character of the barriers , it is most convenient to start from , instead of the schrdinger equation , the equivalent integral equation @xmath136 here the symbol @xmath137 with @xmath138 stands for the green s function of the free radial motion which has outgoing - wave asymptotic . from the second line in ( [ inteq ] ) we find immediately @xmath139 where the scattering amplitude is given by @xmath140 in the second equality we have used matrix notation , with @xmath9 and @xmath141 being @xmath2-dimensional vectors with the components @xmath142 and @xmath143 ( @xmath144=1,2, ... ,@xmath2 ) , respectively , when @xmath145 . according to eq . ( [ inteq ] ) , the @xmath2-dimensional vector @xmath141 satisfies the equation @xmath146 , where @xmath147 and @xmath148 is a symmetric non - hermitian matrix with the matrix elements @xmath149 thus , we obtain finally [ ext ] @xmath150 where we have used the factorized form @xmath151 of the anti - hermitian part of the matrix @xmath148 to pass from the first to the second equality . the @xmath3 matrix propagators @xmath152 are connected with one another by a relation similar to ( [ gt - gr ] ) . the diagonal matrix @xmath153 characterizes the penetrabilities of the barriers @xcite . the poles of the @xmath8 function are defined by the equation @xmath154=0=1+\frac{i}{2}\,k(z)\,.\ ] ] all the matrix elements ( [ eqchi ] ) are entire functions in the complex @xmath109 plane . the same is valid for the determinant in ( [ sroots ] ) . therefore , one can show that this equation has an infinite number of isolated complex roots . further , its roots because of the relation @xmath155^*=-g^0_{ll'}(-z^*)$ ] come in pairs @xmath156 and @xmath157 , @xmath51=1,2 , ... , symmetrically with respect to the imaginary axis , or lie on the latter , @xmath158 . all poles of the first type are situated in the lower part of the complex plane . poles on the positive half of the imaginary axis , @xmath159 , correspond to bound states and can appear only if some number of the constants @xmath134 are negative . those which are situated on the negative part , @xmath160 , correspond to the so - called virtual levels . the total number of purely imaginary poles is finite for any finite @xmath2 . after all poles have been found , the function @xmath13 can be presented in the form @xmath161 where we took into account that @xmath162 , since @xmath4 barriers become transparent for a particle with asymptotically large energy . these expressions are in agreement with the general theory @xcite , which is valid for any potential with a finite radius . each factor in ( [ facts ] ) is singly unitary . in particular , for a pair @xmath163 of conjugate roots we have @xmath164 since @xmath165 , the phase @xmath166 increases when the energy @xmath167 grows and passes the value @xmath48 at the point @xmath168 . the typical energy interval @xmath169 of the main gain of the phase is estimated by the quantity @xmath170 . if this interval is small enough and the latter quantity does not vary appreciably within it , the total gain is close to @xmath171 and the factor @xmath172 receives the standard breit - wigner resonance form @xmath173 with the energy @xmath174 and width @xmath175 of the resonance defined as follows @xmath176 the @xmath16-dependence is neglected in the definition of the width and substitution @xmath177 has been made . such a substitution is well justified when the scattering energy is large enough , but becomes improper near @xmath178 . due to the `` threshold '' @xmath179 dependence of the widths , some specific behavior takes place when in the proximity of this point there exists a number of bound and decaying states @xcite . we conclude that the scattering amplitude @xmath180 $ ] is a meromorphic function in the @xmath109 plane and can therefore also be written down as an infinite sum of the pole contributions @xmath181 where the residues @xmath182 can easily be found from eq . ( [ facts ] ) . another representation of the scattering amplitude sometimes considered in the literature ensues from diagonalization of the ( finite in our case ) matrix @xmath183 at a given fixed real @xmath16 . since this matrix is complex symmetric , such a diagonalization is performed by a complex orthogonal transformation defined together with the _ complex _ eigenvalue matrix @xmath184 from the equation @xmath185\psi(k)=\psi(k)\lambda(k)\,.\ ] ] the scattering amplitude reduces then to a finite sum @xmath186 ^ 2 } { \lambda_l(k)}\,,\ ] ] where @xmath187 is the @xmath144th eigenvector . however , such a representation is of limited practical use inasmuch as the terms of the sum are , generally , extremely complicated . there is no simple interpretation of a given term and all of them can give comparable contributions . the compatibility of the two representations ( [ prept ] ) and ( [ fprept ] ) is very indirect . a particular term of the second one can not be uniquely continued from the real axis at an arbitrary point @xmath109 in the complex plane because both @xmath188 and @xmath189 are multivalued functions in this plane . the same is , of course , also true for the opposite direction . therefore , even if a root @xmath190 of the equation @xmath191 is found and near this point the pole approximation @xmath192 ^ 2/ \lambda'_l(z^l_j)}{z - z^l_j}\ ] ] is valid , this expression can not , generally speaking , be continued on the real axis by simply substituting @xmath193 , since the power expansion @xmath194 has a finite radius of convergence . further , for any fixed @xmath144 a set of roots exists depending on the branch of the function @xmath189 considered . there is no one - to - one correspondence between the set of roots @xmath190 and the manifold of actual poles @xmath195 that appears in the expansion ( [ prept ] ) . many of the roots @xmath190 are false and their contributions must finally cancel out . therefore , diagonalization of the propagator with non - trivial energy dependence is not as a rule useful ; rather it can lead to misleading conclusions . we compare below the closed expressions eq . ( [ ext ] ) found with the formalism described in the previous section . in correspondence with our remark at the end of sec . iii , we fix the separation point @xmath12 by superposing it on the position of the outer barrier , @xmath196 . the appearance of an additional @xmath4 function changes the corresponding boundary conditions . in particular , if the @xmath4-terms are ascribed to the internal region , neumann bc ( [ nint ] ) for the internal problem is replaced by @xmath197 whereas attributing such terms to the external domain yields the following change of the bc of the external problem : @xmath198 we stress that neither of these conditions coincides with the boundary conditions @xmath199 \chi'(a_{+})-\chi'(a_{-})=\kappa_{\textsc{l}}\chi(a)\ , \end{array}\ ] ] satisfied by the exact wave function @xmath200 . in the case ( [ cnext ] ) , the immediate reflection at the point @xmath12 is described instead of @xmath201 by @xmath202 ( recall that @xmath203 . ) the additional factor is due to the influence of the barrier outside the radius of separation @xmath12 . this factor has a pole @xmath204 on the negative part of the imaginary axis , which formally corresponds to a virtual wigner level . actually , such a pole does not exist in the exact solution and disappears due to cancellation with the contribution of the internal region [ see eq . ( [ dlams ] ) below ] . the exact @xmath8 function in the case of the bc ( [ cnint ] ) reads as @xmath205 , where @xmath206 with @xmath207 and @xmath208 the coupling amplitudes are equal to @xmath209 finally , the matrix elements of the operator @xmath210 [ see eq . ( [ calh ] ) ] appear as @xmath211 with the levels @xmath95 being the eigenvalues of the internal problem with the bc ( [ cnint ] ) . for the sake of simplicity , we use the same superscript n as before . in spite of the seemingly similar general structure of the expressions ( [ ext ] ) , ( [ props ] ) on one hand side and ( [ lams])([lamt ] ) on the other , they are , in essence , quite different . the most important distinction shows itself in the dimension of the vectors and matrices which is _ finite _ and coincides with the number of barriers in the first case and _ infinite _ in the second . in addition , the factor @xmath127 of the immediate reflection does not appear explicitly in eq . ( [ ext ] ) . we will analyze below a couple of simple special cases before drawing general conclusions . in this case we find immediately from eqs . ( [ ext ] ) ( @xmath212 and we drop the subscript @xmath2 in the strength of the barrier ) [ 1s ] @xmath213 whereas [ 1r ] @xmath214 in the eqs . ( [ 1r ] ) the phase @xmath215 of the immediate reflection at the point @xmath12 is extracted . the two functions @xmath14 and @xmath15 are related to each other by @xmath216 the positions of the poles of the function @xmath8 coincide with the complex roots of the equation @xmath217 in the right half of the @xmath109 plane they can be searched for in the form @xmath218 , where any @xmath219 is restricted to the strip @xmath220 and satisfies the equation @xmath221=0\,.\ ] ] there also exists the trivial root @xmath222 but this root fully cancels out . even for not very large @xmath51 one can omit @xmath219 in the logarithmic term . this yields the approximate solution @xmath223 which is valid with good accuracy for almost all poles , being asymptotically exact when @xmath80 . however , by no means all of the complex roots @xmath156 correspond to resonances . for a pole to correspond to a long - lived resonance state , the following two additional conditions also must be satisfied : i ) @xmath224 should be a real pole of the function @xmath14 or at least should be close to such a pole ; ii ) the corresponding residue should be small enough for the scattering phase @xmath225 to increase near this point by @xmath171 in an interval @xmath226 . the validity of these conditions depends on the value of the parameter @xmath227 . indeed , the poles of @xmath14 are found from the equation @xmath228 there exists only a _ finite number _ of real roots @xmath229 , @xmath230=1,2, ... ,@xmath231 of this equation . they satisfy the requirement @xmath232 or , equivalently , @xmath233 . ( the trivial solution @xmath16=0 should be dismissed . ) all other roots lie in the complex plane and the phase does not reach its maximal value although can change rather fast if a pole of @xmath234 is still close to the real axis . it is clear that they move away from this axis when @xmath235 grows . in particular , no real solutions exist if @xmath236 @xmath237 . the scattering phase @xmath238 in this case and smoothly depends on @xmath16 . the barrier is too weak to form a long - lived resonance state . when @xmath239 the scattering phase can be calculated as @xmath240 in the framework of perturbation theory . there are two kinds of real roots of eq . ( [ polr1 ] ) when @xmath227 is appreciably less than @xmath241 . the most interesting case @xmath242 , although @xmath243 , can easily be considered analytically . the first set consists of the roots @xmath244 which are close to the points @xmath245 : @xmath246 , ( @xmath247 ) . near such a pole the @xmath234 function manifests typical resonance behavior @xmath248 strictly speaking , a similar contribution of the symmetric root @xmath249 , which corresponds to the same energy @xmath250 , should be added , so we arrive near the @xmath51th resonance at @xmath251 each neighboring pair of resonance roots is separated by a root of the second set : @xmath252(m+\frac{1}{2})\pi$ ] , in a vicinity of which @xmath253 the residues are large in this case and such terms contribute into the smooth background part of the total scattering phase . indeed , the ( dimensionless ) time delay @xmath254=@xmath255=@xmath256 , which measures the wigner time delay @xmath257=@xmath258 in units of the ( negative ) time delay @xmath259=@xmath260 because of the immediate reflection at the point @xmath12(=1 ) , is large near the points @xmath244 but small when @xmath261 . poles of the two different kinds alternate and resonant and smooth contributions are mixed . figure 1 clearly demonstrates all the features described above . [ fig1 ] ( 8,12.5 ) 7.5cm(0,-0.2 ) let us on the other hand consider the properties of the function @xmath15 [ see eq . ( [ 1r ] ) ] . its poles are the roots of the equation @xmath262 instead of eq . ( [ polr1 ] ) . contrary to the latter , _ all _ roots @xmath263 of ( [ poltilr1 ] ) are real . those of them , that satisfy the inequality @xmath264 and , therefore , have numbers @xmath265 , are equal to @xmath266 again and correspond to resonances . the difference from the similar roots of eq . ( [ polr1 ] ) appears only at higher order corrections in the parameter @xmath227 . for larger @xmath51 , when @xmath267 , an infinite number of roots @xmath268 $ ] exists , giving a smooth contribution , which combines with the phase @xmath269 and almost compensates it at large @xmath16 [ see fig . indeed , the @xmath4 barrier becomes almost transparent when the collision energy is large and the total scattering phase can be calculated by applying perturbation theory . the meromorphic character of the function @xmath15 allows us to represent this function in the form of the pole expansion @xmath270 the symmetry connection @xmath271 has been taken into account in the second equality . returning to the poles of @xmath13 , we can expand the expression ( [ appzeta ] ) in the resonant region @xmath272 with respect to this parameter to calculate the poles @xmath273 and the complex energies of the resonance states @xmath274\,.\ ] ] note that the widths @xmath275 of the resonances appear only in the second order in the penetrability parameter @xmath227 when the shifts of their energies are of the first order of magnitude . the resonances are well isolated since the ratio @xmath276 . finally , the remote poles with @xmath277 are given by @xmath278 now we compare our findings with the results ( [ lams])([lamt ] ) of the general formalism . the normalized solutions of the internal problem with bc ( [ cnint ] ) are readily found to be @xmath279 where @xmath244 are the roots of eq . ( [ poltilr1 ] ) which follows directly from the bc ( [ cnint ] ) this time ( we omit the tilde and note that only positive roots @xmath244 , @xmath51=1,2 , ... , are to be kept ) . this fact is quite satisfactory and demonstrates the physical relevance of the choice of bc made . indeed , for such a bc @xmath280 in accordance with the exact boundary condition ( [ exactbc ] ) . the dirichlet bc @xmath281 would be deficient in this sense . with eq . ( [ cnsol ] ) taken into account , we find from eq . ( [ lamr ] ) @xmath282 equivalence to eq . ( [ polexp ] ) is seen from the relation @xmath283^{-1}$ ] , which follows directly from the secular equation ( [ poltilr1 ] ) . [ fig2 ] ( 8,5 ) 7.5 cm ( 0.35,-0.2 ) matrix elements of the operator @xmath0 ( [ lamcalh ] ) are in our case @xmath284 in the resonance domain , @xmath285 , the diagonal matrix element @xmath286 approximates well the complex energy @xmath287 of the resonance which lies near the scattering energy @xmath288 . in this region off - diagonal elements influence only corrections of higher orders . at the same time , the diagonal approximation becomes insufficient for the remote poles , breaking down to reproduce exact asymptotic @xmath289 behavior of the imaginary parts . figure 2 illustrates the consideration . we therefore conclude that the extraction of the phase @xmath290 of the immediate reflection and utilization of the @xmath59function is adequate only in the resonance region where such a reflection is probable . beyond this region the internal and immediate reflection phase shifts almost compensate each other so that the function @xmath14 proves to be a more relevant tool . below , we restrict our attention to the resonance region well above the energy @xmath291 . to simplify the investigation further , we suggest assuming the strength constants to be the same @xmath292 , @xmath293 , for all inner barriers . in accordance with the results of the simple consideration presented at the end of sec . iv c , segregation of the smooth phase of the immediate reflection meets expectations based on physical intuition . to perform this segregation , we write first the scattering amplitude @xmath294 as a block product @xmath295 \mathcal{g}_{\textsc{l}},\end{aligned}\ ] ] where the symbols @xmath296 and @xmath297 , etc . , stand for @xmath298 submatrices and @xmath299-dimensional vectors , respectively , and @xmath300 is the similar upper left block of @xmath148 . to pass from the first to the second line , the relations @xmath301 and @xmath302 have been used which follow , together with the expression @xmath303^{-1}\,e^{-ika}\,,\ ] ] from the equation @xmath304 . we notice now that the function @xmath305 coincides with the amplitude of scattering on the potential ( [ u ] ) with the last barrier being removed : @xmath306 or , in other words , @xmath307 . finally , we obtain after simple transformations @xmath308 , where @xmath309 the poles of the function @xmath310 [ as well as of @xmath13 ] in the complex @xmath16 plane are given by zeros of the denominator . it is convenient to introduce in parallel with this function a sequence of functions @xmath312 which describe the scattering on chains of @xmath144 barriers with the penetrability constant @xmath313 and the bc fixed at the radius @xmath314 . in particular , @xmath315 , @xmath316 , and @xmath317 . these functions are related to each other by a recursion @xmath318 using as before the ansatz @xmath319 in the resonance region @xmath320 , with @xmath321 being a polynomial of the @xmath2th power with respect to the argument @xmath322 . this equation gives a bunch of @xmath2 close complex poles of the @xmath8 function well separated from all the other poles . the corresponding @xmath11 function is readily found from eq . ( [ explams ] ) to be @xmath323 in the limit @xmath324 ( closed interior ) we have @xmath325 and @xmath326 , so that only the immediate reflection at the point @xmath12 survives . obviously , the spectrum of the poles of the function @xmath15 is determined by the equation @xmath327 which should be compared with eq.!([poltilr1 ] ) . it is easy to check that the spectrum exactly coincides with that of the wave numbers of the internal problem for the potential @xmath328 with the bc ( [ cnint ] ) . this potential is perfectly transparent on the separation radius @xmath196 where the latter condition is fixed . this directly follows from the expression @xmath329\ ] ] for the wave function in the region @xmath330 . now we will use the formulaes just found to analyze as an illustrative example resonances in the double - well @xmath4-potential , @xmath331 . the equation for the poles of the @xmath8 function ( [ explams ] ) looks in this case like @xmath332 [ cf . ( [ deq ] ) ] . again , the ansatz @xmath319 with @xmath333 is valid . supposing also the integer number @xmath334 to be appreciably large , we arrive for each fixed @xmath334 at a couple of closed solutions [ compare with ( [ appzeta ] ) ] @xmath335\right\ } \,.\end{aligned}\ ] ] it is worth noting that within the approximation adopted the sum of the imaginary parts @xmath336 , where @xmath337 stand for the arguments of the logarithm in eq . ( [ appzeta2 ] ) , _ does not depend _ on the penetrability constant @xmath313 of the interior barrier . indeed , simple transformations show that @xmath338(1 - 2i\lambda_2 m\pi)$ ] . we fix now the number @xmath339 within the resonance domain , @xmath340 , and consider the doublet of poles close to this point . keeping the terms of the two first orders of magnitude we find @xmath341 in the resonance region , the splitting @xmath342 within a given doublet is much smaller than the distance between adjacent doublets which is @xmath343 . equation ( [ explamr ] ) reads in the case @xmath331 @xmath344 for the @xmath334th resonance doublet @xmath345 , and the small shift @xmath346 satisfies the quadratic equation @xmath347 giving immediately @xmath348 , with @xmath349 from ( [ doublet ] ) . this convinces us that each resonance doublet with a good accuracy can be considered independently of the other poles . near a given @xmath350 within the resonance region the function @xmath351 is decomposed into a sum of two partial fractions @xmath352 the residues being found to be equal to @xmath353 . to pass from the @xmath16 to the energy plane , the term @xmath354 must be added , which yields @xmath355 the two - dimensional vector @xmath356=@xmath357 and matrix @xmath358=@xmath359 of the internal levels @xmath360=@xmath361 of the doublet have been introduced on the last step . the corresponding representation of the scattering amplitude near the scattering energy @xmath362 is as follows : @xmath363 in these formulas @xmath364 , i.e. , we have neglected the change of the scattering energy @xmath6 within the doublet considered . the difference between @xmath365 contributes only at higher orders . this is contrary to the case of widths from different doublets when @xmath366 differ from each other already in the zero order in @xmath367 . in such an approximation , both functions @xmath368 and @xmath369 become meromorphic in the complex energy plane . by the very construction , the complex eigenvalues of the two - dimensional _ energy - independent _ symmetric matrix @xmath370 coincide with the complex energies of the two resonances of the doublet considered . henceforth , we drop the index @xmath334 of the doublet to avoid too bulky notation . the energies @xmath371 are the levels of the internal problem for the potential @xmath372 with the bc ( [ cnint ] ) whereas the amplitudes are given by eq . ( [ lama ] ) , with the wave number @xmath16 substituted by @xmath373 . just the _ energy - independent _ matrix ( [ effh2 ] ) is naturally interpreted as an effective non - hermitian hamiltonian . the notion of the effective hamiltonian is valid , however , only within a fixed doublet . similarly , in the case @xmath374 the `` effective hamiltonian '' coincides with the corresponding diagonal matrix element of the formal operator @xmath0 constructed in sec . ( 8,9.5 ) 7.5 cm ( 0.35,-0.2 ) as we have already mentioned above , the total width @xmath375 of the doublet remains constant as long as the openness of the system is fixed . at the same time , the individual widths depend also on the ratio @xmath376 . a _ remarkable redistribution _ of the total width between members of the doublet takes place when this parameter changes . when the system is almost closed , @xmath377 , @xmath378 e_{-}\approx ( 1-\lambda_2)(m\pi)^2 \ , , \end{array } \quad \gamma_{\pm}\approx 2\lambda^2_2 ( m\pi)^3\,.\ ] ] both levels have similar widths in this case . however , in the opposite limit @xmath379 the state that exists in the outer well appropriates almost the whole total width @xmath380 simultaneously , the energy @xmath381 of the broader resonance gets strongly displaced due to the coupling to the energy continuum . the other resonance turns out to be trapped in the inner well . figure 3 illustrates the behavior just described . for an _ arbitrary _ ratio @xmath376 , the complex energies of the two resonances of a doublet are separated by a distance which is large on the scale of the total width @xcite . for this reason , the off - diagonal matrix elements of the anti - hermitian part of the effective hamiltonian ( [ effh2 ] ) give only corrections of higher order and can therefore be neglected . in spite of the nontrivial behavior of the complex levels just described the resonances may still be considered separately of each other within the given accuracy . extension to the general case of an arbitrary number @xmath2 of the barriers is now straightforward . the effective hamiltonian appears as a @xmath3 block of the operator @xmath0 , which describes a fixed bunch of @xmath2 close resonance states . it is important that energy dependence can be fully neglected within such a bunch . the internal problem with the bc ( [ cnint ] ) does not correspond to a finite motion of a quantum particle even when the coupling matrix elements are neglected . indeed , the internal solution is sensitive to the penetrability of the outer barrier . meanwhile , in numerous applications the intuitively most attractive convention , which goes back to the textbook @xcite , is adopted , using as an internal basis the states of a closed counterpart of the system under consideration . such a counterpart can hardly be defined uniquely but , in our case , it is natural to fix it by choosing dirichlet bc at the point @xmath74 . this corresponds to the internal problem with an infinitely hard wall put at the point @xmath74 . the formalism of sec . ii gives then [ dlams ] @xmath382 } { 1+\frac{i}{\lambda_{\textsc{l}}k } [ 1+\lambda_{\textsc{l } } { { { g_{\textrm{int } } } ^{\textsc{(d ) } } } } ' ' ( a , a ) ] } s_0^{\textsc{(n ) } } \\ \label{dlams0 } & = & \frac{1+\frac{\lambda_{\textsc{l}}}{1+i\lambda_{\textsc{l } } k } { { { g_{\textrm{int } } } ^{\textsc{(d ) } } } } ' ' ( a , a)}{1+\frac{\lambda_{\textsc{l } } } { 1-i\lambda_{\textsc{l } } k } { { { g_{\textrm{int } } } ^{\textsc{(d ) } } } } ' ' ( a , a)}\,\frac{1+i\lambda_{\textsc{l } } k}{1-i\lambda_{\textsc{l } } k}\,s_0^{\textsc{(d)}}.\mbox{\quad}\end{aligned}\ ] ] comparison with the closed solution ( [ lams ] ) and ( [ explams ] ) shows that @xmath383 which implies that the spectrum of the internal problem is given by the equation @xmath384 instead of eq . ( [ tilrspec ] ) . unlike the eigenvalues @xmath385 , the levels @xmath386 _ do not depend _ on @xmath387 . similar to eqs . ( [ lams])([lamt ] ) , we have @xmath388 , with @xmath389 from eq . ( [ vlev ] ) , and @xmath390^{-1}{\bar a}$ ] , where @xmath391 and @xmath392 this should be compared with eq . ( [ lama ] ) and ( [ lamcalh ] ) . not only widths but also hermitian shifts result from the coupling to the external region this time . indeed , in the resonance region @xmath393 the phases that come from the first two factors in eq . ( [ dlams0 ] ) are equal approximately to @xmath394 the two contributions perfectly compensate each other at the points @xmath395 . within a bunch of resonance levels the eigenvalues of the hermitian part of @xmath396 coincide in the main approximation with the corresponding levels @xmath95 . in general , the connection @xmath397 holds for arbitrary @xmath16 . let us first return to the case @xmath331 , for a moment . the amplitudes ( [ bara ] ) for the @xmath334th doublet are easily calculated to be in the main approximation @xmath398 , when the corresponding internal levels are @xmath399 and @xmath400 . therefore , the diagonal elements of the effective hamiltonian of the doublet are equal to @xmath401 \left(1-\lambda_2\right)\,(m\pi)^2-i\lambda_2 ^ 2(m\pi)^3\ , , \end{array}\right.\ ] ] which exactly coincide with eq . ( [ 2<1 ] ) . off - diagonal elements of @xmath396 can be neglected if @xmath377 . these elements become important , however , when the coupling to the continuum becomes strong , @xmath402 . the two resonances ( [ cale0d2 ] ) interfere in this case . diagonalization of the @xmath403-matrix @xmath396 gives now @xmath404\,(m\pi)^2 \nonumber\\ & & -i\lambda_2 ^ 2\left(1\pm\frac{\lambda_2 } { \sqrt{4\lambda_1 ^ 2+\lambda_2 ^ 2}}\right)(m\pi)^3.\end{aligned}\ ] ] in the limit @xmath379 this reduces to ( [ 2>1 ] ) . for an arbitrary number @xmath2 of barriers , the @xmath334th bunch of @xmath2 close resonances is described by the @xmath3-submatrix @xmath396 of the infinite matrix ( [ lamcalhd ] ) , with the wave number @xmath16 being substituted by @xmath405 . the absolute values of the coupling amplitudes within such a resonance bunch are estimated as @xmath406 . the total _ collective _ width @xmath407 of the bunch , which is determined by the trace of the anti - hermitian part of @xmath396 , characterizes the openness of the system and does not depend on the penetrability constant @xmath313 of the internal barriers . the same is valid in the main approximation regarding the collective real energy displacement @xmath408 . when this displacement is small in comparison with the total energy spread of the internal levels , @xmath409 , the bunch consists of @xmath2 independent narrow resonances @xmath410 note that the hermitian shift does not influence the level spacings but changes only the position of the bunch . under the opposite condition , @xmath411 , the doorway basis in which the interaction matrix @xmath412 is diagonal becomes more adequate @xcite than the basis of the internal problem used up to now ( see also @xcite ) . the latter matrix is diagonalized by an orthogonal transformation @xmath413 , where each entry @xmath414 is a real vector in the @xmath2-dimensional part of the total hilbert space , which corresponds to the considered bunch . due to the factorized structure , the interaction matrix possesses the only nonzero eigenvalue @xmath415 , which belongs to the eigenvector @xmath416 . as a result , the effective hamiltonian obtains in the doorway basis the form ( we drop below the index @xmath334 of the bunch ) @xmath417 the following notation has been used here ( @xmath418=2,3, ... ,@xmath2 ) : @xmath419 h_{\mu}=\sum_l\varepsilon_l^{\textsc{(d)}}\,a_l\,\eta_l^{(\mu)}\ , , \end{array}\ ] ] with the @xmath420 vectors @xmath421 and elements @xmath422 of the diagonal matrix @xmath423 being defined by the eigenvalue problem @xmath424 in the @xmath299-dimensional subspace orthogonal to the vector @xmath425 . the quantity @xmath426 is the weighted mean position of the internal levels . the non - hermitian hamiltonian ( [ drh ] ) describes a wide _ doorway _ resonance with width @xmath427 displaced from the bunch by the distance @xmath428 and coupled to a background of @xmath420 stable states by means of the matrix elements @xmath429 . only due to this interaction do such states get access to the continuum via the doorway state existing in the outer well . with the help of the completeness condition @xmath430 one finds from eq . ( [ subeta ] ) @xmath431 the orthogonality condition @xmath432 immediately gives the equation @xmath433 for the new positions of the stable levels . this equation shows that each new level @xmath422 lies between two neighboring old ones @xmath434 and therefore is shifted with respect to the latter only by a distance comparable with the initial mean level spacing . this is much smaller than the displacement @xmath428 of the collective level @xmath435 . the interaction @xmath436 mixes these states and forms @xmath2 final resonance states . the complex energies of the exact states , i.e. the eigenvalues of the effective hamiltonian ( [ drh ] ) , satisfy the secular equation @xmath437 in particular , for the collective doorway state one obtains from this equation @xmath438 \delta\epsilon_c-\frac{i}{2}\ , \left[1- \frac{{\bf h}^2}{(\delta\epsilon_c)^2}\right]\gamma_c\,.\ ] ] equations ( [ collev ] ) and ( [ comp ] ) allow us to express the square length @xmath439 of the mixing vector in terms of the variance of the non - perturbed internal levels , @xmath440 ^ 2\,.\end{aligned}\ ] ] thus , interaction is weak in the doorway basis under the condition @xmath441 and the doorway state keeps almost the whole energy displacement and width . all other states are trapped in the interior region and share small portions @xmath442 ^ 2 $ ] of the collective displacement and width in accordance with the group velocity @xcite . the solution just described has formally very much in common with the schematic model of the so - called nuclear dipole giant resonance developed in @xcite . in this paper we analyzed the relevance of the concept of the non - hermitian effective hamiltonian in finite - range potential scattering . single - channel @xmath9-wave resonance scattering is considered as an example . the scattering function @xmath13 is meromorphic in this case , i.e. , has only isolated poles in the complex plane of the wave numbers . the number of poles is , generally , infinite but only a finite part of them can be interpreted as resonances . we first presented ( secs . ii and iii ) a consistent self - adjoint formulation of the scattering problem , which is based on separation of the configuration space into internal and external segments . in this way , the @xmath13 function is represented in terms of a non - hermitian energy - dependent operator @xmath443 . there exists a wide freedom in choosing the radius @xmath12 of separation as well as the boundary conditions at this point . different choices yield different explicit forms of this operator together with the @xmath13 and @xmath15 functions . this can , in particular , come out strongly if one truncates the matrix @xmath0 to calculate numerically the positions of the poles of @xmath8 function in the complex @xmath16 plane . nevertheless , the true complex poles of the @xmath8 function and this function itself depend , as we explicitly demonstrate , neither on the bc nor on the radius @xmath12 . although all choices are formally allowed , this does not mean that they are equally adequate from the physical point of view . for instance , a fictitious immediate reflection takes place at the ( arbitrarily chosen ) point of separation @xmath12 . the artificial separation of the phase of this reflection results , in turn , in the appearance of an infinite number of remote poles of the function @xmath15 , which describe the smooth contribution from the interior region @xmath444 . the phase of the immediate reflection must be compensated by the contributions of such poles for the radius @xmath12 to disappear finally from the scattering amplitude . this can cause unjustified complications in intermediate stages of calculations . we argue that the harm is minimized if the separation radius matches an outer potential barrier when it is strong enough to make immediate reflection at this point quite probable . the notion of the non - hermitian effective hamiltonian first introduced in the theory of resonance nuclear reactions emerges when a group of very close resonance states strongly overlap and interfere . we stress that , in contrast to the nuclear reactions , in the cases of the potential resonance scattering usually discussed in the literature the density of the energy spectrum is too low and the ordinary breit - wigner approximation of isolated resonances usually suffices . the complex energy of the @xmath334th resonance which dominates in the scattering amplitudes near the scattering energy @xmath445 . there is no room for an effective hamiltonian in this case or , more strictly , it is embodied by the @xmath446 `` matrix '' @xmath447 . we emphasize that the energy dependent operator @xmath443 should not be confused with the effective hamiltonian . energy dependent eigenvalues of this operator are not , generally , in one - to - one correspondence with the complex energies of the actual resonance states . overlap and interference of the resonances can become possible in the cases when the energy spectrum has a band structure . as an example of this kind we investigate a periodically disposed chain of a finite number @xmath2 of radial @xmath4 barriers . all @xmath8 , @xmath234 and @xmath11 functions are found in closed forms in terms of an @xmath3 @xmath16-dependent matrix propagator . this allows us to study in detail all analytical properties in the complex @xmath16 plane and to verify the correspondence with the projection formalism used . there exist a finite number of separated bands of close resonances . within the @xmath334th band which lie near the scattering energy @xmath448 one can neglect all smooth variations with @xmath16 . in this approximation , the mentioned propagator proved to coincide with the resolvent of the @xmath3 block @xmath449 of the matrix @xmath450 _ taken at the fixed value _ just this matrix plays the role of the effective hamiltonian of the system in the energy interval within the band . different choices of bcs yield different patterns of the resonance interference . in particular , the spectrum of the poles of the @xmath11 function is exactly reproduced in the framework of the projection technique with the bc ( [ cnint ] ) of neumann type fixed at the position @xmath196 of the outer barrier . the corresponding energy levels depend on the penetrability constant @xmath387 of the outer barrier . shifts of the levels due to the coupling to the continuum are included in this case from the very beginning . utilization of the dirichlet bc for the intrinsic motion gives another but yet equivalent formulation of the scattering problem considered . the internal problem in this case fixes a closed counterpart of the open system under consideration . this enables us to investigate the change of the regime of the internal motion of the system as its openness grows . the interaction via the continuum shifts the original @xmath387-independent levels of the internal motion along both real and imaginary axes . a transition is explicitly demonstrated from the bunch of @xmath2 similar narrow resonances to the formation of a relatively broad resonance strongly shifted with respect to the band , which exists in the outer well . the other resonances turn out to be trapped in the inner part of the system . this is quite similar to the results of earlier investigations which rely upon the notion of the non - hermitian effective hamiltonian of an open system . we are grateful to y.v . fyodorov for discussions on feshbach s projector technique in an early stage of this work , to h. schanz and v. zelevinsky for interesting discussions , and to g. hackenbroich and c. viviescas for their interest in the work and useful conversations . one of us ( v.v.s . ) greatly appreciates the generous hospitality of the max - planck institute for the complex systems extended to him during his stay in dresden . the financial support by rfbr grant no . 03 - 02 - 16151 ( d.v.s . and v.v.s . ) and sfb 237 `` unordnung and grosse fluktuationen '' ( d.v.s . and is acknowledged with thanks . by a slight shift of the position of the outer barrier and subsequent careful tuning of the two free parameters complete degeneracy of the resonances can be achieved ; see , * * , ( ) ; , , , * * , ( ) . recently , the experimental study of such an _ exceptional point _ has been performed in a microwave cavity setup : , , , , , , , * * , ( ) ; see , however , i. rotter , phys . e * 67 * , 026204 ( 2003 ) .

we study string realizations of split extended supersymmetry , recently proposed in hep - ph/0507192 . supersymmetry is broken by small ( ) deformations of intersection angles of-branes giving tree - level masses of order , where is the string scale , to localized scalars . we show through an explicit one - loop string amplitude computation that gauginos acquire hierarchically smaller dirac masses . we also evaluate the one - loop higgsino mass , , and show that , in the absence of tree - level contributions , it behaves as . finally we discuss an alternative suppression of scales using large extra dimensions . the latter is illustrated , for the case where the gauge bosons appear in representations , by an explicit string model with standard model gauge group , three generations of quarks and leptons and gauge coupling unification .

hep-th0601003

implementing the idea of split supersymmetry @xcite in string theory is straightforward @xcite . the appropriate framework is type i theory @xcite compactified in four dimensions in the presence of constant internal magnetic fields @xcite , or equivalently @xmath1-branes intersecting at angles @xcite in the @xmath8-dual picture . however , in simple brane constructions the gauge group sector comes in multiplets of extended supersymmetry , while matter states are in @xmath9 representations . in ref . @xcite we showed that these economical string - inspired brane constructions reconcile with unification of gauge couplings at scales safe from proton decay problems , and provide a natural dark matter candidate . indeed a simple way to break supersymmetry in the above context is by deforming the intersection angles of the standard model branes from their special values corresponding to a supersymmetric configuration . a small deformation of these angles by @xmath10 breaks supersymmetry via a @xmath1-term vacuum expectation value ( vev ) , associated to a magnetized abelian gauge group factor in the @xmath8-dual picture , @xmath11 with @xmath3 the string scale @xcite . this leads to mass shifts of order @xmath2 for all charged scalar fields localized at the intersections , such as squarks and sleptons , while gauginos ( and higgsinos ) remain massless . alternatively supersymmetry breaking can be communicated to the scalar observable sector by radiative corrections from a supersymmetric messenger sector , with @xmath1-breaking triggered by a magnetized abelian subgroup , or a non - supersymmetric sector with large extra dimensions . in all cases all previously massless scalars in the observable sector are expected to acquire large masses by radiative corrections and a fine - tuning is needed in the higgs sector in order to keep the hierarchy between the electroweak scale and @xmath12 , as required in split supersymmetry . on the other hand fermion ( gaugino and higgsino ) masses are protected by a chiral @xmath13-symmetry and the magnitude of radiative corrections depends on the mechanism of its breaking . in fact @xmath13-symmetry is in general broken in the gravitational sector by the gravitino mass but its value , as well as the mediation of the breaking to the brane ( standard model ) sector , is model dependent and brings further uncertainties . here we will restrict ourselves to possible sources of fermion mass generation due to brane effects described by open string propagation within only global supersymmetry , assuming that gravitational ( closed strings ) corrections are negligible . indeed , @xmath13-symmetry is in general broken by @xmath14-string corrections and gaugino majorana masses can be induced by a dimension - seven effective operator which is the chiral @xmath15-term @xcite : @xmath16 , where @xmath17 and @xmath18 are the magnetic @xmath19 and non - abelian gauge superfields , respectively . its moduli dependent coefficient is given by the topological partition function @xmath20 on a world - sheet with no handles and three boundaries and @xmath18 gauge groups while the third one can be an orientifold . ] . from the effective field theory point of view , it corresponds to a two - loop correction involving massive open string states . upon a vev @xmath21 , the above @xmath15-term generates majorana gaugino masses that are hierarchically smaller than the scalar masses and behave as @xmath22 . in models where the gauge bosons come in multiplets of extended supersymmetry , there exists the possibility of generating dirac gaugino masses that do not require the breaking of @xmath13-symmetry . such a mass can be induced at one - loop via the effective chiral dimension - five operator @xcite : @xmath23 , where @xmath24 denotes the @xmath9 chiral superfield(s ) containing the additional gaugino(s ) . upon the @xmath1-auxiliary vev @xmath25 this term generates dirac gaugino masses that scale as @xmath26 , and are thus much higher than the majorana masses @xmath27 . in ref . @xcite , we studied the renormalization group evolution and showed that this scenario is compatible with one - loop gauge coupling unification at high scale for both cases where the gauge sector is @xmath28 and @xmath7 supersymmetric . the low energy sector of these models contains , besides the standard model fields , just some fermion doublets ( higgsinos ) and eventually two singlets , the binos , if the corresponding corrections to their dirac mass vanish . in fact , the higgsinos must acquire a mass , @xmath5 , of order the electroweak scale in order to provide a dark matter candidate . this can be induced by the following dimension - seven operator , generated at one loop level @xcite : @xmath29 , where @xmath30 are the two @xmath9 higgs supermultiplets . it follows that the induced higgsino mass is of the same order as the gaugino majorana masses , @xmath31 . the appearance of gauginos in multiplets of extended supersymmetry is common in previous attempts to embed the standard model in intersecting brane constructions @xcite . there are several examples in the literature of such models with gauge sectors forming multiplets of either @xmath7 @xcite or @xmath28 extended supersymmetry @xcite . in this work we describe the general string framework realizing the above scenario of split supersymmetry with extended supersymmetric gauge sector and perform an explicit one - loop computation of the dimension - five and seven effective operators needed to produce dirac gaugino and higgsino masses , respectively . the relevant world - sheet diagram is the annulus involving two @xmath1-brane stacks on its boundaries ( or three in the case of higgsinos ) . we find that both dirac gaugino and higgsino masses are in general non - vanishing when the two brane - stacks are parallel in one of the three internal compactification planes . the leading behavior in the supersymmetric limit , @xmath32 , gives the coefficient of the corresponding effective operator , which thus receives non - trivial contributions only from @xmath28 supersymmetric sectors . moreover we find that in this limit the result simplifies and becomes topological ; the non - zero mode determinants cancel and the effective couplings depend only on the momentum lattice of the plane where the two brane - stacks are parallel . in the higgsino case the two fermions should come from an @xmath28 supersymmetry preserving intersection , localized in the remaining two internal planes . finally , for concreteness , we present an explicit string construction with the standard model gauge group , precisely three generations of quarks and leptons , and sharing the desired features described above . moreover , it emerges from an @xmath33 grand unified group and thus satisfies gauge coupling unification , realizing a particular @xmath1-brane configuration proposed in ref . standard model particles live in the intersection of supersymmetric branes while there is a non - supersymmetric brane such that particles living in its intersection with the observable branes are non - chiral and act as messengers of gauge mediated supersymmetry breaking . in this case the hierarchy between the string scale and the masses of supersymmetric partners can be triggered by extra dimensions hierarchically larger than the string length . our paper is organized as follows . in section [ section_susy_breaking ] , we describe the general framework of supersymmetry breaking in @xmath1-brane models intersecting at angles . in section [ section_dirac_mass ] , we perform the one - loop computation of the induced dirac gaugino masses and extract the coefficient of the relevant dimension - five effective operator by evaluating the behavior in the supersymmetric limit . in section [ section_higgsino ] , we perform a similar computation for the higgsino masses and the corresponding dimension - seven effective operator . in section [ model ] , we present an explicit construction of the standard model spectrum in this framework with the desired features . our conclusions are drawn in section [ section_conclusion ] and finally some relevant formulae are presented in appendix a and some technical computational details about the bosonic correlation function of higgsinos in appendix b. we will consider a set of intersecting stacks of branes that can be divided into two subsets : the first one , denoted as @xmath34 ( for observable ) , gives rise in its light spectrum to the observable sector , i.e. a supersymmetric version of the standard model with the gauge sector in @xmath28 or @xmath7 representations . the second subset , which we denote as @xmath35 ( for messenger ) , provides the supersymmetry breaking messengers through its intersections with the branes in @xmath34 . in order to perform explicit computations , we consider a compactification on a six torus factorizable as @xmath36 with appropriate projections and orientifold planes to provide the desired supersymmetric framework . a basis cycles @xmath37 $ ] and @xmath38 $ ] of the corresponding homology classes is defined for every torus @xmath39 . every stack @xmath40 of @xmath41-branes in type iia wraps a 3-cycle @xmath42 $ ] factorizable into the product of 1-cycles : @xmath43=\bigotimes_i [ \pi_a^{(i)}]=\bigotimes_i \left(n_a^{(i)}[{\bf a}^{(i ) } ] + m_a^{(i)}[{\bf b}^{(i)}]\right ) \label{jojo}\ ] ] and forms angles . ] with the cycles @xmath44 $ ] given by : @xmath45 where @xmath46 and @xmath47 are the radii along the horizontal @xmath48 ( @xmath49-cycles ) and vertical @xmath50 ( @xmath51-cycles ) axes , respectively , and @xmath52 is the angle of the tilted torus @xmath53 . in the presence of an orientifold plane along the @xmath48-axis , the angle @xmath52 is fixed to either @xmath54 , which corresponds to rectangular tori , or to @xmath55 which corresponds to tilted tori ( see fig . [ tori ] ) . in these cases , we can write the angles @xmath56 as @xmath57 where we define @xmath58 and @xmath59 ( @xmath60 ) for rectangular ( tilted ) tori . from here on and for simplicity we will remove the tilde from @xmath61 . given a generic couple @xmath62 of stacks of branes they intersect with angles @xmath63 in the @xmath64-th torus : @xmath65 of special importance is the number of such intersections as it measures the number of chiral fermions . it is given by @xmath66\cdot[\pi_b ] = \prod_i \left(n_a^{(i)}m_b^{(i ) } - m_a^{(i)}n_b^{(i)}\right ) \label{intersection_number}\ ] ] and it corresponds in the @xmath8-dual picture to the index theorem for compactification with internal magnetic fields . in the special case where the brane stack @xmath67 is the image of @xmath40 under the orientifold action ( @xmath68 ) , the chiral states given by this formula fall in two categories : they transform either in the antisymmetric ( a ) or in the symmetric ( s ) representation of the gauge group , due to the orientifold projection . their respective multiplicities are given by : @xmath69 in simple toroidal compactifications , states arising from open strings with both ends on the same stack of branes form representations of @xmath7 supersymmetry . for more complicated compactifications , part of these states can be projected out and one remains with representations of lower supersymmetry . in this work , we will focus on the cases where these states are in @xmath7 or @xmath28 multiplets as already stated . open strings stretching between two non - parallel stacks of @xmath70 and @xmath71 branes give rise to states in the bifundamental representation @xmath72 of @xmath73 . the lightest modes contain , in addition to massless fermions , scalars with masses : @xmath74 for parallel branes the first three states become the @xmath7 scalar partners of the gauge vector bosons . on the other hand , if some of them are massless there is a boson fermion degeneracy that indicates that part of the original supersymmetry is preserved . moreover , if any of the @xmath75 angles vanishes , a supersymmetric mass @xmath76 can be generated through separation of the branes by a distance @xmath77 in the corresponding torus . finally , the last mass in ( [ genericmass ] ) contains the contribution of a massive string oscillator that could become the lightest state for some values of the angles . there are three cases corresponding to three different kinds of intersections that must be discussed : @xmath78 the case where branes @xmath40 and @xmath67 are in the observable set @xmath79 with intersections giving rise to chiral @xmath9 multiplets . the lightest states @xmath80 are identified with quarks , leptons and their supersymmetric partners . this can be achieved with the choice @xmath81 which is such that for generic values of @xmath82 and @xmath83 the system only preserves one supersymmetry . for simplicity , we are using here the notation @xmath84 and choosing @xmath85 . @xmath78 the case where branes @xmath40 and @xmath67 are in the observable set @xmath79 , and states in their intersection give rise to @xmath28 supermultiplets which are identified with supersymmetric pairs of higgs doublets . this corresponds to eq . ( [ susymass ] ) with @xmath86 , i.e. @xmath87 we assume that @xmath88 higgs chiral multiplets remain light while the other @xmath28 multiplets get large ( @xmath89 ) supersymmetric masses . actually , this is not the only way to obtain higgs multiplets . one of the two doublets may emerge from a chiral intersection together with the leptons , say between a @xmath90 and a @xmath19 brane , while the other one could arise as a chiral anti - doublet " from the intersection of @xmath90 with the mirror of @xmath19 brane . @xmath78 the case where brane @xmath40 is in the observable set @xmath79 and brane @xmath67 in sector @xmath91 . the states @xmath80 living at such intersections are assumed to be non - chiral and we will call them `` messengers '' for reasons that will be apparent below . they are non - supersymmetric because of mass splitting between fermionic and bosonic modes . through loop effects , they will induce supersymmetry breaking to the observable sector . we will use here the notation @xmath92 and choose again @xmath93 . two different kinds of models can arise at this level . they correspond to the following two possibilities : the first possibility is that the states in the intersection @xmath80 originate as a perturbation around an @xmath28 supersymmetric solution . in fact , in order to have a scale of supersymmetry breaking hierarchically small compared to the string scale we perform a tiny deformation of the intersection angles ( preserving @xmath28 supersymmetry ) : @xmath94 in one of the tori , chosen to be the first one , the branes should remain parallel and separated ) and will make the configuration stable . ] by a distance @xmath95 , i.e. @xmath96 leading to the localized scalar masses : @xmath97 these states will induce at one - loop dirac masses for gauginos and higgsinos , as we will show in next sections . for instance the dirac gaugino masses turn out to be @xmath98 where @xmath99 is the gauge coupling . this provides a stringy realization of gauge mediated supersymmetry breaking in the absence of @xmath13-symmetry breaking . scalars localized at supersymmetric intersections , as those discussed in ( [ susymass ] ) , can be given a small supersymmetry breaking through another tiny deformation of the corresponding angles by @xmath10 . this can be described at the effective field theory level as a fayet - iliopoulos @xmath1-term breaking corresponding to the presence of an anomalous @xmath19 factor @xmath100 @xcite . as a result , scalar masses in the observable sector can acquire different masses depending on whether matter is charged under this @xmath19 or not . in particular : * if the observable sector is charged , bosons of the matter multiplets acquire tree - level masses , @xmath2 . * otherwise the bosons of the observable sector receive masses at the two - loop level @xcite @xmath101 in both cases the transmission of supersymmetry breaking to the gaugino and higgsino sector is mediated through the `` messengers '' since the quark and lepton sectors are chiral ( and do not contribute at this order ) while the higgs sector ( with masses tuned to remain in the tev range ) will contribute negligibly . note that the brane @xmath67 does not have to be necessarily in sector @xmath91 . it could also be part of the observable " sector . consider for instance the example where chiral quark doublets come from the intersection of a @xmath102 and a @xmath90 brane stack . non chiral antiquark doublets , coming from the intersection of @xmath102 with the orientifold image of @xmath90 can play the role of messengers generating dirac gaugino masses for both @xmath102 and @xmath90 gauge groups . moreover , the image of @xmath90 may have non trivial intersection with another @xmath19 stack producing leptons , and thus it is part of sector @xmath79 . the second possibility corresponds to the case where a supersymmetry is preserved by the subset of branes in @xmath79 for some choice of the compactification moduli , but will never be conserved by the whole set of branes @xmath103 . this is for instance the case for the toroidal compactification discussed in section 5 below . let us denote by @xmath104 the relevant coefficients that define in ( [ genericmass ] ) the supersymmetry preserved by the observable sector . the intersection between the @xmath34 and @xmath105 branes leads to states @xmath80 with a supersymmetry breaking mass @xmath106 keeping the branes @xmath62 parallel in one of the tori allows to use again the states @xmath80 at their intersection as messengers to produce gaugino and higgsino dirac masses , as in sections 3 and 4 . however now the desired suppression of @xmath10 will be generated by the presence of a large hierarchy between the size of the compact dimensions . more precisely for @xmath107 one can obtain an @xmath10 parameter hierarchically smaller than one @xmath108 where the sum obviously goes over the tori where branes intersect . in the absence of tree level supersymmetry breaking , superpartners in the observable sector will acquire two - loop mass splitting given by ( [ masaescalar ] ) . a toy model based on these ideas will be presented in section [ model ] . as the gauginos lie in representations of extended supersymmetry ( @xmath109 ) , they come in copies that can pair up to receive a dirac mass at one - loop order in the string genus expansion . the corresponding world - sheet has the topology of a cylinder stretching between two stacks of @xmath70 and @xmath71 branes , respectively ( see fig . [ fig : fig5 ] ) . the vertex operators @xmath110 and @xmath111 associated with the gauginos are inserted on one boundary , for example that corresponding to the @xmath70 branes . the mass is given by the integrated two - point correlation function of the relevant vertex operators : @xmath112 where the integrals are along the boundary of the annulus . = 5.5 in the @xmath28 supersymmetry which relates the gauge vector and gauginos is chosen to be associated with the supercharges that preserve the ( undeformed ) @xmath28 preserving brane intersection ( see eq . ( [ nonsusymass ] ) with @xmath113 ) . these supercharges are given by @xmath114 where @xmath115 denotes helicities in the internal directions and @xmath116 are weyl spinor indices in four dimensions . the supercharges have been grouped into majorana spinors in four dimensions . as the gauginos of @xmath28 supersymmetry have the same internal helicities as the supercharges ( [ conserved_charges ] ) , the vertex operators in the @xmath117-ghost picture are given by : @xmath118 together with their cpt conjugates which are obtained by simply reversing helicities in the internal directions and flipping the space - time chirality . here @xmath119 is the bosonized two - dimensional reparametrization ghost field , @xmath120 are the target space coordinates , @xmath121 are the space - time momenta , @xmath122 arise from bosonization of the spin fields and @xmath123 is the helicity vector in four dimensions which is constrained by the gso projection to be either @xmath124 also @xmath125 is some normalization constant and @xmath126 are chan - paton factors of the associated gauge group factor . on the cylinder , the total @xmath119-charge must vanish . therefore , we must use picture - changing to transform one of the vertex operators to the @xmath127 picture . this is done by operating with the brst charge : @xmath128 where @xmath129 is the world - sheet supercurrent : @xmath130 the fields @xmath131 are real world - sheet fermions while in the second and third terms of ( [ supercurrent ] ) we have used complex coordinates for the internal directions : @xmath132 and analogously for @xmath133 and @xmath134 . note that @xmath135 . after picture - changing , the vertex operator @xmath136 contains several terms , but only one of them contributes to the two - point function @xmath137 . indeed , the internal helicities must cancel between the two vertex operators to yield a non - vanishing correlator . for the choice made in ( [ gaugino_vertices ] ) , @xmath138 must induce a flip of the sign of @xmath139 . we can therefore set @xmath140 inserting these operators into a path integral on the cylinder , with @xmath141 , leads to @xmath142 \times [ fc ] \label{total_correlator}\ ] ] where the factor @xmath71 comes from the trace over the chan - paton degrees of freedom of the second boundary of the cylinder . the factors @xmath143 $ ] and @xmath144 $ ] denote , respectively , correlation functions of world - sheet bosons and world - sheet fermions together with the associated ghosts of opposite statistics . they are given by @xmath145 = \langle\mathbb{i}\rangle_{bc } \langle\mathbb{i}\rangle_{x^\mu } \langle\partial z^1\rangle_{z^1 } \langle\mathbb{i}\rangle_{z^2 } \langle\mathbb{i}\rangle_{z^3 } \label{bc}\ ] ] and @xmath146 = \sum_{a , b } c\begin{bmatrix } a \\ b \end{bmatrix } \langle e^{-\frac{\phi}{2 } } e^{\frac{\phi}{2}}\rangle \langle e^{\frac{i}{2 } \vec s \cdot \vec h } e^{-\frac{i}{2 } \vec s \cdot h}\rangle \prod_{i=1,2,3 } \langle e^{\frac{i}{2 } h_i } e^{-\frac{i}{2 } h_i}\rangle \label{fc}\ ] ] where @xmath147 is the identity operator . in the bosonic correlator , @xmath148 denote bosonic path - integrals over the complex coordinates @xmath149 while @xmath150 is a path integral over the parametrization bc - ghosts . in the fermionic correlation function the sum is over the spin structures for which we use the same notation @xmath151 as for brane stacks but the difference is obvious for the reader . the one - point functions of identity operators are given by the appropriate piece of the partition function on the cylinder for the two stacks of intersecting @xmath41-branes . the bc - ghosts cancel exactly the path - integral over two of the four space - time coordinates @xmath152 , and we are left with the simple expression : @xmath145 = \frac{1}{\eta(it)^2 } \langle\partial z^1(w)\rangle_{z^1 } \frac{\eta(it)}{\vartheta \begin{bmatrix } \frac{1}{2 } + \alpha_2 \\ \frac{1}{2}\end{bmatrix } } \frac{\eta(it ) } { \vartheta \begin{bmatrix } \frac{1}{2}+ \alpha_3 \\ \frac{1}{2}\end{bmatrix } } \label{bc_2}\ ] ] where @xmath153 , @xmath154 are the non - vanishing intersection angles of the @xmath41-branes in the @xmath155 and @xmath156 planes . here , @xmath157 is the open string proper - time parametrizing the world - sheet annulus . to compute the correlation function of @xmath158 , consider the coordinate domain of the annulus to be a square parametrized by two coordinates @xmath159 with ranges @xmath160 , @xmath161 and the identification @xmath162 . the operator @xmath163 is located on the boundary , thus @xmath164 where @xmath165 is the hamiltonian for the @xmath163 coordinate field . only the zero - modes of the mode expansion of @xmath163 contribute to the trace in ( [ z3_correlator ] ) and , therefore , if the branes would intersect with a non - trivial angle in the @xmath163 plane , the correlation function would vanish . we therefore need to impose @xmath166 as anticipated in ( [ alphacond ] ) . this is also a necessary condition for the @xmath1-brane intersection to preserve the supercharges ( [ conserved_charges ] ) . as the two stacks are parallel in the first torus , they can be separated at a distance @xmath167 . for simplicity , this separation is taken along the @xmath168 direction . the mode expansions read : @xmath169 where @xmath170 is the winding number . the momentum @xmath171 is quantized in units of @xmath172 . using the expression for the hamiltonian , @xmath173 shows , as stated above , that the oscillators in ( [ mode_expansions ] ) do not contribute to the trace : @xmath174 the sum over the quantized momenta does not contribute either , since the different terms cancel pairwise . however , the sum over windings is non - vanishing provided that the two stacks are non - coincident , i.e. @xmath175 . inserting ( [ correlator_sum ] ) into ( [ bc_2 ] ) yields the complete bosonic correlator : @xmath145 = \frac{2 i { \alpha'}}{\eta^2 \vartheta\begin{bmatrix}\frac{1}{2 } + \alpha_2 \\ \frac{1}{2}\end{bmatrix}(0 ) \vartheta\begin{bmatrix}\frac{1}{2 } + \alpha_3 \\ \frac{1}{2}\end{bmatrix}(0 ) } \sum_n \left(\frac{n r_2^{(1)}}{{\alpha ' } } + \frac{\ell}{{\alpha'}}\right ) e^{- 2 \pi { \alpha'}t \left(\frac{n r_2^{(1)}}{{\alpha ' } } + \frac{\ell}{{\alpha'}}\right)^2 } \label{bc_final}\ ] ] to compute the fermionic correlator , we will make use of the elementary two - points functions : @xmath176 where @xmath177 labels the two non - compact complexified space - time dimensions and @xmath178 . the difference between ( [ fermion_correlator1 ] ) and ( [ fermion_correlator2 ] ) comes from the fact that in the @xmath179 and @xmath180 planes , the brane intersection angles @xmath153 and @xmath154 are non - vanishing . the @xmath181 dependence was determined for instance in ref . @xcite and the correct normalization is inferred by matching the short distance limit @xmath182 on both sides . for example , from the operator product expansion ( ope ) of bosonized fermions we have @xmath183 as the correlator of the identity operator is the spin - structure dependent partition function of a complex fermion . taking the same limit on the right - hand side of ( [ fermion_correlator1 ] ) and comparing with ( [ normalization ] ) allows to check that the dependence on the cylinder modulus is correctly reproduced . inserting ( [ ghost_correlator])-([fermion_correlator2 ] ) into ( [ fc ] ) , we obtain : @xmath146 = \frac{1}{\eta(it)^4 } \ \frac{\vartheta'_1(0)}{\vartheta_1(z - w ) } \sigma \label{fc_2}\ ] ] where @xmath184 the coefficients @xmath185 can be obtained by the same method as the one used above to deduce the normalization of the elementary correlators . from the ope we deduce that the fermionic contribution @xmath144 $ ] has a pole in @xmath181 whose residue , being the correlator of the identity , should be the full fermionic partition function on the annulus : @xmath186 & \rightarrow & \frac{1}{(z - w ) } z_{f } = \\ & & \frac{1}{z - w } \sum_{a , b } \eta_{a , b } e^{- 2 \pi i b ( \alpha_2 + \alpha_3 ) } \frac{\vartheta\begin{bmatrix } a \\ b \end{bmatrix}^2(0)}{\eta^2 } \frac{\vartheta \begin{bmatrix}a + \alpha_2 \\ b\end{bmatrix}(0)}{\eta } \frac{\vartheta \begin{bmatrix}a + \alpha_3 \\ b\end{bmatrix}(0)}{\eta}\nonumber\end{aligned}\ ] ] where as usual @xmath187 . comparing with the limit @xmath182 of the right - hand side of ( [ fc_2 ] ) , we obtain @xmath188 the spin - structure sum in @xmath189 can now be performed by using the riemann theta - identity ( [ riemann ] ) . to apply this identity we must first use the rearrangement @xmath190 to put the theta functions in the appropriate form . the @xmath191-dependent phase factor in @xmath192 then cancels out . after applying the riemann identity , we can use ( [ rearrangement ] ) again to put @xmath153 and @xmath154 back into the argument of the theta function . the result is : @xmath193 we can now expand this result around the @xmath194 supersymmetric configuration of @xmath1-branes using @xmath195 . expanding ( [ sigma ] ) to lowest order in @xmath10 gives @xmath146 = - 8 \pi^2 \epsilon \eta^2 \vartheta\begin{bmatrix } \frac{1}{2 } + \alpha_2 \\ \frac{1}{2}\end{bmatrix}(0 ) \vartheta\begin{bmatrix } \frac{1}{2 } + \alpha_3 \\ \frac{1}{2}\end{bmatrix } ( 0 ) \label{fc_final}\ ] ] where we made use of the identity @xmath196 . inserting ( [ bc_final ] ) and ( [ fc_final ] ) into ( [ total_correlator ] ) we see that , to lowest order in @xmath10 , all contributions from the string oscillators cancel and only the classical part of the correlation function remains . this piece is independent of @xmath181 and the integrals over the locations of the vertex operators in ( [ amplitude ] ) only contribute a factor @xmath197 . the result is then simply @xmath198 from this we can can obtain the two - gaugino amplitude as @xmath199 where the integration measure has been obtained by comparing with the partition function for intersecting @xmath41-branes . the infinite volume factor @xmath200 arises from momentum conservation . had we chosen @xmath201 , it would be replaced by a @xmath202-function , @xmath203 . the normalization @xmath204 can in principle be determined . however for our purposes we only need to know that @xmath204 is proportional to the open string coupling and therefore @xmath205 , where @xmath206 is the closed string coupling , determined by the vev of the string dilaton @xmath207 . for the dirac mass of the gauginos we then obtain the final result : @xmath208 by a poisson resummation , it can easily be checked that the integral is finite . note also that in the limit where @xmath209 , the mass vanishes . to interpret the meaning of this result , we consider the mass spectrum of open strings which stretch between the two stacks of @xmath1-branes . in the limit where @xmath210 , these string modes form @xmath28 hypermultiplets . the mass of the lightest multiplet is given by @xmath211 . when @xmath10 is non - vanishing , supersymmetry is completely broken and there is a mass splitting between fermions and scalars . the lightest fermions remain at @xmath212 , while their scalar superpartners have their masses lifted to @xmath213 . the result ( [ dirac_mass ] ) then implies the relation , in the limit @xmath214 : @xmath215 where @xmath216 is the string scale . for string size brane separation , such that @xmath217 , we recover the anticipated behavior ( [ masagaugino ] ) . consider two stacks of @xmath41-branes : the first stack of @xmath70 branes aligned with the horizontal @xmath48 axes . ] and the second stack of @xmath218 branes intersecting the horizontal axes in the three tori at angles @xmath219 , @xmath220 . for @xmath221 and @xmath222 , this configuration preserves @xmath194 supersymmetry and the lightest string modes on the intersection describe hypermultiplets . they are made of two @xmath9 chiral multiplets , identified as two higgs bosons and their @xmath9 superpartners , the higgsinos . = 5.5 in = 3.in and @xmath223 . non - supersymmetric states stretching between the branes @xmath62 and @xmath224 induce at one - loop masses for the higgsinos.__,title="fig : " ] a simple way to generate a mass term for these higgsinos is to separate the branes in the first torus where they are parallel . this corresponds to switching on a vacuum expectation value for a particular scalar that parametrizes the location of the brane in the direction @xmath168 . the mass is then given by the tension times the minimal length of the string , which is just the brane separation . this mass generation is not always possible , as for specific constructions such as orbifold models , the corresponding scalar is projected out of the light spectrum . in this section we will investigate the possibility that the higgsino mass is generated at one - loop in a way similar to the dirac gaugino mass . the appropriate string amplitude corresponds to a world - sheet with the topology of an annulus . on the first boundary two vertex operators that create the higgsinos are inserted . the higgsinos arise from open strings with the two endpoints located on different stacks of @xmath1-branes . this means that part of the first boundary of the annulus lies on the stack of @xmath70 branes and the other part lies on the other stack of @xmath218 branes ( see fig . [ fig : fig6 ] ) . this change of boundary conditions is implemented by the insertion of twist fields @xmath225 and @xmath226 . the other boundary of the annulus must be located on a third stack of @xmath71 branes . the two stacks @xmath71 and @xmath70 intersect at angles @xmath191 . as this additional stack must preserve the same supersymmetries as the other stacks , we will start from the simplest possibility @xmath227 and make a small deformation of the angles to break supersymmetry , as described in section [ section_susy_breaking ] . the generalization to @xmath228 is straightforward . the vertex operators of the higgsinos in the @xmath117-ghost picture are : @xmath229 note the similarities with the vertex operators of the gauginos ( [ gaugino_vertices ] ) . the helicities have been shifted by the intersection angles . in addition we have inserted the bosonic twist operators as world - sheet superpartners of the operators @xmath230 . in the limit where @xmath231 , we recover precisely the gaugino vertices since in this limit the bosonic twists become the identity operators . since the annulus has vanishing ghost - charge we need again to transform one of the operators to the @xmath232-picture . this is done precisely in the same way as in section [ section_dirac_mass ] . to obtain a non - vanishing correlator , the world - sheet supercurrent @xmath129 must be used to flip the sign of @xmath139 in one of the exponents . the result is @xmath233 inserting the vertex operators at zero momentum in the correlator leads to an expression of the same form as ( [ total_correlator ] ) , @xmath234 \times [ fc ] \label{total_correlator2}\ ] ] where now @xmath235 & = & \langle\mathbb{i}\rangle_{bc } \langle\mathbb{i}\rangle_{x^\mu } \langle\partial z^1\rangle_{z^1 } \langle\sigma_{1-\beta_2 } \sigma_{\beta_2}\rangle_{z^2 } \langle\sigma_{1-\beta_3 } \sigma_{\beta_3}\rangle_{z^3 } \label{bc_mu } \\ \nonumber \\ ~ [ fc ] & = & \sum_{a , b } c\begin{bmatrix } a \\ b \end{bmatrix } \langle e^{-\frac{\phi}{2 } } e^{\frac{\phi}{2}}\rangle \langle e^{\frac{i}{2 } \vec s \cdot \vec h } e^{-\frac{i}{2 } \vec s \cdot \vec h}\rangle \langle e^{\frac{i}{2 } h_1 } e^{-\frac{i}{2 } h_1}\rangle \nonumber \\ & & \times \prod_{i=2,3}\langle e^{-i ( \beta_{i } - \frac{1}{2 } ) h_i } e^{i ( \beta_{i } - \frac{1}{2 } ) h_i}\rangle \label{fc_mu}\end{aligned}\ ] ] because of the factor @xmath236 in @xmath143 $ ] ( which was computed in section [ section_dirac_mass ] ) , we must impose that @xmath237 and that the branes are separated by a distance @xmath175 in the @xmath168 direction in order to obtain a non - vanishing contribution . the computation of the fermion correlator is similar to the one in section [ section_fermion_correlator ] . the correlator is computed for general @xmath238 then expanded around the ( @xmath7 ) supersymmetric configuration @xmath239 . a slight generalization of ( [ fermion_correlator2 ] ) is given by : @xmath240 with @xmath241 . the remaining elementary correlators are unchanged and are given in ( [ ghost_correlator ] ) and ( [ fermion_correlator1 ] ) . inserting these in ( [ fc_mu ] ) we obtain @xmath146 = \frac{1}{\eta^4 } \left(\frac{\vartheta_1'(0)}{\vartheta_1(z - w)}\right)^{1 - \beta_2 ( 1- \beta_2 ) - \beta_3 ( 1-\beta_3 ) } \times \sigma \label{fc_mub}\ ] ] where @xmath242 the coefficients @xmath185 are given as before by ( [ coefficients ] ) . using the riemann identity ( [ riemann ] ) allows to perform the spin - structure sum and put @xmath189 into a simpler form . after imposing the supersymmetry condition @xmath243 , one finds : @xmath244 this can now be expanded around the supersymmetric configuration @xmath239 . for simplicity , we take @xmath245 and @xmath246 , and expand in powers of @xmath10 : @xmath247 the correlator of world - sheet fermions , eq . ( [ fc_mu_final ] ) , is already of order @xmath248 . since we are interested only in the leading @xmath10 behavior , for the purpose of computing the bosonic correlation function @xmath143 $ ] , we can keep the leading order corresponding to @xmath239 . most pieces of @xmath143 $ ] have already been computed in section [ section_dirac_mass ] . the additional complication is the appearance of correlators of bosonic twist fields . these have been studied for instance for arbitrary genus and any number of twist fields in ref . the complete calculation is presented in appendix b where only the relevant results will be quoted . in fact using eq . ( [ result ] ) in ( [ bc_mu ] ) , leads to & = & i ( 2 i ) | |^-2(1- ) + & & _ n ( + ) e^- 2 t ( + ) ^2 + & & _ v_1^(2 ) , v_2^(2 ) e^- s_cl(v_1^(2 ) , v_2^(2 ) ) _ v_1^(3 ) , v_2^(3 ) e^- s_cl(v_1^(3 ) , v_2^(3 ) ) , [ bc_mu_final ] where @xmath249 is defined as in ( [ period_matrix ] ) for the twist @xmath250 and @xmath251 are defined in eq . ( [ esta ] ) . 0.2 cm finally inserting ( [ fc_mub ] ) , ( [ fc_mu_final ] ) and ( [ bc_mu_final ] ) into ( [ total_correlator2 ] ) gives for the two - point function : v^(1)(z ) v^(2)(w)&= & 32 i ^3 ^2 g_0 ^ 2 tr(^1 ^2 ) n_b + & & _ n ( + ) ( + ) ^2 f(z - w ) [ higgsino_correlator ] where f(z - w ) & = & + & & _ v_1^(2 ) , v_2^(2 ) e^- s_cl(v_1^(2 ) , v_2^(2 ) ) _ v_1^(3 ) , v_2^(3 ) e^- s_cl(v_1^(3 ) , v_2^(3 ) ) the higgsino mass @xmath5 is proportional to the two - higgsino amplitude obtained upon integration of the correlation function ( [ higgsino_correlator ] ) over the position of the vertex operators and inserting the result into ( [ integrated_amplitude ] ) . since the correlator depends only on @xmath181 , one of the integrals is trivial and contributes simply a factor @xmath252 . the final result reads ~g_s ^2 n_1 _ 0^ _ n ( + ) e^- 2 t ( + ) ^2 i where i = _ 0 ^ 2 t dx f(ix ) and it is of order @xmath248 as expected . in this section we will present a simple model as a framework to implement the realization of the extended split supersymmetry scenario described above with gauge bosons in representations of @xmath7 supersymmetry . it consists on a toroidal orientifold based on the factorized six - dimensional torus @xmath253 with an orientifold plane along the @xmath48 axes . this implies for each @xmath254-brane with wrapping numbers @xmath255 the presence of the image brane @xmath256 with wrapping numbers @xmath257 . the minimal setup is given by three intersecting stacks of @xmath41-branes giving rise to the gauge group @xmath258 the branes @xmath259 , @xmath260 and their images , with supersymmetric intersections , provide the observable set @xmath79 while the brane @xmath261 and its image stand for the ( messenger ) sector @xmath91 . the standard model gauge group is embedded in @xmath262 in the following way : * open strings localized at the intersection of @xmath259 and @xmath263 describe three generations transforming as @xmath264 of @xmath33 . their massless modes transform according to the antisymmetric representation and come in three generations , i.e. @xmath265 it is easy to see that an odd number of generations requires the use of tilted tori as those described in fig . 1 . * open strings localized at the intersection of @xmath259 and @xmath266 provide three generations of @xmath267 , i.e. @xmath268 * the @xmath260 brane being parallel to the @xmath259 in one torus , the massless modes at their intersections will be identified with @xmath28 hypermultiplets that contain the supersymmetric standard model higgs doublets . the supersymmetry breaking messengers arise from strings stretching between brane @xmath259 and @xmath269 on one side and between @xmath260 and @xmath270 on the other side . these branes are parallel ( and separated ) along the second and first torus , respectively , and their intersections contain non - chiral matter with non - supersymmetric masses . we will discuss here a simple example given in table [ table2 ] where the wrapping numbers of different @xmath271-branes @xmath272 in the three tori are listed . it is easy to see that the conditions for cancellation of rr - charge tadpole are satisfied . .__wrapping numbers for the three stacks in the model . _ _ [ cols="^,^,^,^",options="header " , ] the association of any stack of branes @xmath271 ( @xmath273 ) with its orientifold image @xmath274 preserves one of the four supersymmetries , that we label as @xmath275 with @xmath276 , if the angles @xmath277 satisfy one of the corresponding relations : @xmath278 it is easy to check that none of these equations can be simultaneously satisfied by the three set of branes . instead , we will look for ratios of radii @xmath279 that allow one of the relations , chosen to be @xmath280 , to be satisfied by the stacks @xmath281 , @xmath282 and their images . this fixes two of the ratios as functions of the third one . for instance , in the limit of small ratios @xmath283 , the supersymmetric conditions read as @xmath284 as a result , the intersections @xmath285 , @xmath286 and @xmath287 preserve @xmath280 , the intersections @xmath288 and @xmath289 preserve @xmath290 , the intersections @xmath291 preserve @xmath292 , and finally the other intersections involving @xmath67 and/or @xmath293 do not preserve any supersymmetry . at these intersections the lightest states have supersymmetry breaking masses of order : @xmath294 in particular , the states at the intersection @xmath295 could act as messengers ( a number of @xmath296 of @xmath33 ) to generate at one - loop level dirac masses for fermions , gauginos and higgsinos , and at two - loop supersymmetry breaking masses for bosons of the observable sector . note that the gauge symmetry @xmath262 can be broken to the standard model one by a discrete or continuous wilson line without introducing extra fields . unification of gauge couplings is thus guaranteed in such particular constructions . actually , this model realizes the particular @xmath1-brane configuration of ref . @xcite , in which the @xmath262 stack is replaced by two separate brane collections , @xmath102 and @xmath90 , describing strong and weak interactions . moreover , the hypercharge is the linear combination @xmath297 of the two corresponding @xmath19 factors @xmath298 and @xmath299 . the antisymmetric representation of @xmath33 is then decomposed in terms of the quark doublets , the up antiquarks and the right - handed lepton , the latter arising as antisymmetric representations of @xmath102 and @xmath90 , respectively . imposing now @xmath300 , the absence of symmetric representations @xmath301 for the two groups , as well as the absence of antiquark doublets @xmath302 , one finds that the strong and weak branes are parallel in all three planes and thus correspond to a wilson line breaking of the gut group @xmath262 studied above . small @xmath10 deformations of the brane intersection angles provide a simple mechanism to break supersymmetry , which in the @xmath8-dual picture corresponds to the introduction of appropriate combinations of magnetic fluxes . at the effective field theory level , this can be described as fayet - iliopoulos @xmath1-term breaking corresponding to the presence of anomalous @xmath19 factor(s ) . charged scalars localized at their intersections acquire tree - level masses of order @xmath303 . scalars in non - chiral sectors can transmit supersymmetry breaking to the observable sector through gauge mediated loop corrections . however , since the @xmath1-term does not break @xmath13-symmetry no gaugino majorana masses are expected ( at least to the lowest order ) . phenomenologically interesting models can appear when the gauge sector has an extended supersymmetry in which case gauginos can get dirac masses . a phenomenological application of this scenario is given in ref . @xcite where a minimal model based on extended split supersymmetry was shown to be consistent with unification and the presence of dark matter . dirac masses for gauginos and higgsinos where obtained from the @xmath1-breaking assuming the presence of some effective higher - dimensional operators . even if such operators are absent at tree - level , the present work shows how they arise at one - loop . the precise string calculation points out important features of the messenger sector in the simplest realization . in particular , the fact that the messengers arise from strings stretching between branes that are separated in one direction . they correspond to deformations of @xmath28 sectors . a simple toroidal compactification , with an orientifold along one direction , does not allow non - trivial supersymmetric intersecting branes . a vacuum to be considered as starting point for the above mentioned @xmath10 deformation is then missing , so we instead propose a different scenario . we identify a subset of supersymmetric intersecting branes with the observable sector . the other branes lead to some non - supersymmetric intersections . the matter living at such non - chiral intersections will , as in the previous scenario , play the role of the supersymmetry breaking messenger sector . in order to obtain masses hierarchically smaller than the string scale , small ratios of radii are required and they make all intersection angles small . we have presented a gut model to illustrate this possibility . finally note that for both scenarios there are two possibilities from the phenomenological point of view : ( i ) to introduce an @xmath10 deformation breaking supersymmetry in the observable sector at tree - level . ( ii ) to keep the observable sector supersymmetric at tree - level . non - supersymmetric branes such that the matter living on their intersection with the observable branes are non - chiral will transmit supersymmetry breaking by gauge interactions . in case ( ii ) both gauginos , squarks , sleptons and higgses have masses of the same order of magnitude ( as in the usual gauge mediated models ) opening up the possibility of having a supersymmetric spectrum in the tev range . this will give rise to a rich phenomenology accessible at lhc energies . work supported in part by the european commission under the rtn contract mrtn - ct-2004 - 503369 , in part by cicyt , spain , under contracts fpa2001 - 1806 and fpa2002 - 00748 , in part by the intas contract 03 - 51 - 6346 and in part by in2p3-cicyt under contract pth 03 - 1 . we wish to thank a. uranga for useful discussions . k.b . wishes to thank the cern theory unit for hospitality . a.d . and m.q wish to thank the lpthe of `` universit de paris vi et vii '' for hospitality . i.a . and k.b . acknowledge the hospitality of the `` universidad autonoma de barcelona '' and ifae . in this short appendix , we establish our conventions for the modular theta functions and list a few useful properties . the theta functions are defined by : @xmath304 where @xmath305 is the ( complex ) modular parameter of the torus , not to be confused with the world - sheet coordinate used in the text . on the cylinder , this parameter is purely imaginary and in the main text we use the definition @xmath306 . alternatively , the theta functions can be defined as an infinite product : @xmath307 where @xmath308 . they satisfy the following periodicity conditions : @xmath309 which is the reason why they are well suited to describe correlators on the torus . defining @xmath310 we can see from the product representation that @xmath311 in particular @xmath312 a fact that was used repeatedly in the main text . we also need the dedekind eta function , which is defined as @xmath313 it is related to the function @xmath314 by the simple identity @xmath315 finally , the theta functions satisfy the following riemann identity : @xmath316 with @xmath317 let us consider the correlator on the torus with twists of an arbitrary angle @xmath318 . this correlator has a quantum piece @xmath319 and a classical piece which takes into account the contributions from world - sheet instantons : @xmath320 where the sum is over classical solutions of the equations of motion . both pieces are expressed in terms of the so - called cut - differentials , which are holomorphic one - forms on a branched covering of the punctured torus . when expressed as functions of the torus coordinate @xmath321 , they become multi - valued with branch - cut singularities at the location of the punctures . in the present context , where the torus has only two punctures corresponding to the insertions of the two twist fields , these cut - differentials are simply given by : @xmath322 and satisfy @xmath323 where @xmath305 is the complex modulus of the torus . the singularities at the insertions @xmath324 depend on the twist - angle @xmath318 . from these cut - differentials , we construct a period matrix " @xmath325 defined as @xmath326 where the paths @xmath327 , @xmath328 , denote the canonical homology basis of the torus . from @xcite we obtain the quantum part of the correlation function : @xmath329 to determine the classical part of the correlator , we need to evaluate the action for all possible classical solutions of the equations of motion . each contribution will have the form @xmath330 where @xmath331 and @xmath332 are classical solutions , i.e. @xmath331 solves @xmath333 on the punctured torus with appropriate boundary conditions ( i.e. branch cut singularities ) at the location of the punctures . in particular , under parallel transport along the canonical homology basis of the torus , @xmath331 is expected to shift as @xmath334 where @xmath335 are given complex numbers . for instance for a toroidal compactification @xmath336 with @xmath337 when @xmath331 is transported along a closed loop on the world - sheet torus , it must return to its original value modulo a lattice vector of the space - time torus . this implies @xmath338 where the integers @xmath339 and @xmath340 correspond to the winding numbers of the closed string propagating in the loop . the classical part of the correlator is written as @xmath341 where @xmath342 is now expressed in terms of the period matrix ( [ period_matrix ] ) : @xmath343 the two - point correlation function of twist fields on the torus is then given by @xmath344 where the sum over the @xmath345 corresponds to a sum over all possible winding modes . the case we are interested here , the annulus two - point correlation function , is obtained from ( [ twist_correlator ] ) by taking the square root " and replacing the complex modulus @xmath305 by @xmath346 . as a result , the twist correlator on the annulus takes the form @xmath347 where now @xmath348 and @xmath349 is a normalization factor which can depend on the modulus . it is now understood that the twist fields are located on one of the boundaries of the cylinder , @xmath350 with @xmath351 real . the values of @xmath335 must also be appropriately modified . instead of ( [ vi_torus ] ) , we now have v_1 = 2 r_1 m v_2 = i 4 ( r_2 n + ) [ esta ] where @xmath352 are the compactification radii of the space - time torus and @xmath353 is the separation between the @xmath1-branes along the @xmath354 direction . taking the limit @xmath355 , allows to check that this expression is correct and to determine the normalization constant @xmath349 . in this limit , the operator product expansion of the twist fields behaves as : @xmath356 where we have used that @xmath357 is the partition function on the annulus of the complex boson @xmath358 . the factor @xmath359 is the contribution of the oscillators while the sum is over the momentum and winding modes of the string . the same result should be obtained by taking the limit @xmath360 on the right - hand side of ( [ twist_correlator_annulus ] ) . the expression ( [ cut_differentials ] ) for the cut - differential implies that @xmath361 in this limit the components of the period matrix and its inverse can be explicitly evaluated : @xmath362 inserting this in the action ( [ classical_actionc ] ) leads to @xmath363 so that the classical piece of the correlator becomes @xmath364 where the second equality is obtained by a poisson resummation . using ( [ theta1_property ] ) and ( [ explicit_period_matrix ] ) , the limit @xmath360 of the quantum part is @xmath365 inserting ( [ explicit_classical_part ] ) and ( [ explicit_quantum_part ] ) into ( [ twist1 ] ) and comparing with ( [ limit_lhs ] ) shows that both expressions agree to each other provided that @xmath366 . n. arkani - hamed and s. dimopoulos , jhep * 0506 * ( 2005 ) 073 [ arxiv : hep - th/0405159 ] ; g. f. giudice and a. romanino , nucl . phys . b * 699 * ( 2004 ) 65 [ erratum - ibid . b * 706 * ( 2005 ) 65 ] [ arxiv : hep - ph/0406088 ] ; n. arkani - hamed , s. dimopoulos , g. f. giudice and a. romanino , nucl . b * 709 * ( 2005 ) 3 [ arxiv : hep - ph/0409232 ] . i. antoniadis and s. dimopoulos , nucl . b * 715 * ( 2005 ) 120 [ arxiv : hep - th/0411032 ] . c. angelantonj and a. sagnotti , phys . * 371 * ( 2002 ) 1 [ erratum - ibid . * 376 * ( 2003 ) 339 ] [ arxiv : hep - th/0204089 ] . c. bachas , arxiv : hep - th/9503030 . c. angelantonj , i. antoniadis , e. dudas and a. sagnotti , phys . b * 489 * ( 2000 ) 223 [ arxiv : hep - th/0007090 ] . m. berkooz , m. r. douglas and r. g. leigh , nucl . phys . b * 480 * ( 1996 ) 265 [ arxiv : hep - th/9606139 ] . r. blumenhagen , l. goerlich , b. kors and d. lust , jhep * 0010 * ( 2000 ) 006 [ arxiv : hep - th/0007024 ] . i. antoniadis , a. delgado , k. benakli , m. quiros and m. tuckmantel , arxiv : hep - ph/0507192 . i. antoniadis , e. gava , k. s. narain and t. r. taylor , nucl . b * 511 * ( 1998 ) 611 [ arxiv : hep - th/9708075 ] ; s. kachru and j. mcgreevy , phys . d * 61 * ( 2000 ) 026001 [ arxiv : hep - th/9908135 ] ; m. mihailescu , i. y. park and t. a. tran , phys . d * 64 * ( 2001 ) 046006 [ arxiv : hep - th/0011079 ] ; e. witten , jhep * 0204 * ( 2002 ) 012 [ arxiv : hep - th/0012054 ] ; d. cremades , l. e. ibanez and f. marchesano , jhep * 0207 * ( 2002 ) 009 [ arxiv : hep - th/0201205 ] . i. antoniadis , k. s. narain and t. r. taylor , nucl . b * 729 * ( 2005 ) 235 [ arxiv : hep - th/0507244 ] . p. j. fox , a. e. nelson and n. weiner , jhep * 0208 * ( 2002 ) 035 [ arxiv : hep - ph/0206096 ] . see for example a. m. uranga , class . quant . grav . * 22 * ( 2005 ) s41 ; and references therein . l. e. ibanez , f. marchesano and r. rabadan , jhep * 0111 * ( 2001 ) 002 [ arxiv : hep - th/0105155 ] ; m. cvetic , g. shiu and a. m. uranga , nucl . b * 615 * ( 2001 ) 3 [ arxiv : hep - th/0107166 ] . r. blumenhagen , l. gorlich and b. kors , jhep * 0001 * ( 2000 ) 040 [ arxiv : hep - th/9912204 ] ; g. aldazabal , s. franco , l. e. ibanez , r. rabadan and a. m. uranga , j. math . * 42 * ( 2001 ) 3103 [ arxiv : hep - th/0011073 ] . g. f. giudice and r. rattazzi , phys . * 322 * ( 1999 ) 419 [ arxiv : hep - ph/9801271 ] . j. j. atick and a. sen , nucl . b * 286 * ( 1987 ) 189 . j. j. atick , l. j. dixon , p. a. griffin and d. nemeschansky , nucl . b * 298 * ( 1988 ) 1 . r. blumenhagen , d. lust and s. stieberger , jhep * 0307 * ( 2003 ) 036 [ arxiv : hep - th/0305146 ] ; t. p. t. dijkstra , l. r. huiszoon and a. n. schellekens , phys . lett . b * 609 * ( 2005 ) 408 [ arxiv : hep - th/0403196 ] and nucl . b * 710 * ( 2005 ) 3 [ arxiv : hep - th/0411129 ] ; a. n. schellekens , private communication .

in the standard model of percolation on multilayer networks , a node is functioning only if its copies in all layers are simultaneously functioning . according to this model , a multilayer network becomes more and more fragile as the number of layers increases . in this respect , the addition of a new layer of interdependent nodes to a preexisting multilayer network will never improve its robustness . whereas such a model seems appropriate to understand the effect of interdependencies in the simplest scenario of a network composed of only two layers , it may seem not reasonable for real systems where multiple network layers interact one with the other . it seems in fact unrealistic that a real system , such a living organism , evolved , through the development of multiple layers of interactions , towards a fragile structure . in this paper , we introduce a model of percolation where the condition that makes a node functional is that the node is functioning in at least two of the layers of the network . the model reduces to the standard percolation model for multilayer networks when the number of layers equals two . for larger number of layers , however , the model describes a scenario where the addition of new layers boosts the robustness of the system by creating redundant interdependencies among layers . we prove this fact thanks to the development of a message - passing theory able to characterize the model in both synthetic and real - world multilayer graphs .

1610.05378

multilayer networks @xcite are emerging as a powerful paradigm for describing complex systems characterized by the coexistence of different types of interactions . multilayer networks represent an appropriate descriptive model for real networked systems in disparate contexts , such as social @xcite , technological @xcite and biological systems @xcite . for example , global infrastructures are formed by several interdependent networks , such as power grids , water supply networks , and communication systems , and studying their properties require to account for the presence of such interdependencies @xcite . cell function and/or malfunction ( yielding diseases ) can not be understood if the information on the different nature of the interactions forming the interactome ( protein - protein interactions , signaling , regulation ) are not integrated in a general multilayer scenario @xcite . similarly , the complexity of the brain is encoded in the different nature of the interactions existing at the functional and the structural levels @xcite . a multilayer networks is composed of a set of networks forming its layers @xcite . nodes can be connected within and across layers . it has been shown that multilayer networks are much more fragile than isolated networks just because of the presence of interdependencies among the layers of the system @xcite . in particular , the fragility of the system increases as the number of layers increases @xcite . such a feature has an intuitive explanation . in the standard percolation model for multilayer networks , the probability that a node is damaged equals to the probability that at least one of its interdependent nodes is damaged . as the number of layers increases , the probability of individual failures grows thus making the system more fragile . this scenario leads , however , to the conundrum : if the fragility of a system is increased by the number of layers of interactions , why are there so many real systems that display multiple layers of interactions ? further , the addition of new layers of interactions in a preexisting multilayer network has generally a cost , so it does nt seem reasonable to spend resources just to make the system less robust . the purpose of the current paper is to provide a potential explanation by introducing a new model for percolation in networks composed of multiple interacting layers . in the model , we will assume that a node is damaged only if all its interdependent nodes are simultaneously damaged . the model is perfectly equivalent to the standard one when the number of layer equals two . additional layers , however , provide the system with redundant interdependencies , generating backup mechanisms against the failure of the system , and thus making it more robust . the robustness of multilayer networks in presence of redundant interdependencies is here investigated using a message - passing theory @xcite ( also known as the cavity method ) . we build on recent advances obtained in standard interdependent percolation theory @xcite to propose a theory that is valid for multilayer networks with link overlap @xcite as long as the multilayer network is locally tree - like . this limitation is common to all message - passing approaches for studying critical phenomena on networks . corrections have been recently proposed @xcite on single networks to improve the performace of message - passing theory and similar approximations valid for loopy multilayer networks might be envisaged in the future . we consider a multilayer network @xmath0 composed of @xmath1 layers @xmath2 with @xmath3 . every layer contains @xmath4 nodes . exactly one node with the same label appears in every individual layer . nodes in the various layers sharing a common label are called _ replica nodes _ , and they are considered as interdependent on each other @xcite . nodes in the network are identified by a pair of labels @xmath5 , with @xmath6 and @xmath3 , the first one indicating the index of the node , and the second one standing for the index of the layer . for every node label @xmath7 , the set of replica nodes is given by the @xmath1 nodes corresponding to pairs of labels @xmath8 with with @xmath3 ( see figure [ fig : multiplex2 ] ) . when at least two replica nodes @xmath5 and @xmath9 are connected to two corresponding replica nodes @xmath10 and @xmath11 we say that the multilayer network displays link overlap . given a multilayer network as described above , we consider a percolation model where some of the nodes are initially damaged . we assume that the interdependencies are redundant , i.e. , every node can be active only if at least one its interdependent nodes is also active . we refer to this model as `` redundant percolation model . '' as an order parameter for the model , we define the so - called redundant mutually connected giant component ( rmcgc ) . the nodes that belong to the rmcgc can be found by following the algorithm : * the giant component of each layer @xmath12 is determined , evaluating the effect of the damaged nodes in each single layer ; * _ every replica node that has no other replica node in the giant component of its proper layer is removed from the network and considered as damaged _ ; * if no new damaged nodes are found at step ( ii ) , then the algorithm stops , otherwise it proceeds , starting again from step ( i ) . the set of replica nodes that are not damaged when the algorithm stops belongs to the rmcgc . the main difference with the standard percolation model @xcite on multilayer networks and the consequent definition of mutually connected giant component ( mcgc ) is that step ( ii ) must be substituted with `` every replica node that has at least a single replica node not in the giant component of its proper layer is removed from the network and considered as damaged , i.e. , if a replica node is damaged all its interdependent replica nodes are damaged '' @xcite . in particular , the rmcgc and the mcgc are the same for @xmath13 layers , but they differ as long as the number of layers @xmath14 . in the latter case , the rmcgc naturally introduces the notion of redundancy among interdependent nodes . as we will see in the following , the main effect of redundancy is to let the robustness of the system increases as the number of layers increases . layers , and @xmath15 nodes is shown . every node @xmath7 has @xmath16 interdependent replica nodes @xmath5 with @xmath17 . in this figure , triplets of replica nodes are also identified by their color . ] we assume that interactions within each layer @xmath12 are described by elements @xmath18}$ ] of the adjacency matrix of the layer , indicating whether the replica nodes @xmath5 and @xmath10 are connected ( @xmath18}=1 $ ] ) or not ( @xmath18}=0 $ ] ) in layer @xmath12 . additionally , we consider a specific realization of the initial damage to the replica nodes indicated by the set @xmath19 . the generic element @xmath20 indicates that the replica node @xmath5 has been initially damaged , whereas @xmath21 indicates that the replica node @xmath5 has not been initially damaged . under these conditions , as long as the multilayer network is locally treelike , the following message - passing algorithm identifies the replica nodes that are in the rmcgc . each node @xmath7 sends to a neighbor @xmath22 a set of messages @xmath23}$ ] in every layer @xmath12 where node @xmath7 is connected to node @xmath22 , i.e. , with @xmath24}_{ij}=1 $ ] . these messages indicate whether ( @xmath23}=1 $ ] ) or not ( @xmath23}=0 $ ] ) node @xmath7 connects node @xmath22 to the rmcgc with links belonging to layer @xmath12 . the message @xmath23}=1 $ ] if and only if all the following conditions are met : * node @xmath7 is connected to node @xmath22 in layer @xmath12 , and both nodes @xmath5 and node @xmath10 are not damaged , i.e. , @xmath25}_{ij}=1 $ ] ; * node @xmath7 is connected to the rmcgc through at least one node @xmath26 in layer @xmath12 ; * node @xmath7 belongs to the rmcgc assuming that also node @xmath22 belongs to the rmcgc . this conditions is satisfied if and only if , assuming that node @xmath22 belongs to the rmcgc , node @xmath7 is connected in at least two layers to the rmcgc . if the previous conditions are not simultaneously met , then @xmath23}=0 $ ] . put together , the former conditions lead to the algorithm for the messages @xmath27}$ ] n_ij^[]=(v_i j,2 ) a^[]_ijs_js_i . [ m1 ] here @xmath28 indicates the set of nodes that are neighbor of node @xmath7 in layer @xmath12 . the term @xmath29}_{\ell\to i}\right)$ ] therefore will equal one if at least one message is arriving to node @xmath7 from a neighboring node @xmath30 , while it will be equal to zero , otherwise . @xmath31 for @xmath32 and @xmath33 , otherwise . @xmath34 indicates in how many layers node @xmath7 is connected to the rmcgc assuming that node @xmath22 also belongs to the rmcgc , i.e. , v_ij&=&_=1^m . [ v ] therefore @xmath35 indicates the number of initially undamaged replica nodes @xmath5 that either receive at least one positive messages from nodes @xmath36 or are connected to the undamaged replica nodes @xmath10 . finally , the replica node @xmath5 belongs to the rmcgc if ( i ) it is not damaged , ( ii ) it is connected to the rmcgc in layer @xmath12 , and ( iii ) it receives at least another positive message in a layer @xmath37 . these conditions are summarized by _ i&=&s_i(1-_n_(i)(1-n^[]_i ) ) + & & \{1-_ } . [ s1 ] the average number @xmath38 of replica nodes belonging to the rmcgc is computed as s=_=1^m_i=1^n _ i. [ s ] the system of eqs . ( [ m1 ] ) , ( [ v ] ) , ( [ s1 ] ) , and ( [ s ] ) represents a complete mathematical framework to estimate the average size of the rmcgc for a given network and a given initial configuration of damage . the solution can be obtained by first iterating eqs . ( [ m1 ] ) and ( [ v ] ) to obtain the values of the messages @xmath27}$ ] . those values are then plugged into eqs . ( [ s1 ] ) to compute the values of the variables @xmath39 , and finally these variables are used into eq . ( [ s ] ) to estimate the average size of the rmcgc . we stress that , being valid for a given network and for a given configuration of damage , the values of the variables @xmath27}$ ] and @xmath39 are either @xmath40 or @xmath41 . the variables @xmath34 can assume instead integer values in the range @xmath42 $ ] . the mathematical framework works properly also in presence of edge overlap among layers . this is an important feature that can change dramatically change the robustness properties of multilayer networks @xcite . in absence of link overlap , a given pair of nodes @xmath7 and @xmath22 may be linked exclusively along a single layer @xmath12 . nontrivial messages potentially different from zero will therefore exist only on a specific layer for every pair of connected nodes @xmath7 and @xmath22 . it can be easily seen that the message - passing eqs . ( [ m1 ] ) and ( [ v ] ) reduce to n_ij^[]&=&s_is_ja_ij^ [ ] + & & \{1-_}. [ m2 ] we further notice that in this situation the result of the message - passing algorithm does not change if we consider messages that depend exclusively on the state @xmath43 of the node @xmath7 that sends the message . even if we drop the factor @xmath44 in eq . ( [ m2 ] ) , the message will be allowed anyways to propagate further at the next iteration step , if the replica node @xmath45 is not initially damaged . therefore , we can further simplify eq . ( [ m2 ] ) and consider n_ij^[]&=&s_ia_ij^ [ ] + & & \{1-_}. [ m3 ] eqs . ( [ m3 ] ) replace eqs . ( [ m1 ] ) and ( [ v ] ) in the case of a multilayer network without link overlap . the rest of the framework is identical , so that eqs . ( [ s1 ] ) and ( [ s ] ) remain unchanged . ( [ m3 ] ) , ( [ s1 ] ) , and ( [ s ] ) determine the average size of the rmcgc in a multilayer network without link overlap for a given realization of the initial damage @xmath19 . these equations can be , however , extended to make predictions in the case of a random realization of the initial damage when the replica nodes are damaged independently with probability @xmath46 , i.e , . such that the the initial damage @xmath19 is a random configuration obeying the probability distribution ( \{s_i})=_i=1^n_=1^m p^s_i(1-p)^1-s_i . [ ps ] to this end , we denote the probability that node @xmath7 sends a positive message to node @xmath22 in layer @xmath12 by @xmath47}={\left\langle{n_{i\to j}^{[\alpha]}}\right\rangle}$ ] , and the probability that the replica node @xmath8 belongs to the rmcgc by @xmath48 . the message - passing algorithm determining the values of @xmath47}$ ] and @xmath49 is given by _ ij^[]&=&a_ij^[]p + & & \{1-_ } , + _ i&=&p(1-_n_(i)(1-^[]_i ) ) + & & \{1-_ } . this algorithm can be applied to a given network , and provides the average number of replica nodes @xmath38 belonging to the rmcgc for a random realization of the initial damage obeying eq . specifically the value of @xmath51 is related to @xmath49 by = _ i=1^n _ = 1^m _ i. a multilayer network where every layer is a sparse network generated according to the configuration model is a major example of a multilayer network without link overlap in the limit of large network sizes . it is therefore natural and important to characterize the rmcgc in this case . we assume that every network layer @xmath2 is a random graph taken from the probability distribution ^[](g_)=_i=1^n ( k_i^[],_j=1^n a_ij^ [ ] ) , [ ens ] where @xmath52}$ ] indicates the preimposed degree of node @xmath7 in layer @xmath12 , @xmath53 if @xmath54 and @xmath55 , otherwise , and @xmath56 is the normalization factor indicating the total number of networks in the ensemble . averaging over the network ensemble allows us to translate the message - passing equations into simpler expressions for the characterization of the percolation transition . let us consider a random multilayer network obeying the probability of eq . ( [ ens ] ) , and a random realization of the initial damage described by the probability of eq . ( [ ps ] ) . the average message in layer @xmath12 , namely @xmath57}=1}\right\rangle}$ ] , and the average number of replica nodes of layer @xmath58 that are in the rmcgc , denoted by @xmath59 , obey the equations s_&=&p_*k*p(*k * ) + & & \{1-_ } , + s_&=&p_*k*p(*k * ) , + & & \{1-_ } , [ su1 ] where @xmath60 indicates the probability that a generic node @xmath7 has degrees @xmath61 , i.e. @xmath62}_i , k^{[2]}_i,\ldots , k^{[m]}_i)=(k^{[1]},k^{[2]},\ldots , k^{[m]})$ ] . if there are no correlations between the degrees of a node in different layers , the degree distribution @xmath60 can be factorized as p(*k*)=_p^[](k^ [ ] ) , [ eq : pk ] where @xmath63}(k)$ ] is the degree distribution in layer @xmath12 . in this case , @xmath64 can be expressed in terms of the generating function of the degree distribution in each layer . specifically , we have s_&=&p + & & \{1-_ } , + s_&=&p + & & \{1-_ } , [ su2 ] where the generating functions @xmath65}(z)$ ] and @xmath66}(z)$ ] of the degree distribution @xmath63}(k)$ ] of layer @xmath12 are given by h_0^[](x)&=&_kp^[](k ) x^k , + h_1^[](x)&=&_kp^[](k ) x^k-1.[slmcgc ] finally the average number @xmath38 of replica nodes in the rmcgc is given by s= _ s_. if we consider the case of equally distributed poisson layers with average degree @xmath67 , we have that eq . ( [ eq : pk ] ) is p^[](k)=z^ke^-z for every layer @xmath68 . then , using eqs . ( [ su2 ] ) , one can show that @xmath69 , @xmath70 , and @xmath38 is determined by the equation s = p(1-e^-zs)\{1-[1-p+pe^-zs]^m-1}. [ p1 ] this equation has always the trivial solution @xmath71 . in addition , a nontrivial solution @xmath72 indicating the presence of the rmcgc , emerges at a hybrid discontinuous transition characterized by a square root singularity , on a line of points @xmath73 , determined by the equations h_z , p(s_c)&=&0 , + .|_s = s_c&=&0 , where h_z , p(s)&=&s - p(1-e^-zs ) + & & \{1-[1-p+pe^-zs]^m-1}=0 . for @xmath74 there is a rmcgc , for @xmath75 there is no rmcgc . the entity of the discontinuous jump at @xmath76 in the fraction @xmath38 of replica nodes in the rmcgc is given by @xmath77 . the percolation threshold @xmath78 as a function of the average degree @xmath67 of the network is plotted in figure @xmath79 for @xmath80 . it is shown that as the number of layers @xmath1 increases the percolation threshold decreases for every value of the average degree @xmath67 . additionally also the discontinuous jump @xmath81 decreases as the number @xmath1 of layer increases for very given average degree @xmath67 ( see figure @xmath82 ) . therefore as the number of layers increases the multilayer networks becomes more robust . is plotted versus the average degree @xmath67 of each layer for poisson multilayer networks with @xmath80 layers indicated respectively with with blue solid , red dashed , green dot - dashed and orange dotted lines . ] of the rmcgc at the percolation threshold @xmath76 , is plotted versus the average degree @xmath67 of each layer for poisson multilayer networks with @xmath80 layers indicated respectively with blue solid , red dashed , green dot - dashed and orange dotted lines . ] in this section , we compare the robustness of multilayer networks in presence of ordinary interdependencies and in presence of redundant interdependencies . to take a concrete example , we consider the case of a multilayer network with @xmath1 poisson layers , each layer having the same average degree @xmath67 . in this case the fraction @xmath38 of replica nodes in the rmcgc is given by the solution of eqs . ( [ p1 ] ) while the fraction of replica nodes in the mcgc is given by s=(1-e^-zs)^m . [ o ] in eq . ( [ o ] ) , it is assumed that every replica node @xmath5 of a given node @xmath7 is damaged simultaneously ( with probability @xmath83 ) . on the contrary , in presence of redundant interdependencies it is natural to assume that the initial damage is inflicted to each replica node independently ( with probability @xmath84 ) . therefore , in order to compare the robustness of the multilayer networks in presence and in absence of redundant interdependencies , we set @xmath85 , i.e. , replica nodes are not initially damaged , and compare the critical value of the average degree @xmath86 at which the percolation transition occurs respectively for the rmcgc and for the mcgc . additionally we will characterize also the size @xmath87 of the jump in the size of the rmcgc and the mcgc at the percolation transition . in fig . [ fig : comp_m ] , we display the values of @xmath88 and @xmath89 as a function of the number of layers @xmath1 for the rmcgc and the mcgc . for @xmath13 , the two models give the same results as they are identical . for @xmath90 , differences arise . in presence of redundant interdependencies , multilayer networks become increasingly more robust as the number @xmath1 of layers increases . this phenomenon is apparent from the fact that the rmcgc emerges for multilayer networks with an average degree of their layers @xmath88 which decreases as the number of layers @xmath1 increases . on the contrary , in ordinary percolation the value of @xmath88 for the emergence of the mcgc is an increasing function of @xmath1 . additionally , the size of the discontinuous jumps @xmath89 at the transition point decreases with @xmath1 for the rmcgc , while increases with @xmath1 for the mcgc showing that the avalanches of failures have a reduced size for the rmcgc . of the average degree as a function of the number of network layers @xmath1 . results for the rmcgc model are displayed as red diamonds . results for the mcgc model are denoted by blue triangles . ( b ) height of the jump @xmath89 at the transition point as a function of the number of network layers . ] in this section , we compare the results obtained with eqs . ( [ m1 ] ) , ( [ v ] ) , ( [ s1 ] ) , and ( [ s ] ) on a single instance of damage with the predictions the message - passing algorithm described in eq . ( [ su2 ] ) characterizing the size @xmath38 of the rmcgc in an ensemble of networks . specifically , we consider the case of a multilayer network with @xmath16 poisson layers with the same average degree @xmath67 . in order to draw the percolation diagram for single instances of initial damage as a function of the probability of damage @xmath46 , we associate each replica node @xmath5 with a random variable @xmath91 drawn from a uniform distribution and we set s_i=\ { ccc 1 & & r_ip + 0 & & r_i > p . [ rs ] fig . [ fig : mp_single ] displays the comparison between the two approaches , showing an almost perfect agreement between them . poisson layers poisson with average degree @xmath67 and no link overlap , and the message - passing results over single network realization and given configuration damage . we consider different values of the average degree @xmath92 . points indicate results of numerical simulations : blue circles ( @xmath93 ) , red squares ( @xmath94 ) , green diamonds ( @xmath95 ) , and orange triangles ( @xmath96 ) . message - passing predictions are denoted by lines with the same color scheme used for numerical simulations . simulations results are performed on a single instance of a multilayer network with @xmath97 nodes . ] additionally in fig . [ fig : single_average ] , we compare simulation results averaged over several realizations of the initial damage and several instances of the multilayer network model with the theoretical predictions given by the numerical solution of eqs . ( [ su1])-([p1 ] ) , obtaining a very good agreement . , but for averages over @xmath98 instances of the multilayer network model and configurations of random initial damage . ] in isolated networks , two nodes can be either connected or not connected . in multilayer networks instead , the complexity of the structure greatly increases as the ways in which a generic pair of nodes can be connected is given by @xmath99 possibilities . a very convenient way of accounting for all the possibilities with a compact notation is to use the notion of multilink among pairs of nodes @xcite . multilinks @xmath100},m^{[2]},\ldots , m^{[m]}\right)$ ] with @xmath101}=0,1 $ ] , describe any of the possible patterns of connections between pairs of nodes in a multilayer network with @xmath1 layers . specifically , @xmath101}=1 $ ] indicates that a connection exists in layer @xmath12 , whereas @xmath101}=0 $ ] indicates that the connection in layer @xmath12 does not exists . in particular , we can say that , in a multilayer network with @xmath1 layers , two nodes @xmath7 and @xmath22 are connected by the multilink _ ij=(a_ij^[1],a_ij^[2 ] , , a_ij^[m ] ) . in order to distinguish the case in which two nodes are not connected in any layer with the case in which in at least one layer the nodes are connected , we distinguish between the trivial multilink @xmath102 and the nontrivial multilinks @xmath103 . the trivial multilink @xmath102 indicates the absence of any sort of link between the two nodes . using the concept of multilinks , one can define multiadjacency matrices @xmath104 whose element @xmath105 indicates whether ( @xmath106 ) or not @xmath107 a node @xmath7 is connected to node @xmath22 by a multilink @xmath108 . the matrix elements @xmath105 of the multiadjacency matrix @xmath104 are given by a_ij^=_=1^m(m^[],a_ij^ [ ] ) . using multiadjacency matrices , it is straightforward to define multidegrees @xcite . the multidegree of node @xmath7 indicated as @xmath109 is the sum of rows ( or columns ) of the multiadjacency matrix @xmath104 , i.e. , k_i^=_j a_ij^ , and indicates how many multilinks @xmath108 are incident to node @xmath7 . using a multidegree sequence @xmath110 , it is possible to build multilayer network ensembles that generalize the configuration model . this way , overlap of links is fully preserved by the randomization of the multilayer network . these ensembles are specified by the probability @xmath111 attributed to every multilayer network @xmath112 of the ensembles , where @xmath111 is given by ( ) = _ i=1^n _ ( k_i^,_j=1^n a_ij^ ) , with @xmath113 normalization constant equal to the number of multilayer networks with given multidegree sequence . our goal here is to generalize the message - passing algorithm already given by eqs . ( [ m1 ] ) , ( [ v ] ) , ( [ s1 ] ) , and ( [ s ] ) for a generic single instance of a multilayer network and single realization of initial damage to the cases of ( i ) random multilayer networks with given multidegree sequence and/or ( ii ) random realizations of the initial damage . the extensions for both cases has been already considered for the case of multilayer networks without link overlap . in presence of link overlap , however , the approach is much more cumbersome . for two nodes @xmath7 and @xmath22 in fact , the messages @xmath114}$ ] given by eq . ( [ m1 ] ) and sent from node @xmath7 to node @xmath22 over the different layers @xmath115 are correlated because they all depend on the value of the variable @xmath34 given by eq . ( [ v ] ) . such correlations require particular care when averaging the messages to treat the percolation transition for random initial damages . similar technical challenges are also present in the treatment of the standard mcgc model where interdependencies are not redundant @xcite . in presence of redundant interdependencies there is an additional precaution that needs to be taken . in fact , the messages @xmath114}$ ] are explicitly dependent on the state of all replicas @xmath11 of node @xmath22 . this state is indicated by the variables @xmath116 where @xmath117 specifies whether the replica node @xmath11 is initially damaged or not . as a consequence of this property , when averaging over random realizations of initial damage , message - passing equations are written in terms of the messages @xmath118 explicitly accounting for the probability that node @xmath7 is sending to node @xmath22 the set of messages @xmath119},n_{i\to j}^{[2]}\ldots n_{i\to j}^{[\alpha ] } , \ldots n_{i\to j}^{[m]})$ ] , given that node @xmath22 is in state @xmath120 and node @xmath7 and node @xmath22 are connected by a multilink @xmath121 . we have derived these equations for a general multilayer network with @xmath1 layers . however , the message - passing algorithm has a very long expression . to make the paper more readable , we decided to place the exact treatment of the general case in the sm , and consider here only the special case of ensembles of random multilayer networks with overlap . for these ensembles in fact , the message - passing equations are written in terms of average messages sent between nodes with given multilinks @xmath108 , i.e. , @xmath122 , and the equations greatly simplify . two specific cases of multilayer network ensembles are discussed below , for the cases of @xmath13 and @xmath16 layers . in this case , every replica node is in the rmcgc if and only if also its interdependent node in the other layer is in the rmcgc . therefore , the only messages that are different from zero are the messages @xmath123 sent to nodes @xmath22 in state @xmath124 . specifically , we consider the case of a random multilayer network with poisson multidegree distributions characterized by the averages & = & z_2 , + & = & = z_1 . the messages @xmath123 only depend on the multiplicity of overlap of the multilinks @xmath108 and @xmath125 given respectively by & = & _ = 1^m m^ [ ] , + & = & _ = 1^m n^ [ ] . the fraction @xmath38 of replica nodes in the rmcgc is determined by the variables x_,=s^,(_j=(1,1 ) ) . the value of @xmath126 indicates the probability that node @xmath7 to sends a message @xmath127 to its neighbor @xmath22 with @xmath124 connected by a multilink @xmath128 . this fact occurs if and only if node @xmath7 has both replica nodes that are not initially damaged ( which occurs with probability @xmath129 ) and if at least one positive message in each layer @xmath12 is reaching node @xmath7 from neighbors different from @xmath22 . the value of @xmath130 indicates the probability that node @xmath7 sends a message @xmath131 to its neighbor @xmath22 with @xmath124 connected by a multilink @xmath132 or equivalently sends a message @xmath133 to its neighbor @xmath22 with @xmath124 connected by a multilink @xmath134 . this fact occurs if and only if node @xmath7 has both replica nodes that are not initially damaged ( which occurs with probability @xmath129 ) and if at least one positive message in each layer @xmath12 is reaching node @xmath7 from neighboring nodes different from @xmath22 . the latter is a necessary condition to have @xmath135 . the value @xmath136 indicates the probability that node @xmath7 is sending a message @xmath131 to its neighbor @xmath22 in state @xmath124 and connected by a multilink @xmath128 or equivalently sends a message @xmath133 to its neighbor @xmath22 in state @xmath124 and connected by a multilink @xmath128 . this fact occurs if only if node @xmath7 has both replica nodes that are not initially damaged ( which occurs with probability @xmath129 ) and if at least one positive message is reaching node @xmath7 in the layer for which @xmath137}=1 $ ] from neighbors different from @xmath22 and no positive message is reaching node @xmath7 in the layer where @xmath137}=1 $ ] from neighboring nodes different from node @xmath22 . finally , @xmath38 is the probability that a replica node @xmath138 is in the rmcgc which implies that ( i ) it is not initially damaged , ( ii ) its replica node in the other layer is not initially damaged , and ( iii ) at least one positive message reaches node @xmath7 in both layers . the values of the variables @xmath139 and @xmath38 are therefore determined by the following set of equations @xmath140 \\ x_{2,1}= & p^2\left[e^{-z_1x_{1,1}-z_2(x_{2,2 } + x_{2,1})}\right . \\ & \left . -e^{-2z_1x_{1,1}-z_2(x_{2,2}+2x_{2,1})}\right ] \\ s= & x_{1,1}=x_{2,2 } \end{array } \;.\ ] ] these equations are the same equations as those that determine the value of the mcgc as long as the fact that the damage in each replica node is independent is taken into account , which can be done by making the substitution @xmath141 @xcite . we consider now the case of a random multilayer network with @xmath16 layers . the network has poisson multidegree distributions and averages given by & = & z_3 , + & = & = = z_2 , + & = & = = z_1 . in this case , the messages @xmath142 only depend on the multiplicity of overlap of the multilinks @xmath108 and @xmath125 and the number of layers where @xmath143 and @xmath101}=1 $ ] . therefore , messages depend only on & = & _ = 1^m m^ [ ] , + & = & _ = 1^m n^ [ ] , + & = & _ = 1^m s_jm^ [ ] . $ ] in a multilayer network with @xmath16 layers and @xmath144 , @xmath145 and @xmath146 . filled circles indicate initially undamaged replica nodes @xmath21 , whereas empty circles indicate initially damaged replica replica nodes @xmath20 . the message are sent along the direction indicated by the arrows . a solid line reaching node @xmath7 in layer @xmath12 indicates that at least one positive message is reaching node @xmath7 from nodes different from node @xmath22 in layer @xmath12 . dotted lines joining node @xmath7 in layer @xmath12 indicate that no positive message reaches node @xmath7 from nodes different from node @xmath22 in layer @xmath12 . a solid ( dotted ) line between node @xmath7 and node @xmath22 in layer @xmath12 indicates @xmath101}=1 $ ] ( @xmath147}=0 $ ] ) . ] the fraction of replica nodes in the rmcgc @xmath38 is determined by the variables x_,^()=s^,(_j ) . let us explicitly describe the equations that one of these variables needs to satisfy , and introduce a symbolic way to describe the equations . specifically , we consider @xmath148 as the probability @xmath149 that a node @xmath7 , connected to a node @xmath22 by a multilink @xmath150 , sends to node @xmath22 a message @xmath151 provided that node @xmath22 is in the state @xmath152 ( see fig . [ fig : diagram_example ] ) . this probability is equal to the sum of ( i ) the probability that node @xmath7 is in the state @xmath153 [ which occurs with probability @xmath154 and it sends the message @xmath151 to node @xmath22 and ( ii ) the probability that node @xmath7 is in the state @xmath155 ( which occur with probability @xmath156 ) and sends the same message to node @xmath22 . node @xmath7 sends the message @xmath151 only if the following conditions are met : * if node @xmath7 is in the state @xmath153 , node @xmath7 must receive at least one positive message from nodes different from node @xmath22 in layers @xmath157 and @xmath158 . * if node @xmath7 is in the state @xmath155 , node @xmath7 must receive at least one positive message from nodes different from node @xmath22 in layers @xmath157 and @xmath158 and must not receive any positive message from nodes different from node @xmath22 in layer @xmath159 . these requirements are summarized by the diagram of fig . [ fig : diagram_example ] . diagrams that describe the equations to determine the value of all the other variables @xmath160 are presented in fig . [ fig : diagrams ] . these equations read as x_3,3^(3)&=&p^3 + x_3,2^(3)&=&p^2(1-p)+p^3 + x_2,2^(2)&=&p^2(1-p)+p^3 + x_3,2^(2)&=&x_2,2^(2 ) + x_1,1^(1)&=&2p^2(1-p ) + & & + p^3 + x_2,1^(2)&=&p^2(1-p)+p^2(1-p ) + & & + p^3 + x_2,1^(1)&=&x_11^(1 ) + x_3,1^(2)&=&x_2,1^(2 ) + x_3,1^(3)&=&2p^2(1-p)+p^3 + s&=&x_1,1^(1 ) [ m_3l ] where h_1,3&=&e^-z_1x_1,1^(1)-z_2(2x_2,2^(2)+2x_2,1^(2))-z_3(x_3,3^(3)+2x_3,2^(3)+x_3,1^(3 ) ) + h_2,3&=&e^-2z_1x_1,1^(1)-z_2(3x_2,2^(2)+4x_2,1^(2))-z_3(x_3,3^(3)+3x_3,2^(3)+2x_3,1^(3 ) ) + h_3,3&=&e^-3z_1x_1,1^(1)-z_2(3x_2,2^(2)+6x_2,1^(2))-z_3(x_3,3^(3)+3x_3,2^(3)+3x_3,1^(3 ) ) + h_1,2&=&e^-z_1x_1,1^(1)-z_2(x_2,2^(2)+x_2,1^(2)+x_2,1^(1))-z_3(x_3,2^(2)+x_3,1^(2 ) ) + h_2,2&=&e^-2z_1x_1,1^(1)-z_2(x_2,2^(2)+2x_2,1^(2)+2x_2,1^(1))-z_3(x_3,2^(2)+2x_3,1^(2 ) ) . we note that , in absence of overlap , i.e. , for @xmath161 , @xmath162 and @xmath163 , eqs . ( [ m_3l ] ) reduce eqs . ( [ p1 ] ) . by defining a suitable order of the variables @xmath160 , it is possible to introduce a vector @xmath164 whose elements are the variables @xmath160 , and rewrite the eqs . ( [ m_3l ] ) in a matrix form as = * g*(*x * ) . [ mg ] the hybrid discontinuous phase transition can be found by imposing that the system of eqs . ( [ mg ] ) is satisfied together with the condition that the determinant of the jacobian @xmath165 of @xmath166 equals one , that is = * g*(*x * ) = 1 . layers and poisson multidegree distribution with @xmath144 , @xmath145 and @xmath146 . we consider here a single network instance and a given configuration of damage . data are shown for @xmath167 ( blue ) , @xmath168 , ( red ) , and @xmath169 ( green ) . symbols stand for results from numerical simulations , whereas lines represent results for the numerical solution of the message - passing equations . simulations results are performed on networks with @xmath97 nodes . ] , with the difference that results of the simulations stand for averages obtained over @xmath98 instances of the multilayer network models and configurations of the initial damage . ] we perform a comparison between results obtained from the solution of eqs . ( [ m1 ] ) , ( [ v ] ) , ( [ s1 ] ) , and ( [ s ] ) , and those obtained from the solution of eqs . ( [ m_3l ] ) for poisson multilayer networks composed of @xmath16 layers and different values of the averages @xmath170 , @xmath171 , and @xmath172 . the percolation transition is studied for single and multiple instances of the multilayer network model as already described in sec . results are presented in figs . [ fig : mp_3layers ] and [ fig:3layers_average ] , and they provide clear evidence of a perfect agreement between the two approaches . in this paper , we introduced and fully characterized an alternative percolation model for multilayer networks . the model serves to quantify the robustness of networks with redundant interdependencies . according to the model , interdependencies make a system more fragile than it would be by considering each layer independently . this fact is consistent with the original model used to study percolation in multilayer networks @xcite , and it is apparent from the fact that the transition is abrupt for any number of network layers considered in the interdependent model . on the other hand , redundancy of interdependencies across multiple layers favors system robustness , as the height of the discontinuous jump and the location of the transition point decrease as the number of layers increases . this is a fundamental difference with respect to the model currently adopted to study the robustness of multilayer networks , where instead increasing the number of layers generates more and more fragile networks @xcite . we believe that having a model where system robustness is augmented by the number of layers is generally more appropriate . often , new interdependent layers are indeed created to provide backup options . for example , adding a new mode of transportation in a preexisting multimodal transportation system should make the system more reliable against eventual failures . similarly in a living organism , the development of new types of interactions among constituents should increase the stability of the same organism against possible mutations . in the current setting , the model assumes that the functioning of individual nodes requires that nodes are correctly operating on at least two interdependent layers . the model can be , however , generalized to deal with a variable number of minimal functioning layers to describe more realistic scenarios in specific situations of interest . 99 s. boccaletti , g. bianconi , r. criado , c. i. del genio , j. gmez - gardees , m. romance , i. sendia - nadal , z. wang , and m. zanin , phys . rep . * 544 * , 1 ( 2014 ) . m. kivel , a. arenas , m. barthelemy , j. p. gleeson , y. moreno , and m. a. porter , j. complex netw . * 2 * , 203 ( 2014 ) . lee , b. min , and k .- goh , eur . b * 88 * , 1 ( 2015 ) . m. szell , r. lambiotte , and s. thurner , proc . usa * 107 * , 13636 ( 2010 ) . g. menichetti , d. remondini , p. panzarasa , r. j. mondragn , and g. bianconi , plos one * 9 * , e97857 ( 2014 ) . a. cardillo , j. gmez - gardees , m. zanin , m. romance , d. papo , f. del pozo , and s. boccaletti , sci . rep . * 3 * , 1344 ( 2013 ) . m. de domenico , a. sol - ribalta , s. gmez , and a. arenas , proc . usa * 111 * , 8351(2014 ) . v. nicosia and v. latora , phys . e * 92 * 032805 ( 2015 ) . j. menche , et al . science * 347 * , 1257601 ( 2015 ) . e. bullmore and o. sporns , nat . * 10 * , 186 ( 2009 ) . s. d. s. reis , y. hu , a. babino , j. s. andrade jr . , s. canals , m. sigman , and h. a. makse , nature phys . * 10 * , 762 ( 2014 ) . s. v. buldyrev , r. parshani , g. paul , h. e. stanley , and s. havlin , nature * 464 * , 1025 ( 2010 ) . s. watanabe and y. kabashima , phys . e. * 89 * , 012808 ( 2014 ) . g. j. baxter , s. n. dorogovtsev , a. v. goltsev , and j. f. f. mendes , phys . 109 * , 248701 ( 2012 ) . b. min , s. d. yi , k .- lee , and k .- goh , phys . e * 89 * , 042811 ( 2014 ) . s .- w . son , g. bizhani , c. christensen , p. grassberger , and m. paczuski , epl * 97 * , 16006 ( 2012 ) . r. parshani , c. rozenblat , d. ietri , c. ducruet , and s. havlin , epl * 92 * , 68002 ( 2010 ) . r. parshani , s. v. buldyrev , and s. havlin , phys . lett . * 105 * , 048701 ( 2010 ) . d. cellai , s.n . dorogovtsev , g.bianconi , phys . e 94 , 032301 ( 2016 ) . g. bianconi , s. n. dorogovtsev , and j. f. f. mendes , physical review e * 91 * , 012804 ( 2015 ) . g. bianconi and s. n. dorogovtsev , phys . e * 89 * , 062814 ( 2014 ) . f. radicchi and a. arenas , nature phys . * 9 * , 717720 ( 2013 ) . s. n. dorogovtsev , a. goltsev , and j. f. f. mendes , rev . * 80 * , 1275 ( 2008 ) . b. karrer , m. e. j. newman , and l. zdeborov , phys . lett . * 113 * , 208702 ( 2014 ) . m. mezard and a. montanari , _ information , physics and computation _ ( oxford university press , oxford , 2009 ) . a. k. hartmann and m. weigt , _ phase transitions in combinatorial optimization problems _ , ( wiley - vch , weinheim , 2005 ) . g. j. baxter , g. bianconi , r. a. da costa , s. n. dorogovtsev , and j. f. f. mendes , phys . e 94 , 012303 ( 2016 ) . d. cellai , e. lpez , j. zhou , j. p. gleeson , and g. bianconi , phys . e * 88 * , 052811 ( 2013 ) . f. radicchi , nature phys . * 11 * , 597 ( 2015 ) . b. min , s. lee , k .- lee , and k - i . goh , chaos , solitons & fractals * 72 * 49 ( 2015 ) . g. bianconi , phys . e * 87 * , 062806 ( 2013 ) . f. radicchi and c. castellano , phys . e * 93 * , 030302(r ) ( 2016 ) . let us consider a given multilayer network @xmath112 with @xmath1 layers . each layer @xmath68 of the multilayer network has adjacency matrix @xmath173}$ ] . in this multilayer network , each pair of nodes @xmath7 and @xmath22 is connected by a multilink _ ij=(a_ij^[1],a_ij^[2 ] , a_ij^ [ ] , , a_ij^[m ] ) . any two nodes @xmath7 and @xmath22 are connected by a nontrivial multilink is @xmath174 implying that at least one link between the two nodes is present across the @xmath1 layers . we assume that the initial damage configuration is known and that it is given by the set of variables @xmath19 where @xmath43 indicates if a replica @xmath5 is initially damaged ( @xmath21 ) or not ( @xmath20 ) . the message passing algorithm given in sec . iii of the main text allows us to determine for any given initial damage configuration , if any replica node @xmath5 is in the rmcgc ( @xmath175 ) or not ( @xmath176 ) as long as the multilayer network is locally tree - like . specifically the variables @xmath177 are determined in terms the set of messages _ ij=(n_ij^[1],n_ij^[2], ,n_ij^[], n_ij^[m ] ) going from any node @xmath7 to any node @xmath22 joined by a nontrivial multilink @xmath178 . the messages @xmath179 are determined according to the following recursive equation n_ij^[]=(v_ij,2)a^[]_ijs_js_i , [ rm1 ] where @xmath28 indicates the set of nodes that are neighbor of node @xmath7 in layer @xmath12 and where @xmath180 is the step function with values @xmath31 for @xmath32 and @xmath33 for @xmath181 . here the variable @xmath34 indicates in how many layers node @xmath7 is connected to the rmcgc assuming that node @xmath22 also belongs to the rmcgc , v_i j&=&_=1^m\{s_i+s_is_ja^[]_ij_n_(i)j(1-n^[]_i)}. [ rv ] finally the variables @xmath177 are expressed in terms of the messages @xmath179 and are given by _ i&=&s_i\{1-_ } . [ rs1 ] in many situations , however , the initial configuration of the damaged @xmath19 is not known , and instead it is only known the probability distribution @xmath182 of the initial damage configuration . in this case , one aims to know the probability @xmath48 that a replica node @xmath5 is in the rmcgc for a random configuration of the initial damage . the value of @xmath49 , on a locally treelike multilayer network is determined by a distinct message passing algorithm that can be derived from the message passing algorithm valid for single realization of the initial damage , by performing a suitable average of the messages . particular care should be taken when one aims to perform this average . in fact @xmath177 depends on all the messages @xmath183}$ ] sent by node @xmath7 to node @xmath22 in all the layers @xmath12 . these messages are correlated and therefore they can not be averaged independently . an alternative formulation of the eqs . @xmath184 provides the necessary framework for deriving in few steps the message passing algorithm to predict @xmath49 . this alternative formulation is written terms of the variables @xmath185 indicating whether ( @xmath186 ) or not ( @xmath187 ) node @xmath7 send to node @xmath22 a message @xmath188 given that node @xmath22 is connected to node @xmath7 by a multilink @xmath121 . using eqs . @xmath189 it is easy to see that the value of the variables @xmath185 is determined by the following equations : * if @xmath190}>1 $ ] and @xmath121 , ^,_ij&=&_=1^m^n^[]_=1^m^(1-n^[])m^[]s_j , [ ra1 ] * if @xmath190}=1 $ ] and @xmath121 , ^,_ij=\{_=1^m^n^[]}\{1-_ |n^[]=0 } , [ ra2 ] * if @xmath190}=0 $ ] and @xmath121 , ^,_ij=1-_^,_ij , where @xmath179 is determined in terms of the messages @xmath191 as _ ij=_^_ij,_ij . finally a replica node @xmath5 is in the rmcgc ( @xmath175 ) or not ( @xmath176 ) depending on the messages it receives from its neighbors , i.e. _ i&=&s_i\{1-_ } . by averaging eqs . @xmath192 we can derive the message passing algorithm predicting the probability @xmath49 that a replica node @xmath5 is in the rmcgc when the initial damage @xmath193 is randomly drawn for the probability distribution @xmath182 . assuming that each replica node is damaged independently the probability distribution @xmath182 is given by ( \{s_i})=_i=1^n_=1^m p^s_i(1-p)^1-s_i . [ rps ] the message passing algorithm valid for a random distribution of the initial disorder , is written in terms of the messages @xmath194 . the messages @xmath194 take real values between zero and one . they indicate the probability that node @xmath7 send to node @xmath22 a message @xmath188 given that node @xmath22 is connected to node @xmath7 by a multilink @xmath121 and that node @xmath22 has initial damage configuration @xmath195 , i.e. @xmath196 . let us indicate with @xmath197 the probability of a local initial damage configuration given by ( ) = _ = 1^m p^s_(1-p)^1-s _ [ rlps ] and let us indicate with @xmath198 the vector = ( r^[1],r^[2 ] , , r^ [ ] , , r^[m])of elements @xmath199}=0,1 $ ] . using this notation , the messages @xmath194 are determined by the following algorithm ( see last section of this supplementary information for the derivation of these results ) : * if @xmath190}>1 $ ] and @xmath121 , ^,_ij()&= & _ _ i|_s_i>1(_i)_|r^[]=0 ( n^[]+(1-n^[])m^[]s_)=0 c^,(_i , , ) + & & , [ rb1 ] where ^,(_i,,)=_=1^m , [ rc ] * if @xmath200}=1 $ ] , @xmath137}=1 $ ] and @xmath121 , ^,_ij()&= & _ _ i|_ s_i>1(_i)s_is_a_ij^[]\ { 1-_n(i)j ( 1-_|(n^)^[]>0_i^_i(_i ) ) . + & & -_|r^[]=0_(1-s_i)^(1-r^)(s_i)^r^_n(i)j(1-_|_(n^)^r^>0_i^_i(_i ) ) + & & + ._|r^[]=0_(1-s_i)^(1-r^)(s_i)^r^_n(i)j(1-_|_(n^)^>0_i^_i(_i ) ) } , [ rb2 ] * if @xmath190}=0 $ ] and @xmath121 , ^,_ij()=1-_^,_ij ( ) . finally the probability @xmath49 that a replica node @xmath5 is in the rmcgc is given by _ i|_ s_i>1(_i)s_ia_ij^[]\ { 1-_n(i ) ( 1-_|(n^)^[]>0_i^_i(_i ) ) . + & & -_|r^[]=0_(1-s_i)^(1-r^)(s_i)^r^_n(i ) ( 1-_|_(n^)^r^>0_i^_i(_i ) ) + & & + ._|r^[]=0_(1-s_i)^(1-r^)(s_i)^r^_n(i)(1-_|_(n^)^>0_i^_i(_i))}. [ rb3 ] in order to derive the phase diagram of the percolation transition in presence of redundant interdependencies over given multilayer network ensembles , it is useful to consider a further average of the messages @xmath201 . to this end we consider the multilayer network ensemble that preserves the multidegree sequence @xmath110 . every multilayer network @xmath112 in this ensemble has probability ( ) = _ i=1^n _ , where @xmath113 is a normalization constant equal to the number of multilayer networks with the given multidegree sequence . in this multilayer network ensemble the average messages @xmath202 indicate the probability that a message @xmath125 is sent toward a node with initial damage configuration @xmath203 over a multilink @xmath108 . these average messages can be found by solving the following recursive equations : * if @xmath190}>1 $ ] s^,()&= & _ \{k^}p(\{k^})__i|_s_i>1(_i)_|r^[]=0 ( n^[]+(1-n^[])m^[]s_)=0 c^,(_i , , ) + & & , where ^,(_i,,)=_=1^m , * if @xmath200}=1 $ ] , @xmath137}=1 $ ] s^,()&= & _ \{k^}p(\{k^ } ) _ _ i|_ s_i>1(_i)s_is_a_ij^[]\ { 1-_ ( 1-_|(n^)^[]>0s^(_i))^k^-(, ) . + & & -_|r^[]=0_(1-s_i)^(1-r^)(s_i)^r^_(1-_|_(n^)^r^>0s^(_i))^k^-(, ) + & & + ._|r^[]=0_(1-s_i)^(1-r^)(s_i)^r^_(1-_|_(n^)^>0s^(_i))^k^-(, ) } , * if @xmath190}=0 $ ] s^,()=1-_s^ , ( ) . finally the probability @xmath204 that a replica node in layer @xmath12 is in the rmcgc in the multilayer network ensemble is given by s_&= & _ \{k^ } p(\{k^ } ) _ _ i|_ s_i>1(_i)s_ia_ij^[]\ { 1-_ ( 1-_|(n^)^[]>0s^(_i))^k^. + & & -_|r^[]=0_(1-s_i)^(1-r^)(s_i)^r^_(1-_|_(n^)^r^>0s^(_i))^k^ + & & + ._|r^[]=0_(1-s_i)^(1-r^)(s_i)^r^_(1-_|_(n^)^>0s^(_i))^k^}. in this section , we will discuss in detail the derivation of eq . @xmath205 from eq.(@xmath206 ) . a similar derivation ( that we omit here ) can be performed to derive eqs . @xmath207 from eqs . @xmath208 . we start from eq . ( [ ra1 ] ) written for the messages @xmath191 sent from a node @xmath7 to a node @xmath22 with @xmath125 satisfying @xmath190}>1 $ ] and @xmath209},a_{ij}^{[2]}\ldots , a_{ij}^{[m]})$ ] . this equation is rewritten here for convenience , ^,_ij&=&_=1^m^n^[]_=1^m^(1-n^[])m^[]s_j , [ ra1b ] we given a set of variables @xmath210}=0,1 $ ] we can use the following identity & & _ = 1^m ( y_+z_)^p^[]=_|p^[]>0(y_+z _ ) = _ |r^[]=0 p^[]=0 _ = 1^m , [ r ] where in the last expression we perform a sum over all the @xmath1-dimensional vectors @xmath198 = ( r^[1],r^[2 ] , , r^ [ ] , , r^[m ] ) , with @xmath199}=0,1 $ ] if @xmath210}=1 $ ] and @xmath199}=0 $ ] if @xmath210}=0 $ ] . using this expansion for the products in eq . @xmath211 we obtain ^,_ij&= & _ |r^[]=0 ( n^[]+(1-n^[])m^[]s_j)=0c^,(_i,,)_n(i)j , [ rd2 ] where @xmath212 is given by eq . @xmath213 . by using the fact that the messages @xmath191 take only values zero or one , that that out of all the messages @xmath191 from node @xmath7 to node @xmath22 only one is actually equal to one , and all the others are zero , we can rewrite eq . @xmath214 as ^,_ij= _ |r^[]=0 ( n^[]+(1-n^[])m^[]s_j)=0c^,(_i,,)_n(i)j(1-_|_(n^)^[]r^[]>0_i^_i).finally , averaging over the probability distribution @xmath215 of the configuration @xmath216 of the initial damage of node @xmath7 , in the locally treelike approximation we obtain for the messages @xmath217 the eq . @xmath205 that we rewrite here for convenience , ^,_ij(_j)&= & _ _ i|_s_i>1p(_i)_|r^[]=0 ( n^[]+(1-n^[])m^[]s_j)=0c^,(_i , , ) + & & _ n(i)j(1-_|_(n^)^[]r^[]>0_i^_i(_i ) ) .

we consider a generalized multi - hop mimo amplify - and - forward ( af ) relay network with multiple sources / destinations and arbitrarily number of relays . we establish two dualities and the corresponding _ dual transformations _ between such a network and its dual , respectively under single network linear constraint and per - hop linear constraint . the result is a generalization of the previous dualities under different special cases and is proved using new techniques which reveal more insight on the duality structure that can be exploited to optimize mimo precoders . a unified optimization framework is proposed to find a stationary point for an important class of non - convex optimization problems of af relay networks based on a _ local lagrange dual method _ , where the _ primal algorithm _ only finds a stationary point for the inner loop problem of maximizing the lagrangian w.r.t . the primal variables . the input covariance matrices are shown to satisfy a _ polite water - filling structure _ at a stationary point of the inner loop problem . the duality and polite water - filling are exploited to design fast primal algorithms . compared to the existing algorithms , the proposed optimization framework with duality - based primal algorithms can be used to solve more general problems with lower computation cost . multi - hop mimo networks , amplify and forward , relay , duality , mimo precoder optimization

1301.4729

the amplify and forward ( af ) relay technique is useful in wireless systems for cost - effective throughput enhancement and coverage extension . it has attracted many research works recently . the optimization of the mimo ( multiple - input multiple - output ) af relay system has been well studied ( see the tutorial paper @xcite and the reference there in ) . the optimal relay precoding matrix was established in @xcite for a two - hop mimo relay system and the result was later extended to multi - hop mimo relay system in @xcite , where the optimal source and relay precoding matrices are shown to diagonalize the source - relay - destination channel . it was shown in @xcite that this diagonalization property also holds for power minimization of mimo af relay system under qos constraints . however , the power minimization problem is still difficult to solve due to its non - convex nature@xcite . a low complexity algorithm is proposed in @xcite to closely approximate the solution of this non - convex problem for the two - hop case . the source and relay power allocation problem of the multicarrier two - hop af relay system was studied in @xcite . all of the above works focus on the case with a single relay at each hop . in @xcite , the power minimization under qos constraints was addressed for a two - hop mimo relay system with multiple relays . there are also some works on the optimization of multiuser mimo af relay systems . for example , the multiuser mimo mac ( multiaccess channel ) af relay networks are addressed in @xcite , while the mimo bc ( broadcast channel ) af relay networks are investigated in @xcite . since the optimal structure of the source and relay precoding matrices is still unknown for multi - user mimo af relay networks , most of the algorithms in @xcite are based on standard optimization methods such as geometric programming @xcite and logarithmic barrier method @xcite . for mac / bc af relay networks with single - antenna source / destination nodes , two dualities with respect to a total network power constraint and per - hop power constraint have been established in @xcite . the results were later generalized to two - hop mimo mac / bc af relay networks @xcite and multi - hop mimo af relay systems @xcite . in some cases , the duality can be used to simplify the network optimization problem , e.g. , some applications of the duality can be found in @xcite . in this paper , we consider a generalized multi - hop mimo af relay network called the _ b - mac af relay network _ , which has multiple sources , multiple destinations and arbitrarily number of relays . the b - mac af relay network topology includes almost all the existing af relay networks as special cases and the optimization of such general network has not been addressed in the literature . we derive structural properties of the optimization problem ( _ duality _ and _ polite water - filling ( pwf ) _ ) and exploit the duality and pwf structure to design efficient source and relay precoding algorithms for general b - mac af relay networks . for the special case of the single - hop _ b - mac interference network _ ( b - mac ifn ) , the duality and pwf results have been established in @xcite . however , due to the key difference in topology ( single - hop versus multi - hop ) , there are various technical challenges and we can not extend the existing results and algorithms trivially from @xcite . specifically , due to the multi - hop topology , the achievable rate is a complicated function of the input covariance and relay precoding matrices . this makes both the proof of dualities and the optimization of b - mac af relay networks much more challenging than the previous works with simpler network topology . furthermore , the optimization constraints in the multi - hop b - mac af relay network is more complicated than that in b - mac ifn . for example , the _ per - hop power constraint _ and _ per - relay power constraint _ are two practically important constraints that only occur in the multi - hop network topology . we propose several new techniques to solve these challenges . the main contributions are listed below . * * structural properties ( duality ) of mimo af relay network optimization : * we show that the achievable regions of a b - mac af relay network and its dual are the same under _ single network linear constraint _ ( type i duality ) or _ per - hop linear constraint _ ( type ii duality ) . we also derive explicit _ dual transformations _ to calculate the dual input covariance and relay precoding matrices . it is difficult to use the brute force computation approach in @xcite to prove the duality for general b - mac af relay networks . we propose two new techniques , namely the _ network equivalence _ and the _ network dual scaling _ technique , to prove the type i and type ii dualities . the proof is not only simpler but also reveals more insight on the duality structure . the duality and the network equivalence results are useful for mimo precoder optimization in b - mac af relay networks . first , they are used in section [ sub : pwf_structure ] to derive the pwf structure of the optimal input covariance matrices , which can be exploited to design efficient input covariance optimization algorithm as illustrated in section [ sub : polite - water - filling - based ] . second , the duality is used in section [ sub : algmacbc ] and [ sub : algorithm - for - multi - hop ] to simplify the optimization of relay precoding matrices in bc af relay networks . * * efficient mimo precoder optimization algorithm for af relay networks : * based on the duality structure established above , we propose efficient algorithms for mimo precoder optimization in general mimo af relay networks . since the duality is only established for single network / per - hop linear constraint , we can not directly exploit the duality structure to handle general power constraints such as per - relay power constraint . to solve this problem , we propose a unified optimization framework to find a stationary point for an important class of b - mac af relay network optimization problems . in such a framework , we only need to design a primal algorithm to find a stationary point for the _ unconstrained _ inner loop problem of maximizing the lagrangian . using the network equivalence and duality results , we then show that at a stationary point of the inner loop problem , the input covariances have the pwf structure , which is a generalization of the pwf structure in b - mac ifn @xcite to the b - mac af relay network . finally , we propose several efficient primal algorithms based on duality and pwf structure for two - hop , three - hop , and multi - hop af relay networks . the proposed algorithms have significant advantages w.r.t . conventional step - size based iterative algorithms ( such as the gradient ascent method @xcite and the logarithmic barrier algorithms in @xcite ) in terms of complexity and convergence speed . it can also be used to solve a more general class of problems than the existing af relay network optimization algorithms in @xcite . the rest of the paper is organized as follows . section [ sec : system model ] describes the system model , the achievable scheme and the preliminary results on rate duality for b - mac ifn . section [ sec : duality - results ] presents the main results on the two types of dualities for multi - hop b - mac af relay networks . in section [ sec : applications - on - network ] , the duality is applied to study the non - convex optimization of b - mac af relay networks . a unified optimization framework together with several primal algorithms are proposed to find a stationary point for an important class of b - mac af relay network optimization problems . the performance of the proposed algorithms is verified by simulation in section [ sec : simulation - results ] . section [ sec : conlusion ] concludes . a _ b - mac af relay network _ is a general multi - hop af relay network with multiple sources ( transmitters ) , multiple destinations ( receivers ) and @xmath0 relay clusters as illustrated in fig . [ fig : b - mac_relay ] . each relay cluster can have multiple relays . each node can have multiple antennas . we follow the widely used orthogonal assumption @xcite : [ asm : ifhop]due to propagation loss and proper channel reuse , the node in the @xmath1 cluster only receive the signals from the @xmath2 cluster , where the sources and destinations are considered as the @xmath3 and @xmath4 cluster respectively . we consider @xmath5 data links between the sources and destinations . each source can have independent data for different destinations , and each destination may want independent data from different sources . assume the sources are labeled by elements in @xmath6 and the destinations are labeled by elements in @xmath7 . define @xmath8 ( @xmath9 ) as the label of the source ( destination ) of the @xmath10 data link . the numbers of antennas at @xmath8 and @xmath9 are @xmath11 and @xmath12 respectively . the total number of antennas at the @xmath1 relay cluster is @xmath13 . an example of b - mac af relay network is illustrated in fig . [ fig : b - mac_relay ] . there are 2 sources , 2 destinations , and 3 data links , of which @xmath14 , @xmath15 , @xmath16 , @xmath17 , @xmath18 , and @xmath19 . the figure also illustrates how to properly allocate sub - channels to achieve the orthogonality among different hops . in fig . [ fig : b - mac_relay ] , we assume that due to path loss , the aggregate interference from the nodes which are @xmath20 or more hops away from the receiver is negligible . hence we can avoid interference among the transmissions of different hops by dividing the whole bandwidth into @xmath21 orthogonal sub - channels and allocating different sub - channels to adjacent clusters . for convenience , we use link based notation for signals . denote @xmath22 the channel matrix between source @xmath8 and the first relay cluster , @xmath23 the overall channel matrix between the @xmath1 and @xmath24 relay cluster , and @xmath25 the channel matrix between the @xmath26 relay cluster and destination @xmath9 . let @xmath27 denote the transmit signal of link @xmath28 and assume it is a circularly symmetric complex gaussian ( cscg ) vector with covariance @xmath29 . let @xmath30 denote the overall cscg noise vector at the @xmath1 relay cluster with covariance @xmath31 . then the composite received signal at the first relay cluster is @xmath32 . the @xmath33 relay in the @xmath1 cluster applies linear transformation @xmath34 to the received signal and the resulting composite transmit signal from the @xmath1 relay cluster is @xmath35 , where @xmath36\in\mathbb{c}^{l_{q}\times l_{q}}$ ] is the composite relay precoding matrix , @xmath37 is the number of relays in the @xmath1 relay cluster , and @xmath38 is the composite received signal at the @xmath1 relay cluster . for convenience , define @xmath39 then using induction , it can be shown that the received signal at @xmath9 is @xmath40 where @xmath41 is the cscg noise vector at @xmath9 with covariance @xmath42 . as an example , the equivalent channel model for the b - mac af relay network in fig . [ fig : b - mac_relay ] is illustrated in the lower sub - figure . we consider a centralized optimization scheme in which a central node computes all the input covariance and relay precoding matrices and then transmits them to the source and relay nodes . [ asm : csi]the central node has the knowledge of the global channel state information ( csi ) @xmath43 s @xmath44 s and @xmath45 s . in two - hop af relay systems , both the source - relay and relay - destination channels can be estimated directly at the destination node by transmitting pilot signals @xcite . however , acquiring global csi in multi - user multi - hop af relay networks is difficult and requires further study . in this paper , our main focus is to understand the first order optimization problem and solution properties in multi - user multi - hop af networks . an interesting future work is to study efficient decentralized solution which does not require global csi knowledge . the centralized solution in this paper provides the basis for studying the decentralized solutions . for fixed relay precoding matrix @xmath46 s , the source nodes and destinations see an equivalent channel in ( [ eq : recvsignal ] ) , which is essentially a b - mac ifn studied in @xcite . therefore , we can adopt the same set of transmit / receive schemes as in @xcite . [ asm : txrxscheme]@xmath47 1 . the transmit signal is cscg . each signal is decoded by no more than one destination and can be cancelled at this destination for the decoding of other signals . the interference among the source - destination pairs can be specified by some binary coupling matrix @xmath48 . that is , @xmath49 , after ( possible ) interference cancellation , the interference from @xmath50 to @xmath51 is given by @xmath52 , where @xmath53 is the @xmath54 element of @xmath55 . because @xmath56 , each interference is either completely cancelled or treated as noise . the set of allowed transmit / receive schemes includes most interference management techniques as special cases , such as spatial interference reduction through beamforming , dirty paper coding ( dpc ) @xcite at sources and successive interference cancellation ( sic ) at destinations @xcite . for example , in the b - mac af relay network in fig . [ fig : b - mac_relay ] , suppose dpc is used at @xmath57 with @xmath58 encoded after @xmath59 , and sic is used at @xmath60 with @xmath61 decoded after @xmath59 . then @xmath62 , @xmath63 , @xmath64 and @xmath65 are 1 and all other elements of @xmath55 are 0 . for given input covariance matrices @xmath66 and relay precoding matrices @xmath67 , the covariance matrix of the transmit signal at the @xmath1 relay cluster is @xmath68 the interference - plus - noise covariance at @xmath9 is @xmath69 where @xmath70 and @xmath71 the achievable rate of the @xmath10 data link is @xcite @xmath72 on which the achievable regions in this paper are based . we consider two types of linear constraints : define the _ single network linear constraint _ as @xmath73 where @xmath74 and @xmath75 are positive semidefinite ; the _ _ _ per - hop linear constraint _ is defined as @xmath76 the above linear constraints are more general than the sum / individual power constraints considered in @xcite . the per - hop linear constraint is more relevant to practical systems . however , single network linear constraint is also important in practice because it can be used to handle the more general multiple linear constraints using lagrange multiplier method as will be shown in section [ sec : applications - on - network ] . [ dual of a b - mac af relay network][def : dualnetdef ] the dual of a b - mac af relay network is obtained by the following operations . 1 ) reverse the transmission directions , i.e. , in the dual network , @xmath9 s become the sources , @xmath8 s become the destinations , and the transmit signal of @xmath9 travels through the @xmath26 relay cluster , the @xmath77 relay cluster , ... , the @xmath78 relay cluster , and finally reaches the destination @xmath79 2 ) replace all the channel matrices by their conjugate transposes , i.e. , the channel matrix between source @xmath9 and the @xmath26 relay cluster is @xmath80 , the channel matrix between the @xmath81 relay cluster and the @xmath1 relay cluster is @xmath82 , and the channel matrix between the @xmath78 relay cluster and destination @xmath8 is @xmath83 . 3 ) in the dual network , the covariance of the noise at the @xmath1 relay cluster is set as @xmath75 and the covariance of the noise at the destination @xmath8 is set as @xmath74 , where @xmath75 s and @xmath74 s are the linear constraint matrices in ( [ eq : defsnlc ] ) of the original network . 4 ) the coupling matrix for the dual network is set as the transpose of that for the original network , i.e. , @xmath84 . as an example , the dual of the b - mac af relay network in fig . [ fig : b - mac_relay ] is illustrated in fig . [ fig : b - mac_relay_dual ] . we use the notation @xmath85 to denote the terms in the dual network . given input covariance matrices @xmath86 at the sources and relay precoding matrices @xmath87 at the relays , the covariance matrix of the transmit signal at the @xmath1 relay cluster is @xmath88 where @xmath89 the interference - plus - noise covariance at @xmath8 is @xmath90 where @xmath91 . hence , the achievable rate of the @xmath10 data link in the dual network is @xcite @xmath92 similarly , in the dual network , the two types of linear constraints are defined as @xmath93 @xmath94 the proof of the duality results in this paper relies on the rate duality for b - mac ifn @xcite , which is a special case of the b - mac af relay network when @xmath95 . let @xmath96,{\textstyle \sum_{l=1}^{l}}\textrm{tr}\left(\mathbf{\sigma}_{l}\hat{\mathbf{w}}_{l}\right)\leq p_{t},\left[\mathbf{w}_{l}\right]\right),\label{eq : net - color - linear - constraint}\ ] ] denote a b - mac ifn where the channel matrix between @xmath97 and @xmath9 is @xmath98 ; the input covariance satisfies @xmath99 ; the noise covariance at @xmath9 is @xmath100 . for fixed coupling matrix @xmath55 , the interference - plus - noise covariance and the rate of link @xmath101 is given by @xmath102 and @xmath103 respectively . by definition , the dual network or _ reverse links _ is @xmath104,{\textstyle \sum_{l=1}^{l}}\textrm{tr}\left(\hat{\mathbf{\sigma}}_{l}\mathbf{w}_{l}\right)\leq p_{t},\left[\hat{\mathbf{w}}_{l}\right]\right).\label{eq : net - forward - color - dual}\ ] ] the interference - plus - noise covariance and the rate of reverse link @xmath101 is given by @xmath105 and @xmath106 respectively . the key of the rate duality is a _ covariance transformation _ @xcite defined as follows . [ covariance transformation for b - mac ifn][def : the - covariance - transformation ] the covariance transformation of the input covariance @xmath107 for a b - mac ifn with parameters @xmath108,\left[\mathbf{w}_{l}\right],\left[\hat{\mathbf{w}}_{l}\right]\right\ } $ ] as in ( [ eq : net - color - linear - constraint ] ) is defined by 4 steps . * step * 1 : decompose the signal of each link @xmath101 to @xmath109 streams as @xmath110 where @xmath111 is a transmit vector with @xmath112 and @xmath113 s are the transmit powers . * step * 2 : @xmath114 , compute the mmse - sic receive vector as @xcite @xmath115 where @xmath116 is chosen such that @xmath117 . calculate the sinr @xmath118 as a function of @xmath119 @xcite . * step * 3 : in the reverse links , use @xmath120 as transmit vectors and @xmath121 as receive vectors . apply the sinr duality @xcite to obtain the reverse transmit powers @xmath122^{t}$ ] : @xmath123 where @xmath124 and the @xmath125 row and @xmath126 column of @xmath127 is @xmath128 @xmath129 is a diagonal matrix with the @xmath130 diagonal element given by @xmath131 @xmath132^{t}$ ] with @xmath133 . * step * 4 : the _ _ covariance transformation _ _ from @xmath107 to @xmath134 is @xmath135 we restate the rate duality in @xcite as follows . [ rate duality for b - mac ifn][thm : linear - color - dual ] for any @xmath107 satisfying @xmath99 and achieving a rate point @xmath136 in the network ( [ eq : net - color - linear - constraint ] ) , its covariance transformation @xmath137 achieves a rate point @xmath138 in the dual network ( [ eq : net - forward - color - dual ] ) and satisfies @xmath139 . thus , the achievable regions in the forward and reverse links are the same under a single linear constraint . in this section , we establish two types of duality for multi - hop b - mac af relay network respectively under single network linear constraint and per - hop linear constraint . the type i duality can be established using the following _ network equivalence _ result . [ network equivalence][thm : network - equivalence]fixing the relay precoding matrices @xmath140 , the b - mac af relay network in section [ sub : b - mac - af - relay ] under constraint ( [ eq : defsnlc ] ) is equivalent to the following b - mac ifn @xmath141,{\textstyle \sum_{l=1}^{l}}\textrm{tr}\left(\mathbf{\sigma}_{l}\hat{\mathbf{w}}_{l}^{'}\right)\leq p_{t}-p_{c},\left[\mathbf{w}_{l}^{'}\right]\right),\label{eq : equibmac}\ ] ] where @xmath142 is defined in ( [ eq : ehlk ] ) , @xmath143 is defined in ( [ eq : wlp ] ) , and @xmath144 similarly , fixing the dual relay precoding matrices @xmath145 , the dual b - mac af relay network in definition [ def : dualnetdef ] under constraint ( [ eq : defdualsnlc ] ) is equivalent to the dual network of ( [ eq : equibmac ] ) given by @xmath146,{\textstyle \sum_{l=1}^{l}}\textrm{tr}\left(\hat{\mathbf{\sigma}}_{l}\mathbf{w}_{l}^{'}\right)\leq p_{t}-p_{c},\left[\hat{\mathbf{w}}_{l}^{'}\right]\right).\label{eq : dual_equibmac}\ ] ] please refer to appendix [ sub : proof - of - theoremne ] for the proof . define the _ type i dual transformation _ as follows . [ type i dual transformation][def : typeitf ] for any input covariance and relay precoding matrices @xmath147 , let @xmath134 be the covariance transformation of @xmath107 in definition [ def : the - covariance - transformation ] with parameters @xmath148,\left[\mathbf{w}_{l}^{'}\right],\left[\hat{\mathbf{w}}_{l}^{'}\right]\right\ } $ ] . then @xmath134 and @xmath149 is called the _ type i dual transformation _ of @xmath147 . then it follows from theorem [ thm : linear - color - dual ] and [ thm : network - equivalence ] that the following theorem holds . [ type i duality][thm : typeiduality]for any input covariance and relay precoding matrices @xmath147 satisfying @xmath150 and achieving a rate point @xmath136 in a b - mac af relay network , its type i dual transformation @xmath151 achieves a rate point @xmath138 in the dual network and satisfy @xmath152 . thus , the achievable regions in a b - mac af relay network under constraint ( [ eq : defsnlc ] ) and that in the dual network under constraint ( [ eq : defdualsnlc ] ) are the same . type ii duality can be proved from type i duality using the conception of _ network dual scaling_. [ scaled dual network]a scaled dual network with scaling vector @xmath153^{t}$ ] , where @xmath154 and @xmath155 , is obtained by scaling the covariance of the noise vectors in the dual network defined in section [ sub : the - dual - network ] as follows : the covariance of the noise at the @xmath1 relay cluster is scaled to be @xmath156 , and the covariance of the noise at the destination @xmath8 is scaled to be @xmath157 . furthermore , the relay precoding matrices is given by @xmath145 . the following corollary follows immediately from theorem [ thm : typeiduality ] . [ duality for scaled dual network][cor : scalednet ] for any input covariance and relay precoding matrices @xmath147 achieving a rate point @xmath136 in a b - mac af relay network , let @xmath158 be the covariance transformation of @xmath107 obtained by definition [ def : the - covariance - transformation ] with parameters @xmath148,\left[\mathbf{w}_{l}^{'}\right],\left[\hat{\mathbf{w}}_{l}^{'}\left(\mathbf{d}\right)\right]\right\ } $ ] , where @xmath159 then @xmath158 and @xmath145 achieves a rate point @xmath138 in the scaled dual network with scaling vector @xmath160 and satisfy the constraint @xmath161 where @xmath162 another theorem is needed to prove the type ii duality . obtain @xmath121 , @xmath120 , @xmath127 and @xmath163 using ( [ eq : decomsig ] ) , ( [ eq : mmserev1 g ] ) , ( [ eq : faig ] ) and ( [ eq : dg ] ) in definition [ def : the - covariance - transformation ] with parameters @xmath148,\left[\mathbf{w}_{l}^{'}\right],\left[\hat{\mathbf{w}}_{l}^{'}\right]\right\ } $ ] . define @xmath164 where @xmath165^{t}$ ] and @xmath114 , @xmath166 @xmath167^{t}$ ] and @xmath114 , @xmath168 define @xmath169 whose @xmath170 element is @xmath171 [ network dual scaling][thm : network - dual - scaling]consider the following eigensystem @xmath172 where @xmath173 is the dominant eigenvector and @xmath174 is the maximum eigenvalue of @xmath175 . then @xmath174 and @xmath173 must be strictly positive . let @xmath176 $ ] be the dominant eigenvector of @xmath175 with the last component scaled to one . then in the scaled dual network with scaling vector @xmath177 , @xmath178 defined in corollary [ cor : scalednet ] and @xmath145 satisfies @xmath179 the proof is given in appendix [ sub : proof - of - theorem_nds ] . define the _ type ii dual transformation _ as follows . [ type ii dual transformation]for any input covariance and relay precoding matrices @xmath180 , let @xmath181 be the covariance transformation of @xmath107 in definition [ def : the - covariance - transformation ] with parameters @xmath148,\left[\mathbf{w}_{l}^{'}\right],\left[\hat{\mathbf{w}}_{l}^{'}\left(\mathbf{\tilde{d}}\right)\right]\right\ } $ ] , where @xmath176 $ ] is the dominant eigenvector of the eigensystem in ( [ eq : keyeigensys ] ) . then @xmath134 and the dual relay precoding matrices @xmath182 , where @xmath183 , @xmath184 , is called the _ type ii dual transformation _ of @xmath180 . for the same input covariance @xmath181 , the rate point in the scaled dual network with scaling vector @xmath177 and relay precoding matrices @xmath185 is equal to that in the original dual network with relay precoding matrices @xmath186 . furthermore , it follows from ( [ eq : indlconsdualtx ] ) that the type ii dual transformation @xmath187 satisfies ( [ eq : defdualilc ] ) . combining the above and corollary [ cor : scalednet ] , the following theorem is proved . [ type ii duality][thm : typeiiduality]for any input covariance and relay precoding matrices @xmath180 satisfying @xmath188 @xmath189 , and achieving a rate point @xmath136 in the b - mac af relay network defined in section [ sub : b - mac - af - relay ] , its type ii dual transformation @xmath187 achieves a rate point @xmath138 in the dual network and satisfy @xmath190 @xmath191 . thus , the achievable regions in a b - mac af relay network under constraint ( [ eq : defilc ] ) and that in the dual network under constraint ( [ eq : defdualilc ] ) are the same . [ generality of type i / ii duality][rem : generality - of - typei_ii]the previous duality results for various special cases are compared to illustrate the generality of the type i / ii duality . in @xcite and @xcite , the duality was established respectively for multi - hop mac / bc af relay networks with single - antenna source / destination nodes and two - hop mimo mac / bc af relay networks . however , the approach in @xcite can not be easily extended to the general b - mac af relay network . * * the duality for multi - hop mimo af relay system was established in @xcite . although the proof can be extended to multi - user af relay networks by using block diagonal precoding / receiving matrices and the notion of independent streams , the duality in [ 3 ] can not cover the duality in this paper as elaborated below . \1 ) the b - mac af relay network is not captured by the system model in @xcite . in @xcite , there are @xmath192 independent data streams in one transmission . similar to the coupling matrix defined in assumption [ asm : txrxscheme ] , we can use a binary _ inter - stream coupling matrix _ @xmath193 to specify the interference among the data streams . in @xcite , @xmath194 can only have two forms : a ) if linear transceivers are used at the source and destination nodes , all the diagonal elements of @xmath194 are zero , and all the off - diagonal elements of @xmath194 are one ; b ) if non - linear transceivers are used , @xmath194 is a triangular matrix . however , for general b - mac af relay networks , if we decompose the equivalent mimo links into independent data streams , the inter - stream coupling matrix @xmath194 can be any binary matrix with zero diagonal elements . \2 ) explicit dual transformations are not part of @xcite for the multi - user case with general inter - stream coupling matrix @xmath194 . the dual transformation between the dpc - based and sic - based mimo af relay systems in @xcite can not be generalized to b - mac af relay networks because the matrix `` @xmath55 '' in ( 14 ) of @xcite is no longer upper - triangle . furthermore , the above special cases consider power constraints only . by using the techniques of _ network equivalence _ and _ network dual scaling _ , we are able to establish the dualities and explicit dual transformations for b - mac af relay networks under more general linear constraints . the duality proof based on these new techniques is not only simpler but also reveals more insight on the duality structure that can be exploited to design mimo precoder optimization algorithms as shown in the next section . we propose an unified optimization framework to find a stationary point for a class of af relay network optimization problems based on the local lagrange dual method ( ldm ) @xcite , where the _ primal algorithm _ only finds a stationary point for the inner loop problem of maximizing the lagrangian . the duality is first used to characterize the pwf structure of the input covariance matrices at a stationary point . then the duality and pwf are exploited to design efficient primal algorithm for general _ b - mac _ af relay networks and structured primal algorithms for _ bc _ af relay network . a general optimization problem in a b - mac af relay network can be expressed as @xmath195 where @xmath196 and @xmath197 are real valued functions of @xmath180 . in this paper , we consider a class of problems whose lagrangian can be written in the following form @xmath198 where @xmath199_{n=1, ... ,n}$ ] are lagrange multipliers , @xmath200 are positive semidefinite , @xmath201 are positive definite , and @xmath202 . the lagrangian of many important optimization problems can be written in the form of ( [ eq : lag ] ) . two typical problems are listed below . p1 : maximize the weighted sum - rate under multiple linear constraints : @xmath203 where @xmath204 , @xmath205 s and @xmath206 s are positive semidefinite and @xmath207 . p2 : minimize a single linear cost under individual rate constraints : @xmath208 where @xmath209 is the rate constraint for the @xmath10 data link . [ per - relay power constraint]the multiple linear constraints in p1 include the per - relay power constraint as special cases . let @xmath37 denote the number of relays in the @xmath1 relay cluster , and let @xmath210 denote the number of antennas at the @xmath33 relay of the @xmath1 relay cluster . then the power constraint @xmath211 for the @xmath33 relay in the @xmath1 relay cluster can be expressed in the form of the general multiple linear constraints in p1 as @xmath212 by setting @xmath213 , @xmath214 , @xmath215 , and @xmath216 $ ] , where @xmath217 is an @xmath218 identity matrix and @xmath219 is an @xmath220 zero matrix . it is well known that the problems in ( [ eq : p1 ] ) and ( [ eq : p2 ] ) are usually non - convex problems with non - zero duality gap . hence , the standard lagrange dual method ( ldm ) @xcite can not be used to solve these problems . in @xcite , we proposed a _ local ldm _ to find a stationary point for a non - convex problem . we apply the local ldm to obtain the following algorithm for problem ( [ eq : generalpf ] ) . _ algorithm lldm _ ( for finding a stationary point of problem ( [ eq : generalpf ] ) ) : * * initialization * * : choose initial @xmath221 such that @xmath222 . let @xmath223 . * * step 1 * * ( primal update in the inner loop ) : for fixed @xmath224 , find a stationary point @xmath225 of @xmath226 using some primal algorithm with initial point is randomly generated . ] @xmath227 . * * step 2 * * ( dual update in the outer loop ) : update @xmath228 as @xmath229 where @xmath230 is the step size , and @xmath231 is the optimal solution of the following quadratic programming problem : @xmath232 where @xmath233^{t}$ ] with @xmath234 , and @xmath235 is positive definite . * return to step 1 until convergence . * @xmath236 is obtained by the well known bfgs update as follows @xcite @xmath237 @xmath238 the initial @xmath239 , where @xmath240 is the smallest positive integer such that @xmath241 , and @xmath242 $ ] is the optimal solution of the problem @xmath243 . set @xmath244 , where @xmath245 is an integer which is initialized as @xmath246 and is incremented until one of the following conditions is satisfied : 1 ) @xmath247 . 2 ) @xmath248 , where @xmath249 is a small positive integer , e.g. , we let @xmath250 in the simulations . 3 ) @xmath251 , where @xmath252 is defined as the _ _ residual error _ _ after the @xmath253 iteration . finally , @xmath254 is given by @xmath255 where @xmath256 , @xmath257 and are set as @xmath258 in the simulations . algorithm lldm converges to a stationary point of problem ( [ eq : generalpf ] ) under mild conditions @xcite . in the subsequent sections , we first characterize the pwf structure at a stationary point of problem ( [ eq : mainpro ] ) . then we design several efficient primal algorithms based on the pwf and duality . at any stationary point of problem ( [ eq : mainpro ] ) with fixed relay precoding matrices @xmath140 , the input covariance matrices @xmath107 must satisfy a pwf structure as stated in the following corollary . [ pwf for fixed relay precoding][cor : pwf_mlc ] let @xmath259 be a stationary point of problem ( [ eq : mainpro ] ) with fixed @xmath140 , i.e. , @xmath259 satisfies the following kkt conditions @xmath260 then the type i dual transformation of @xmath261 is @xmath262 , where @xmath263 with @xmath264 obtain the interference - plus - noise covariance matrices @xmath265 s from ( [ eq : whitemg ] ) using @xmath261 and the dual ones @xmath266 s from ( [ eq : whitemg-1 ] ) using @xmath267 . for each @xmath101 , create an equivalent channel : @xmath268 . perform the thin svd @xmath269 , where @xmath270 , and @xmath271 . then , @xmath259 must have a polite water - filling structure , i.e. , @xmath272 furthermore , @xmath273 also possesses the polite water - filling structure , i.e. , @xmath274 on the other hand , if certain @xmath107 has the above polite water - filling structure for a given @xmath140 , it must be a stationary point of problem ( [ eq : mainpro ] ) for this fixed @xmath140 . corollary [ cor : pwf_mlc ] follows straightforward from the pwf results for b - mac ifn in theorem 3 of @xcite and the network equivalence result in theorem [ thm : network - equivalence ] . the detailed proof is omitted due to limited space . in some cases , better algorithms can be designed by considering the following _ dual network problem_. [ dual network problem]the _ dual network problem _ of ( [ eq : mainpro ] ) is defined as @xmath275 [ equivalence of problem ( [ eq : mainpro ] ) and its dual][thm : dualnetpro ] for any stationary point @xmath276 of dual network problem ( [ eq : mainpro_dual ] ) , its type i dual transformation @xmath277 must also be a stationary point of problem ( [ eq : mainpro ] ) . please refer to appendix [ sub : proof - of - theoremdualpro ] for the proof . we present several primal algorithms for two - hop b - mac af relay networks . in this case , the relay precoding matrices is @xmath278 $ ] , where @xmath279 is the relay precoding matrix at the @xmath33 relay , @xmath280 is the number of antennas at the @xmath33 relay , and @xmath281 is the number of relays . the gradient ascent ( ga ) algorithm with the step size determined by armijo rule @xcite can be used to find a stationary point of problem ( [ eq : mainpro ] ) as summarized in table [ tab : alggp ] . in algorithm ga , the gradient update for source node is performed over the precoding matrices @xmath282 s , where @xmath283 . for convenience , we group the gradients over @xmath284 as a block diagonal matrix @xmath285 $ ] . then the gradients required in algorithm ga are given by @xcite @xmath286 @xmath287\circ\nabla_{\mathbf{f}}l_{\lambda}\left(\mathbf{\sigma},\mathbf{f}\right),\label{eq : gradf}\ ] ] where @xmath288 ; the notation @xmath289 denotes the pointwise product of two matrices ; @xmath290 is an @xmath291 matrix of which all the elements are ones ; and @xmath292 is given by @xmath293 where @xmath294 @xmath295 , and @xmath296 . .[tab : alggp]algorithm ga ( gradient ascent method for finding a stationary point of problem ( [ eq : mainpro ] ) ) [ cols= " < " , ] consider sum rate maximization under multiple linear constraints ( i.e. , problem ( [ eq : p1 ] ) with @xmath297 ) in a two - hop bc af relay network applying dpc . we apply the duality and corollary [ cor : pwf_mlc ] to design a structured primal algorithm to find a stationary point of the inner loop problem ( [ eq : mainpro ] ) for this special case . note that in this case , @xmath298 , @xmath299 are the same , and @xmath200 are the same . for convenience , let @xmath300,\mathbf{w}_{\textrm{b}}^{\textrm{r}},\left[\mathbf{w}_{\textrm{b}}^{l}\right],\mathbf{\hat{w}}_{\textrm{b}},\mathbf{\hat{w}}_{\textrm{b}}^{\textrm{r}}\right),\label{eq : bcrelay}\ ] ] denote a two - hop bc af relay network where the channel matrix between source and relay is @xmath301 , the channel matrix between relay and the @xmath10 destination is @xmath302 , the covariance of the noise at the relay is @xmath303 , the covariance of the noise at the @xmath10 destination is @xmath304 , and the constraint matrices in lagrangian ( [ eq : lag ] ) is @xmath305 , @xmath306 . the dual network of ( [ eq : bcrelay ] ) is @xmath307,\mathbf{h}_{\textrm{m}},\mathbf{w}_{\textrm{m}}^{\textrm{r}},\mathbf{w}_{\textrm{m}},\left[\mathbf{\hat{w}}_{\textrm{m}}^{l}\right],\mathbf{\hat{w}}_{\textrm{m}}^{\textrm{r}}\right),\label{eq : macrelay}\ ] ] which is a two - hop mac af relay network where the channel matrix between the @xmath10 source and relay is @xmath308 , the channel matrix between relay and the destination is @xmath309 , the covariance of the noise at the relay is @xmath310 , the covariance of the noise at the @xmath10 destination is @xmath311 , and the constraint matrices in lagrangian ( [ eq : lag ] ) is @xmath312 , @xmath313 . by theorem [ thm : dualnetpro ] , problem ( [ eq : mainpro ] ) for the bc relay network ( [ eq : bcrelay ] ) can be equivalently solved by solving the dual network problem ( [ eq : mainpro_dual ] ) , whose objective function , in this special case , is given by @xmath314 where @xmath315 . by exploiting the specific structure of @xmath316 , we propose an improved algorithm over pwf . _ algorithm pwfi _ : * * initialization * * : choose proper initial @xmath187 such that @xmath317 . * step * 1 : for fixed @xmath318 and @xmath319 , problem ( [ eq : mainpro_dual ] ) reduces to the single - user optimization problem : @xmath320 where @xmath321 , @xmath322 , and @xmath323 . applying corollary [ cor : pwf_mlc ] to the single - user optimization problem in ( [ eq : maxlagl ] ) , the optimal solution is given by @xmath324 where @xmath325 and @xmath326 are obtained by performing the thin svd @xmath327 . update each @xmath328 for once using ( [ eq : gwfsolution ] ) when fixing @xmath318 and @xmath319 . * * step 2 * * : for fixed @xmath329 , @xmath318 is updated by solving @xmath330 , which is equivalent to the following problem @xmath331 where @xmath332 , @xmath333 and @xmath334 . define @xmath335 . perform the svd @xmath336 and the eigenvalue decomposition @xmath337 . let @xmath338 denote the @xmath339 largest singular values of @xmath340 with descending order and @xmath341 denote the semi - unitary matrix formed by the singular vectors corresponding to @xmath342 s . let @xmath343 denote the @xmath339 largest eigenvalues of @xmath344 with descending order and @xmath345 denote the semi - unitary matrix formed by the eigenvectors corresponding to @xmath346 s . then the optimal @xmath347 is given by @xcite @xmath348 where @xmath349 with @xmath350 given by @xmath351^{+}.\ ] ] * return to step 1 until convergence . * since the objective @xmath352 is upper bounded and is increased after each update , algorithm pwfi must converge to a fixed point @xmath353 . it can be shown that the fixed point @xmath353 must be a stationary point of the dual network problem ( [ eq : mainpro_dual ] ) . then by theorem [ thm : dualnetpro ] , the corresponding type i dual transformation @xmath277 is a stationary point of problem ( [ eq : mainpro ] ) in the bc relay network ( [ eq : bcrelay ] ) . algorithm ga / pwf can be easily extended to solve problem ( [ eq : mainpro ] ) for b - mac af relay networks with more than two hops . the only difference is that the gradient update of the relay precoding matrices is generalized to multi - hop case as @xmath354 where the gradient @xmath355 can be calculated similar to the one in ( [ eq : gradf ] ) . the duality can be used to simplify the the sum rate maximization under multiple linear constraints for a three - hop bc relay network . we design _ algorithm pwf3 _ to find a stationary point of the inner - loop problem ( [ eq : mainpro ] ) for this special case , where @xmath356 , @xmath357 and @xmath358 . algorithm pwf3 switches the optimization between the original network and the dual network so that at each time , we only need to consider a two - hop network optimization problem solved in section [ sub : algmacbc ] . _ _ algorithm pwf__3 : * * initialization * * : choose proper initial @xmath359 such that @xmath360 . * * step 1 * * : for fixed @xmath361 , the inner loop problem reduces to problem ( [ eq : mainpro ] ) in the two - hop bc af relay network ( [ eq : bcrelay ] ) with @xmath362 , @xmath363 , @xmath364 , @xmath365 , @xmath366 and @xmath367 . obtain the input covariance and relay precoding matrices @xmath368 for the dual two - hop mac af relay network ( [ eq : macrelay ] ) by the type i dual transformation of @xmath369 applied to network ( [ eq : bcrelay ] ) . update each @xmath370 for once using ( [ eq : gwfsolution ] ) when fixing @xmath371 and @xmath372 . then update @xmath371 using ( [ eq : updatefhed ] ) for fixed @xmath373 . finally , obtain the updated @xmath369 by the type i dual transformation of @xmath374 applied to network ( [ eq : macrelay ] ) . * * step 2 * * : obtain @xmath375 by the type i dual transformation of @xmath180 . * * step 3 * * : for fixed @xmath376 , update @xmath377 by improving the objective value of the dual network problem ( [ eq : mainpro_dual ] ) , which reduces to problem ( [ eq : mainpro ] ) in the two - hop mac af relay network ( [ eq : macrelay ] ) with @xmath378 , @xmath379 , @xmath380 , @xmath381 , @xmath382 , and @xmath383 . treating @xmath377 as the input covariance and relay precoding matrices in network ( [ eq : macrelay ] ) , update @xmath377 using a single iteration of algorithm pwfi applied to network ( [ eq : macrelay ] ) . * * step 4 * * : obtain @xmath180 by the type i dual transformation of @xmath375 . * return to step 1 until convergence . * using theorem [ thm : typeiduality ] , it can be shown that algorithm pwf3 monotonically increases the objective after each iteration and converges to a stationary point of problem ( [ eq : mainpro ] ) . [ rem : comppwfifc]for the special case of b - mac ifn , the duality and pwf have been established in @xcite . however , we can not extend the existing results and algorithms trivially from @xcite . we proposed new techniques , namely the network equivalence technique in theorem [ thm : network - equivalence ] , and the network dual scaling technique in theorem [ thm : network - dual - scaling ] , to generalize the duality and pwf to b - mac af relay networks . the proof of theorem [ thm : network - dual - scaling ] is non - trivial . the algorithm design also relies on the new network equivalence result . moreover , theorem [ thm : dualnetpro ] ( equivalence of problem ( [ eq : mainpro ] ) and its dual ) is non - trivial and can not be re - derived from our previous results in @xcite . based on theorem [ thm : dualnetpro ] , we proposed structured primal algorithms ( pwfi and pwf3 ) which are entirely different from the single - hop pwf algorithms in @xcite . block fading channel is assumed and each channel matrix has zero - mean i.i.d . gaussian entries with unit variance . the simulation results in all figures are averaged over 100 random channel realizations . the parameters of problem ( [ eq : mainpro ] ) are set as @xmath384 . for accuracy comparison , define the _ convergence error _ as the sum of the norms of the gradients over @xmath140 and all @xmath282 s . convergence comparison for a two - hop mac af relay network , width=302 ] convergence comparison for a two - hop mac af relay network , width=302 ] we first demonstrate the advantages of algorithm pwf over ga described in section [ sub : baseline - algorithm-(gradient ] . consider the b - mac af relay network in fig . [ fig : b - mac_relay ] with @xmath298 ( two - hop ) , @xmath385 relay , and @xmath386 , @xmath387 antennas at each node . in fig . [ fig : twohop_bmac_af ] , we plot the average objective value and the convergence error versus the number of iterations . it can be observed that pwf has faster convergence speed than ga . it also achieves a higher objective value . the complexity order per iteration of pwf and ga is similar . however , the overall complexity of pwf is lower because it avoids the linear search for the gradient update of @xmath107 , and requires less iterations to achieve the same accuracy . in fig . [ fig : twohop_mac_af ] , we compare the convergence of algorithm ga , pwf and pwfi for a two - hop mac relay network with @xmath388 sources , @xmath385 relay and @xmath389 , @xmath387 , @xmath390 antennas at each node . the results show that pwfi has the fastest convergence speed . pwfi also has the lowest complexity because it avoids the linear search in the gradient update . finally , we verify the convergence of algorithm pwf3 for a three - hop bc relay network with @xmath391 relay at each relay cluster , @xmath388 destinations and @xmath392 , @xmath393 , @xmath394 antennas at each node . fig . [ fig : threehop_bc_af ] plots the average objective value and _ objective error _ , defined as the gap ( in logarithmic scale ) from the stationary point that the algorithm converges to , versus the number of iterations . algorithm pwf3 quickly converges to a stationary point of problem ( [ eq : mainpro ] ) with high accuracy . average sum - rate / cpu time versus power constraint at relay for a two - hop bc af relay network , width=302,height=253 ] average sum - rate / cpu time versus power constraint at relay for a two - hop bc af relay network , width=302 ] average sum power / cpu time versus individual rate constraint for a two - hop mac af relay network , width=302 ] average sum power / cpu time versus individual rate constraint for a two - hop mac af relay network , width=302 ] in fig . [ fig : threehop_bc_af ] and [ fig : twohop_bc_lldm ] , we consider problem ( [ eq : p1 ] ) with @xmath297 ( i.e. , sum - rate maximization ) and illustrate the advantages of the proposed local ldm with duality - based primal algorithms ( i.e. , algorithm pwf or pwfi ) . the following baseline algorithms are compared . * baseline 1 ( algorithm 3 in @xcite ) : algorithm 3 in @xcite is based on logarithmic barrier method @xcite and is designed for sum - rate maximization under power constraints in two - hop bc relay networks . * baseline 2 ( logarithmic barrier algorithm ) : this is an extension of the algorithm 3 in @xcite to b - mac af relay networks using the logarithmic barrier method @xcite . * baseline 3 : local ldm with ga as the primal algorithm . we first evaluate the performance in a two - hop bc relay network with @xmath385 relay , @xmath388 destinations and @xmath392 , @xmath387 , @xmath394 antennas at each node . assume a sum power constraint at the source : @xmath395 , and a power constraint at the relay : @xmath396 . in fig . [ fig : twohop_bc_lldm ] , we plot the average sum - rate , and the cpu time required to achieve the same accuracy , versus the power constraint at relay @xmath397 . all algorithms achieve similar sum - rate . however , lldm with both pwf and pwfi clearly outperform the algorithm 3 in @xcite in terms of cpu time . in fig . [ fig : twohop_bmac_af_ldm ] , we evaluate the sum - rate and cpu time performance in the b - mac af relay network in fig . [ fig : twohop_bmac_af ] with @xmath298 ( two - hop ) , @xmath398 relays , @xmath399 antennas at each relay , and @xmath386 antennas at each source / destination node . consider individual power constraints at each node : @xmath400 , @xmath401 , @xmath402\right)\leq p_{r}$ ] and @xmath403\right)\leq p_{r}$ ] . all algorithms achieve similar sum - rate . however , lldm with pwf requires the least cpu time . in fig . [ fig : twohop_mac_af_qos ] , we consider problem ( [ eq : p2 ] ) with @xmath404 ( i.e. , sum power minimization ) , and @xmath405 for a two - hop mac relay network with @xmath406 sources , @xmath385 relay and @xmath389 , @xmath387 , @xmath390 antennas at each node . the proposed lldm with pwf is compared to baseline 3 and baseline 4 : algorithm 9 in @xcite ( joint search based on logarithmic barrier method ) . we plot the average sum power and the cpu time versus the individual rate constraint ( @xmath407 ) . all algorithms achieve similar sum power . however , lldm with pwf requires the least cpu time . we show that the achievable regions of a multi - hop mimo b - mac af relay network and its dual are the same under _ single network linear constraint _ or _ per - hop linear constraint_. two dual transformations are provided to calculate the dual input covariance and relay precoding matrices . these results include the dualities in @xcite as special cases . furthermore , our proof is simpler and reveals more structural property of the duality . based on the established duality structure , we propose efficient algorithms for mimo precoder optimization in b - mac af relay networks . first , a unified optimization framework is proposed based on the _ local lagrange dual method _ in @xcite so that we only need to focus on designing a primal algorithm to find a stationary point of the unconstrained inner loop problem . using duality , we characterize the polite water - filling ( pwf ) structure of the input covariance matrices at a stationary point of the inner loop problem . then , the duality and pwf are exploited to design efficient primal algorithms . the proposed local ldm with duality - based primal algorithms has lower computation cost and faster convergence speed than the conventional step - size based iterative algorithms . [ [ proof - of - theorem - thmnetwork - equivalence - subproof - of - theoremne ] ] proof of theorem [ thm : network - equivalence ] [ sub : proof - of - theoremne ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ it follows from @xmath408 that @xmath409 . hence the equivalent channel of the dual network is @xmath410 . by ( [ eq : whitemg-1 ] ) , the covariance of the equivalent noise of dual link @xmath101 is @xmath411 . note that @xmath412 hence , the linear constraint in ( [ eq : defdualsnlc ] ) can be expressed as @xmath413 . the above proves the equivalence between the dual b - mac af relay network under constraint ( [ eq : defdualsnlc ] ) and the b - mac ifn in ( [ eq : dual_equibmac ] ) . the equivalence between the b - mac af relay network under constraint ( [ eq : defsnlc ] ) and the b - mac ifn in ( [ eq : equibmac ] ) can be proved similarly . [ [ subproof - of - theorem_ndsproof - of - theorem - thmnetwork - dual - scaling ] ] [ sub : proof - of - theorem_nds]proof of theorem [ thm : network - dual - scaling ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ recall that @xmath414 , @xmath415 , @xmath127 and @xmath163 are obtained using ( [ eq : decomsig ] ) , ( [ eq : mmserev1 g ] ) , ( [ eq : faig ] ) and ( [ eq : dg ] ) in definition [ def : the - covariance - transformation ] with parameters @xmath148,\left[\mathbf{w}_{l}^{'}\right],\left[\hat{\mathbf{w}}_{l}^{'}\right]\right\ } $ ] . since @xmath158 is the covariance transformation of @xmath107 obtained by definition [ def : the - covariance - transformation ] with parameters @xmath148,\left[\mathbf{w}_{l}^{'}\right],\left[\hat{\mathbf{w}}_{l}^{'}\left(\mathbf{d}\right)\right]\right\ } $ ] , using the fact that @xmath414 , @xmath415 , @xmath127 and @xmath163 only depends on @xmath107 and @xmath148,\left[\mathbf{w}_{l}^{'}\right]\right\ } $ ] , we have @xmath416 where @xmath417 is given by @xmath418 and @xmath419 s are defined in ( [ eq : defnqhead ] ) . note that @xmath420 where ( [ eq : indscale1]-a ) follows from ( [ eq : defnq ] ) and ( [ eq : qbar ] ) , and ( [ eq : indscale1]-b ) follows from the definition of @xmath421 s in ( [ eq : defaq ] ) . hence the first linear constraint in ( [ eq : indlconsdualtx ] ) can be expressed as @xmath422 similarly , @xmath423 , @xmath424 where ( [ eq : indscaleq ] ) follows from ( [ eq : defaq ] ) and ( [ eq : defbq ] ) , and thus the other linear constraints in ( [ eq : indlconsdualtx ] ) can be expressed as @xmath425 now replace @xmath426 s in ( [ eq : indscale1ina ] ) and ( [ eq : indscaleqina ] ) with @xmath427 s , we obtain another set of equations @xmath428 it can be verified that the equations in ( [ eq : lamdequ ] ) forms the eigensystem @xmath429 in ( [ eq : keyeigensys ] ) , where @xmath430 is defined in ( [ eq : defa ] ) , @xmath431^{t}$ ] . note that @xmath432 due to nonsingularity of @xmath42 s and @xmath74 s . furthermore , since @xmath31 s and @xmath75 s are assumed to be non - singular and @xmath433 , we have @xmath434 . using these facts , the following lemma can be proved . [ lem : propertya]the following is true for @xmath175 defined in ( [ eq : defa ] ) . 1 ) @xmath435 and @xmath436 . 2 ) for any @xmath437 and @xmath160 satisfying @xmath429 , if any element of @xmath160 is zero , then @xmath438 . it follows from lemma [ lem : propertya ] and the perronfrobenius theorem ( * ? ? ? * chp . 8) that the maximum eigenvalue @xmath174 and the associated dominant eigenvector @xmath173 of @xmath175 satisfies @xmath439 and @xmath440 . then we can always obtain a scaled eigenvector @xmath441^{t}$ ] that satisfies @xmath442 . on the other hand , it follows from ( [ eq : snlcfixd ] ) that @xmath174 is equal to 1 , which indicates that @xmath178 and @xmath443 satisfies ( [ eq : indlconsdualtx ] ) . by corollary [ cor : pwf_mlc ] , @xmath259 satisfies the pwf structure for fixed @xmath444 and the kkt condition in ( [ eq : kktsigma ] ) with @xmath445 . to prove that @xmath446 is a stationary point , we only need to further show that @xmath446 satisfies @xmath447 since @xmath276 is a stationary point of the dual problem ( [ eq : mainpro_dual ] ) , we have @xmath448 . in the following , we prove that @xmath449 , from which ( [ eq : kktf ] ) follows immediately . note that for fixed @xmath450 and @xmath451 , problem ( [ eq : mainpro ] ) and ( [ eq : mainpro_dual ] ) can be equivalent to the inner loop problem and its dual for a two - hop b - mac af relay network . hence , without loss of generality , we only need to consider the two - hop case ( i.e. , @xmath298 ) and prove that @xmath452 . using ( [ eq : sigmhead1 ] ) in corollary [ cor : pwf_mlc ] and the expression of @xmath453 in ( [ eq : gf ] ) , it can be shown that @xmath454 where @xmath455 and @xmath456 . similar , @xmath457 it follows from ( [ eq : pwf_mgl ] ) and ( [ eq : pwf_mlc_dual ] ) in corollary [ cor : pwf_mlc ] that @xmath458 combining ( [ eq : gf1]-[eq : keyeqgfequ ] ) and @xmath459 , we have @xmath452 . this completes the proof . s. a. jafar , k. s. gomadam , and c. huang , `` duality and rate optimization for multiple access and broadcast channels with amplify - and - forward relays , '' _ ieee trans . theory , _ , vol . 53 , no . 10 , pp . 3350 3370 , oct . 2007 . a. liu , y. liu , h. xiang , and w. luo , `` duality , polite water - filling , and optimization for mimo b - mac interference networks and itree networks , '' _ submitted to ieee trans . info . theory _ 2010 ; revised oct . [ online ] . available : http://arxiv.org/abs/1004.2484 a. liu , y. liu , h. xiang , and w. luo , `` mimo b - mac interference network optimization under rate constraints by polite water - filling and duality , '' _ ieee trans . signal processing _ , vol . 59 , no . 1 263 276 , jan . 2011 . a. liu , v. k. n. lau , and y. liu , `` local dual method for optimization of parallel mimo b - mac interference networks under multiple linear constraints , '' _ submitted to ieee trans . signal processing _ 2011 ; revised apr . 2012 . [ online ] . available : http://www.ee.ust.hk/~eeknlau/hkust-office-homepage/publications.html

nonrelativistic formalism is developed , which allows describing systems with internal degrees of freedom in the scalar potential field , which is a function both on relative coordinates and time , and on relative speed and accelerations . the equation for energy , which is an integral of motion when satisfies to certain differential condition , is derived for the general case . for a free mass point all solutions of the equations of motion in the center - of - inertia reference frame , moving with constant velocity , are found . as a result , the center of mass follows along helical and more complicated trajectories round a direction of motion of the center of inertia . this motion can be interpreted as trembling movement ( zitterbewegung ) . on this basis a conclusion is done that zitterbewegung has purely classical origin , arising even in a nonrelativistic case if internal degrees of freedom are taken into account . the general equation of motion for a spin which can be interpreted from positions of a classical mechanics is written down . application of the obtained results to the electron leads to new conception of electric charge , sign of which corresponds to a sign of spin polarization . pacs numbers : 14.60cd , 45.20.d , 45.50.j keywords : classical mechanics , internal degrees of freedom , equations of motion , energy conservation , electron

1201.4965

experimental data on high energy physics say that elementary particles are complex systems with internal degrees of freedom . fundamental particle which in certain relation can pretend to simplicity , on the one hand , and underlie all known interactions , on the other hand , is the electron . nonexistence of the electron , being a source of electromagnetic field , in the maxwell theory and its generalizations is paradoxical . it may be explained by that the maxwell theory describes macroscopical averaged fields created by systems consisting of large quantity of moving charges . therefore the electron theory , whose foundations were laid by lorentz , is necessary for the description of the majority of the electromagnetic phenomena in material systems . the electron theory has been developing in classical trend after discovery of the electron by j.j.thomson in cathode rays . the theory by abraham @xcite , based on the maxwell - lorentz electrodynamics , became the first theory considering internal structure of the electron . during two decades preceding experience by stern - gerlach and the kronig - uhlenbeck - goudsmit hypothesis about electron spin , there were appeared a few alternative theories and guesses about electron structure . it should be mentioned here works by compton @xcite- @xcite , influenced by both the paper by parson @xcite and stereoscopic photos of tracks of the @xmath1-rays , made by wilson , some of which had almost ideal form of a helix . compton has come to a conclusion that the model of the electron , `` spinning like a tiny gyroscope '' , can eliminate difficulties in an explanation of curvature of these tracks , as well as of richardson - barnett effect and diffraction of x - rays by magnetic crystals . nevertheless , the parson - compton theory continuing early idea of vortex atoms , going back to greek atomists , kepler , descartes , leibnitz , svedenborg , boskovich , ampere , kelvin , and theories of some other researchers had rather natural philosophic and empirical nature , than they gave any mathematical instrument for future theory . the first mathematical realization of the parson - compton theory was the work by frenkel @xcite , where the electron was considered as a point with six - vector of magnetic moment , what has allowed to explain the anomalous zeeman effect . after creation of quantum mechanics , especially after works by pauli and dirac , the electron theory began to develop chiefly in quantum direction . in 1930 schrdinger has shown that the trembling motion ( zitterbewegung ) of the electron takes place in the dirac theory where eigenvalues of any component of the velocity operator are equal to @xmath2 . microscopic trembling motion with the velocity of light of imaginary center of the cloud of charge , whose amplitude is about half compton wave length , is imposed on translational macroscopic motion of the center of mass of the electron . as shrdinger talks , `` exclusively entangled relations , which are present according to the dirac equation already at free movement of a mass point , seem to me worthy of enunciating though i can not present some completed result of this research '' @xcite . in 1952 huang has shown that zitterbewegung of a free dirac electron may be looked upon as a circular motion about the direction of the electron spin that in turn may be interpreted as the `` orbital angular momentum '' of this motion . as a result the electric current produced by zitterbewegung is seen to give rise to the intrinsic magnetic moment of the electron @xcite . however , as early as in 1937 mathisson has written down general relativistic equations of motion for systems possessed multipole momenta @xcite . he has shewed also , that their application to a particle with spin treated as a dipole gives rise to equations of motion , describing zitterbewegung @xcite . mathisson has considered the motion of a free uncharged particle with spin and free electron taking account of the reaction of radiation . here he has assumed spin to be a constant ( pseudo-)vector , what was rather stringent assumption , for spin direction in general can change . nevertheless , having associated trembling motion of the electron with the de broglie s wave , he has obtained the well - known value of a spin @xmath3 . following basic articles by schrdinger and mathisson a lot of works were appeared , developing both quantum and classical electron theory and establishing connection between them . sufficiently full list of references one may find in the books @xcite , @xcite . despite a considerable quantity of researches devoted to the theory of the electron , there are many unclear matters associated both with radiation of the electron and with dependence of its trajectory from its spin ( see , e.g. , @xcite ) . an interest in classical theory of the electron has especially increased at the recent years , and this work contains some arguments in favour of a classical origin of zitterbewegung . relating of the electron spin with its proper rotation allows considering the electron as non - inertial object which can be described as a mass point with internal degrees of freedom @xcite , @xcite . the formalism of such description based on generalization of the second newton s law is considered in 2 . here the equations of motion of the point interacting with an external field described by the potential function depending on relative variables are obtained . consequence of the equation of motion is the equation of balance of energy which is integral of motion only if a certain condition is fulfilled . non - conservation of the energy , in general , seems to be caused by the fact that mass point in question is non - inertial system . it is supposed that internal degrees of freedom are described by the pseudo - vectors @xmath4 and @xmath5 connected both with internal structure of the point , and with its interaction with external fields . it is shown in 3 that equation of motion of free mass point reduces to a conservation of the velocity of the center of inertia . to obtain solutions natural equations of motion are introduced for internal degrees of freedom describing precession of pseudo - vectors @xmath4 and @xmath5 round the direction of the velocity of the center of inertia . all solutions of the equation of motion for a free mass point ( when @xmath6 and @xmath7 ) in the center - of - inertia reference frame are found in 4 . the center of inertia proves to be does not coincide with the center of mass . as a result , the center of mass moves by a complicated trajectory round the direction of motion of the center of inertia . some solutions in the center - of - inertia reference frame are infinite . it is shown that they become finite at zero energy of the mass point . it is of interest that equation of motion admits also solutions for zero mass and transversal polarization . in 5 equation of moments is considered and speculations are contained about the physical sense of pseudo - vectors @xmath8 and @xmath9 and their relation with spin whose equation of motion in general case we deal with in 6 , which contains also conclusive remarks on possible interpretation of obtained solutions . a mass point with internal degrees of freedom can be considered as a non - inertial system whose equation of motion taking into account its interaction with an external field has the form of the newton s second law ( @xcite-@xcite ) @xmath10 where @xmath11 \ ; , \eqno{(2.2)}\ ] ] is a dynamical momentum of the point , @xmath12 is its rest mass , @xmath13 \ ; , \eqno{(2.3)}\ ] ] is a force , acting to the point . expressions ( 2.2 ) and ( 2.3 ) follow from the definition of the elementary work of force , @xmath14 , if potential function @xmath0 depends on velocity . potential function characterizes both medium in which the point moves and interaction of the point with physical objects which are in this medium . therefore it should be assumed that @xmath0 generally can depend on time @xmath15 , relative coordinates @xmath16 , velocity @xmath17 and accelerations @xmath18 , @xmath19 , so that @xmath20 . internal degrees of freedom are characterized by pseudo - vectors @xmath4 and @xmath5 connected with both internal structure of mass point and interaction . hence , they can be represented as sums @xmath21 where @xmath22 and @xmath23 are connected exclusively with interaction and depend on the same variables as potential function ; @xmath24 and @xmath25 are connected exclusively with internal structure of mass point and during its motion they can change only in the direction but not in the module provided an interaction is neglected . if the function @xmath0 is represented in the form @xmath26 ) = u_0 + ( { \bf v } \cdot [ { \bf r}\times{\bf c } ] ) = u_0 - ( [ { \bf r } \times { \bf v } ] \cdot { \bf c } ) \ ; , \eqno{(2.5)}\ ] ] then eqs . ( 2.2 ) and ( 2.3 ) take the form @xmath27 - [ { \bf r}\times { \bf c } ] + ( [ \mathbf{r } \times \mathbf{v } ] \cdot \frac{\partial \mathbf{c}^{ext}}{\partial { \bf v } } ) \ ; , \eqno{(2.6)}\ ] ] @xmath28 \cdot \frac{\partial \mathbf{c}^{ext}}{\partial { \bf r } } ) \ ; , \eqno{(2.7)}\ ] ] where @xmath29 \cdot \frac{\partial{\bf c}^{ext}}{\partial{\bf r}})_i = \varepsilon_{klm } r^{k } v^{l } \frac{\partial({\bf c}^{ext})^m}{\partial{r^i } } \ ; , \eqno{(2.8)}\ ] ] @xmath29 \cdot \frac{\partial{\bf c}^{ext}}{\partial{\bf v}})_i = \varepsilon_{klm } r^{k } v^{l } \frac{\partial({\bf c}^{ext})^m}{\partial{v^i } } \;. \eqno{(2.9)}\ ] ] it follows from eq.(2.1 ) , which is reduced to equation @xmath30 - [ { \bf r}\times{{\bf c}_0 } ] } \right ) = - \frac{\partial u_0}{\partial { \bf r } } + ( [ { \bf r}\times { \bf v } ] \cdot \frac{\partial{\bf c}^{ext}}{\partial{\bf r } } ) + \ ] ] @xmath31 \cdot \frac{\partial{\bf c}^{ext}}{\partial{\bf v } } ) - [ { { \bf s}^{ext}}\times { \bf w } ] + [ { \bf r}\times{{\bf c}^{ext } } ] \right ) \ ; , \eqno{(2.10)}\ ] ] that there take place an equation for energy @xmath32 \cdot \frac{\partial \mathbf{c}^{ext}}{\partial t } ) + \ ] ] @xmath33 \cdot \frac{\partial \mathbf{c}^{ext}}{\partial { \bf w}^{(k ) } } ) \cdot { \bf w}^{(k + 1 ) } ) } \;,\eqno{(2.11)}\ ] ] where @xmath34 \cdot { \bf s}_0 ) - ( [ { \bf v } \times { \bf w } ] \cdot { \bf s}^{ext } ) - ( { \bf v } \cdot \frac{\partial u_0}{\partial { \bf v}})+\ ] ] @xmath35 \cdot \frac{\partial \mathbf{c}^{ext}}{\partial \mathbf{v } } ) ) + u_0 \ ; , \eqno{(2.12)}\ ] ] @xmath36 \cdot \frac{\partial \mathbf{c}^{ext}}{\partial \mathbf{v } } ) ) = \varepsilon_{klm } r^{k } v^{l } v^{i } \frac{\partial ( \mathbf{c}^{ext})^{m}}{\partial v^{i } } \ ; , \eqno{(2.13)}\ ] ] @xmath37 \cdot \frac{\partial \mathbf{c}^{ext}}{\partial \mathbf{w}^{(k ) } } ) \cdot { \bf w}^{(k+1 ) } ) = \varepsilon_{klm } r^{k } v^{l } ( \mathbf{w}^{(k+1)})^{i } \frac{\partial ( \mathbf{c}^{ext})^{m}}{\partial ( \mathbf{w}^{(k)})^{i } } \;. \eqno{(2.14)}\ ] ] obviously , the energy ( 2.12 ) is integral of motion , if r.h.s . of eq . ( 2.11 ) becomes zero . in this section we will consider a motion of free mass point m with internal degrees of freedom , and will not concern its interaction and physical sense of quantities @xmath4 and @xmath5 . mass point in question will be free , if @xmath38 , @xmath39 , @xmath40 . then equation ( 2.10 ) gives rise to conservation of the vector @xmath41 - [ { \bf r}\times { \bf c}_0 ] = m{\bf v}_{\mathrm{c } } \ ; , \eqno{(3.1)}\ ] ] where @xmath42 is kinetic momentum of the point m being _ a center of mass _ , @xmath43 is an effective mass . it is reasonably to term the vector ( 3.1 ) as the kinetic momentum associated with the point m. here @xmath44 is a velocity of some point c specified by radius - vector @xmath45 where @xmath46 is a radius - vector of initial position of the point c. it follows from eq.(2.7 ) that the vector @xmath47 is a constant vector , if @xmath48 does not depend on the relative radius - vector . according to eq.(3.2 ) the point c moves inertially with velocity @xmath49 . hence , it is _ a center of inertia _ , which in general does not coincide with the center of mass m , moving along some trajectory round the direction of @xmath49 . figure 1 shows parameters of this trajectory . parameters of the trajectory of the mass point m eq.(3.1 ) may be rewritten in the form @xmath50 - [ { \bf r}\times { \bf c}_0 ] = ( m - m_0){\bf v_{\mathrm{c}}}\ ; , \eqno{(3.3)}\ ] ] where @xmath51 are radius - vector , velocity and acceleration of the center of mass m relative to the center of inertia c , respectively . the energy ( 2.12 ) can be expressed as @xmath52 where @xmath53 is a kinetic energy of the center of inertia as though the mass of the point m was in the point c , @xmath54 ) \ ; \eqno{(3.7)}\ ] ] is additional kinetic energy stipulated by both the motion of the center of inertia and the motion of the point m relative to the center of inertia c , @xmath55 ) \ ; \eqno{(3.8)}\ ] ] is a kinetic energy of the point m stipulated by its motion relative to the center of inertia . according to equation of motion ( 3.1 ) radius - vector @xmath56 should be determined by pseudo - vectors @xmath57 and @xmath58 , which are precessing with the same velocity about the direction of the vector @xmath59 . therefore @xmath57 and @xmath58 ought to satisfy to following equations of motion @xmath60 \ ; , \frac{d \textbf{c}_0}{dt } = [ \mathbf{\omega}_{0 } \times \textbf{c}_{0 } ] \ ; , \eqno{(3.9)}\ ] ] where @xmath61 is an angular velocity of precession , @xmath62 has a dimension of inverse length . if we choose the axis z to be coincided with the direction of @xmath49 , i.e. @xmath63 , then @xmath64 can be both positive , and negative quantity . unit vectors @xmath65 and @xmath66 may be chosen in the following form @xmath67 \ ; , \eqno{(3.11)}\ ] ] @xmath68 \ ; , \eqno{(3.12)}\ ] ] where @xmath69^{-2 } = \frac{1}{\mathbf{r}_0^{2 } \mathbf{\omega}_{0}^{2 } \sin^{2 } \theta } \ ; , \eqno{(3.13)}\ ] ] @xmath70^{2 } = \mathbf{\omega}_{0}^{2 } k^{-1 } \ ; , \eqno{(3.14)}\ ] ] @xmath71 is a radius - vector of the center of mass m relative to the center of inertia c at initial time @xmath72 ; @xmath73 corresponds to that the vector @xmath74 lies in a plane , perpendicular to a direction of motion . equations ( 3.9 ) have solutions @xmath75 @xmath76 where @xmath77 , @xmath78 are constant angles between @xmath79 , @xmath80 and direction of @xmath81 , respectively . it is convenient to solve equations ( 3.3 ) in the center - of - inertia reference frame , where @xmath82 , @xmath83 , @xmath84 , @xmath85 . let us introduce dimensionless variable @xmath86 and denotations @xmath87 @xmath88 @xmath89 then , representing @xmath90 in the form @xmath91 z(\xi ) \ ; , \eqno{(3.20)}\ ] ] where functions @xmath92 and @xmath93 satisfy to initial condition @xmath94|}{r_{0 } \omega_{0 } } = \sin{\theta } \ ; , \eqno{(3.21)}\ ] ] we reduce equation of motion ( 3.3 ) and equation of energy ( 3.8 ) to following system @xmath95 @xmath96 @xmath97 ' = 0 \;. \eqno{(3.24)}\ ] ] @xmath98 @xmath99 the detailed analysis of equations ( 3.22)-(3.25 ) leads to following possible solutions ( @xcite ) . * i. * @xmath100 , @xmath101 , @xmath102 . @xmath103 , @xmath104 , @xmath105 , @xmath106 , @xmath107 , @xmath108 . @xmath109 @xmath110 the equation of a trajectory looks like @xmath111 \ ; , \eqno{(4.3)}\ ] ] i.e. the point m moves on a circle round z - direction with angular velocity @xmath112 . the direction of pseudo - vector @xmath79 is given by eq.(4.1 ) , and the direction of pseudo - vector @xmath80 is determined by eq.(4.2 ) , following from the condition @xmath113 . an angle @xmath1 between @xmath79 and @xmath80 may be found from the relation @xmath114 @xmath115^{\frac{1}{2 } } \;. \eqno{(4.4)}\ ] ] * i.2 . * @xmath103 , @xmath104 , @xmath116 , @xmath106 , @xmath107 , @xmath117 , @xmath118 . this case is obtained from the previous one at @xmath119 . @xmath120 @xmath121 the trajectory is described by eq.(4.3 ) and represents a circle of radius @xmath122 , lying in a plane , perpendicular to a direction of the motion of the center of inertia c. a direction of the motion of the point m in the cases i.1 , i.2 is determined by sign of angular velocity of precession @xmath112 , consistent with equations ( 4.1 ) and ( 4.2 ) . for example , for @xmath119 and @xmath123 we have @xmath124 , whence it follows @xmath125 . hence , such a point moves by left spiral directed along @xmath81 . an angle @xmath1 between @xmath79 and @xmath80 may be found from the relation @xmath126^{\frac{1}{2 } } - \frac{m_{0}^{2}}{4 s_{0 } c_{0 } } \;. \eqno{(4.7)}\ ] ] * i.3 . * @xmath103 , @xmath127 , @xmath116 , @xmath128 , @xmath107 , @xmath117 , @xmath129 . @xmath130 @xmath131 @xmath132 the equation of a trajectory looks like @xmath111 ( \pm \omega_{\varepsilon } t + 1 ) \ ; , \eqno{(4.11)}\ ] ] where @xmath133 . in the center - of - inertia reference frame the trajectory is a plane helix , perpendicular to the direction of @xmath81 . in the laboratory reference frame the trajectory is convergent and then divergent conical spiral . at @xmath119 the trajectory ( 4.11 ) becomes finite and takes the form ( 4.3 ) . @xmath103 , @xmath116 , @xmath128 , @xmath107 , @xmath134 . @xmath137 @xmath138 @xmath139 the equations of a trajectory are @xmath111 \cos{\omega_{\varepsilon } t } \ ; , \ ; e_{0 } < 0 \ ; , \eqno{(4.17)}\ ] ] @xmath111 \cosh{\omega_{\varepsilon } t } \ ; , \ ; e_{0 } \geq 0 \;,\eqno{(4.18)}\ ] ] here , as in the preceding case infinite trajectories ( 4.18 ) become finite ones at @xmath119 . * i.5 . * @xmath103 , @xmath116 , @xmath128 , @xmath140 , @xmath141 . @xmath144 @xmath145 @xmath146 the equations of a trajectory are @xmath147 \cos{(\omega_{c } t \mp \varphi_{c } ) } \ ; , \ ; 0 < \mu_{c } < -\varepsilon_{0 } \ ; , \ ; e_{0 } < 0 \ ; , \eqno{(4.24)}\ ] ] @xmath148 \cosh{(\omega_{c } t \pm \varphi_{c } ) } \ ; , \ ; \mu_{c } \leq 0 \ ; , \ ; \mu_{c } \leq -\varepsilon_{0 } \ ; , \eqno{(4.25)}\ ] ] where @xmath149 @xmath150 infinite trajectories ( 4.25 ) can be excluded , having imposed a condition @xmath151 . then ( 4.25 ) reduces to ( 4.3 ) . @xmath152 , @xmath116 , @xmath153 , @xmath140 , @xmath117 , @xmath154 . @xmath155 @xmath156 @xmath157 @xmath158 the equation of a trajectory is @xmath111 \biggl[\pm \frac{\sqrt{-2m_{0}e_{0}}}{2r_{0 } \omega_{0}^{2 } s_{0 } } \omega_{0 } t + 1 \biggr ] \pm\ ] ] @xmath159 reducing to finite form ( 4.3 ) at @xmath119 . * @xmath152 , @xmath116 , @xmath160 , @xmath140 , @xmath161 . @xmath162 @xmath163 @xmath164 @xmath165 @xmath166 the equations of a trajectory are @xmath147 \cos{(\omega_{c } t \pm \varphi_{c } ) } + \ ] ] @xmath167 \mathbf{e}_{z } \ ; , \ ; 0 < \mu_{c } < -\varepsilon_{0 } \ ; ; \eqno{(4.38)}\ ] ] @xmath148 \cosh{(\omega_{c } t \pm \varphi_{c } ) } + \ ] ] @xmath168 \mathbf{e}_{z } \ ; , \ ; \mu_{c } \leq 0 \ ; , \ ; \mu_{c } \leq -\varepsilon_{0 } \ ; , \eqno{(4.39)}\ ] ] where @xmath169 is determined in eqs.(4.26)-(4.27 ) , @xmath170 infinite trajectories ( 4.39 ) becomes finite ones , when @xmath171 . then the equation ( 4.39 ) takes the form @xmath111 - r_{0 } \sqrt{\frac{4 \omega_{0}^{2 } s_{0}^{2}}{m_{0}^{2 } } -1 } \mathbf{e}_{z } \ ; , \ ; \eqno{(4.41)}\ ] ] and it follows from ( 4.34 ) that @xmath172 * ii . * @xmath100 , @xmath101 , @xmath173 . in this case equation ( 2.10 ) is a generalization of non - relativistic frenkel - mathisson - weyssenhoff equation ( @xcite , @xcite , @xcite ) , describing the motion of point particle with constant spin @xmath174 . it may be deduced from the cases i.5 and i.7 at @xmath175 , @xmath140 . as a result we have following variants . * @xmath176 , @xmath116 , @xmath128 , @xmath140 , @xmath175 . @xmath177 @xmath178 the equation of a trajectory looks as @xmath179 \cos{(\omega_{0 } t \mp \varphi ) } \ ; , \ ; e_{0 } < 0 \ ; , \eqno{(4.45)}\ ] ] where @xmath180 * ii.2 . * @xmath152 , @xmath116 , @xmath181 , @xmath140 , @xmath175 . @xmath182 @xmath183 the equation of a trajectory is @xmath179 \cos{(\omega_{0 } t \cos{\alpha_{s } \mp \varphi ) } } + \ ] ] @xmath184 \mathbf{e}_{z } \ ; , \ ; e_{0 } < 0 \;. \eqno{(4.49)}\ ] ] the case ii.1 is deduced from the case ii.2 at @xmath128 , i.e. if conditions ( 4.43 ) fulfill . assuming for the electron @xmath185 , @xmath186 , we obtain @xmath187 , or @xmath188 is the frequency corresponding to generally accepted rest energy of the electron or its compton wavelength @xmath189 . the energy of the electron in the center - of - inertia reference frame is negative , @xmath190 . polarization , or spin projection to the direction of motion , is determined by the value @xmath191 value @xmath192 corresponds to @xmath193 , @xmath124 , i.e. to counter - clockwise motion , whereas @xmath194 corresponds to @xmath195 , @xmath196 , i.e. to clockwise motion . it suggests associating these motions with motions of the electron and positron , which hence should differ from each other by the type of motion rather than by charge . more strictly it it is possible to prove or refuse this hypothesis , having considered two - body problem taking into account their interaction with each other , as well as their motion in constant electric and magnetic fields . * @xmath100 , @xmath197 , @xmath102 . in this case @xmath128 , @xmath198 , and equations ( 3.22)-(3.25 ) give solutions @xmath176 , @xmath199 , and relations @xmath200 , @xmath201 , or @xmath202 it is easy to see that this case turns out from case i.1 at @xmath128 . the equation of a trajectory looks like eq.(4.3 ) . the system of equations ( 3.22)-(3.25 ) admits the solution corresponding to zero mass , @xmath203 . in this case @xmath175 , @xmath140 , @xmath199 , and equation of motion gives following variants . * @xmath203 , @xmath101 , @xmath102 . @xmath204 in the center - of - inertia the equation of a trajectory looks like eq.(4.3 ) . * iv.2 . * @xmath203 , @xmath101 , @xmath173 . @xmath205 in the center - of - inertia the equation of a trajectory looks like @xmath111 + r_{0 } \left[-\frac{\cot{\alpha_{s}}}{2 } \omega_{0}^{2 } t^{2 } + b_{1 } \omega_{0 } t \right ] \mathbf{e}_{z } \;. \eqno{(4.55)}\ ] ] to eliminate such divergent trajectories , it is sufficient to assume @xmath206 , @xmath207 , what corresponds to transversal polarization of pseudo - vector @xmath79 . * iv.3 . * @xmath203 , @xmath197 , @xmath102 . in this case we have @xmath176 , @xmath208 , @xmath209 , what corresponds to transversal polarization of pseudo - vector @xmath80 . the equation of a trajectory looks like eq.(4.3 ) . summarizing the results obtained , it is possible to assert that all finite trajectories of free mass points with internal degrees of freedom are subdivided into three types . trajectories of the first type are right or left helix along the direction of motion of the center of inertia . they are specific for the cases i.1-i.3 , i.6 , i.7 ( at @xmath119 ) , iii and iv , and there is no restrictions in @xmath210 only for the cases i.1 and iv.1 , whereas for the rest cases we have @xmath119 . figure 2 shows trajectories for the case i.3 for polarization @xmath194 ( clockwise motion along z - axis , figure 2a ) and @xmath192 ( counter - clockwise motion along z - axis , figure 2b ) . trajectories of the second type in the center - of - inertia reference frame are plane multi - petal rosettes . they are specific for the cases i.4 , i.5 , i.7 ( at @xmath211 ) and ii.1 . they are closed @xmath212-petal plane rosettes , when frequencies @xmath213 , @xmath214 and @xmath215 are multiple to the frequency @xmath112 and represented in figure 3 for the case i.4 ( @xmath216 , @xmath217 ) . as in preceding cases the motion is clockwise for @xmath194 ( figure 3a ) and counter - clockwise one for @xmath192 ( figure3b ) . the direction of pseudo - vector @xmath80 in figures 2 and 3 does not pointed out . figure 2 . figure 3 . figure 4 . trajectories of the third type are specific for the cases i.7 and ii.2 . they are multi - petal symmetric space curves in the center - of - inertia reference frame . the number of petals is defined by a condition of the maximum removal of the mass point from the center of inertia that takes place at @xmath218 , i.e. at @xmath219 these curves are closed @xmath212-petal space rosettes , when frequency @xmath215 is multiple to the frequency @xmath112 , @xmath220 , or @xmath221 the motion is going clockwise for @xmath222 and counter - clockwise for @xmath222 . figure 4 shows samples of trajectories for @xmath222 , @xmath223 ( figure 4a ) , @xmath224 ( figure 4b ) and for the nonintegral @xmath225 ( figure 4c ) in the interval @xmath226 . as it is known , one of internal property of particles is spin , associated classically with proper angular momentum of particle . therefore a temptation arises to connect pseudo - vectors @xmath4 and @xmath5 with spin . for the sake of it we will consider the equation of moments @xmath227 where @xmath228 = m_{0 } [ \mathbf{r } \times \mathbf{v } ] - [ \mathbf{r } \times \frac{\partial u}{\partial \mathbf{v } } ] + [ \mathbf{r } \times [ \mathbf{s } \times \mathbf{w } ] ] = \ ] ] @xmath229 - [ \mathbf{r } \times \frac{\partial u_0}{\partial \mathbf{v } } ] + [ \mathbf{r } \times [ \mathbf{s } \times \mathbf{w } ] ] - [ \mathbf{r } \times [ \mathbf{r } \times \mathbf{c } ] ] + [ \mathbf{r } \times ( [ \mathbf{r } \times \mathbf{v } ] \cdot \frac{\partial \mathbf{c}^{ext}}{\partial \mathbf{v } } ) ] \eqno{(5.2)}\ ] ] is a dynamical angular momentum , @xmath230 = - [ \mathbf{r } \times \frac{\partial u}{\partial \mathbf{r } } ] + [ \mathbf{r } \times [ \mathbf{c } \times \mathbf{v } ] ] = \ ] ] @xmath231 + [ \mathbf{r } \times ( [ \mathbf{r } \times \mathbf{v } ] \cdot \frac{\partial \mathbf{c}^{ext}}{\partial \mathbf{r } } ) ] \eqno{(5.3)}\ ] ] is a moment of force , acting at the mass point , @xmath232 = - [ \mathbf{v } \times \frac{\partial u}{\partial \mathbf{v } } ] + [ \mathbf{v } \times [ \mathbf{s } \times \mathbf{w } ] ] = \ ] ] @xmath233 + [ \mathbf{v } \times [ \mathbf{s } \times \mathbf{w } ] ] - [ \mathbf{v } \times [ \mathbf{r } \times \mathbf{c } ] ] + [ \mathbf{v } \times ( [ \mathbf{r } \times \mathbf{v } ] \cdot \frac{\partial \mathbf{c}^{ext}}{\partial \mathbf{v } } ) ] \eqno{(5.4)}\ ] ] is an additional twisting moment , or torque . the equation of moments ( 5.1 ) follows from equation of motion ( 2.1 ) . therefore , having only definition ( 5.2 ) for angular momentum it is impossible to define a concept of proper angular momentum for the point with internal degrees of freedom . indeed , let us consider free mass point , for which @xmath234 , @xmath235 , @xmath236 . then the moment of force ( 5.3 ) becomes zero , and the dynamical angular momentum ( 5.2 ) and torque ( 5.4 ) are equal to @xmath237 + [ \mathbf{r } \times [ \mathbf{s}_{0 } \times \mathbf{w } ] ] - [ \mathbf{r } \times [ \mathbf{r } \times \mathbf{c}_{0 } ] ] \ ; , \eqno{(5.5)}\ ] ] @xmath238 - [ \mathbf{v } \times [ \mathbf{r } \times \mathbf{c}_{0 } ] ] \;. \eqno{(5.6)}\ ] ] respectively . introducing variables relative to the center of inertia ( 3.4 ) , we have @xmath239 @xmath240 where @xmath241 - [ \mathbf{r}_{\mathrm{c } } \times [ \mathbf{r}_{\mathrm{c } } \times \mathbf{c}_{0 } ] ] \eqno{(5.9)}\ ] ] is an angular momentum of the center of inertia c relative to the origin o , as if total rest - mass @xmath12 was in the center of inertia c , @xmath242 + m_{0 } [ \mathbf{r}_{\mathrm{c } } \times \mathbf{v } ] + [ \mathbf{r}_{\mathrm{c } } \times [ \mathbf{s}_{0 } \times \mathbf{w } ] ] - [ \mathbf{r}_{\mathrm{c } } \times [ \mathbf{r } \times \mathbf{c}_{0 } ] ] - [ \mathbf{r } \times [ \mathbf{r}_{\mathrm{c } } \times \mathbf{c}_{0 } ] ] \eqno{(5.10)}\ ] ] is an angular momentum of the center of mass m relative to the origin o , related with both its motion relative to the center of inertia c and a motion of the latter one in absolute reference frame , @xmath243 + [ \mathbf{r } \times [ \mathbf{s}_{0 } \times \mathbf{w } ] ] - [ \mathbf{r } \times [ \mathbf{r } \times \mathbf{c}_{0 } ] ] \eqno{(5.11)}\ ] ] is an angular momentum of the center of mass m relative to the center of inertia c ; @xmath244 \eqno{(5.12)}\ ] ] is a torque relative to the origin o , acting upon the center of mass m , @xmath245 - [ \mathbf{v}_{\mathrm{c } } \times [ \mathbf{r } \times \mathbf{c}_{0 } ] ] - [ \mathbf{v } \times [ \mathbf{r}_{\mathrm{c } } \times \mathbf{c}_{0 } ] ] \eqno{(5.13)}\ ] ] is additional torque relative to the origin o , acting upon the center of mass m and related with both its motion relative to the center of inertia c and a motion of the latter one in absolute reference frame , @xmath246 - [ \mathbf{v } \times [ \mathbf{r } \times \mathbf{c}_{0 } ] ] \eqno{(5.14)}\ ] ] is a torque relative to the center of inertia c , acting upon the center of mass m. in the center - of - inertia reference frame we have @xmath247 , @xmath248 , @xmath249 , @xmath250 , so that the equation of moments ( 5.1 ) takes the form @xmath251 or @xmath252 -[\mathbf{r } \times \mathbf{c}_{0 } ] ] ] = \mathbf{0 } \;. \eqno{(5.16)}\ ] ] in the reference frame in question @xmath253 -[\mathbf{r } \times \mathbf{c}_{0 } ] = \mathbf{0}$ ] , therefore equations ( 5.15 ) and hence ( 5.1 ) are identities . on the other hand , since @xmath254 , @xmath255 , @xmath256 , in the center - of - mass reference frame , then @xmath257 . therefore @xmath258 can not play a role of proper angular momentum ( spin ) of the point m to which we want to relate pseudo - vectors @xmath4 and @xmath5 . for definition of their physical sense additional reasons are necessary . point m with internal degrees of freedom should be considered as non - inertial extended object rotating with angular velocity @xmath259 and possessing the proper angular momentum ( spin ) . to take into account internal rotational degrees of freedom it is necessary to introduce the total moment of momentum instead of angular moment ( 5.7 ) ( see , for example , @xcite ) @xmath260 which is defined as spin @xmath261 in the center - of - mass reference frame , and equals to @xmath262 in the center - of - inertia reference frame . when interaction is missing , equation of motion ( 2.1 ) reduces to eq.(3.1 ) . consequently , pseudo - vectors @xmath263 , @xmath258 and @xmath261 in the center - of - inertia reference frame should be precessing round the direction of the center of inertia with the same angular velocity @xmath264 , as pseudo - vectors @xmath79 and @xmath80 , i.e. they have to satisfy to equations of motion of the same form @xmath265 = \sigma [ \mathbf{v}_{\mathrm{c } } \times \mathbf{j}_{0 } ] \ ; , \eqno{(5.19)}\ ] ] @xmath266 = \sigma [ \mathbf{v}_{\mathrm{c } } \times \mathbf{l}_{0 } ] \ ; , \eqno{(5.20)}\ ] ] @xmath267 = \sigma [ \mathbf{v}_{\mathrm{c } } \times \mathbf{s } ] \;. \eqno{(5.21)}\ ] ] comparison of eq.(5.20 ) with eq.(5.15 ) and taking into account eq.(5.14 ) gives @xmath268 = \mathbf{t}_0 = [ \mathbf{v } \times ( [ \mathbf{s}_{0 } \times \mathbf{w } ] - [ \mathbf{r } \times \mathbf{c}_0 ] ) ] = - [ \mathbf{v } \times m \mathbf{v } ] = \mathbf{0 } \ ; , \eqno{(5.22)}\ ] ] whence it follows that @xmath269 is parallel to the angular velocity of precession @xmath270 , @xmath243 + [ \mathbf{r } \times [ \mathbf{s}_{0 } \times \mathbf{w } ] ] - [ \mathbf{r } \times [ \mathbf{r } \times \mathbf{c}_{0 } ] ] = i_{0 } \mathbf{\omega}_{0 } = m_{0 } r_{0}^{2 } \mathbf{\omega}_{0 } \ ; , \eqno{(5.23)}\ ] ] whereas spin @xmath261 is parallel to the angular velocity @xmath271 of proper rotation of extended point m , if it is defined as its proper angular momentum . then pseudo - vector @xmath258 gets a sense of orbital angular momentum of the point m relative to the center of inertia . if @xmath272 is a proper tensor of inertia of the point m , the spin is defined as follows @xmath273 where @xmath274 is eigenvalue of @xmath275 , i.e. proper moment of inertia of the point m relative to the axis of its rotation . due to eq.(5.23 ) and eq.(5.24 ) the total moment of momentum ( 5.18 ) is equal to @xmath276 where @xmath277 is the tensor of inertia of the point m relative to the center of inertia c , @xmath278 is the radius of zitterbewegung . applying the expression ( 5.24 ) to electron roughly represented as a rigid sphere of radius @xmath279 , consisting of structureless mass points , we have @xmath280 , @xmath281 , whence it follows @xmath282 substituting here a value of the electron radius @xmath283 ( dehmelt , @xcite ) , we will obtain an estimate @xmath284 i.e. frequency of proper rotation of the electron is at least 19 orders greater than the frequency of zitterbewegung ( 4.50 ) , @xmath285 , whereas a velocity on the electron surface is @xmath286 , what is 9 orders greater than the speed of light . at high frequency of proper rotation in electron volume there should be arising huge centrifugal forces of inertia , which relocate interior substance of the electron to periphery . on the other hand , stability of the electron implies that centrifugal forces of inertia should be balanced by interior forces so that the equilibrium shape of the electron represented something like a toroid or a ring with a rigid surface . it is consistent with both earlier idea by parson - compton @xcite , @xcite , and modern toroidal or ring model ( see , e.g. , @xcite , @xcite ) or dumbbell model of the electron ( @xcite ) . ring ( or dumbbell ) is characterized , at least , by two sizes , by its radius ( length ) and thickness . therefore it is difficult to say what value @xmath287 obtained by dehmelt concerns . even if @xmath288 is a classical electron radius , @xmath289 , we will obtain instead of ( 5.27 ) an estimate for the frequency of rotation @xmath290 and for the velocity @xmath291 , that also is greater than the speed of light . in due time lorentz has refused an idea of extended electron as equatorial velocity of a surface of spinning electron has turned out to be more than the speed of light . however as long as we are in the frameworks of classical mechanics , we have not any restriction on speed . whatever the electron would be actually arranged , its internal structure should determine both field , created by it , and a type of its motion , depending on a spin being integral property of this structure . if the motion of free electron reduces to equations ( 3.3 ) , ( 3.9 ) , then in the center - of - inertia reference frame we obtain trajectories , described in 4 . a motion of free extended electron relative to the center of inertia means that , on the one hand , its interior substance is acted upon by centrifugal forces of inertia , and , on the other hand , by centripetal forces which twist a trajectory . resultant of these forces can be represented as @xmath292 $ ] , where @xmath293 is some coefficient with dimension of mass . for the point mass particle there takes place a kind of equivalence principle , @xmath294 , where the rest mass @xmath12 play a role of the measure of inertness of inertially moving mass point , and a mass @xmath293 is a measure of non - inertiality of such a point moving with acceleration , whereas a mass @xmath43 , entering to right hand side of eq.(3.1 ) , is a measure of inertness of extended object , which center of inertia moves inertially . therefore the hypothesis @xmath294 is not obvious for such extended particle as the electron , the more so mass of micro - object depends on its interaction with external fields and is not an additive quantity . inasmuch as @xmath295 $ ] is resultant force acting upon free electron , it follows from equation ( 2.3 ) , written in the center - of - inertia reference frame , @xmath296 \right)_{u = 0 , \mathbf{v}_{\mathrm{c } } = \mathbf{0 } } = [ \mathbf{c}_{0 } \times \mathbf{v } ] = m ' [ \mathbf{v } \times \mathbf{\omega}_0 ] \;. \eqno{(5.28)}\ ] ] from here pseudo - vector @xmath80 considering eq.(3.10 ) may be determined as @xmath297 where @xmath298 and @xmath299 are constant pseudo - scalars . it should be noted that transformation ( 3.4 ) is a special case of the galileo transformation @xmath300 where @xmath301 is a velocity of inertial system @xmath302 relative to absolute system k ( here system @xmath302 is the center - of - inertia reference frame , i.e. @xmath303 ) . if the galileo s relativity principle is valid , equation ( 2.1 ) should be covariant relative to transformations ( 5.30 ) , i.e. in the system @xmath302 it should be @xmath304 , where @xmath305 = m_{0 } \mathbf{v ' } + m_{0 } \mathbf{v}_{\mathrm{k } ' } - \frac{\partial u}{\partial \mathbf{v ' } } + [ \mathbf{s } \times \mathbf{w ' } ] = \mathbf{p ' } + m_{0 } \mathbf{v}_{\mathrm{k } ' } \ ; , \eqno{(5.31)}\ ] ] @xmath306 = - \frac{\partial u}{\partial \mathbf{r ' } } + [ \mathbf{c } \times \mathbf{v ' } ] + [ \mathbf{c } \times \mathbf{v}_{\mathrm{k ' } } ] = \mathbf{f ' } + [ \mathbf{c } \times \mathbf{v}_{\mathrm{k } ' } ] \ ; , \eqno{(5.32)}\ ] ] whence it follows relation @xmath307 = \mathbf{0 } \;. \eqno{(5.33)}\ ] ] for free electron we have @xmath7 and it follows from eq.(5.33 ) and eq.(5.29 ) that @xmath308 . thus , finally @xmath309 writing down expressions for @xmath80 , corresponding to cases i.1- i.7 ( @xmath101 ) , and comparing them with eq.(5.34 ) , we find that the condition ( 4.6 ) , corresponding to @xmath117 and @xmath119 , should satisfied , whence it follows @xmath310 , and due to eqs.(3.15 ) , ( 4.1 ) , ( 4.5 ) , ( 4.8 ) , ( 4.12 ) , ( 4.19 ) , ( 4.28 ) and ( 4.33 ) pseudo - vector @xmath79 looks like @xmath311 i.e. two kinds of motion , corresponding to polarizations @xmath312 , are possible . combining eqs.(5.34 ) and ( 5.35 ) , we obtain relation @xmath313 the case ii ( @xmath173 , @xmath314 ) corresponds to @xmath315 . expression for @xmath80 , corresponding to the case iii ( @xmath197 , @xmath316 ) , which is similar to eq.(5.34 ) , gives @xmath317 . it may be assumed that the state with @xmath318 is a bound state of two particles with opposite polarizations , contribution of every of which in @xmath80 is @xmath319 . more strictly it can be confirmed after a detailed solution of the two - body problem for interacting mass points with internal degrees of freedom . trajectory of mass point in cases i and iii is a circle ( 4.3 ) of radius @xmath278 in the center - of - inertia reference frame . the negative value of mass @xmath293 means , that the force ( 5.28 ) is centripetal , rather than centrifugal one . the unique reason of such strange behavior is existence of internal rotational degrees of freedom , described by pseudo - vector @xmath8 . it is reasonably to express @xmath8 in terms of spin ( 5.24 ) , as follows @xmath320 where @xmath321 is some constant with dimension of velocity . we obtain from eqs.(5.24 ) , ( 5.35 ) and ( 5.37 ) @xmath322 equations of motion ( 3.9 ) give @xmath323 , implying conservation of spin direction when interaction is negligible . substituting ( 5.38 ) in ( 5.25 ) , we obtain for the total moment of momentum relative to the center of inertia @xmath324 introducing a denotation @xmath325 we have for spin @xmath326 . here a question remains open whether @xmath321 and @xmath327 be the velocity of light and planck constant , respectively . its solution will be determined by behavior of particles in external fields and their interaction with each other . for obtaining complete solution of a problem about a motion of the system in question it is necessary to add the equations for internal degrees of freedom . in the previous paragraph we have found out that the equation of the moments ( 5.1 ) is a consequence of the equation ( 2.1 ) , and it can not be considered as the additional equation . for free system we have the equation for spin ( 5.21 ) which means that there exists a preferred direction , namely , a direction of motion of the center of inertia , round which the spin is precessing with constant angular velocity @xmath270 . generally at every given instant spin is precessing round any instantaneous direction , simultaneously moving in space together with the center of mass . if @xmath328 is a vector pointing out in this direction the spin equation of motion can be written as @xmath329 + \mathbf{m}(t ) = \sigma_{\mathbf{n}}(t ) [ \mathbf{n } \times \mathbf{s } ] + \mathbf{m}(t ) \ ; , \eqno{(6.1)}\ ] ] where @xmath330 is some pseudo - vector having a sense of the moment of force or torque acting to extended point . the structure of @xmath330 , apparently , can be determined on specifying of an interaction of internal substance of the point with external fields . it follows from eq.(6.1 ) the constancy of absolute value of spin if @xmath331 . otherwise spin changes not only over the direction , but also over absolute value . assuming the interaction of internal substance with external fields to be much weaker than the interaction of the point as a whole object we will consider @xmath332 as first approximation . besides , taking into account that equation ( 6.1 ) should be reduced to eq.(5.21 ) , when interaction is negligible , it is necessary to take as a vector @xmath328 the vector @xmath333 + \ ] ] @xmath334 - [ \mathbf{r } \times \mathbf{c}^{ext } ] + ( [ \mathbf{r } \times \mathbf{v } ] \cdot \frac{\partial \mathbf{c}^{ext}}{\partial \mathbf{v } } ) \ ; , \eqno{(6.2)}\ ] ] which can be treated as a definition of the kinetic momentum related to a particle , specifying the motion of the center of inertia of the latter . in the presence of interaction the momentum ( 6.2 ) is not conserved , and spin is precessing round instantaneous direction of the momentum with instant angular velocity @xmath335 . thus , the equation of motion of the spin can be finally written down in the form @xmath336 + \mathbf{m}(t ) = \ ] ] @xmath337 - \frac{\sigma_{\mathbf{p}}(t)}{m } [ \frac{\partial u_0}{\partial \mathbf{v } } \times \mathbf{s } ] + \frac{\sigma_{\mathbf{p}}(t)}{m c^2 } [ \mathbf{s } \times [ \mathbf{s } \times ( \mathbf{w } - \omega^{2}_{0 } \mathbf{r } ) ] ] -\ ] ] @xmath338 + \frac{\sigma_{\mathbf{p}}(t)}{m } [ \mathbf{s } \times [ \mathbf{r } \times \mathbf{c}^{ext } ] ] - \frac{\sigma_{\mathbf{p}}(t)}{m } [ \mathbf{s } \times ( [ \mathbf{r } \times \mathbf{v } ] \cdot \frac{\partial \mathbf{c}^{ext}}{\partial \mathbf{v } } ) ] + \mathbf{m}(t ) \;. \eqno{(6.3)}\ ] ] all known equations of motion of spin in non - relativistic approximation have structure of eq.(6.3 ) with @xmath339 . system of equations ( 2.1 ) and ( 6.3 ) with momentum ( 6.2 ) and force ( 2.7 ) allow to solve a lot of problems about the motion of spinning particles in various fields and to compare these solutions with well - known results . obtained here non - relativistic equations of motion suppose the relativistic generalization @xcite , however there are subtleties which should be considered carefully . being founded on the stated above it is possible to assert that the trembling motion of objects with internal degrees of freedom has origin in classical mechanics . this circumstance does not contradict numerous modern researches in which the classical theory of spin is developing ( see , e.g. , @xcite ) . experimental observation of the motion of individual micro - objects with internal structure is for now impossible at present status of experimental technique . nevertheless , the quantum phenomena of zitterbewegung type , which though are not experimentally observable now , but they can be simulated , and the first results are presented in work @xcite . however , the results obtained say that this phenomenon should take place , in principle , for such classical objects as spinning top whose spin ( proper moment of momentum ) is not parallel to the velocity of translational motion of its center of inertia . other examples are the projectile ( bullet ) , which is shot through a rifle trunk , and a boomerang . it is possible to hope also that equations similar to the equations obtained in 4 will allow explaining such little - understandable phenomenon , as dzhanibekov s effect . for micro - objects zitterbewegung is manifested so as they have the wave nature , whereas this phenomenon is imperceptible for macro - objects because of both their large mass , and smallness of radius ( amplitude ) @xmath122 of trembling motion . thus , there appears a possibility of classical interpretation of quantum phenomena . moreover , two types of motion , corresponding to opposite polarizations , give rise to new look at the origin of electric charge . the solution of non - relativistic problem of the motion of the point with internal degrees of freedom , given in 4 - 5 , and its application to free electron leads to new interpretation of the charge of elementary particle which sign is determined by its helicity . right helicity @xmath340 , @xmath195 , @xmath196 ( figure 2a ) , corresponds to right polarization of spin for free antiparticles ( positrons ) , charged positively , whereas left helicity @xmath341 , @xmath193 , @xmath342 ( figure 2b ) , corresponds to left polarization of spin for free particles ( electrons ) , charged negatively . for interacting particles helicity can be distinct from @xmath343 , but its sign as before corresponds to the sign of charge . hence , _ the charge is the conventional concept characterizing type of the motion of spinning particle , corresponding to its helicity_. in this connection it would like to mention w. ritz s opinion , according to which `` these latter ( electric charges ) only playing , like the masses in mechanics , the role of coefficients , conveniently chosen and invariable for a given ion or electron . in a certain sense it is a mechanical theory of electricity '' ( @xcite , p. 149 ) . abraham m. prinzipien der dynamik des elektrons . d. physik , 1902 , * 10 ( 315 ) * , h. 1 , 105 - 179 . compton a.h . the size and shape of the electron . , 1918 , * 11 * , no . 4 , 330 . compton a.h . the size and shape of the electron . , 1919 , * 14 * , no . 1 , 20 - 43 . compton a.h . the size and shape of the electron . ii . the absorption of high frequency radiation . // , 1919 , * 14 * , no . 3 , 247 - 259 . compton a.h . the magnetic electron . // j. franklin inst . , 1921 , * 192 * , no . 2(1148 ) , 145 - 155 . parson a.l . magneton theory of the structure of the atom . // smithsonian miscellaneous collection , 1915 , * 65 * , no . 11 ( pub . no frenkel j. die elektrodynamik des rotierenden elektrons . // z. phys . , 1926 , * 37 * , h. 4 - 5 , 243 - 262 . schrdinger e. ber die krftefreie bewegung in die relativistischen quantenmechanik . wiss . , phys .- math . kl . , 1930 , 418 - 428 . huang k. on the zitterbewegung of the dirac electron . j. phys . , 1952 , * 20 * , no . 8 , 479 - 484 . mathisson m. neue mechanik materieller systeme . // acta phys . polon . , 1937 , * 6 * , no . 3 , 163 - 200 . mathisson m. das zitternde elektron und seine dynamik . // acta phys . polon . , 1937 , * 6 * , no . 3 , 218 - 227 . corben h.c . classical and quantum theories of spinning particles . san francisco : holden - day , inc . , 1968 . xvi+279 pp . rivas m. kinematical theory of spinning particles : classical and quantum mechanical formalism of elementary particles ( fundamental theories of physics , vol . new york - boston - dordrecht - london - moscow : kluwer academic publs , 2002 . 360 pp . pomeranskii a.a . , senkov r.a . , khriplovich i.b . spinning relativistic particles in external fields . // physics - uspekhi , 2000 , * 43 * , no . 10 , 1055 - 1066 . tarakanov a.n . on the dynamics of the mass point with internal degrees of freedom . // in : foundations & advances in nonlinear science . proc . of the 15th intern . conf .- school , 20 - 23 , 2010 , minsk , belarus . minsk : publishing center of bsu , 2010 . 123 - 132 . tarakanov a.n . generalized dynamics of the mass point with internal degrees of freedom . // tarakanov a.n . classical analog of zitterbewegung . // in : kovariantnye metody v teoreticheskoi fizike . fizika elementarnyh chastiz i teoriya otnositelnosti . minsk : institute of physics , 2011 . 156 - 164 ( in russian ) . weyssenhoff j. , raabe a. relativistic dynamics of spin fluids and spin particles . // acta phys . pol . , 1947 , * 9 * , no . 1 , 7 - 18 . dehmelt h.g . experiments with an isolated subatomic particle at rest . nobel lecture , 1989 . // physica scripta , 1991 , * t34 * , 47 - 51 . bergman d.l . , wesley j.p . spinning charged ring model of electron yielding anomalous magnetic moment . // galilean electrodynamics , 1990 , * 1 * , no . 5 , 63 - 67 . matora i.m . realnyi electron . dubna : jinr , 2006 ( in russian ) . griffiths d.j . , szeto e.w . dumbbell model for the classical radiation reaction . , 1978 , * 46 * , no . 3 , 244 - 248 . gerritsma r. , kirchmair g. , zhringer f. , solano e. , blatt r. , roos c.f . quantum simulation of the dirac equation . // nature , 2010 , * 463 * , no . 7277 , 68 - 71 . ritz w. recherches critiques sur lelectrodynamique gnrale . et de phys . , ser . 8 , 1908 , * 13 * , 145 - 275 ; in : ritz w. gesammelte werke . paris : gauthier - villars , 1911 . s. 317 - 426 .

we employ an ensemble of 24 hydrodynamic cluster simulations to create spatially and spectrally resolved images of quality comparable to _ chandra _ s expected performance . emission from simulation mass elements is represented using the xspec ` m`ekal program assuming 0.3 solar metallicity and the resulting spectra are fit with a single - temperature model . despite significant departures from isothermality in the cluster gas , single - temperature models produce acceptable fits to 20,000 source photon spectra . the spectral fit temperature is generally lower than the mass weighted average temperature due to the influence of soft line emission from cooler gas being accreted as part of the hierarchical clustering process . the nature of this deviation depends on the bandpass used for spectral fitting . in a _ chandra_-like bandpass of 0.5 to 9.5 kev we find a nearly uniform fractional bias of , although smaller clusters sometimes demonstrate much greater deviations . if the minimum energy threshold is raised to 2 kev , however , the effect of line emission on the spectrum is greatly decreased and becomes a nearly unbiased estimator of for smaller clusters . the fractional deviation in relative to is scale - dependent in this bandpass and follows the approximate relation . this results in an observed relationship for the simulations with slope of about 1.6 , intermediate between the virial relation and the observed relation . tracking each cluster in the ensemble at 16 epochs in its evolutionary history , we catalogue merger events with mass ratios exceeding 10% in order to investigate the relationship between spectral temperature and proximity to a major merger event . clusters that are very cool relative to the mean mass - temperature relationship lie preferentially close to a major merger , suggesting a viable observational method to cull a subset of dynamically young clusters from the general population .

astro-ph0004309

galaxy clusters are the youngest and largest organized structures in the universe , and as such provide us with a wealth of cosmological information . the most massive clusters draw their substance from cosmologically significant volumes of linear scale @xmath7 mpc . these scales are large enough that no known coherent process competes against gravity , so rich cluster contents are thought to comprise a fair sample of the universe s ingredients ( white 1993 ) . because clusters are rare nonlinear excursions of the cosmic density field , the statistical properties of their population are quite sensitive to both cosmological model and slope of the primordial fluctuation spectrum . unfortunately , their relative youth can also make interesting physical properties difficult to measure : about 50% of the local population bears evidence of ongoing mergers , and the canonical `` relaxed '' cluster is a relatively rare beast . nearly all interesting cosmological tests depend on accurate measurement of cluster virial masses . observations of the intracluster medium ( icm ) have shown promise in this regard : the icm s high x - ray luminosity and large spatial extent make it possible to probe the content and structure of clusters in great detail . bulk properties of the icm such as luminosity , temperature , mass , and gas density profile shape have been found to display highly significant correlations with each other ( edge & stewart 1991 ; david 1993 ; mohr & evrard 1997 ; mushotzky & scharf 1997 ; markevitch 1998 ; allen & fabian 1998 ; mohr , mathiesen & evrard 1999 ; arnaud & evrard 1999 ) . in contrast to the noisy correlations of early x - ray data ( sarazin 1986 and references therein ) , many correlations now display scatter at the @xmath8 level , indicating that a high degree of physical uniformity exists even in this structurally diverse population . hydrodynamic simulations of cluster evolution predict tight relationships between observable quantities and between those quantities and the cluster binding mass ( evrard 1990 ; kang 1994 ; navarro , frenk & white 1995 ; evrard 1996 ; bryan & norman 1998 ) , even when some members of the sample are far from dynamical equilibrium . the existence of both observed and theoretical correlations implies that the prevalence of cluster substructure is not a fundamental barrier to interpreting the properties of the population . however , moderate biases caused by the presence of substructure are likely to be present , and we explore the role of substructure in temperature measurements of the icm in this paper . a previous paper ( mathiesen , evrard & mohr 1999 ) demonstrated that a icm clumping leads to a modest ( @xmath9 ) overestimate of icm masses derived under the typical assumptions of spherical symmetry and isothermality . as the x ray data improve , the limits of simplifying assumptions such as these become clearer . high - resolution x ray images reveal secondary peaks and strong asphericities in many clusters and x ray spectra indicate the presence of multiple temperature components within the cores of many clusters ( fabian 1994 , holzapfel 1997 , allen 2000 ) . chandra _ and _ xmm _ satellite missions will provide the most detailed maps of the icm emission and temperature structure yet obtained and will allow more precise definition of the limitations of the current models . three - dimensional hydrodynamical simulations of cluster formation can help bridge the gap between the new generation of data and traditional methods and results . while simulated clusters often do not include many processes thought to be important to icm evolution ( e.g. radiative cooling and galactic winds ) , they excel at the creation of populations with realistic merger histories ( mohr 1995 ; tsai & buote 1996 ) . ensembles of simulated clusters can therefore be used to investigate the effects of accretion events , major mergers , clumping , and substructure on measurements of the icm . in this paper , we analyze the spectral properties of an ensemble of 24 simulated clusters using a realistic plasma emission model and assuming a uniform metallicity @xmath10 times the solar abundance . we find that even minor accretion events can significantly bias our measurements of the mean , mass - weighted cluster temperature , and that clusters undergoing a major merger can sometimes be identified as extreme examples of this bias . section 2 of this paper describes the simulations , the cluster ensemble , and the process of creating our spectral images . section 3 discusses the various measures of cluster temperature which have seen frequent use and explores the relationships between them . in particular we explore the relationship between spectrally determined temperatures and the mass - weighted mean temperature . the latter is found in simulations to follow most closely the virial relationship . section 4 then delves into cluster dynamics , investigating the effects of a major merger on the icm and looking for observable signatures of the merging process . finally , section 5 summarizes our conclusions . we use an ensemble of 24 hydrodynamical cluster simulations , divided evenly between two reasonable cold dark matter ( cdm ) cosmological models . these models are ( i ) ocdm ( @xmath11 , @xmath12 , @xmath13 , @xmath14 ) ; and ( ii ) @xmath15cdm ( @xmath16 , @xmath17 , @xmath18 , @xmath13 , @xmath19 ) . here the hubble constant is @xmath20 km s@xmath21 mpc@xmath21 , and @xmath22 is the linearly evolved , present power spectrum normalization on @xmath23 mpc scales . the initial conditions are gaussian random fields consistent with a cdm transfer function with the specified @xmath24 ( e.g. bond & efstathiou 1984 ) . the baryon density is set in each case to @xmath25 of the total mass density ( @xmath26 ) . the simulation scheme is p3msph ; first a p@xmath27 m ( dm only ) simulation is used to find cluster formation sites in a large volume , then a hydrodynamic simulation is performed on individual clusters to resolve their dm halo and icm structure in detail . the baryonic component is modeled with @xmath28 particles , providing a typical mass resolution of 0.01% within the virial radius . the resulting cluster sample covers a little more than a decade in total mass , ranging from about @xmath29 to @xmath30 . further details on these simulations can be found in mohr & evrard ( 1997 ) . the simulations model the dynamical and thermodynamical effects of gravitation , shock heating and adiabatic work on the icm . several potentially important pieces of physics are neglected . radiative cooling is one ; our clusters can not produce cooling flows in their cores . these are inferred to be quite common in the population ( edge & stewart 1991 , white 1997 ) , and many methods have been presented in the literature for removing the effects of cooling flows from measurements of bulk icm properties . we expect that the new satellites _ chandra _ and _ xmm _ will be able to directly observe the spatial extent of cooling flows and simply remove those pixels if desired . we thus analyze our simulations under the assumption that they are comparable to real icm observations correct for the presence of cooling flows through excision ( e.g. markevitch 1998 ) or explicit modeling ( e.g. allen 2000 ) of the extra emission component . we also lack any treatment of icm evolution through galaxy interactions such as supernova enrichment . instead , we assume that the abundance of heavy elements is constant throughout a cluster s history . the heating effects of galactic winds are difficult to parameterize accurately , and we postpone this investigation to a future set of numerical experiments including an approximate version of galactic feedback . it also has been proposed that the icm electron temperature lags behind the ion temperature to a significant degree , an effect which occurs due to the long equipartition timescale for plasmas with typical icm densities and temperatures . this view has been supported by observation of asca temperature profiles ( markevitch 1996 ) , analytical modeling of accretion shocks ( fox & loeb 1997 ) , and hydrodynamical simulation of the formation of a rich cluster ( chize , alimi & teyssier 1998 ; takizawa 1999 ) . the latter show that the effect is small ( @xmath31 ) except in the outer regions of the most massive clusters . our simulations , which treat the icm as a single thermodynamic fluid , do not take into account this effect . allowing cooler electron temperatures in the outskirts of clusters would increase the magnitude of the spectral effects calibrated below . from the simulation output , observables such as x ray spectra and surface brightness maps are obtained as follows . each simulation particle is given a model for its x ray emission based on its local ion density @xmath32 and thermodynamic temperature @xmath33 . previous works have traditionally used a simple bremsstrahlung - based model such as @xmath34 , but we employ the more sophisticated ` m`ekal model from the xspec utility ( mewe 1986 ; a url for the xspec user s manual is also in the bibliography ) . the ` m`ekal spectra are created using 0.3 solar abundance for elements heavier than hydrogen . each cluster is `` viewed '' along a single axis , and a subvolume of the simulation containing the cluster is divided into image pixels . each particle s emission is collected into the image and smoothed over neighboring pixels with a gaussian function of the same width as that particle s sph smoothing kernel . the resultant `` data cube '' contains a spectrum at each `` sky '' position . to our knowledge , this is the first such spectral visualization of a simulated intracluster medium . the particle data are output at twenty time intervals over the entire simulation , equally spaced in cosmic time . groups are identified at each epoch using a friends - of - friends algorithm with linking parameter 0.15 the mean interparticle separation . this procedure connects particles having separations corresponding to roughly @xmath35 times the background density ( lacey & cole 1994 ) . the position of each group is taken to be that of the dark matter particle with the minimum gravitational potential ; this point is always used as the image center . we have created a catalogue of `` major merger '' events for each cluster by identifying all instances of groups with two or more progenitors having a mass ratio @xmath36 . the entire sample contains 75 such events . images are created for the last 16 outputs along a single viewing direction , resulting in a total of @xmath37 observable physical states . in certain specified situations we use a subset of these outputs , focusing on the images corresponding to redshifts about equal to 0 , 0.5 , and 1 for each cluster . the full evolutionary histories of the clusters are used to enhance the variation of substructure in our sample and provide insight into the effects of mergers on the icm s observable characteristics . our primary database consists of two spectral cubes for each output differing only in radial scale . we choose half - widths of @xmath38 and @xmath39 in the plane of the sky , and line - of - sight depth of @xmath40 . the scale @xmath41 is the radius at which the mean interior density contrast falls to @xmath42 times the critical density @xmath43 at the viewing epoch . the smaller radius @xmath38 corresponds roughly to the readily visible portion of current observations , while @xmath39 is the theoretical preference for the virial radius of clusters in an @xmath44 cosmology . we used scaled radii rather than a fixed metric radius in order to probe regions with similar dynamical character . each image uses the ` m`ekal spectral model for the icm x ray emission , and contains a 256-pixel spatial image of the cluster for each energy bin . the model spectra have a bin size of 100 ev and bandpass of [ 0.1,20 ] kev , allowing a great deal of flexibility in producing spectra comparable to observations . the spectral cubes created for this work have a bin size of 150 ev and a bandpass of [ 0.5,9.5 ] kev in order to simulate the quality of data expected from _ chandra s _ advanced ccd imaging spectrometer ( acis ) . a color figure ( table 2 ) displaying x - ray surface brightness and spectral temperature images in two different bandpasses is presented and discussed in section 4.1 , but the reader may wish to glance at it now to get a visual impression of the data used in this paper . as we will see , our use of a spectral model for icm emission has important consequences . previous x ray images derived from simulations use a simple bremsstrahlung model , but line emission can make up as much as 20% of the cluster luminosity for gas at relatively low ( @xmath45 kev ) temperatures . while continuum emission in the low energy rosat passband is a weak function of temperature above this threshold ( mohr 1999 ) , the introduction of line emission into the model brings a strong temperature dependence back into the soft x rays . this effect can be seen in figure [ spectra ] , which plots isothermal ` m`ekal spectra with 0.3 solar abundance at temperatures typical of the icm . since soft ( @xmath46 kev ) line emission is progressively more pronounced at cooler temperatures , even a small amount of cold gas in the observation window can make a spectrum appear significantly cooler . this effect is central to the results presented in this paper . 6.0 in 6.0 in cluster atmospheres are only approximately , never identically , isothermal . description of the icm by a single temperature @xmath47 is not only incomplete but also dependent on the measure employed to derive the value of @xmath47 . common definitions employed in the literature use the mass - weighting and an approximate emission - weighting methods described below . with lagrangian simulations , volume integrals tranform to sums over particles within the volume @xmath48 where @xmath49 is the mass and @xmath50 and @xmath33 the density and temperature of the @xmath51 particle . our simulations employ equal mass gas particles , so the mass - weighted temperature within @xmath41 is a simple average over @xmath52 sph particles @xmath53 while the approximate emission - weighted temperature employs a @xmath54 volume weighting that transforms to @xmath55 in a lagrangian implementation . somewhat more sophisticated implementations of emission - weighting have been employed ; for example , evrard , metzler & navarro ( 1996 ) use a rosat emission weighting measure that incorporates the temperature dependence of bremsstrahlung emission within an @xmath56 kev energy passband . within the simulations , we determine both @xmath1 and @xmath57 within spherical boundaries of radii @xmath38 and @xmath39 . if cluster atmospheres within @xmath41 are hydrostatic and in virial equilibrium , then the thermal energy of the icm will reflect the net gravitational potential energy of the cluster . this leads to the oft - cited virial relation between mass - weighted temperature and binding mass @xmath58 where @xmath59 is the total mass within a sphere of radius @xmath41 and @xmath60 is a dimensionless form factor . assuming that @xmath60 is a constant independent of mass and redshift ( strict self similarity ) and recalling that @xmath61 with @xmath62 , we arrive at a compact form for the virial mass - temperature relation @xmath63 with @xmath64 . the factor @xmath65 takes into account the cosmological scaling of this relationship , and can be derived from the friedmann equation . for the flat universe modeled in our simulations , @xmath66 $ ] , where @xmath67 . this factor is equivalent to the evolution factor @xmath68 used by bryan & norman ( 1998 ) . this relationship determined at @xmath69 is presented in figure [ virmasstemp ] using three simulation outputs for each cluster at the redshifts @xmath70 , @xmath71 and 0 . the data agree well with the virial expection ; the best - fit power law @xmath72\ ] ] exhibits only 14% scatter in mass at fixed @xmath1 . the data within @xmath73 have a similar slope and slightly larger scatter . 6.0 in 6.0 in results such as these have led many cosmologists to use the virial relation , calibrated by simulations or a well - studied nearby cluster , to estimate binding masses from cluster temperatures . even if current simulations are complete and exact in their physical description of the icm , this procedure need not be correct , since the mass - weighted temperature is not a directly obervable quantity . computational cosmologists have often employed the simplified emission - weighted temperature @xmath57 in an effort to move closer to observations . what is actually measured , however , is a spectral temperature derived from fitting a single raymond - smith or ` m`ekal model to the spectral data of a cluster . papers analyzing a single cluster or a small sample sometimes employ two temperature components and/or a model cooling flow spectrum in order to improve the fit . these variations are physically well - motivated but often unnecessary in the sense that a single temperature component usually provides an acceptable fit to even high quality spectral data ( holzapfel 1997 ) . from the simulations , we produce spectral temperatures by fitting isothermal models to sample photon spectra . the latter are produced by summing emission from gas within a cylinder of appropriate radius and depth @xmath40 centered on the cluster . this collection volume includes most of the line - of - sight gas in the simulation , but projection effects from large - scale structure ( represented in the high - resolution simulation by massive dark matter particles with no associated baryons ) are not taken into account . the combined spectrum is rebinned into channels of width 150 ev ( comparable to the _ chandra _ acis energy resolution ) , convolved with the _ chandra _ effective area function , and assigned poisson error bars appropriate to the number of photons in each bin , assuming a total photon count of 20,000 . bins with a low count rate are grouped together until they contain at least twenty photons . the result is then fit to a single - temperature ` m`ekal spectrum with 0.3 solar metallicity . no poisson _ noise _ is added to the photon counts , as we are trying to isolate the systematic effects of substructure , but the error bars are used in fitting to assign appropriate relative weights to the data . the resulting spectral temperature @xmath0 which minimizes the chi - squared ( the normalization of the spectrum is also a free parameter , of course ) is a direct analog of observed temperatures obtained by the _ chandra _ observatory . other satellites can readily be simulated by changing the bin width , bandpass , and effective area function . a sample of this process is shown in figure [ specfit ] , which displays the combined spectrum , best - fit spectrum , and residuals for a typical cluster of the ensemble . although this spectrum is formally well fit by a single - temperature model , there are clear trends in the residuals : the strength of the iron line is somewhat too low for a continuum with this slope , the width of the iron line is increased somewhat from the lower - energy transitions in this complex , the 1.0 kev bin is much too high , and the overall shape of the spectrum is slightly different . with the exception of the iron line and the 1.0 kev bin these excursions are very small , and we would not expect them to be visible in a real spectrum with this photon count and typical noise levels . deep exposures from _ chandra _ might be able to pick them out , however . increasing the photon count to 100,000 for this cluster brings the reduced chi - squared up to 1.64 for 54 degrees of freedom , enough to detect the difference in spectral shape even with typical levels of random error . the unusual coincidence that a realistic combined spectrum has nearly the same shape as an isothermal spectrum may be explained by our mediocre energy resolution , which keeps us from constraining the structure of the emission lines to a useful degree . 6.0 in 6.0 in we have cut off the spectrum at 9.5 kev due to a lack of photons above this range . the spectral temperatures are significantly lower than both the emission- and mass - weighted temperatures for this cluster . the reasons for these differences are related to this particular cluster s dynamical state . in figure [ ob1phase ] , we show a phase diagram of the icm gas within @xmath39 the cluster center . the tongue of low temperature material is a small merging subclump that is barely visible in the x ray surface brightness . we have drawn a horizontal line at 2.2 kev ( log @xmath47 [ k ] = 7.4 ) to visually separate the core of this subclump . particles below this threshold ( and especially those in the clump , which is nearly as dense as the cluster core ) produce significant line emission in the [ 0.5,2.0 ] kev band and enhance the spectrum in this region . only 4% of the particles in the cluster lie below this line , but it turns out that this is more than enough to significantly alter the shape of the spectrum . the dashed lines in figure [ ob1phase ] represent the minimum entropy thresholds inferred by lloyd - davies , ponman , & cannon ( 2000 ) from x ray observations of galaxy groups , and will be discussed further at the end of section 3.2 . this is an important point : minor mergers such as this are just as likely as major events to cause deviations between the spectral and mass - weighted temperatures . not only are they much more common , their gas is colder and therefore more luminous in the soft x ray band . 6.0 in 6.0 in we find that this bias is nearly universal in our sample , and are motivated to construct the mutual correlations of the three temperatures . these relationships are displayed in figure [ tvst ] . the spectral temperatures occur at discrete intervals because we tabulated ` m`ekal outputs at temperatures with spacing @xmath74 in all our analysis ( at considerable savings in cpu time ) . this interval is much smaller than the uncertainties associated with our spectral temperatures . the constants @xmath75 and @xmath76 displayed in the plots result from fits @xmath77 obtained by least - squares linear regression using logarithmic variables . 6.0 in 6.0 in we can see from these plots that the mass- or emission - weighted temperatures are almost universally higher than the spectral temperatures . why is this the case ? the mass - weighted temperatures , as we have seen , follow the virial expectation quite closely and can be used as a benchmark for the cluster s evolution . the emission - weighted temperatures strongly favor the core region , so also tend to mark the potential well depth rather well . the spectral temperature , on the other hand , is weighted by the fitting process towards whichever regions emit the most photons . this favors both high - density regions and cool regions , and the cores of accreting subclumps satisfy both criteria . accreting clumps of cool gas can therefore create an excess of soft photons for some time during and after a merger event . this process will be discussed in more detail in section 4 , but for now it suffices to realize that accretion and substructure can have a rather large effect on spectral temperatures . @xmath0 does nt change a great deal when we go from @xmath39 to @xmath38 , but the mass - weighted temperature increases ; this is why we see a larger bias within @xmath38 despite the fact that the former window includes a larger volume of cool gas . is there evidence of this phenomenon in the literature ? accurate temperature determinations for clusters are available from the satellites _ einstein _ , _ exosat _ , _ ginga _ , and _ asca _ , each of which uses a different bandpass in fitting their x ray spectra . we can test for the presence of the effect described here by looking for a systematic difference between spectral temperatures in different energy bands . all of these satellites have effective upper limits comparable to the spectrum shown in figure [ specfit ] thanks to the relative scarcity of icm photons above 10 kev . the _ einstein _ mpc temperatures of david ( 1993 ) are available for many clusters and use a consistent lower limit of 2 kev , so we use this data set as our basis . each of the other three satellites makes some use of photons below this threshold , and should therefore return cooler temperatures than the mpc . the _ exosat _ temperatures ( edge & stewart 1991 ) used here are derived from raymond - smith spectral fits in the range 1.5 - 9.0 kev , with an additional constraint coming from unresolved soft photon counts in the approximate range 0.6 - 1.0 kev . _ ginga _ spectra ( hayakude 1989 ) have an energy range of 1 - 20 kev for cluster observations ( arnaud 1991 ) and typically obtain a single significant bin below 2 kev . some comparisons between the results of these satellites have already been made in the literature . david ( 1993 ) found reasonable agreement between mpc , _ exosat _ , and _ ginga _ temperatures , and note that _ exosat _ temperatures are typically 0.7 kev less than their mpc temperatures for clusters between 5 and 8 kev . they also find reasonable agreement between _ ginga _ and mpc temperatures , but fail to note that out of 8 clusters only one of the _ ginga _ temperatures is higher than the mpc determination . more telling evidence comes from an analysis of _ asca _ data . markevitch ( 1998 ) state that _ asca _ temperatures determined in the [ 0.5,11 ] kev bandpass were typically lower by about 0.5 kev than temperatures determined in other ranges . markevitch use the [ 1.5,2]+[2.5,11 ] bandpass for their published temperatures in an effort to avoid this systematic ( the hole in the spectrum avoids a region with a difficult point spread function ) , and found good agreement between their temperatures and those of other satellites . if an excess of soft photons is really making the icm appear cooler than it is , we should expect that the mpc has the hottest spectra because it has the highest minimum energy . we thus collect temperatures from the literature ( _ exosat _ : edge & stewart 1991 ; _ ginga _ : hatsukade 1989 via david 1993 ; mpc : david 1993 ; _ asca _ : markevitch 1998 ) and make our own comparison to see if this is true . there is good overall agreement between _ asca _ and _ mpc _ temperatures , but the gap in the _ asca _ spectrum makes comparison of the techniques difficult ; we therefore do not include these data in our analysis . the _ exosat _ and _ ginga _ temperatures used here , however , each make use of the full [ 2,10 ] kev band as well as including constraints from photons below 2 kev . these temperatures are plotted in figure [ sattemps ] , and are for the most part _ lower _ than the mpc measurements . while in most cases the error bars intersect the line of equality , the probability that so many of these observations scatter below the line is quite small and argues for some systematic error . assuming that the clusters are drawn from an unbiased distribution in which each object has equal probability to fall above or below the line of equality , this uneven arrangement of clusters ( 4 above , 12 below ) has only a 4% chance of occuring at random . it is also interesting to note that all the clusters whose vertical error bars fall short of the line are in the `` cool '' direction predicted by these simulations . 6.0 in 6.0 in the observational evidence , though far from conclusive , suggests that the influence of line emission in the [ 0.5,2 ] kev range is not negligible . if true , a conservative response would be to adopt the [ 2 - 10 ] kev band as a standard for spectral temperature determinations . judging from figures [ spectra ] and [ specfit ] , this would remove most of the excess soft photons from the spectrum and reduce the bias due to cool gas . we test the effectiveness of this technique by fitting our cluster spectra in the [ 2.0,9.5 ] bandpass while maintaining a 20,000 photon normalization over the entire energy range . the results of this analysis are shown in figure [ tvst_g2kev ] ; throwing out all the photons below 2 kev results in temperatures driven primarily by the bremsstrahlung component , and the spectral temperatures are now much closer to the mass - weighted measure . 6.0 in 6.0 in another view of this analysis is given in figure [ dtvsts ] , which shows the fractional differenct @xmath78 plotted against the spectral temperature for four methods . this plot makes it easier to estimate the size of the errors and their scale - dependence . we see that the [ 0.5,9.5 ] kev spectral temperatures almost universally underestimate the mass - weighted temperatures , typically by 10 - 20% but sometimes by as much as a factor of two . we see similar results within @xmath38 and @xmath39 for these spectra . the picture looks brighter in the [ 2.0,9.5 ] kev bandpass , with almost all the errors confined within @xmath79 , but the scale - dependence of the effect is stronger . the panel which corresponds most closely to real measurements is the [ 2.0,9.5 ] bandpass temperature for emission within @xmath38 , in the lower right of the figure . the best - fit line drawn through these points follows the equation @xmath80-(0.016\pm 0.010).\ ] ] the uncertainties are at the one - sigma confidence level . we thus find that even in this bandpass the temperatures of hotter clusters are underestimated by 10 - 20% . although gross deviations are rarer , there are still a few cases where the mass - weighted temperature is 40% higher than the spectral temperature . from here on , when we use the term `` spectral temperature '' we refer to the temperature measured within @xmath38 in the [ 2.0,9.5 ] kev bandpass , unless otherwise noted . 6.0 in 6.0 in before going on , we point out a few effects which are potentially important but beyond the scope of this work . first , it seems a likely possibility that electron temperatures are much lower than ion temperatures in the cluster outskirts . chize ( 1998 ) , who performed an eulerian simulation of the icm for a coma - like cluster , found little difference in the two temperatures within @xmath38 but about a 20% difference between the two at @xmath39 . takizawa ( 1999 ) performed head - on collisions of merging clusters designed to produce a final state of 8 kev and found similar results . because the coulomb relaxation time scales as @xmath81 , any nonequlibrium thermodynamic signal would affect the scale - dependence of the temperature errors described in figure [ dtvsts ] . however , the magnitude of this effect is likely to be small , especially within the higher density @xmath38 window . a second effect which may be important in some clusters is the absorption of soft x rays by galactic hydrogen ; if this absorption factor is strong enough the influence of bins below 2 kev on the fit can be greatly reduced . applying a photon absorption model appropriate for a column density of @xmath82 to our spectra and rederiving figures 5 , 7 , and 8 , we find no significant change in the best - fit parameters for these correlations . however , a column density of @xmath83 is sufficient to change the behavior of the spectral temperatures . the best - fit parameters for figure [ tvst ] under this level of absorption become @xmath84 and @xmath85 within @xmath39 ; and @xmath86 and @xmath87 within @xmath38 . the results for the [ 2.0,9.5 ] kev band spectral temperatures are unchanged under both absorption models . we close this section with a brief discussion of preheating . the slope of the @xmath88@xmath47 relation in cooling - flow corrected clusters is well - constrained observationally to be 2.64 + /- 0.27 ( markevitch 1998 ) , a value significantly different from the slope of 2 predicted by models of self - similar gravitational collapse . simulations which preheat the baryons at an early epoch impose a minimum entropy on the icm during cluster formation and have been able to better reproduce the observed slope ( cavaliere 1998 ; bialek , evrard & sulkanen in preparation ) . if preheating proves to be important , the density ( and therefore emissivity ) of very cool infalling gas will be limited and its influence on the spectrum reduced . two such thresholds are plotted in figure [ ob1phase ] to illustrate this effect . these lines represent minimum entropies ( defined as @xmath89 ) of 60 and 120 kev @xmath90 , reported in a paper by lloyd - davies , ponman , & cannon ( 2000 ) which directly measures the entropy in 20 cluster cores , including 4 groups with spectral temperatures below 1 kev . because the baryon fraction in these simulations is about a factor of two higher than observed icm mass fractions , we rescale these entropy thresholds by a factor of @xmath91 to 40 and 80 kev @xmath90 . the effect of preheating on our results is difficult to assess without recourse to another simulation ensemble , but we can estimate the magnitude of this effect by assigning each particle violating the upper threshold the maximum allowed density at its temperature . applying this technique to the cluster displayed in figure [ ob1phase ] results in a spectrum with best - fit temperatures @xmath92 ( an increase of 15% ) in the [ 0.5,9.5 ] kev band and @xmath93 ( an increase of 4% ) in the [ 2.0,9.5 ] kev band . we can conclude that a large degree of preheating has the potential to reduce the deviation between spectral and mass - weighted temperatures , especially in the [ 0.5,9.5 ] kev band . we defer further analysis of this question in anticipation of a preheated simulation ensemble calibrated to reproduce the observed scaling relations ( evrard , mohr , in preparation ) . the results of the previous section make clear two important points . first , spectral temperature determinations tend to be lower than the underlying mass - weighted temperature . because of this bias , an analysis of the population which makes use of spectral temperatures and the virial relationship will underestimate the binding masses of rich clusters by 15 - 30% . second , the exact nature of the systematic is dependent on the bandpass used in spectral fitting . a simple test of this dependence can be easily made with chandra data by comparing [ 0.5,9.5 ] kev band temperatures to [ 2.0,9.5 ] kev band temperatures . predictions for this effect from the simulation ensemble are shown in figure [ ts_tsg2 ] . the solid line in this figure is the linear least - squares best fit to the points , and follows the relation @xmath94 } } = ( 0.81 \pm 0.01)\ kt_{s , { \rm [ 2.0 - 9.5 kev ] } } ^{1.09\pm0.01 } .\ ] ] 6.0 in 6.0 in an interesting side effect of this spectral modeling becomes apparent when we create azimuthally averaged surface brightness profiles for our simulations in the [ 0.5,2.0 ] kev bandpass . analysis of _ rosat _ position sensitive proportional counter ( pspc ) images has shown that cluster emission profiles are generally well fit by a beta - model with an average slope parameter @xmath95 of 0.64 ( mohr 1999 , this paper is referred to as mme in this section ) ; this value is consistent with previous analyses and the canonical value @xmath96 . the value of this parameter has been a challenge for simulations , which have traditionally produced much steeper emission profiles . an earlier generation of surface brightness images created from this ensemble , for example , produced profiles which were well fit by a beta - model but with @xmath97 ; such results are typical in the field ( e.g. eke 1998 ) , but not often discussed . simulated clusters have moderate temperature profiles ; the ( mass - weighted ) icm temperature at @xmath38 is typically between 80 and 90% of the mean temperature within this radius . the gas in the outer regions also tends to exhibit wider variations in density and temperature ( mathiesen 1999 ) , and pockets of gas much cooler than this can be found in most clusters . the luminosity of simulated clusters is generally modeled with a bremsstrahlung spectrum , which is rather insensitive to temperature in a soft x ray band such as the [ 0.5,2.0 ] kev used to analyze pspc images in mohr when a spectral model including line emission is used instead , we find that cool gas in the outskirts of the cluster ( including the very cool gas near the virial radius which lies along our line - of - sight ) significantly enhances the luminosity in this band . the mean @xmath95-parameter for our ensemble s spectral profiles is 0.59 , and corresponds much more closely ( in both value and process ) to observed clusters . this sudden reversal of a long - standing discrepancy between simulations and reality emphasizes the importance of interpreting simulated data with a realistic `` observational '' framework . the scale dependence of deviations in the spectral temperature will tilt the virial mass - temperature relationship . the relationship between spectral temperature and total cluster mass derived from three output epochs of the simulations is displayed in figure [ specmasstemp ] . this correlation has a significantly higher slope and somewhat larger scatter than that displayed in figure [ virmasstemp ] . the best - fit @xmath98@xmath47 relation for these data has a slope of @xmath99 ( 90% confidence level ) . the baryon fractions of our simulated clusters within @xmath38 fall in the range @xmath100 ( the global fraction is 0.2 ) and are independent of mass , so for us @xmath98 and @xmath101 are essentially equivalent . this result can therefore be compared directly to the recent result of mme , which measured the slope and scatter of the intracluster medium @xmath102@xmath47 relationship in nearby clusters and found a slope of @xmath103 . the degree of scatter in the two relations is similar ; mme found a scatter of 17% , and the simulations display a scatter of 20% . the result displayed in figure [ specmasstemp ] implies that some of the observed deviation from the virial relationship can be explained by invoking a normal incidence of both obvious and hidden substructure , including both major mergers and the minor accretion events which are typical of any epoch . since both figure [ specmasstemp ] and the result of mme depend on the spectral temperature , their slopes can be compared directly . combining these two results , we can infer a mild temperature dependence in the icm mass fraction of the form @xmath104 . this is wholly consistent with the directly observed variation @xmath105 reported by mme . to put it another way , the variation in @xmath106 with temperature observed by mme is not strong enough to account for the observed difference between the @xmath101@xmath47 relation and the virial relation . it is , however , strong enough to match the @xmath98@xmath0 relation found in this paper . 6.0 in 6.0 in we report the best - fit @xmath98@xmath47 relationships from these simulations for all our definitions of temperature in table [ mttable ] . .mass - temperature relationships , defined as @xmath107}$ ] , for all definitions of temperature used in this paper . the parameters @xmath108 ( mass in units of @xmath109 ) and @xmath75 are calcuated from the linear least - squares equation based on the scatter @xmath110 in @xmath111 around the relation . the uncertainties represent a one - sigma confidence interval . [ cols="<,^,^,^,^,^,^ " , ] [ mttable ] while the mass - weighted and [ 0.5,9.5 ] kev band spectral temperatures are consistent with the virial relation , the [ 2.0,9.5 ] kev band and emission - weighted temperatures are not . it is important to emphasize that the emission - weighted temperature used in this paper is not an observable quantity , and there is no evidence to suggest that it is comparable to the observer s definition . further work with these simulations needs to be done to reproduce an observable version , for example the flux - weighted average of spectral temperatures observed in each image pixel . this analysis seems to imply that the best way to measure temperatures with _ chandra _ is to limit spectral fits to photons of 2 kev or greater energy ; the temperature that results will not only be closer to the mass - weighted temperature , but will also be comparable to previous determinations . there is a price to pay , however , for rejecting the abundant low - energy photons . as we will see in the next section , the simulations lead us to expect the presence of very cool gas not just in cooling flows but also in the outskirts of the cluster . it is well known that clusters are continually accreting smaller concentrations of matter . since they are less massive , the gas in these subsystems is naturally much cooler than that in the cluster . these clumps of remain visible in our simulated spectral temperature maps for up to two billion years after the virial radii of the two systems intersect . including the soft photons decreases the accuracy of spectral temperatures , but also increases the contrast between these hot and cold regions . thus , spectral temperatures in the full _ chandra _ bandpass will be more useful in determining a cluster s dynamic state . it is logical to ask several questions at this point . first , there are several instances where the spectral temperature underestimates the mass - weighted temperature by a considerable degree . if these excursions are the result of cluster dynamics , we should be able to see evidence of unusual structure in the cluster at these points . second , we have claimed that the reason spectral temperatures typically underestimate the mass - weighted temperature is that most clusters contain significant amounts of cold gas in their outskirts . if accretion events and mergers are driving the difference between the two temperatures , we should be able to discover a time dependence in the effect , and perhaps even relate the scale of the errors to typical accretion rates . both of these questions are really asking the question named in the title of this paper : how are temperature measurements affected by a cluster s dynamical state ? fortunately , the history of the simulations gives us a means to answer this question . as stated in 2 , we have created spectral temperature maps for each cluster at 16 points during its evolutionary history . we have also kept track of the trajectories of dark matter groups during this evolution and used this information to determine the times at which major merger events occur . using this information , we can create objective definitions of the cluster s dynamic age and attempt to correlate these quantities with the biases described in the previous section . the mergers themselves are easy to find . at each timestep we identify dark matter groups with a mass of at least 100 particles ( the real mass corresponding to this threshold varies but is always a fixed fraction of the total mass in the simulation volume ) . by keeping track of the particle indices associated with each group , we can identify at later times which groups were absorbed into a given cluster . we identify merging groups with a mass ratio of at least 10% as `` major mergers '' and build up a catalogue of these events . this results in a catalogue of 75 merger events for the 24 clusters . the limiting mass ratio was chosen to strike a balance between frequency and significance ; a merging subclump of 10% the primary mass can be easily seen in our temperature maps ( when the viewing angle is right ) as a region of cool emission behind a bow shock , but is not so common that simultaneous events are typical . a fine example , which also happens to be one of the two exceptions to this rule , is shown in table [ 3waytmap ] , which shows the spectral temperature map of a three - way merger in the two bandpasses discussed earlier . although the merging mass ratio is small , it is clear in this illustration that soft photons from the merging subclumps dominate the spectrum over a fairly large volume . a minor shock front is visible between the two accreting systems , but these few pixels have very little effect on the overall temperature of the cluster . cc + & + + & + 0.25 kev & 4.0 kev + + the time at which a merger occurs is a more nebulous concept than the merger s existence . the individual outputs are separated by about 0.6 gy , so merely identifying this time with the first output in which two groups are merged introduces a large random error . we can , however , calculate the center of mass of each dark matter group in previous outputs and create an interpolating function for the groups separation . this function can be used to create more accurate and objective definitions of the merger time . our intuitive expectation for the effect of a merger event on the icm temperature is as follows : as the two groups approach one another , their icms begin to interact at a fairly large separation and we will see a shock begin to form . the gas in a typical shock is much hotter than the rest of the icm , and should bias measurements towards higher temperatures . as the event proceeds , the main body of the merging system penetrates the virial radius of the cluster and introduces a large mass of cold gas into the image . the subclump will be more massive and more luminous than the gas in the shock , so its weight will be greater under any reasonable definition of the temperature . we thus expect to see a rapid cooling in the combined spectrum shortly after the shock forms , even though the shock may still be clearly visible in temperature maps . after some time the icm will equilibrate and approach the virial temperature again . most merger events involve very small mass fractions , so this process should not greatly affect mass - weighted temperatures . we can hope to observe the process of heating and cooling through a merger much more clearly in the spectral and emission - weighted measures . practically , we interpret this expectation by identifying a merger time @xmath112 for each event , defined as the time at which the virial radii ( @xmath39 ) of the two groups intersect for the first time . identifying the merger time in this manner corresponds to the requirement that the mergers have a uniform free - fall timescale , coming together in a constant fraction of the hubble time @xmath113 . further analysis of the dark matter halo dynamics reveals that the centers of the two groups typically coincide about @xmath114 after @xmath112 . another relevant time scale in this problem is the time required for the icm to approach hydrostatic equilibrium after a merger . the equilibration time is often approximated by the time required for a sound wave to traverse the cluster , @xmath115 where @xmath116 is measured in kev and d is the diameter of the cluster ( sarazin 1986 ) . the clusters in our ensemble have sound crossing times between 1 and 4 gy , with a mean of 2.3 gy . the mean time between merger events , on the other hand , is about 2.7 gy . it would seem from this that our clusters are out of hydrostatic equilibrium for most of their total lifespan . armed with a list of 75 merger times @xmath112 , the hubble time @xmath117 at each output frame , and the sound crossing time @xmath118 in each output frame , we can proceed to look for correlations of temperature with the cluster dynamical states . we start by defining a parameter which measures the cluster s proximity to a major merger : @xmath119 where for each output frame @xmath112 is taken to be the time of the _ closest _ merger event . @xmath120 can thus be either positive or negative , a freedom which reflects our uncertainly as to when the merger process will begin to influence the spectrum . the relationship between spectral temperature errors and @xmath120 is striking , and is displayed in figure [ dtvst10 ] . this figure is constructed for clusters observed in the @xmath38 window ; the corresponding figure for @xmath39 looks similar . the four different point sizes correspond to categories in the merger mass ratio : from smallest to largest they represent the ranges @xmath121 , and @xmath122 . a glance at this plot demonstrates that the mass ratio of the merger does nt have much impact on the spectral temperature measurements . this can be understood as evidence for a balance in the roles of the subclump s temperature and luminosity in causing the bias ; in a major merger event the subclump has more influence on the spectrum but its temperature is also closer to the parent system s . the fact that all of the large deviations are clustered near the @xmath123 axis indicates that these excursions are indeed related to the merger process , and suggests a method for culling dynamically young clusters from our sample which we will return to later . the horizontal line is a visual aid to the scale of error which may be isolated from the sample , drawn so that all the points above it are isolated within a small range in @xmath120 . there is a single cluster in our sample which goes through one and only one major merger in its lifetime ; its points are connected with a solid line in figure [ dtvst10 ] to serve as a benchmark for our physical intuition . unfortunately , this cluster is not entirely typical ; it is rather small and somewhat cool for its mass ( @xmath124 , @xmath125 4 kev , @xmath126 = 2.8 kev ) . still , the process of apparent cooling as it goes through a merger with mass ratio @xmath127 is rather obvious . 6.0 in 6.0 in we can reiterate this point by defining a second dynamical time @xmath128 with respect to the sound crossing time of the cluster : @xmath129 although this equation looks very similar to our definition of @xmath120 , there is an important difference : we do not allow this timescale to take on negative values . thus the merger time used is now _ always _ the last merger to occur before the output time @xmath130 . for outputs where no merger has yet occurred , we take the numerator to be the age of the cluster . we define the time @xmath128 in this manner because we are interested in time required for our clusters to come to equilibrium after a merger . we plot the bias in spectral temperature against @xmath128 in figure [ dtvsteq ] . the result is similar in that the large errors mostly occur close to a merger event and typically less than one sound crossing time distant . this suggests that the measure given in equation [ tcross ] is not just a sound dimensional estimate but actually a very good measure of the equilibration timescale . again , the results for other bandpasses and density contrasts look quite similar . 6.0 in 6.0 in it is interesting to note from figures [ dtvsts ] and [ dtvst10 ] that the clusters which have large temperature differences are different for the two bandpasses . in the [ 0.5,9.5 ] kev range , the spectral temperatures are more likely to carry large errors in the low - temperature clusters . in the [ 2.0,9.5 ] bandpass , we see the largest errors associated with the hottest clusters . despite this difference , however , we see in figure [ dtvst10 ] and its equivalents that the largest errors in both bandpasses are associated with merging clusters , and moreover are concentrated in an interval about @xmath131 wide after @xmath132 . this tells us that merging clusters , and only merging clusters , are likely to carry large temperature errors , but the different bandpasses select clusters in different mass ranges . in the [ 2.0,9.5 ] band the fit is driven almost entirely by the effective slope of the bremsstrahlung component . this slope varies rapidly at low temperatures , and the spectrum of a blob of colder gas will be swamped by that of the parent . at higher temperatures ( for example , a 9 kev parent and a 5 kev satellite ) the mixing of exponential tails is more likely to produce a significantly cooler spectrum . in the [ 0.5,9.5 ] kev band , however , there is more balance between the soft and hard photons . between 0.5 and 2.0 kev the spectrum will be driven by the merging subclump , while in the brehmsstrahlung tail the spectrum will be driven by the parent cluster . the tug - of - war between the two regimes is decided by the number of bins in the spectrum , so in this bandpass the hottest clusters are better able to resist the influence of a cool component . we have seen in the previous sections that a simulated cluster ensemble , with realistic incidence of substructure , displays spectral temperatures with a significant and scale - dependent systematic bias . the largest deviations between spectral and mass - weighted temperatures come from clusters which are dynamically young : they are either about to undergo or have just undergone a major merger and the presence of cool gas in the observation window changes the shape of the combined spectrum . some of these mergers will be easily identifiable with _ chandra _ through the temperature structure of the gas , displaying images similar to those displayed in table [ 3waytmap ] . others will be invisible , however , hidden behind the main body of the cluster but still significantly altering the spectra . we would like to develop a method for identifying some or all of these , so that mass - weighted temperatures and cluster binding masses can be determined through x ray observations with a minimum of error . our expectation is that a large sample of cluster observations , culled of all of obvious mergers and some inobvious mergers , will reflect the virial relations better and allow us to more accurately measure variations in baryon fraction , deviations from self - similarity , and constraints on the power spectrum . the results of the previous section tell us that the clusters with spectral temperatures much colder than the mass - weighted temperature always lie close to a merger , although the reverse is certainly not true . we are left with little else to go on : the deviation in spectral temperature is uncorrelated with either merger mass fraction ( a bit suprising ) and shows no dependence on the cosmological model . it does show some dependence on the mass of the parent , which leads to a change in slope of the observed mass - temperature relation . figures [ dtvst10 ] and [ dtvsteq ] of the previous section are not observables , since we ca nt yet reliably identify merger events . we would like to cull the largest excursions from a real cluster sample , but we need to do it some other way . the solution is to build up an observable mass - temperature correlation ( or icm mass , if you prefer ; remember that the fixed baryon fraction in our simulations makes the two equivalent ) and cull out the clusters which are coolest _ relative to the mean relationship_. we have already plotted the cluster mass vs. spectral temperature for the [ 2.0,9.5 ] bandpass within @xmath38 in figure [ specmasstemp ] . our analysis leads us to expect that the clusters which fall far below the best - fit line may be merging clusters . to test this we treat mass as the independent variable , and calculate the difference between @xmath0 as measured and the predicted @xmath0 based on the best - fit correlation with cluster mass . this quantity is then plotted against @xmath120 , as defined in the previous section . the results for the [ 0.5,9.5 ] bandpass within @xmath38 are displayed in figure [ dtmvst10 ] . we ve switched bands here because the culling technique we propose seems to work best when temperatures are measured over the entire energy range ( due to the increased contrast mentioned above ) . this plot is harder to interpret , but by drawing a horizontal line we see that we can isolate the 40 ( 10% ) coolest ( relative to the expected temperature at that mass ) images and be reasonably sure that these clusters are dynamically young , i.e. @xmath133 . in real life , we could apply this technique by ( 1 ) measuring icm gas masses with precision in a soft bandpass such as that of the _ rosat _ pspc where the luminosity is a weak function of temperature ( e.g. mohr 1999 ) ; ( 2 ) measuring spectral temperatures with _ chandra _ ; and ( 3 ) culling clusters from the sample which deviate by more than @xmath134 from the mean relation . needless to say , this would work best with a large survey and would require a large time investment but the remainder of the sample would suffer less bias from clusters which are dynamically young . 6.0 in 6.0 in note that the definition of `` dynamical youth '' defined above is empirically driven . at low redshifts , 40% of the hubble time is a generous 3 - 5 billion years , typically about twice as long as the freefall ( @xmath135 ) and sound crossing timescales . we adopt this definition solely because all of the clusters with large temperature errors fall within this range , without further justification , since our main interest lies in removing the least accurate measurements from our sample . out of the whole ensemble of 384 images , fully 70% lie within this range of dynamical states . it is also interesting to note that the hottest spectra occur either long after a merger event or at times @xmath136 . the free fall time from @xmath39 is about @xmath114 , so the time interval from @xmath132 to @xmath137 represents the time required for a merging system to completely enter the observation window . the equivalent plot for @xmath128 is not given here , but again shows that most of the clusters below this cut are dynamically young . ( @xmath138 ) . a similar result is found in constructing these plots with the emission - weighted temperature , except there is also mild evidence for shock heating in temperatures determined previous to @xmath132 . finally , we apply this analysis to the three - redshift subset of our images and their mass - temperature relationship . figure [ cullmasstemp ] plots spectral temperature against total mass in the [ 0.5,9.5 ] band within @xmath38 , draws the best - fit line and a cut on clusters 15% cooler than this line , and identifies those clusters which are dynamically young with a different symbol . we can see from this plot that the proposed cut preferentially selects out merging clusters , although there are still many left in the sample . a previous analysis of these simulations ( mathiesen 1999 ) determined that icm masses of regular clusters are measurable with only @xmath139 scatter and @xmath140 bias using the parametric technique of mohr ( 1999 ) , and more importantly that the degree of error in @xmath101 is uncorrelated with mass and temperature . this culling method would therefore be unchanged if we attempted to apply it to the observable @xmath101@xmath0 relationship . this cut does not change the slope of the observed @xmath102@xmath0 relation and has a small ( about 0.5 sigma ) effect on the normalization . this method is thus primarily useful for identifying a subset of dynamically young clusters in a sample , presumably in the hopes that some will be found which are nt obvious in other ways . we have selected out ten merging clusters and a single relaxed object ; analysis of figure [ dtmvst10 ] implies that this ratio would be similar in a larger sample . 6.0 in 6.0 in the first ( and perhaps the most suprising ) result of this paper is the revelation that the emission spectrum of a realistically complex intracluster medium is barely distinguishable from that of an isothermal gas , even over a broad spectrum . this correspondence arises because we have created spectra with rather coarse ( 150 ev ) bins in an effort to match our observation model to _ chandra s _ acis instrument . we provide relationships between spectral , emission , and mass - weighted temperatures in two different bandpasses and within two density constrasts ; in particular , we find that spectral temperatures determined in the [ 2.0,9.5 ] kev range within @xmath38 are very similar in nature to those determined by the _ mpc , _ ginga _ , and _ exosat_. the other bandpass , [ 0.5,9.5 ] kev , is designed to be similar to _ chandra _ observations , and can be used to better interpret its results . we find that realistically determined spectral temperatures are commonly 10 - 20% lower than the mass - weighted temperatures in these simulations , a fact which has important implications for cluster physics . the bias arises through the natural and frequent occurence of minor accretion events : small clumps of cool gas which merge into the icm and produce an excess of soft x - rays , biasing the spectrum towards cooler temperatures until they are fully assimilated . because the mass - weighted temperature follows the virial relationship , it is a more accurate indicator of the binding mass . previous measurements of the cluster mass function power spectrum need to account for this source of error if they make use of spectral temperature data . we have calibrated the relationship between spectral and mass - weighted temperature for the [ 2.0,9.5 ] kev bandpass within @xmath38 for this purpose ; this should allow an appropriate correction for clusters observed by satellites with a similar energy range . the scale - dependent nature of this bias changes the slope of the observed @xmath101@xmath0 relationship in a direction consistent with recent determinations , but it does not account for the whole difference . additional physics , such as a variation in the icm mass fraction with temperature , is still needed to explain the observed slope . although we are stuck with a discrepance between observed temperatures and mass - weighted temperatures , this bias can be useful in identifying clusters which are dynamically young . not all of these events will be obvious , even with _ chandra s _ high - resolution surface brightness and temperature maps ; some of the mergers will be occuring on an axis near our line - of - sight . in such cases the shocks and cool regions will probably be masked by core emission , but the presence of a cool subclump can still produce an unusually large deviation in the spectral temperature . we should not fall into the trap of assuming that a cluster is relaxed because it looks spherically symmetric ; rather , we should do everthing we can to determine whether or not it is truly relaxed . the simulations show that all of the clusters with temperatures falling far ( @xmath141 ) below the mean mass - temperature relation are dynamically young ; a subset of merging clusters can thus be identified as outliers in the observed relation . the results described in this paper are a necessary step towards accurately measuring cluster masses under the assumption of virial equilibrium , as well as towards a reliable technique for determining a cluster s dynamical state observationally . it is not , however , the only step : first the astronomical community must recognize that the various definitions of temperature in common use are _ not _ equivalent . further work needs to be done in modeling observed temperatures ( e.g. core - excised spectral temperatures and flux - weighted temperatures averaged over an image ) and the effects of substructure on observable quantities . spectral modeling of simulations is computationally more expensive , but not difficult , and the level of systematic error caused by inaccurate modeling is now comparable to the observational constraints on cosmological parameters . the spectral cubes described in this paper will be made avaiable to the public on mathiesen s research web site along with higher - resolution images for the subset of images used in fitting the mass - temperature relationships , the accretion history of each cluster , and documentation describing the file format and how to model other satellite responses . our simulations of spatially resolved emission spectra can be used to help interpret _ chandra _ and _ xmm _ observations of real clusters , and we are sure that enterprising researchers can mine these datasets for other interesting results . we would like to thank nancy brickman , john raymond , randall smith , and martin sulkanen for their help in implementing the raymond - smith and ` m`ekal spectral models . this research was supported by nasa grants nag5 - 2790 , nag5 - 7108 , and nag5 - 8458 ; and nsf grant ast-9803199 . 99 allen s. w. & fabian a. c. , 1998 , mnras , 297 , 57 allen s. w. , fabian a. c. , johnstone r. m. , arnaud k. a. , & nulsen p. e. j. , mnras , preprint ( astro - ph/9910188 ) arnaud , m. & evrard , a.e . , 1999 , mnras , 305 , 631 bond j. r. & efstathiou g. , 1984 , apjl , 285 , l45 bryan g. l. & norman m. l. , 1998 , apj , 495 , 80 buote d. a. & canizares c. r. , 1996 , apj , 457 , 565 cavaliere a. & fusco - femiano r. , 1978 , a&a , 70 , 677 cavaliere a. , menci n. , & tozzi p. , 1998 , apj , 501 , 493 chize j .- , alimi j .- m . & teyssier r. , 1998 , apj , 495 , 630 david l. p. , slyz a. , jones c. , forman w. , & vrtilek s. d. , 1993 , apj , 412 , 479 edge a. c. & stewart g. c. , 1991 , mnras , 252 , 414 & 428 eke v. r. , navarro j. f. , & frenk c.s . , 1998 , apj , 503 , 569 evrard a. e. , 1990 , apj , 363 , 349 evrard a. e. , metzler c. a. , & navarro j. f. , 1996 , apj , 469 , 494 fabian a. c. , hu e. m. , cowie l. l. , & grindlay j. , 1981 , apj , 248 , 47 fabian a. c. , arnaud k. a. , bautz m. w. , & tawara y. , 1994 , apjl , 436 , l63 fox d. c. & loeb a. , 1997 , apj , 491 , 459 gunn k. f. & thomas p. a. , 1996 , mnras , 281 , 1133 hatsukade i. , 1989 , ph.d . thesis , osaka university holzapfel w. l. , arnaud m. , ade p. a. r. , church s. e. , fischer m. l. , mauskopf p. d. , rephaeli y. , wilbanks t. m. , & lange a. e. , 1997 , apj , 480 , 449 kang h. , cen r. , ostriker j. p , & ryu d. , 1994 , apj , 428 , 1 knopp g. p. & henry j. p. , 1996 , apj , 472 , 125 lacey c. & cole s. , 1994 , mnras , 271 , 676 lima neto g. b. , pislar v. , durret f. , gerbal d. & slezak e. , 1997 , a&a , 327 , 81 lloyd - davies e. j. , ponman t. j. & cannon d. b. , 2000 , mnras in press ( astro - ph/0002082 ) markevitch m. , 1998 , apj , 504 , 27 markevitch m. , forman w. r. , sarazin c. l. , & vikhlinin a. , 1998 , apj , 503 , 77 markevitch m. , mushotzky r. , inoue h. , yamashita k. , furuzawa a. , & tawara y. , 1996 , apj , 456 , 437 . mathiesen b. , evrard a. e. , & mohr j. j. , 1999 , apjl , 520 , 21 mewe r. , gronenschild e. h. b. m. , & van den oord g. h. j. 1985 , a&as , 62 , 197 mohr j. j. & evrard a. e. , 1997 , apj , 491 , 38 mohr j. j. , evrard a. e. , fabricant d. g. , & geller m. j. , 1995 , apj , 447 , 8 mohr j. j. , mathiesen b. & evrard a. e. , 1999 , apj , 517 , 627 mushotzky r. f. & scharf c. a. , 1997 , apjl , 482 , l13 nagai d. , evrard a. e. , & sulkanen m. e. , 1999 , mnras in press ( astro - ph/9903308 ) navarro j. f. , frenk c. s. , & white s. d. m. , 1995 , mnras , 275 , 720 sarazin c. l. , 1986 , _ reviews of modern physics _ , vol . 58 , no . 1 thomas p. a. , 1998 , mnras , 299 , 349 takizawa , m. 1999 , apj , 520 , 514 tsai , j.c . & buote , d.a . , 1996 , mnras , 282 , 77 waxman e. & miralda - escude j. , 1995 , apj , 451 , 451 white d. a. , jones c. , & forman w. , 1997 , mnras , 292 , 419 white s. d. m. , navarro j. f. , evrard a. e. , & frenk c. s. , 1993 , _ nature _ , 366 , 429 xspec user s manual : http://legacy.gsfc.nasa.gov/docs/xanadu/xspec/index.html

we investigate brane inflation driven by two stacks of mobile branes in a throat . the stack closest to the bottom of the throat annihilates first with antibranes , resulting in particle production and a change of the equation of state parameter . we calculate analytically some observable signatures of the collision ; related decays are common in multi - field inflation , providing the motivation for this case study . the discontinuity in enters the matching conditions relating perturbations in the remaining degree of freedom before and after the collision , affecting the power - spectrum of curvature perturbations . we find an oscillatory modulation of the power - spectrum for scales within the horizon at the time of the collision , and a slightly redder spectrum on super - horizon scales . we comment on implications for staggered inflation .

1004.1417

constructing models of inflation in string theory is an active field , for reviews see e.g. @xcite . because string theory naturally has many dynamical degrees of freedom , multi - field models of inflation are also common . examples are inflation from axions @xcite , tachyons @xcite , m5-branes @xcite or multiple d - branes @xcite to name a few . if many degrees of freedom are present , it is natural for them to decay or stabilize during inflation one after the other . a phenomenological approach to recover some of the effects caused by decaying fields was proposed in @xcite and dubbed staggered inflation . @xcite relies on coarse graining , assuming that many fields decay in any given hubble time , which may or may not be the case in concrete models . in this paper , we set out to investigate a single decay in detail to extract analytically the consequences for cosmological perturbations . as a simple and yet non - trivial framework , we choose the kklmmt @xcite brane inflation @xcite set up . in this proposal , inflation is driven by a brane - antibrane pair , located in a warped throat generated by the background branes and fluxes @xcite . inflation ends when the distance between the mobile brane and the antibrane reaches the string scale and a tachyon is formed . after the brane - antibrane collision , the energy stored in their tensions is released into light closed string modes , which can ignite reheating . we extend this proposal by including two mobile stacks with @xmath1 and @xmath2 branes respectively and @xmath3 antibranes at the tip of the throat . we assume that the stacks are separated from each other so the stack closer to the tip , say stack one , annihilates first during inflation . this collision results in particle production during inflation and a sudden change in the inflationary potential . we would like to examine some effects of the produced particles via the change in the equation of state parameter on the power - spectrum of cosmological perturbations . inflation ends when the remaining stack of @xmath2 branes annihilates with the background s @xmath2 antibranes . this extension does not improve upon the back - reaction and fine - tuning issues of the kklmmt proposal @xcite , but provides a well motivated toy model in string theory . to handle the analysis , we assume that perturbations in the effective field describing the first stack , @xmath4 , are taken over by the radiation bath , decaying quickly after the annihilation event . hence , we focus on perturbations in the remaining field only , @xmath5 . for this simplification to be consistent , we assume that @xmath6 . we derive asymptotic solutions in the two slow - roll regimes before and after the collision for this field , which we match at the collision time . treating the mechanism of tachyon formation and the collision as instantaneous events , we use the israel junction conditions @xcite to provide the matching conditions , which boil down to the continuity of the bardeen potential @xmath7 and the curvature perturbation @xmath8 in the case at hand . it is in these matching conditions that the jump in the equation of state parameter enters . we then compute the power - spectrum of curvature perturbations and find a slightly redder spectrum on scales that were super - horizon at the time of the first stack s annihilation . on sub - horizon scales , we find an oscillatory modulation of the power spectrum . the concrete outline of this paper is as follows : we review and extend the kklmmt setup in sec . [ kklmmt ] before deriving the background , slow - roll solution in sec . [ sec : bgr ] . perturbations are discussed in sec . [ sec : pertub ] , with the matching conditions in sec . [ sec : matching ] and the bogoliubov coefficients in sec . [ sec : bogol ] . these coefficients enable the computation of the power - spectrum in sec . [ sec : powersp ] , which we expand for super- and sub - horizon scales in sec . [ sec : suph ] and sec . [ sec : subh ] . we conclude in sec . [ sec : concl ] . here we present our set up which is an extension of the kklmmt proposal of warped brane inflation @xcite . we start with a warped throat given by the geometry [ throat ] ds^2 = h^-1/2(r ) dx^ dx^ + h^1/2(r ) ( dr^2 + r^2 ds_5 ^ 2 ) where @xmath9 is the warp factor h(r)= . here @xmath10 is the ads scale of the throat , which is created by @xmath11 coincident background branes located at @xmath12 [ l - eq ] l^4= c g_s n ^2 , where @xmath13 is the perturbative string coupling , @xmath14 is the string theory length scale and @xmath15 is a geometric factor depending on the internal geometry , @xmath16 . for example , for @xmath17 we have @xmath18 while for @xmath19 we get @xmath20 . alternatively , one may imagine that the background ( [ throat ] ) is created by turning on fluxes @xcite . as usual in models of brane inflation , it is assumed that the throat is smoothly glued to the bulk of the compactification . and @xmath21 while @xmath22-branes are located at the bottom of the throat at @xmath23 . the first annihilation happens when the physical distance between @xmath24 and @xmath23 becomes equal to string scale @xmath25 , which occurs shortly after @xmath26.,width=192 ] [ throat1 ] . once the @xmath1 d3-branes come within @xmath27 to the antibranes , they annihilate . their decay products , such as light closed string loops denoted by @xmath28 , cause a jump in the equation of state parameter @xmath0 ( see sec . [ sec : eosparam ] ) . we require @xmath29 in this study , so that @xmath28 is subdominant after the collision.,width=432 ] [ throat2 ] the metric in ( [ throat ] ) represents the background geometry , on top of which we add two stacks of @xmath1 and @xmath2 d3-branes within the throat at positions @xmath24 and @xmath21 respectively . at the bottom of the throat are @xmath3 @xmath30-branes . in accordance to the kklmmt proposal , the inflationary period arises from the mutual attraction of the mobile d3 branes and anti d3-branes plus the interaction of the @xmath30-branes with the background fluxes . we work in the probe brane approximation , meaning , we assume that the added stacks of branes and antibranes do not destroy the background . for a schematic view of our set up see * fig . [ throat1]*. a crucial feature of our model is the annihilation of brane - antibrane pairs during inflation . to illustrate , during inflation , the mobile stack of branes closest to the tip of the throat , comprised of @xmath1 d3-branes , annihilates with @xmath1 @xmath22-branes . the effective field corresponding to this stack drops out of the dynamics , leading to a sudden jump in the potential . after this annihilation , inflation proceeds driven by the remaining stack of branes . inflation ends when the second stack annihilates with the @xmath2 antibranes located at the tip . it is straightforward to generalize this setup to multiple stacks , @xmath31 , and multiple collisions . however , since our primary goal is the investigation of a single collision event , we consider only two stacks . for a schematic of the collision , see * fig . [ throat2]*. after the annihilation event , the kinetic energies of the branes as well as potential energies originating from the branes tensions are transferred into closed string modes . in order to simplify the analysis , we assume that these are in the form of massless modes so that they behave as radiation . as we shall see in our analysis , it will not change the results if one assumes that the produced particles are in the form of dust . the important effect is the sudden change in the equation of state parameter due to the collision . furthermore , we assume that the brane annihilation and tachyon formation is sudden , leading to an efficient method of converting decay products of the brane - antibrane annihilation into closed string modes . for a list of phenomenological studies of reheating in models of brane inflation see e.g. @xcite and the reference therein . to calculate the inflationary potential , we treat the @xmath32-branes as probes in the geometry created by the stacks of @xmath11 background branes as well as the additional two stacks of d3-branes @xcite . this leads to the condition @xmath33 . the warp factor in this configuration is then given by [ h2 ] h(r)= + + . this form of the metric indicates that the multiple stacks of d3-branes in the absence of antibranes and fluxes are supersymmetric . the action of the @xmath34-branes located at the bottom of the throat , @xmath23 , is @xcite [ actiond ] s&=&-2 ( p_1 + p_2 ) t_3 d^4x h(r_0)^-1 , where @xmath35 is the d3-brane tension and the factor of @xmath36 in eq . ( [ actiond ] ) originates from the combined tension and charge of the antibranes in this background . defining @xmath37 for @xmath38 , @xmath39 where the subscript @xmath40 denotes antibranes , and noting that @xmath41 @xcite , we can write the potential for the two scalar fields @xmath42 before the collision as [ potential ] v^-= v_0 ^ - , where v_0 ^ - 2 ( p_1+p_2 ) t_3 , b_i . if @xmath43 the field @xmath44 drops out of the dynamics first . after the annihilation of the first stack , the potential reduces to [ v2 ] v^+(_2 ) v_0^+ , where @xmath45 leading to a jump in the potential . after coupling the system to four - dimensional gravity , the total action is [ totals ] s= & & d^4 x + & & + d^4 x ( _ ra - v^+(_2 ) ) ( t - t_c ) where @xmath46 is the planck mass related to newton s constant @xmath47 via @xmath48 and @xmath49 indicates the collision time . in order to respect conservation of energy at @xmath49 we are forced to include an additional contribution to the action , @xmath50 , corresponding to radiation created during the annihilation event . for self - consistency , we require the initial distance between branes and antibranes to be greater than the string scale , @xmath51 , otherwise , the appearance of tachyons leads to the immediate annihilation of branes and antibranes . the first inflationary stage terminates at @xmath52 where the distance between the first stack at @xmath24 and the antibranes at the tip is @xmath25 . defining @xmath53 and using the metric in ( [ throat ] ) , we obtain @xmath54 . the onset of tachyon formation and brane annihilation results in the violation of the slow - roll conditions at the collision time . practically , as in @xcite , the onset of the brane collision is indicated by @xmath55 where @xmath56 , defined in ( [ slowrollparameters ] ) , is a slow - roll parameter . in this section , we derive the background dynamics in a homogeneous and isotropic universe where the four dimensional metric is ds^2 = -dt^2 + a(t)^2 d ^2 , @xmath57 is the scale factor and @xmath58 is cosmic time . the friedmann and klein gordon equations are h^2 = , _ i + 3 h _ i + = 0 , where @xmath59 is the total energy density containing contributions of scalar fields and radiation , @xmath60 , after the collision . the background solutions for @xmath61 before the collision are easily obtained if the system is in the slow - roll regime , that is if the slow - roll parameters [ slowrollparameters ] _ i ( ) ^2 , _ i m_p^2 , _ 1 + _ 2 , _ 12 = , are all much smaller than one . since there is no cross term in the potential , the slow - roll parameter @xmath62 vanishes . here , we used the short - hand notation @xmath63 . during slow - roll we can approximate [ phi - eq ] 3 h_bc _ i + 0 , where we used the potential in ( [ potential ] ) and @xmath64 is the expansion rate in the slow - roll limit before the collision . ( [ phi - eq ] ) can be solved to _ i in ^6 - _ i c^6 t_c , where we set the initial time at the start of inflation to zero and @xmath65 as well as @xmath66 . we can replace cosmic time in terms of the number of e - folds @xmath67 , so that [ phii - n - eq ] _ i in^6- _ i c^6 24 m_p^2 b_i ( n_t - n_e ) . here @xmath68 represents the total number of e - folds before the end of inflation , which needs to be about sixty in order to solve the flatness and horizon problem , and @xmath69 is the number of e - folds from the time of the first stacks s collision till the end of inflation , which we assume to be close to @xmath70 throughout this article . after the first stack s collision , the field @xmath71 drops out of the dynamics and the system consists of the field @xmath72 and radiation , which is diluted quickly . thereafter , @xmath72 resumes its slow - roll evolution , that is the dynamics of @xmath72 is given by [ phi2-eq ] 3 h_ac _ 2 + 0 , where @xmath73 is the expansion rate in the slow - roll limit after the collision , [ hac ] h_ac . ( [ phi2-eq ] ) is solved by [ phi2c ] _ 2 c^6- _ 2 f^6 24 m_p^2 b_2 n_e . here @xmath74 is the final value of @xmath72 when inflation ends , corresponding to the time when the second stack annihilates with the remaining @xmath2 antibranes at the bottom of the throat . the first stack s collision is instigated when @xmath75 , as in @xcite , where @xmath76 represents the slow - roll parameters before and after the collision . the @xmath77 can be calculated to [ eta- ] _ i&= & -20 + & = & -()^6 . consequently , the condition @xmath75 results in [ phi1-c - in ] _ 1 c^6 = 20 b_1 m_p^2 , _ 1 in^6 = 24 b_1 m_p^2 ( n_t- n_e + 5/6 ) . similarly , @xmath78 , the time of the second stack s collision and the end of inflation , is set by @xmath79 , which results in @xmath80 . combining this with ( [ phi2c ] ) and ( [ phii - n - eq ] ) we obtain [ phi2-f - c ] _ 2 in^6 24 m_p^2 b_2 ( n_t + 5/6 ) , _ 2 c^6 24 m_p^2 b_2 ( n_e + 5/6 ) . for later reference it is instructive to express the slow - roll parameters in terms of the number of e - folds @xmath69 and @xmath81 . using eqs . ( [ phi1-c - in ] ) and ( [ phi2-f - c ] ) we get [ eta-1 - 2 ] ^-_1 in , ^-_2 in , ^+_2 c . here @xmath82 and @xmath83 indicate the slow - roll parameters at the start of inflation whereas @xmath84 indicates the corresponding slow - roll parameter after the first stack s collision . to obtain @xmath84 we assume that radiation is quickly diluted so that the remaining stack resumes its slow - roll motion quickly . regarding the other slow - roll parameters , we note that @xmath85 , since ( ) ^4 1 , due to the large value of @xmath86 and @xmath87 . this is common in warped brane inflation models such as @xcite , since the warp factor reduces the height of the potential . this implies that @xmath88 which is the main reason why gravitational wave production in these models is suppressed . since @xmath71 drops out of the dynamics after the collision , the energy associated with the tensions of the @xmath1 antibranes as well as the kinetic energy of the @xmath1 mobile branes are transferred to radiation . during the annihilation event the velocity @xmath89 is continuous , but due to the step in the potential ( [ v2 ] ) , the acceleration @xmath90 makes a jump . energy conservation , @xmath91 , dictates the value of @xmath60 just after the brane annihilation , _ r(t_c ) = - _ , where we defined @xmath92_{\pm } \equiv f(t_{c}^{+ } ) - f(t_{c}^{-})$ ] . assuming slow - roll inflation before the collision , one can neglect the kinetic energy of the mobile branes compared to the tension of the antibranes and obtains _ r(t_c ) ( v_0 ^ - - v_0^+ ) = 2 p_1 t_3 r_0 ^ 4/l^4 . we are interested in the limit where the energy transferred into radiation is small compared to the background inflationary potential , that is @xmath93 or @xmath94 . next , we would like to solve for the evolution of the scale factor after the collision ; noting that @xmath60 scales like radiation _ r(t ) ( v_0 ^ - - v_0^+ ) ( ) ^4 , where @xmath95 is the value of the scale factor at @xmath49 , we can approximate the friedmann equation for @xmath96 by 3 m_p^2 ( ) ^2 v_0^+ + ( v_0 ^ - - v_0^+ ) ( ) ^4 , which can be solved analytically for @xmath57 , [ a - biger ] a(t ) = ( ) ^1/2 ^1/2 a_c e^h_ac ( t - t_c ) . as explained before , @xmath73 , given by eq . ( [ hac ] ) , is the expansion rate after the collision , when radiation is diluted and the approximate de - sitter background is recovered from the interaction of the remaining @xmath2 d3 and @xmath22-branes . comparing @xmath73 with the expansion rate before the collision , we get @xmath97 . note that @xmath98 due to the presence of radiation although @xmath99 itself is continuous . asymptotically , the scale factor in ( [ a - biger ] ) approaches [ a - biger2 ] a ( tt_c ) ( ) ^1/2 a_c e^h_ac ( t - t_c ) . an interesting implication of ( [ a - biger2 ] ) is the increase of the scale factor s amplitude by a factor of @xmath100 , caused by the transient presence of radiation . in this section , we consider linear perturbations before and after the collision at @xmath101 . for @xmath102 perturbations are carried by the two scalar fields ( we assume a flat field space metric ) and after the collision , by the remaining scalar field and radiation . perturbations in @xmath71 are carried over predominantly by the radiation bath for @xmath96 and decay rapidly in the subsequent inflationary phase . thus , we do not need to follow their evolution . our goal in this section is to derive and solve the relevant equations of motion for perturbations in the surviving field before and after the collision , which need to be matched at @xmath103 . the most general line element including scalar perturbations is @xmath104 dx^i dx^j\,,\end{aligned}\ ] ] where we did not remove any gauge modes yet . the equations of motion of the perturbations in the scalar fields include couplings to the metric degrees of freedom and read @xcite @xmath105\,,\end{aligned}\ ] ] with @xmath38 . one can define gauge invariant metric perturbations , for instance the two bardeen potentials @xcite @xmath106 that coincide in the absence of anisotropic stress @xmath107 . further , we can define the gauge invariant sasaki - mukhanov variables @xmath108 these coincide with the field perturbations in the spatially flat gauge @xmath109 . in the absence of any other components in the energy momentum tensor , such as radiation , and using the perturbed einstein equations as well as the background equations of motion one gets ( see @xcite for details ) @xmath110 thus , the equations of motion for the sasaki - mukhanov variables decouple from the perturbed einstein equations for the bardeen potentials , and we need not be concerned about the latter . in deriving ( [ q - eq ] ) we neglected perturbations in the radiation bath . our philosophy is to keep the setup as simple as possible , while retaining some crucial effects onto perturbations of the annihilation event . following @xcite , one can show that the equations of motion simplify during slow - roll to [ ueq ] u_i + ( k^2 - ) u_i _ j m_ij u_j , where a prime denotes a derivative with respect to conformal time @xmath111 , @xmath112 and the matrix @xmath113 is defined as m_ij ( cc + 2 _ 1 - _ 11 & 2 _ 12 + 2 _ 12 & + 2 _ 2 - _ 22 ) , where we used the slow - roll parameters defined in ( [ slowrollparameters ] ) . since @xmath88 in warped brane inflation , the matrix @xmath114 becomes diagonal and the equations for the @xmath115 separate , @xmath116 where _ i^- - _ i^- and @xmath117 is defined in eq . ( [ eta- ] ) . we impose the minkowski vacuum state @xmath118 in the far past ( @xmath119 ) , corresponding to the choice ( see i.e. the review @xcite ) [ u- ] u_i^- = e^i ( _ i^- + 1/2)/2 h__i^-^(1 ) ( - k ) * e*_i , where @xmath120 is the hankel function of the first kind of order @xmath121 and @xmath122 are independent unit gaussian random fields with < * e*_i > = 0 , = _ ij ^3 ( * k * - * k * ) . since perturbations originate during the early stages of inflation well before the collision , we can use @xmath123 so that @xmath124 from ( [ eta-1 - 2 ] ) , resulting in [ mu-12 ] _ 1 ^ - = + , _ 2 ^ - = + . after the collision , when radiation is diluted away and @xmath72 resumes its slow - roll motion , the equation for @xmath125 is given by ( [ ueq ] ) with @xmath126 and @xmath127 , which can be integrated to [ u+ ] u_2^+ = e^i ( _ 2^+ + 1/2)/2 ( h__2^+^(1 ) ( - k ) + h__2^+^(2 ) ( - k ) ) . here @xmath128 , while @xmath129 and @xmath130 , often referred to as bogoliubov coefficients , need to be determined by an appropriate matching procedure near the collision . using eq . ( [ eta-1 - 2 ] ) we have [ mu2 + ] _ we note that @xmath131 since @xmath132 . in deriving the asymptotic solutions in ( [ u- ] ) and ( [ u+ ] ) we treat the slow roll parameters as small and constant . this approximation is common when dealing with inflationary slow - roll models and its validity has been shown numerically . in our model , slow roll is briefly violated during the collision , but due to the short time scale of this collision , we expect this approximation to remain valid . the time scale of tachyon formation and brane and anti - brane annihilation at the bottom of the throat is given by the inverse of the warped string scale . to trust our four - dimensional effective field theory approximation , the hubble expansion rate has to be much smaller than the warped string scale . this indicates that branes annihilate and slow - roll is violated for less than an e - fold and the slow - roll approximation is restored quickly . our goal in the next sections is to calculate @xmath129 and @xmath130 to compute the power - spectrum . for later use , we define [ alpha - beta ] _ 1 * e*_1 + _ 2 * e*_2 , _ 1 * e*_1 + _ 2 * e*_2 . we treat the collision in a sudden approximation , infusing the entire energy of @xmath71 into the radiation bath @xmath60 at @xmath103 . we are interested in the limit where the energy transferred into radiation is small compared to the background inflationary potential and @xmath93 . to perform the matching of cosmological perturbations , we demand that both the intrinsic and extrinsic curvatures are continuous on the hyper - surface separating the two inflationary phases @xcite . as in standard hybrid inflation models @xcite the hyper - surface of the first stack s collision , and thus the transition between the two phases , is set by a certain value of an inflaton field , @xmath133 for us . treating the transition as instantaneous , that is assuming @xmath134 , the cosmological matching conditions relating perturbations before and after the collision become @xcite @xmath135_\pm = 0 \quad , \quad \label{second - b.c.}[\mathcal{r}]_{\pm}=0\ , , \end{aligned}\ ] ] where the absence of anisotropic stress was used ( @xmath107 ) . these matching conditions are valid for modes with @xmath136 , which includes modes that are within the hubble horizon at @xmath103 . in the above , the comoving curvature perturbation is defined as @xmath137 the matching conditions in ( [ match1 ] ) for perturbations are accompanied by the continuity of the scale factor and the hubble parameter @xmath138 ( and thus the total energy ) at the background level , @xmath139_\pm=0 } \;\;\;\ ; , \;\;\;\ ; { [ \mathcal{h}]_\pm=0 } \ , , \end{aligned}\ ] ] which also follow from the continuity of the induced metric as well as the extrinsic curvature on the hyper - surface set by the first stack s collision . to make contact with observations , we need to compute the power - spectrum of the curvature perturbation @xmath8 , which is related to the sasaki - mukhanov variables via @xcite @xmath140 if no additional contributions to the energy momentum tensor are present . to impose the matching conditions , we need to know @xmath141 as a function of the @xmath115 fortunately , @xmath142 is already given in terms of the @xmath115 in eq . ( [ defr2 ] ) . to obtain @xmath7 as a function of the @xmath115 , we use the perturbed einstein equations @xcite @xmath143 with the gauge invariant perturbations of the energy momentum tensor @xmath144 and the gauge invariant field perturbation @xmath145 in the flat gauge we have @xmath109 and we can use @xmath146 as well as @xmath147 . plugging these into the perturbed einstein equations and using @xmath148 as well as the background equations of motion , we can solve ( [ peeq2 ] ) for @xmath149 , insert it into ( [ peeq1 ] ) and arrive after some algebra at @xmath150 where we used the slow - roll approximation and worked in leading order of the slow - roll parameters , that is we used @xmath151 and @xmath152 before and after the collision . it is at this point where we make an important approximation to simplify the matching procedure : based on the fact that perturbations in @xmath71 are predominantly carried over by perturbations in @xmath28 , which rapidly redshift , and that @xmath29 so that @xmath153 , we ignore @xmath154 before the collision just like we ignored @xmath155 after the collision . thus , in order to perform the matching , we approximate ( [ defr2 ] ) and ( [ defphi ] ) by @xmath156 so that the matching conditions in ( [ match1 ] ) become [ match2 ] _ = 0 , _ = 0 , where we used that @xmath157 is continuous . if we now use the asymptotic solutions before and after the collision respectively , ( [ u- ] ) and ( [ u+ ] ) , we can read off the bogoliubov coefficients @xmath129 and @xmath130 . our approximation of neglecting perturbations in radiation and @xmath4 is equivalent to having @xmath158 and @xmath159 in eq . ( [ alpha - beta ] ) . before we proceed , we would like to review our approximations . our first assumption is that the system is in the slow - roll limit before the collision . a further crucial assumption in our analysis is that radiation is quickly diluted after the collision so that the system resumes its slow - roll evolution quickly . in this limit , ( [ defr2 ] ) and ( [ defphi ] ) can be used , which are valid for systems composed of scalar fields only . we note that there have been many phenomenological studies of brane annihilations and reheating in models of brane inflation e.g. @xcite . however , we do not know the exact stringy details of tachyon formation and the brane collisions , nor the mechanism of energy transfer into radiation , especially in a non - trivial background with background fluxes present and collisions between multiple branes and anti - braens , see for example @xcite . nevertheless , we expect that perturbations in @xmath71 are predominantly carried over by perturbations in radiation so that we can use ( [ defr2approx ] ) and ( [ defphiapprox ] ) in order to perform the matching . we still keep some effects of radiation via the change in the equation of state parameter @xmath0 ; by means of the bogoliubov coefficients we can relate the solutions in the two slow - roll inflationary regimes by using the matching conditions ( [ match2 ] ) . one shortcoming of this treatment is that we do not take into account perturbations in radiation and @xmath71 , which would enter into the matching conditions . nevertheless , even with this simplified setup , we find interesting effects that we expect to prevail in a more rigorous treatment , which we leave to future study . before the collision , the branes move slowly and the slow - roll conditions are satisfied for both fields so that @xmath160 . close to the annihilation event , @xmath161 becomes of order one ; however , since the annihilation takes place shortly thereafter , the field @xmath71 has no time to speed up and we may use its slow - roll value , even though slow - roll is violated . particularly , we can neglect its kinetic energy compared to the potential one just before and after the collision . similarly , we may use the slow - roll value for @xmath72 . we would like to relate the outgoing solution in the slow - roll approximation in ( [ u+ ] ) to the incoming solution in ( [ u- ] ) . we already specified the parameters @xmath121 and @xmath162 in ( [ mu-12 ] ) and ( [ mu2 + ] ) , but we still need to calculate the equation of state parameter for the incoming and outgoing solutions . for @xmath102 we have [ 1+w- ] 1 + w_-= , where @xmath163 is the hubble expansion rate in the slow - roll approximation from ( [ hbc ] ) . after the collision , assuming that radiation is diluted quickly , the slow - roll conditions are satisfied again and the equation of state parameter is given by [ 1+w+ ] 1 + w_+ , so that [ w - ratio ] . here we defined [ gamma - def ] ( 1 + n_e ) ^5/6 , where we used ( [ phi1-c - in ] ) and ( [ phi2-f - c ] ) . similarly , the fractional change in @xmath0 is [ deltaw ] . in the limit @xmath164 , corresponding to no brane collision during inflation , or @xmath165 , we obtain @xmath166 , @xmath167 and @xmath168 as expected . finally , we note that during slow - roll inflation @xmath169 . in our approximation two slow - roll inflationary regimes are glued to each other with a continuous hubble expansion rate so there is no jump in conformal time ; , so that the difference between @xmath170 and @xmath171 is not only slow - roll suppressed , but also suppressed by @xmath172 ; as we shall see shortly , @xmath173 is needed to guarantee that additional features do not dominate the power - spectrum . ] thus @xmath174 is continuous on the hyper - surface of the first stack s collision . at last we are in a position to match the two phases of inflation and calculate the bogoliubov coefficients @xmath129 and @xmath130 , which we need in order to compute the power - spectrum after the first stack s collision . starting with the matching conditions ( [ match2 ] ) we obtain q_2|_- & = & q_2|_+ , + q_2 |_- & = & q_2 |_+ . in the limit where perturbations in radiation and @xmath4 are ignored , we have @xmath175 and @xmath176 in eq . ( [ alpha - beta ] ) . with @xmath177 we find [ alpha - beta-2 ] _ 2 & = & e^i , + _ 2 & = & e^i , where we defined x & & -k_c , + & & ( _ 2 ^ - - _ 2^+ ) = ( _ 2^+ - _ 2 ^ - ) . to derive ( [ alpha - beta-2 ] ) we used the identity @xmath178 > from here on , all hankel functions depend on @xmath179 and a prime denotes a derivative with respect to @xmath179 . since @xmath180 is the wave number for which perturbations leave the hubble horizon at the time of the collision , we see that @xmath181 smaller or bigger than one discriminates between super- and sub - horizon modes . with the above coefficients and after replacing the hankel functions in terms of bessel functions as well as using the corresponding identity for bessel - functions , @xmath182 , we obtain @xmath183 we note that @xmath184 , since the perturbations in @xmath71 encoded in @xmath185 and @xmath186 are `` lost '' at the collision ( we assume that they are taken over by radiation and redshifted away rapidly ) . in the limit @xmath187 we recover the usual normalization of the bogoliubov coefficients . however , for arbitrary values of @xmath188 we checked that by including the contributions of @xmath185 and @xmath186 the relation @xmath189 holds . we are interested in the power spectrum @xmath190 of the curvature perturbation @xmath8 , which is defined by ( see @xcite for a review ) @xmath191 to relate @xmath8 to the outgoing perturbations after the collision in ( [ u+ ] ) we use ( [ defr2 ] ) , yielding @xmath192 at late times ( @xmath193 ) we can neglect the decaying part of the hankel functions in ( [ u+ ] ) , that is @xmath194 consequently , we have @xmath195 which results in @xmath196 the overall amplitude @xmath197 around sixty e - folds before the end of inflation is set by the cobe normalization @xmath198 , which in turn sets the hubble scale during inflation . the important feature in this expression is the transfer function @xmath199 , which encodes some effects of the brane annihilation in terms of the bogoliubov coefficients . the scalar spectral index , defined as @xmath200 , becomes @xmath201 where @xmath202 is the usual slow - roll result given by n_2 1- . as in @xcite and @xcite one can tune the parameters of the model ( @xmath11 , @xmath203 and @xmath204 ) to guarantee a sufficiently long inflationary phase and @xmath205 . compared to standard single field inflationary models , the power - spectrum is altered by the transfer function @xmath206 , which encodes some of the effects of the brane collision and particle production during inflation . additional contributions to the power - spectrum and higher order correlation functions originate via back - scattering of the produced particles onto the inflaton condensate , leading to ir - cascading @xcite ; to estimate this effect , which can dominate the power - spectrum @xcite and is not retained in our approach , a better understanding of the brane annihilation process is needed . with @xmath207 and @xmath208 from ( [ alpha - beta-2 ] ) , we get [ alpha - beta-22 ] | _ 2 - _ 2|^2 & = & ( ) ^2 j__2^+^2 ( j__2 ^ -^2 + y__2 ^ -^2 ) + & & + x + & & + x^2 . we are interested in the shape of the transfer function in different limits , namely @xmath209 , @xmath210 and @xmath211 , corresponding to modes leaving the horizon before , during and after the collision . the full expression is plotted in * fig . [ pic : fulltransfer ] * , but it is instructive to derive approximate analytic expressions of the transfer function in the limiting cases @xmath209 and @xmath211 . from ( [ alpha - beta-22 ] ) plotted over @xmath212 for @xmath213 , @xmath214 , @xmath215 and @xmath216 . [ pic : fulltransfer],width=384 ] [ power ] the case @xmath217 represents perturbations which cross the hubble radius long before the collision , @xmath218 . in the absence of entropy perturbations these modes are constant during the two slow - roll phases before and after the collision and we do not expect strong changes as a result of the collision . to see this , we use the small @xmath179 limit of the bessel functions in ( [ alpha - beta-22 ] ) , which results in |_2 - _ 2|^2 & & ( ) ^2 ( _ 2^+ -_2 ^ - ) ^2 [ transfersmallx ] + & & ( ) ^2 ( _ 2^+ -_2 ^ - ) ^2 . with @xmath219 from ( [ mu-12 ] ) and @xmath162 from ( [ mu2 + ] ) we get @xmath220 where @xmath221 , @xmath222 is the number of e - folds from the collision until the end of inflation and we expanded in the last step , see * fig . [ pic : smallxtransfer ] * for a plot . we note a mild suppression on super - horizon scales proportional to @xmath223 . this is in agreement with the finding of @xcite , whose authors investigated the consequences of a discontinuity in the second derivative of the potential in a single inflaton model , which also caused a jump in @xmath224 . the scalar spectral index in ( [ ns ] ) picks up an extra contribution on super - horizon scales , @xmath225 which is small compared to the leading order slow - roll contribution . in the absence of a collision , that is for @xmath226 and @xmath227 , we recover @xmath228 , as expected . in ( [ alpha - beta-22 ] ) , solid line , is compared to the small @xmath179 limit in ( [ transfersmallx ] ) , dashed line , for @xmath213 , @xmath214 , @xmath215 and @xmath216 . [ pic : smallxtransfer],width=384 ] in the limit @xmath230 we obtain |_2 - _ 2 |^2 & & ^2 ( x - - ) + ( ) ^2 ^ 2 ( x - - ) [ transferlargex ] + & & 1 - 2 ^ 2 ^2(x - ) , where we expanded for @xmath231 . again , we recover @xmath232 in the limit @xmath187 . we observe that within the horizon , perturbations are modulated and the power - spectrum picks up an unsuppressed oscillatory mode with amplitude 2 ^ 22 ( 1+n_e)^5/3 , see * fig . [ pic : largextransfer]*. if @xmath233 , these oscillations are within the observational window of cmb experiments . in order to prevent them from dominating the power - spectrum , we require @xmath234 . as a consequence , we need @xmath235 . on the other hand , the mobile stacks of @xmath1 and @xmath2 branes are assumed to be probe branes . for consistency , we require large enough background branes ( flux charges ) corresponding to @xmath236 . this can be easily obtained in the light of developments in string theory flux compactifications @xcite , although , the large hierarchy @xmath237 may look problematic . however , if several branes annihilate successively , we expect these oscillations to be averaged out to some extent , alleviating this bound . thus , it would be interesting to extend the model examined in this paper to include several stacks of branes that annihilate one after the other . such a setup falls within the framework of staggered inflation , as proposed in @xcite . we plan to investigate such a model numerically in a forthcoming publication @xcite , to bridge the gap between the analytic results of this paper , valid for a single annihilation event , and the analytic results of @xcite , which are expected to hold in the presence of many annihilation events . small oscillations on top of the power - spectrum , as in * fig . [ pic : largextransfer ] * , have recently caught the attention of the community @xcite , since it is possible to mimic outliers in the cmb power - spectrum @xcite . in @xcite the ringing pattern is damped with increasing @xmath179 whereas in our model , similar to @xcite , we find no such damping . this indicates that all modes within the horizon are affected by the annihilation event , not only those with @xmath238 close to the horizon scale . however , on the physical grounds one may expect that very small scale modes which effectively live in the minkowski background should not feel the effect of brane annihilation and particle creation . our result of eq . ( [ transferlargex ] ) with constant amplitude power spectrum modulation may be due to the sudden brane annihilation approximation used in our analysis . in practice , the onset of tachyon formation and brane annihilation happens at a time scale given by the inverse of the warped string scale . this time scale is short enough , i.e. much shorter than @xmath239 , so our approximation for super - horizon perturbations is valid . however , for wavelengths shorter than the inverse of the warped string mass scale the transition should be smoothed out . as a result , this can remove the power spectrum modulation for wavelength shorter than the inverse warped string scale . a word of caution regarding the scalar spectral index in ( [ ns ] ) might be in order : if one inserts the transfer function ( [ alpha - beta-22 ] ) into ( [ ns ] ) and evaluates @xmath240 for large @xmath179 one finds a linearly increasing mode this mode is an artifact arising from the attempt to describe an oscillatory power - spectrum by means of a simple power law , which is inappropriate . in ( [ alpha - beta-22 ] ) , solid line , is compared to the large @xmath179 limit in ( [ transferlargex ] ) , dashed line , for @xmath213 , @xmath214 , @xmath215 and @xmath216 . [ pic : largextransfer],width=384 ] [ power ] in order to derive the power - spectrum analytically we have made a series of approximations . one important assumption in our analysis is the neglection of perturbations in the decaying field and radiation , which would both enter in the matching conditions . we plan to relax this approximation in a forthcoming publication preliminary results show the same oscillations in the power - spectrum with a comparable amplitude ( up to factors of order one ) , resulting in a similar bound on the number of branes . we ignored other decay channels , for instance the production of cosmic strings at the collision . the oscillatory corrections to the power - spectrum that we find are caused by the jump in the equation of state parameter and the potential . these jumps are present regardless of the exact type of decay products ; hence , we expect only minor quantitative changes to the power - spectrum if they were included . we also worked in a sudden decay approximation , matching two asymptotic solutions at the collision - time . in reality , the decay takes some time which smoothes out all discontinuities while slow roll is briefly violated for the decaying field . as explained previously , the time scale of tachyon formation and brane annihilation is controlled by the inverse of the warped string scale which is much smaller than @xmath239 during inflation in order for our effective four - dimensional approximation to be valid . in conclusion , our result for super - hubble perturbations with @xmath241 is trusted . however , for wavelengths shorter than the inverse of the warped string scale the transition should be smoothed out . this in turn may remove the unwanted power spectrum modulation for very small scales . motivated by the recent interest in inflationary models with more than one dynamical degree of freedom , we investigated the consequences for cosmological perturbations of a decaying field . as a model we chose one of the best understood models of inflation in string theory , the kklmmt proposal , and extended it slightly : instead of driving inflation by one brane that annihilates with an antibrane at the end of inflation , we allowed for two stacks of branes with @xmath1 and @xmath2 branes respectively . these two stacks correspond to two inflaton fields at the level of four dimensional effective field theory . if the two stacks are separated initially , the first stack can annihilate with antibranes during inflation , creating closed string loops ( radiation ) in the process ( we ingored other decay channels ) . we focused on @xmath29 , in which case the produced radiation does not interrupt inflation , but causes a slight jump in the equation of state parameter . this setup has all the shortcomings of the original kklmmt proposal , i.e. the back reaction of the volume modulus on the inflaton field @xcite , but it provides a well understood framework to investigate the effects of a brane annihilation and particle creation in detail . at the perturbed level , we focused on fluctuations of the surviving inflaton field , with the assumption that perturbations in the field corresponding to the first stack of branes are taken over by the produced radiation , which subsequently decay rapidly . this is our main simplifying assumption . we also ignored contributions from ir - cascading . the asymptotic solutions for fluctuations in the surviving field during the two inflationary regimes , before and after the collision , were matched at the time of the collision according to the israel junction conditions . it is in this matching procedure that the jump in the equation of state parameter enters . the resulting power - spectrum contains a transfer function , determined by the bogoliubov coefficients . on scales that are super - horizon during the collision , we find a slightly redder spectrum . on sub - horizon scales , the power - spectrum is modulated by oscillations , or a ringing , with an amplitude of @xmath242 . such a ringing pattern might help explain glitches in the cmb power - spectrum . since these oscillations should not yield corrections bigger than a few percent , we deduce that the remaining stack of branes needs to be comprised of many more branes than the first stack , @xmath243 . for our probe brane approximation to be valid , we require a large enough background charge , @xmath236 . it is possible that if several annihilation events were to be superimposed , the amplitude of the oscillations would be reduced and the hierarchy between @xmath244 become milder . thus , we are naturally lead to the framework of staggered inflation , where many annihilation events are assumed to occur in any given hubble time . the inclusion of many annihilation events renders a numerical study a necessity , which we plan to come back to in a forthcoming publication . to summarize , the presence of several dynamical degrees of freedom can lead to additional large contributions to the power - spectrum , such as a ringing pattern , if they start to decay during inflation . we would like to thank bruce bassett , cliff burgess , jim cline , misao sasaki , ashoke sen and henry tye for valuable discussions . we also thank amjad ashoorioon and shahin sheikh - jabbari for collaboration at the early stages of this work and stephon alexander , damien easson , anupam mazumdar , enrico pajer , paul steinhardt and brett underwood for comments . is supported by the council on science and technology at princeton university and is thankful for hospitality at the pi , where this work was instigated . d.b . is thankful for hospitality at princeton university . s. h. henry tye , `` brane inflation : string theory viewed from the cosmos , '' lect . notes phys . * 737 * , 949 ( 2008 ) [ arxiv : hep - th/0610221 ] . j. m. cline , `` string cosmology , '' arxiv : hep - th/0612129 . c. p. burgess , `` lectures on cosmic inflation and its potential stringy realizations , '' pos * p2gc * , 008 ( 2006 ) [ class . * 24 * , s795 ( 2007 ) ] [ arxiv:0708.2865 [ hep - th ] ] . l. mcallister and e. silverstein , `` string cosmology : a review , '' gen . grav . * 40 * , 565 ( 2008 ) [ arxiv:0710.2951 [ hep - th ] ] . d. baumann and l. mcallister , `` advances in inflation in string theory , '' arxiv:0901.0265 [ hep - th ] . a. mazumdar and j. rocher , `` particle physics models of inflation and curvaton scenarios , '' arxiv:1001.0993 [ hep - ph ] . s. dimopoulos , s. kachru , j. mcgreevy and j. g. wacker , `` n - flation , '' jcap * 0808 * , 003 ( 2008 ) [ arxiv : hep - th/0507205 ] . y. s. piao , r. g. cai , x. m. zhang and y. z. zhang , `` assisted tachyonic inflation , '' phys . d * 66 * , 121301 ( 2002 ) [ arxiv : hep - ph/0207143 ] . m. majumdar and a. c. davis , `` inflation from tachyon condensation , large n effects , '' phys . d * 69 * , 103504 ( 2004 ) [ arxiv : hep - th/0304226 ] . k. becker , m. becker and a. krause , `` m - theory inflation from multi m5-brane dynamics , '' nucl . b * 715 * , 349 ( 2005 ) [ arxiv : hep - th/0501130 ] . a. ashoorioon and a. krause , `` power spectrum and signatures for cascade inflation , '' arxiv : hep - th/0607001 . a. ashoorioon , a. krause and k. turzynski , `` energy transfer in multi field inflation and cosmological perturbations , '' jcap * 0902 * , 014 ( 2009 ) [ arxiv:0810.4660 [ hep - th ] ] . j. m. cline and h. stoica , `` multibrane inflation and dynamical flattening of the inflaton potential , '' phys . d * 72 * , 126004 ( 2005 ) [ arxiv : hep - th/0508029 ] . a. ashoorioon , h. firouzjahi and m. m. sheikh - jabbari , `` m - flation : inflation from matrix valued scalar fields , '' jcap * 0906 * , 018 ( 2009 ) [ arxiv:0903.1481 [ hep - th ] ] . a. ashoorioon , h. firouzjahi and m. m. sheikh - jabbari , `` matrix inflation and the landscape of its potential , '' arxiv:0911.4284 [ hep - th ] . d. battefeld , t. battefeld and a. c. davis , `` staggered multi - field inflation , '' jcap * 0810 * , 032 ( 2008 ) [ arxiv:0806.1953 [ hep - th ] ] . t. battefeld , `` exposition to staggered multi - field inflation , '' nucl . suppl . * 192 - 193 * , 128 ( 2009 ) [ arxiv:0809.3242 [ astro - ph ] ] . s. b. giddings , s. kachru and j. polchinski , `` hierarchies from fluxes in string compactifications , '' phys . rev . d * 66 * , 106006 ( 2002 ) , hep - th/0105097 . k. dasgupta , g. rajesh and s. sethi , `` m theory , orientifolds and g - flux , '' jhep * 9908 * , 023 ( 1999 ) [ arxiv : hep - th/9908088 ] . d. baumann , a. dymarsky , i. r. klebanov , j. m. maldacena , l. p. mcallister and a. murugan , `` on d3-brane potentials in compactifications with fluxes and wrapped d - branes , '' jhep * 0611 * , 031 ( 2006 ) [ arxiv : hep - th/0607050 ] . c. p. burgess , j. m. cline , k. dasgupta and h. firouzjahi , `` uplifting and inflation with d3 branes , '' jhep * 0703 * , 027 ( 2007 ) [ arxiv : hep - th/0610320 ] . d. baumann , a. dymarsky , i. r. klebanov and l. mcallister , `` towards an explicit model of d - brane inflation , '' jcap * 0801 * , 024 ( 2008 ) [ arxiv:0706.0360 [ hep - th ] ] . f. chen and h. firouzjahi , `` dynamics of d3-d7 brane inflation in throats , '' jhep * 0811 * , 017 ( 2008 ) [ arxiv:0807.2817 [ hep - th ] ] . j. m. cline , l. hoi and b. underwood , `` dynamical fine tuning in brane inflation , '' jhep * 0906 * , 078 ( 2009 ) [ arxiv:0902.0339 [ hep - th ] ] . w. israel , `` singular hypersurfaces and thin shells in general relativity , '' nuovo cim . b * 44s10 * , 1 ( 1966 ) [ erratum - ibid . b * 48 * , 463 ( 1967 nucia , b44,1.1966 ) ] . g. shiu , s. h. h. tye and i. wasserman , `` rolling tachyon in brane world cosmology from superstring field theory , '' phys . d * 67 * , 083517 ( 2003 ) [ arxiv : hep - th/0207119 ] . j. m. cline , h. firouzjahi and p. martineau , `` reheating from tachyon condensation , '' jhep * 0211 * , 041 ( 2002 ) [ arxiv : hep - th/0207156 ] . j. h. brodie and d. a. easson , `` brane inflation and reheating , '' jcap * 0312 * , 004 ( 2003 ) [ arxiv : hep - th/0301138 ] . n. barnaby , c. p. burgess and j. m. cline , _ warped reheating in brane - antibrane inflation _ , jcap * 0504 * , 007 ( 2005 ) , hep - th/0412040 . l. kofman and p. yi , _ reheating the universe after string theory inflation _ , hep - th/0507257 . d. chialva , g. shiu and b. underwood , _ warped reheating in multi - throat brane inflation _ , hep - th/0508229 . a. r. frey , a. mazumdar and r. myers , _ stringy effects during inflation and reheating _ , hep - th/0508139 . h. firouzjahi and s. h. tye , `` the shape of gravity in a warped deformed conifold , '' jhep * 0601 * , 136 ( 2006 ) [ arxiv : hep - th/0512076 ] . x. chen and s. h. tye , `` heating in brane inflation and hidden dark matter , '' jcap * 0606 * , 011 ( 2006 ) [ arxiv : hep - th/0602136 ] . a. mazumdar and h. stoica , `` exciting gauge field and gravitons in a brane - anti - brane annihilation , '' phys . * 102 * , 091601 ( 2009 ) [ arxiv:0807.2570 [ hep - th ] ] . k. enqvist , a. jokinen , a. mazumdar , t. multamaki and a. vaihkonen , `` cosmological constraints on string scale and coupling arising from tachyonic instability , '' jhep * 0508 * , 084 ( 2005 ) [ arxiv : hep - th/0502185 ] . s. sarangi and s. h. h. tye , `` cosmic string production towards the end of brane inflation , '' phys . b * 536 * , 185 ( 2002 ) [ arxiv : hep - th/0204074 ] . h. firouzjahi and s .- h . h. tye , `` brane inflation and cosmic string tension in superstring theory , '' jcap * 0503 * , 009 ( 2005 ) , hep - th/0501099 . v. f. mukhanov , h. a. feldman and r. h. brandenberger , `` theory of cosmological perturbations . classical perturbations . part 2 . quantum theory of perturbations . extensions , '' phys . rept . * 215 * , 203 ( 1992 ) . c. gordon , d. wands , b. a. bassett and r. maartens , `` adiabatic and entropy perturbations from inflation , '' phys . d * 63 * , 023506 ( 2001 ) [ arxiv : astro - ph/0009131 ] . a. taruya and y. nambu , `` cosmological perturbation with two scalar fields in reheating after inflation , '' phys . b * 428 * , 37 ( 1998 ) [ arxiv : gr - qc/9709035 ] . c. t. byrnes and d. wands , `` curvature and isocurvature perturbations from two - field inflation in a slow - roll expansion , '' phys . d * 74 * , 043529 ( 2006 ) [ arxiv : astro - ph/0605679 ] . b. a. bassett , s. tsujikawa and d. wands , `` inflation dynamics and reheating , '' rev . phys . * 78 * , 537 ( 2006 ) [ arxiv : astro - ph/0507632 ] . n. deruelle and v. f. mukhanov , `` on matching conditions for cosmological perturbations , '' phys . d * 52 * , 5549 ( 1995 ) [ arxiv : gr - qc/9503050 ] . j. martin and d. j. schwarz , `` the influence of cosmological transitions on the evolution of density perturbations , '' phys . rev . d * 57 * , 3302 ( 1998 ) [ arxiv : gr - qc/9704049 ] . a. d. linde , `` hybrid inflation , '' phys . d * 49 * , 748 ( 1994 ) [ arxiv : astro - ph/9307002 ] . e. j. copeland , a. r. liddle , d. h. lyth , e. d. stewart and d. wands , `` false vacuum inflation with einstein gravity , '' phys . d * 49 * , 6410 ( 1994 ) [ arxiv : astro - ph/9401011 ] . i. zaballa and m. sasaki , `` boosted perturbations at the end of inflation , '' arxiv:0911.2069 [ astro-ph.co ] . d. h. lyth , k. a. malik , m. sasaki and i. zaballa , `` forming sub - horizon black holes at the end of inflation , '' jcap * 0601 * , 011 ( 2006 ) [ arxiv : astro - ph/0510647 ] . i. zaballa , a. m. green , k. a. malik and m. sasaki , `` constraints on the primordial curvature perturbation from primordial black holes , '' jcap * 0703 * , 010 ( 2007 ) [ arxiv : astro - ph/0612379 ] . n. t. jones , l. leblond and s. h. h. tye , `` adding a brane to the brane anti - brane action in bsft , '' jhep * 0310 * , 002 ( 2003 ) [ arxiv : hep - th/0307086 ] . n. t. jones and s. h. h. tye , `` an improved brane anti - brane action from boundary superstring field theory and multi - vortex solutions , '' jhep * 0301 * , 012 ( 2003 ) [ arxiv : hep - th/0211180 ] . n. barnaby , z. huang , l. kofman and d. pogosyan , `` cosmological fluctuations from infra - red cascading during inflation , '' phys . d * 80 * , 043501 ( 2009 ) [ arxiv:0902.0615 [ hep - th ] ] . n. barnaby and z. huang , `` particle production during inflation : observational constraints and signatures , '' phys . d * 80 * , 126018 ( 2009 ) [ arxiv:0909.0751 [ astro-ph.co ] ] . m. joy , v. sahni and a. a. starobinsky , `` a new universal local feature in the inflationary perturbation spectrum , '' phys . d * 77 * , 023514 ( 2008 ) [ arxiv:0711.1585 [ astro - ph ] ] . m. joy , a. shafieloo , v. sahni and a. a. starobinsky , `` is a step in the primordial spectral index favored by cmb data ? , '' jcap * 0906 * , 028 ( 2009 ) [ arxiv:0807.3334 [ astro - ph ] ] . d. battefeld , t. battefeld , j. t. giblin jr and e.pease , in preparation . a. e. romano and m. sasaki , `` effects of particle production during inflation , '' phys . d * 78 * , 103522 ( 2008 ) [ arxiv:0809.5142 [ gr - qc ] ] . m. zarei , `` short distance physics and initial state effects on the cmb power spectrum and cosmological constant , '' phys . d * 78 * , 123502 ( 2008 ) [ arxiv:0809.4312 [ hep - th ] ] . r. flauger , l. mcallister , e. pajer , a. westphal and g. xu , `` oscillations in the cmb from axion monodromy inflation , '' arxiv:0907.2916 [ hep - th ] . t. biswas , a. mazumdar and a. shafieloo , `` wiggles in the cosmic microwave background radiation : echoes from non - singular cyclic - inflation , '' arxiv:1003.3206 [ hep - th ] . e. komatsu _ et al . _ , `` seven - year wilkinson microwave anisotropy probe ( wmap ) observations : cosmological interpretation , '' arxiv:1001.4538 [ astro-ph.co ] . k. ichiki , r. nagata and j. yokoyama , `` cosmic discordance : detection of a modulation in the primordial fluctuation spectrum , '' arxiv:0911.5108 [ astro-ph.co ] . j. hamann , a. shafieloo and t. souradeep , `` features in the primordial power spectrum ? a frequentist analysis , '' arxiv:0912.2728 [ astro-ph.co ] .

radio frequency ( rf ) energy transfer and harvesting techniques have recently become alternative methods to power the next generation wireless networks . as this emerging technology enables proactive energy replenishment of wireless devices , it is advantageous in supporting applications with quality of service ( qos ) requirements . in this paper , we present an extensive literature review on the research progresses in wireless networks with rf energy harvesting capability , referred to as rf energy harvesting networks ( rf - ehns ) . first , we present an overview of the rf - ehns including system architecture , rf energy harvesting techniques and existing applications . then , we present the background in circuit design as well as the state - of - the - art circuitry implementations , and review the communication protocols specially designed for rf - ehns . we also explore various key design issues in the development of rf - ehns according to the network types , i.e. , single - hop networks , multi - antenna networks , relay networks , and cognitive radio networks . finally , we envision some open research directions . _ index terms- rf energy harvesting , simultaneous wireless information and power transfer ( swipt ) , receiver operation policy , beamforming , communication protocols , rf - powered cognitive radio network_.

1406.6470

recently , there has been an upsurge of research interests in radio frequency ( rf ) energy harvesting / scavenging technique ( see @xcite and references therein ) , or rf harvesting in short , which is the capability of converting the received rf signals into electricity . this technique becomes a promising solution to power energy - constrained wireless networks . conventionally , the energy - constrained wireless networks , such as wireless sensor networks , have a limited lifetime which largely confines the network performance . in contrast , an rf energy harvesting network ( rf - ehn ) has a sustainable power supply from a radio environment . therefore , the rf energy harvesting capability allows the wireless devices to harvest energy from rf signals for their information processing and transmission . consequently , rf - ehns have found their applications quickly in various forms , such as wireless sensor networks @xcite , wireless body networks @xcite , and wireless charging systems . with the increasingly emerging applications of rf energy harvesting / charging , the wireless power consortium is also making the efforts of establishing an international standard for the rf energy harvesting technique . in rf energy harvesting , radio signals with frequency range from 300ghz to as low as 3khz are used as a medium to carry energy in a form of electromagnetic radiation . rf energy transfer and harvesting is one of the wireless energy transfer techniques . the other techniques are inductive coupling and magnetic resonance coupling . inductive coupling @xcite is based on magnetic coupling that delivers electrical energy between two coils tuned to resonate at the same frequency . the electric power is carried through the magnetic field between two coils . magnetic resonance coupling @xcite utilizes evanescent - wave coupling to generate and transfer electrical energy between two resonators . the resonator is formed by adding a capacitance on an induction coil . both of the above two techniques are near - field wireless transmission featured with high power density and conversion efficiency . the power transmission efficiency depends on the coupling coefficient , which depends on the distance between two coils / resonators . the power strength is attenuated according to the cube of the reciprocal of the distance @xcite , specifically , 60 db per decade of the distance , which results in limited power transfer distance . besides , both inductive coupling and resonance coupling require calibration and alignment of coils / resonators at transmitters and receivers . therefore , they are not suitable for mobile and remote replenishment / charging . in contrast , rf energy transfer has no such limitation . as the radiative electromagnetic wave can not retroact upon the antenna that generated it ( by capacitive or inductive coupling ) at a distance of above @xmath0 @xcite , rf energy transfer can be regarded as a far - field energy transfer technique . thus , rf energy transfer is suitable for powering a larger number of devices distributed in a wide area . the signal strength of far - field rf transmission is attenuated according to the reciprocal of the distance between transmitter and receiver , specifically , 20 db per decade of the distance . table [ wet ] shows the comparison between the three major wireless energy transfer techniques . we can see that rf energy transfer technique has clear advantages in effective energy transfer distance . however , it has low rf - to - dc energy conversion efficiency especially when the harvested rf power is small . the readers can refer to @xcite for more detailed introduction of wireless energy transfer techniques . in this article , we focus on wireless networks with the rf energy harvesting technique . wireless power transfer has caught research attention since long ago , as a separate problem with wireless information transmission . traditionally , free - space beaming and antennas with large apertures were used to overcome propagation loss for large power transfer . for example , in 1960 s , the authors in @xcite demonstrate a small helicopter hovering at an height of 50-feet , powered by an rf source with a dc power supply of 270w operating on 2.45ghz on the ground . in @xcite , the authors demonstrate a space - to - earth power transfer system using gigantic transmit antenna arrays at a satellite and receive antenna arrays at a ground station . for transmit power of 2.7gw , the power transfer efficiency is estimated to be @xmath1 over a transfer distance of @xmath2 . during the past decade , with the development in rf energy harvesting circuit , low power transfer for powering mobile terminals in wireless communication systems began to attract increasing attention @xcite . the authors in @xcite propose a network architecture for rf charging stations , overlaying with an uplink cellular network . in @xcite , a harvest - then - transmit protocol is introduced for power transfer in wireless broadcast system . moreover , various modern beamforming techniques are employed to improve power transfer efficiency @xcite for mobile applications . [ cols="<,<,<,<,<,<",options="header " , ] table [ routing ] shows the comparison of the existing routing protocols for rf - ehns . it can be observed that all the protocols work in the systems with a dedicated rf charger , because of which the majority of them consider out - of - band charging to avoid interference . though in @xcite , the authors perform an experiment of the system where a sensor node and rf charger work on the same frequency , no interference management scheme is taken into account . therefore , this is a room for the investigation of routing protocols in the system adopting swipt . furthermore , given the routing metric , information exchange among the network devices through broadcasting is required during route selection . due to hardware limitation as aforementioned , network devices can not harvest rf energy from the same carrier for information decoding . as a result , the network devices working in rf harvesting mode may miss the broadcast information . thus , an efficient message broadcasting mechanism for the time - switching based receiver architecture is also required for routing protocols . besides , an rf - ehn operates on ism band ( e.g. , wifi , zigbee and bluetooth ) may overlap with the frequency band for wireless charging ( e.g. , the system in @xcite ) . in this context , an rf charger , if not well controlled , can cause severe interference to the network communication , as its power is usually much higher than that of network devices . thus , there is a need for efficient spectrum allocation mechanisms to coordinate communication and charging . in this section , we discuss about open research issues . distributed energy beamforming enables a cluster of distributed energy sources to cooperatively emulate an antenna array by transmitting rf energy simultaneously in the same direction to an intended energy harvester for better diversity gains . the potential energy gains at the receiver from distributed energy beamforming are expected to be the same as that from the well - known information beamforming . however , challenges arise in the implementation , e.g. , time synchronization among energy sources and coordination of distributed carriers in phase and frequency so that rf signals can be combined constructively at the receiver . existing interference management techniques , e.g. , interference alignment and interference cancellation , attempt to avoid or mitigate interference through spectrum scheduling . however , with rf energy harvesting , harmful interference can be turned into useful energy through a scheduling policy . in this context , how to mitigate interference as well as facilitate energy transfer , which may be conflicting , is the problem to be addressed . furthermore , the scheduling policy can be combined with power management schemes for further improvement in energy efficiency . in rf - ehns , rf energy becomes a valuable resource . the rf energy market can be established to economically manage this energy resource jointly with radio resource . for example , wireless charging service providers may act as rf energy suppliers to meet the energy demand from network nodes . the wireless energy service providers can decide on pricing and guarantee the quality of charging service . one of the efficient approaches in this dynamic market is to develop demand side management , which allows the service providers and network nodes to interact like in smart grid , to guarantee energy - efficiency and reliability . however , the issues related to the amount of rf energy and price at which they are willing to trade while optimizing the tradeoff between the revenue and cost must be investigated . network nodes , rf sources , and information gateway can be mobile . therefore , mobility becomes an important factor for rf energy harvesting and information transmission . the major issue is due to the fact that the energy harvesting and information transmission performances become time - varying , and resource allocation has to be dynamic and adaptive . a recent work @xcite investigates the impact of mobile rf source under two different mobility models , namely _ center - to - center mobility ( cm ) model _ and _ around edges moving ( em ) model _ with the focus on the energy gain at receivers . the tradeoff between transmit power and distance is explored , taking the energy loss during movement into account . it is found that cm yields better network performance in small networks with high node density . by contrast , em yields better performance in large networks with low node density . network coding @xcite is well - known to be energy efficient in information transmission . with network coding , senders are allowed to transmit information simultaneously . this property , especially in large - scale network , increases the amount of rf energy that can be harvested . during the time slots when relays or senders are not transmitting , they can harvest ambient rf signals . a pioneer study in @xcite analyzes the network lifetime gain for a two - way relay network with network coding . it is found that the lifetime of the network can be increased up to @xmath3 by enabling rf energy harvesting . from the perspective of network lifetime , more diverse network models and network coding schemes , such as physical - layer network coding and analogy network coding , are worth to be explored . additionally , the energy gain of network coding has been proved to be upper bounded by @xmath4 in the literature . intuitively , taking advantage of the broadcast nature of rf signals to reuse some of the dissipated energy can lead to energy saving . however , theoretically , whether rf energy harvesting will increase the upper bound of energy gain or not and how much exactly the bound will increase still require further investigation . it has long been recognized that intense rf exposure can cause heating of materials with finite conductivity , including biological tissues @xcite . the studies in @xcite focus on the effects of electromagnetic waves particularly from mobile phones and cellular networks . most of the measurements conclude that rf exposure from radio communication is safe . however , investigations in @xcite and @xcite show that some effects to genes are noticed when the rf power reaches the upper bound of international security levels . although there are many existing studies on the health risks of mobile phones , little effort has been made for investigation on health effect caused by a dedicated rf charger , which can release much higher power . thus , there is a need to address the safety concerns on deploying rf chargers . * due to the inverse - square law that the power density of rf waves decreases proportionally to the inverse of the square of the propagation distance , practical rf energy transfer and harvesting that complies to fcc regulations is limited to a local area . for example , the fcc allows operation up to 4w equivalent isotropically radiated power . however , as shown in @xcite , to realize 5.5@xmath5w energy transfer rate with a 4w power source , only the distance of 15 meters is possible . * other than transfer distance , rf energy harvesting rate is also largely affected by the direction and gain of the receive antenna(s ) . therefore , to improve the energy harvesting efficiency , devising a high gain antenna ( e.g. , based on materials and geometry ) for a wide range of frequency is an important research issue . * impedance mismatching occurs when the input resistance and reactance of the rectifier do not equal to that of the antenna . in this context , the antenna is not able to deliver all the harvested power to the rectifier . thus , impedance variations ( e.g. , introduced by on - body antennas ) can severely degrade the energy conversion efficiency . there is a need to develop circuit design techniques that automatically tune the parameters to minimize impedance mismatch . * the rf - to - dc conversion efficiency depends on the density of harvested rf power . improving the rf - to - dc conversion efficiency at low harvested power input is important . moreover , realizing a high - efficient low - power dc - to - dc converter , which converts a source of dc from a voltage level to another , would be another effort to achieve highly efficient rf energy harvesting . * rf energy harvesting components need to be small enough to be embedded in low - power devices . for example , the size of an rf - powered sensor should be smaller than or comparable to that of a battery - power sensor . as introduced above , an rf energy harvesting component may require an independent antenna , matching network and rectifier . the antenna size has a crucial impact on an energy harvesting rate . additionally , high voltage at the output of a rectifier requires very high impedance loads ( e.g. , 5m@xmath6 ) , which is a function of the length of the impedance . thus , it is challenging to reduce the size of embedded devices while maintaining high energy harvesting efficiency . * without line - of - sight for rf waves from an rf source to an energy harvester , the considerable energy transfer loss is expected . therefore , the rf energy source must be optimally placed to support multiple receivers to be charged . moreover , in a mobile environment , the mobility of receivers and energy sources can affect the rf energy transfer significantly . * the sensitivity of an information receiver is typically much higher than that of an rf energy harvester . consequently , a receiver located at a distance away from an rf transmitter may be able only to decode information and fail to extract energy from the rf signals . in this case , any swipt scheme can not be used efficiently . therefore , improving the sensitivity of rf energy harvesting circuit is crucial . * for rf - powered devices , as the transmit power is typical low , multiple antennas can be adopted to improve the transmission efficiency . however , larger power consumption comes along when the number of antennas increases . thus , there exists a tradeoff between the transmission efficiency and power consumption . the scheme to optimize this tradeoff needs to be developed . this issue becomes more complicated in a dynamic environment , e.g. , with varying energy harvesting rate . * as rf - powered devices typically have a strict operation power constraint , it is not practical to support high computation algorithms . any schemes , such as modulation and coding , receiver operation policy and routing protocol , to be adopted need to be energy - efficient and low - power . hence , power consumption is always a serious concern in rf - powered devices , which may require the re - design of existing schemes and algorithms for conventional networks . we have presented a comprehensive survey on rf energy harvesting networks ( rf - ehns ) . firstly , we have provided an overview of rf - ehns with the focus on architecture , enabling techniques and existing applications . then , we have reviewed the background in circuit design and state - of - the - art circuitry implementations . afterwards , we have surveyed various design issues related to resource allocation in rf - ehns , and the up - to - date solutions . finally , we have discussed on the future directions and practical challenges in rf energy harvesting techniques . h. j. visser and r. j. m. vullers , rf energy harvesting and transport for wireless sensor network applications : principles and requirements , " _ proceedings of the ieee _ , vol . 101 , no . 6 , 1410 - 1423 , june 2013 . x. zhang , h. jiang , l. zhang , c. zhang , z. wang , and x. chen , an energy - efficient asic for wireless body sensor networks in medical applications , " _ ieee transactions on biomedical circuits and systems _ , vol . 11 - 18 , feb . 2010 . a. kurs , a. karalis , r. moffatt , j. d. joannopoulos , p. fisher , and m. soljacic , `` wireless power transfer via strongly coupled magnetic resonances , '' _ science _ , vol . 5834 , pp . 83 - 86 , june 2007 . j. o. mur - miranda , w. franklin , g. fanti , y. feng , k. omanakuttan , r. ongie , a. setjoadi , and n. sharpe , wireless power transfer using weakly coupled magnetostatic resonators , " in _ proc . of ieee energy conversion congress and exposition ( ecce ) _ , atlanta , ga , sept . 2010 . n. shinohara , the wireless power transmission : inductive coupling , radio wave , and resonance coupling , " _ wiley interdisciplinary reviews : energy and environment _ , vol . 1 , no . 3 , pp . 337 - 346 , sept . 2012 . k. huang and v. k. n. lau , enabling wireless power transfer in cellular networks : architecture , modeling and deployment , " _ ieee transactions on wireless communications _ , vol 13 , no . 902 - 912 , feb . 2014 . x. chen , x. wang , and x. chen , energy - efficient optimization for wireless information and power transfer in large - scale mimo systems employing energy neamforming , " _ ieee wireless communications letters _ 2 , no . 6 , pp . 667 - 670 , dec . 2013 . grover and a. sahai , shannon meets tesla : wireless information and power transfer , " in _ proc . of ieee international symposium on information theory proceedings ( isit ) _ , pp . 2363 - 2367 , austin , tx , june 2010 . t. k. sarkar , j. zhong , k. kim , a. medouri , m. salazar - palma , a survey of various propagation models for mobile communication , " _ ieee antennas and propagation magazine _ , vol . 51 - 82 , june 2003 . m. erol - kantarci and h. t. mouftah , mission - aware placement of rf - based power transmitters in wireless sensor networks , " in _ proc . of ieee symposium on computers and communications ( iscc ) _ , pp . 12 - 17 , cappadocia , july 2012 . m. erol - kantarci and h.t . mouftah , drift : differentiated rf power transmission for wireless sensor network deployment in the smart grid , " in _ proc . of ieee globecom workshops _ 1491 - 1495 , anaheim , ca , dec . i. flint , x. lu , n. privault , d. niyato , and p. wang , performance analysis of ambient rf energy harvesting : a stochastic geometry approach , " in _ proc . of ieee global communications conference ( globecom ) _ , austin , tx , usa , december 2014 . t. le , k. mayaram , and t. fiez , efficient far - field radio frequency energy harvesting for passively powered sensor networks , " _ ieee journal of solid - state circuits _ , vol . 1287 - 1302 , may 2008 . m. stoopman , s. keyrouz , h. j. visser , k. philips , and w. a. serdijn , a self - calibrating rf energy harvester generating @xmath9 at @xmath10 , " in _ proc . of ieee symposium on vlsi circuits ( vlsic ) _ , pp . 226 - 227 , kyoto , june 2013 . m. stoopman , s. keyrouz , h. j. visser , k. philips , and w. a. serdijn , co - design of a cmos rectifier and small loop antenna for highly sensitive rf energy harvesters , " _ ieee journal of solid - state circuits _ , vol . 622 - 634 , march 2014 . powercast documentation [ online ] available : http://powercastco.com/ a. p. sample , a. n. parks , s. southwood , and j. r. smith , wireless ambient radio power , " _ wirelessly powered sensor networks and computational rfid _ , pp . 223 - 234 . springer new york , 2013 . g. papotto , f. carrara , a. finocchiaro , and g. palmisano , a @xmath11 cmos @xmath12 crystal - less rf - powered transceiver for wireless sensor network nodes , " _ ieee journal of solid - state circuits _ , vol . 335 - 346 , feb . 2014 . d. dondi , s. scorcioni , a. bertacchini , l. larcher , and p. pavan , an autonomous wireless sensor network device powered by a rf energy harvesting system , " in _ proc . of ieee annual conference on ieee industrial electronics society ( iecon ) _ , pp . 2557 - 2562 , oct . d. pavone , a. buonanno , m. durso , and f. g. d. corte , design considerations for radio frequency energy harvesting devices , " _ progress in electromagnetics research b _ , vol . 19 - 35 , october 2012 . f. zhang , y. zhang , j. silver , y. shakhsheer , m. nagaraju , a. klinefelter , j. pandey , j. boley , e. carlson , a. shrivastava , b. otis , and b. calhoun , a batteryless 19@xmath13 mics / ism - band energy harvesting body area sensor node soc , " in _ proc . of ieee international solid - state circuits conference _ , pp . 298 - 300 , san francisco , ca , feb . 2012 . a. al - khayari , h. al - khayari , s. al - nabhani , m. m. bait - suwailam , and z. nadir , design of an enhanced rf energy harvesting system for wireless sensors , " in _ proc . of ieee gcc conference and exhibition ( gcc ) _ , pp . 479 - 482 , nov . 2013 . m. al - lawati , m. al - busaidi , and z. nadir , rf energy harvesting system design for wireless sensors , " in _ proc . of ieee international multi - conference on systems , signals and devices ( ssd ) _ , pp . 1 - 4 , chemnitz , german , march 2012 . t. b. lim , n. m. lee , and b. k. poh , feasibility study on ambient rf energy harvesting for wireless sensor network , " in _ proc . of ieee mtt - s international microwave workshop series on rf and wireless technologies for biomedical and healthcare applications ( imws - bio ) _ , pp . 1 - 3 , singapore , dec . 2013 . k. kaushik , d. mishra , s. de , s. basagni , w. heinzelman , k. chowdhury , and s. jana , experimental demonstration of multi - hop rf energy transfer , " in _ proc . of ieee international symposium on personal indoor and mobile radio communications ( pimrc ) _ , pp . 538 - 542 , london , uk , sept . 2013 . j. p. olds and w. k. g. seah , design of an active radio frequency powered multi - hop wireless sensor network , " in _ proc . of ieee conference on industrial electronics and applications ( iciea ) _ , pp . 1721 - 1726 , singapore , july 2012 . w. k. g. seah and j. p. olds , data delivery scheme for wireless sensor network powered by rf energy harvesting , " in _ proc . of ieee wireless communications and networking conference ( wcnc ) _ , pp . 1498 - 1503 , shanghai , china , april 2013 . n. barroca , h. m. saraiva , p. t. gouveia , j. tavares , l. m. borges , f. j. velez , c. loss , r. salvado , p. pinho , r. goncalves , n. borgescarvalho , r. chavez - santiago , and i. balasingham , antennas and circuits for ambient rf energy harvesting in wireless body area networks , " in _ proc . of ieee international symposium on personal indoor and mobile radio communications ( pimrc ) _ , pp . 532 - 537 , london , uk , sept . 2013 . l. xia , j. cheng , n. e. glover , and p. chiang , 0.56 v , 20dbm rf - powered , multi - node wireless body area network system - on - a - chip with harvesting - efficiency tracking loop , " _ ieee journal of solid - state circuits _ , 49 , no . 6 , 1345 - 1355 , june 2014 . v. kuhn , f. seguin , c. lahuec , and c. person , a multi - tone rf energy harvester in body sensor area network context , " in _ proc . of ieee antennas and propagation conference ( lapc ) _ , pp . 238 - 241 , loughborough , uk , nov . 2013 . y. zhang , f. zhang , y. shakhsheer , j.d . silver , a. klinefelter , m. nagaraju , j. boley , j. pandey , a. shrivastava , e. carlson , a. wood , b. calhoun , and b.p . otis , a batteryless 19@xmath13 mics / ism - band energy harvesting body sensor node soc for exg applications , " _ ieee journal of solid - state circuits _ , vol . 199 - 213 , jan . 2013 . s. mandal , l. turicchia , and r. sarpeshkar , a battery - free tag for wireless monitoring of heart sounds , " in _ proc . of ieee international workshop on wearable and implantable body sensor networks _ , 201 - 206 , berkeley , ca , june 2009 . u. olgun , c. c. chen , and j. l. volakis , low - profile planar rectenna for batteryless rfid sensors , " in _ proc . of ieee antennas and propagation society international symposium ( apsursi ) _ , toronto , on , june 2010 . u. olgun , c. c. chen , and j. l. volakis , wireless power harvesting with planar rectennas for 2.45 ghz rfids , " in _ proc . of ieee international symposium on electromagnetic theory ( emts ) 329 - 331 , berlin , aug . s. s. b. hong , r. ibrahim , m. h. m. khir , h. daud , and m. a. zakariya , rectenna architecture based energy harvester for low power rfid application , " in _ proc . of ieee international conference on intelligent and advanced systems ( icias ) 382 - 387 , kuala lumpur , malaysia , june 2012 . p. kamalinejad , k. keikhosravy , s. mirabbasi , and v. c. m. leung , an efficiency enhancement technique for cmos rectifiers with low start - up voltage for uhf rfid tags , " in _ proc . of ieee international green computing conference ( igcc ) _ , arlington , va , june 2013 . m. r. shokrani , m. n. hamidon , m. khoddam , and v. najafi , a uhf micro - power cmos rectifier using a novel diode connected cmos transistor for micro - sensor and rfid applications , " in _ proc . of ieee international conference on electronics design , systems and applications ( icedsa ) _ , pp . 234 - 238 , kuala lumpur , malaysia , nov . s. scorcioni , l. larcher , and a. bertacchini , optimized cmos rf - dc converters for remote wireless powering of rfid applications , " in _ proc . of ieee international conference on rfid ( rfid ) _ , pp . 47 - 53 , orlando , fl , april 2012 . s. amini , and c. plett , design and analysis of very low voltage charge pumps for rfid tags , " in _ proc . of ieee microsystems and nanoelectronics research conference _ , pp . 9 - 12 , ottawa , canada , oct . s. shabana , c. thej , h. sankar , r. pandava , a. dutta , and s. g. singh , design of highly efficient charge pump for energy harvesting rfid applications , " in _ proc . of ieee asia pacific conference on postgraduate research in microelectronics and electronics ( primeasia ) _ , pp . 46 - 50 , dec . 2012 . d. d. donno , l. catarinucci , and l. tarricone , an uhf rfid energy - harvesting system enhanced by a dc - dc charge pump in silicon - on - insulator technology , " _ ieee microwave and wireless components letters _ , vol . 23 , no . 6 , pp . 315 - 317 , june 2013 . j. wilas , k. jirasereeamornkul , and p. kumhom , power harvester design for semi - passive uhf rfid tag using a tunable impedance transformation , " in _ proc . of ieee international symposium on communications and information technology _ , pp . 1441 - 1445 , icheon , sept . a. costanzo , m. fabiani , a. romani , d. masotti , and v. rizzoli , co - design of ultra - low power rf / microwave receivers and converters for rfid and energy harvesting applications , " in _ proc . of ieee international microwave symposium digest ( mtt ) _ , pp . 856 - 859 , anaheim , ca , may 2010 . a. chasin , v. volskiy , m. libois , m. ameys , m. nag , m. rockele , k. myny , s. steudel , s. schols , g. a. e. vandenbosch , w. d. raedt , j. genoe , g. gielen , and p. heremans , integrated uhf a - igzo energy harvester for passive rfid tags , " in _ proc . of ieee international electron devices meeting ( iedm ) _ , washington , dc , dec . 2013 . u. olgun , c. chen , and j. l. volakis , efficient ambient wifi energy harvesting technology and its applications , " in _ proc . of ieee antennas and propagation society international symposium ( apsursi ) _ , pp . 1 - 2 , chicago , il , july 2012 . m. arrawatia , m. s. baghini , and g. kumar , rf energy harvesting system from cell towers in @xmath14 band , " in _ proc . of ieee national conference on communications ( ncc ) _ , pp . 1 - 5 , bangalore , jan . 2011 . m. pinuela , p. d. mitcheson , and s. lucyszyn , ambient rf energy harvesting in urban and semi - urban environments , " _ ieee transactions on microwave theory and techniques _ , vol . 61 , no . 7 , pp . 2715 - 2726 , july 2013 . w. m. d. r. gunathilaka , g. g. c. m. gunasekara , h. g. c. p. dinesh , k. m. m. w. n. narampanawe , and j. v. wijayakulasooriya , ambient radio frequency energy harvesting , " in _ proc . of ieee international conference on industrial and information systems ( iciis ) _ , chennai , china , august 2012 . u. batool , a. rehman , n. khalil , m. islam , m. u. afzal , and t. tauqeer , energy extraction from rf / microwave signal , " in _ proc . of ieee international multi topic conference ( inmic ) _ , pp . 165 - 170 , islamabad , pakistan , dec . p. nintanavongsa , m. y. naderi , and k. r. chowdhury , a dual - band wireless energy transfer protocol for heterogeneous sensor networks powered by rf energy harvesting , " in _ proc . of ieee international computer science and engineering conference ( icsec ) _ , pp . 387 - 392 , nakorn pathom , thailand , sept . 2013 . c. mikeka , h. arai , a. georgiadis , and a. collado , dtv band micropower rf energy - harvesting circuit architecture and performance analysis , " in _ proc . of ieee int . rfid - technol . 561 - 567 , sitges , spain , sept . 2011 . h. nishimoto , y. kawahara , and t. asami , prototype implementation of wireless sensor network using tv broadcast rf energy harvesting , " in _ proc . of the 12th acm international conference adjunct papers on ubiquitous computing _ , pp . 373 - 374 , kona , hawaii , usa , nov . 2010 . r. vyas , h. nishimoto , m. tentzeris , y. kawahara , and t. asami , a battery - less , energy harvesting device for long range scavenging of wireless power from terrestrial tv broadcasts , " in _ proc . of ieee international microwave symposium digest ( mtt ) _ , pp . 1 - 3 , montreal , qc , canada , june 2012 . s. kitazawa , h. ban , and k. kobayashi , energy harvesting from ambient rf sources , " in _ proc . of ieee international microwave workshop series on innovative wireless power transmission : technologies , systems , and applications ( imws ) _ , pp . 39 - 42 , kyoto , japan , may 2012 . b. g. karthik , s. shivaraman , and v. aditya , wi - pie : energy harvesting in mobile electronic devices , " in _ proc . of ieee global humanitarian technology conference ( ghtc ) _ , pp . 398 - 401 , seattle , wa , nov . 2011 . h. jabbar , y. s. song , and t. t. jeong , rf energy harvesting system and circuits for charging of mobile devices , " _ ieee transactions on consumer electronics _ , vol . 247 - 253 , february 2010 . j. yi , w. h. ki , and c. y. tsui , analysis and design strategy of uhf micro - power cmos rectifiers for micro - sensor and rfid applications , " _ ieee transactions on circuits and systems i _ , vol . 54 , , no . 153 - 166 , jan . 2007 . a. shameli , a. safarian , a. rofougaran , m. rofougaran , and f. d. flaviis , power harvester design for passive uhf rfid tag using a voltage boosting technique , " _ ieee transactions on microwave theory and techniques _ , 55 , no . 6 , pp . 1089 - 1097 , june 2007 . y. yao , w. jie , s. yin , and f. f. dai , a fully integrated 900-mhz passive rfid transponder front end with novel zero - threshold rf dc rectifier , " _ ieee transactions on industrial electronics _ 56 , no . 7 , pp . 2317 - 2325 , july 2009 . g. andia vera , a. georgiadis , a. collado , and s. via , design of a 2.45 ghz rectenna for electromagnetic ( em ) energy scavenging , " in _ proc . of ieee radio wireless symp . new orleans , la , jan . 2010 , o. h. seunghyun , and d. d. wentzloff , a 32dbm sensitivity rf power harvester in @xmath15 cmos , " in _ proc . of ieee radio frequency integrated circuits symposium ( rfic ) _ , pp . 483 - 486 , montreal , qc , june 2012 . j. masuch , m. delgado - restituto , d. milosevic , and p. baltus , an rf - to - dc energy harvester for co - integration in a low - power 2.4 ghz transceiver frontend , " _ ieee international symposium on circuits and systems ( iscas ) _ , pp . 680 - 683 , seoul , south korea , may 2012 . s. scorcioni , a. bertacchini , and l. larcher , a @xmath16 cmos rf - dc power converter with @xmath17 input power sensitivity and efficiency higher than @xmath18 over @xmath19 input power range , " in _ proceedings of the ieee european conference solid - state circuits ( esscirc ) _ , pp . 109 - 112 , bordeaux , france , sept . 2012 . s. scorcioni , l. larcher , and a. bertacchini , optimized cmos rf - dc converters for remote wireless powering of rfid applications , " in _ proc . of ieee international conference on rfid _ , pp . 47 - 53 , orlando , fl , april 2012 . h. sun , y. guo , m. he , and z. zhong , design of a high - efficiency 2.45-ghz rectenna for low - input - power energy harvesting , " _ ieee antennas and wireless propagation letters _ , vol . 929 - 932 , aug . 2012 . d. karolak , t. taris , y. deval , j. b. begueret , and a. mariano , design comparison of low - power rectifiers dedicated to rf energy harvesting , " in _ proc . of ieee international conference on electronics , circuits and systems ( icecs ) _ , pp . 524 - 527 , seville , spain , dec . 2012 . m. roberg , t. reveyrand , i. ramos , e. a. falkenstein , and z. popovic , high - efficiency harmonically terminated diode and transistor rectifiers , " _ ieee transactions on microwave theory and techniques _ , vol . 4043 - 4052 , dec . 2012 . p. nintanavongsa , u. muncuk , d. r. lewis , and k. r. chowdhury , design optimization and implementation for rf energy harvesting circuits , " _ ieee journal on emerging and selected topics in circuits and systems _ , 24 - 33 , march 2012 . b. r. franciscatto , v. freitas , j. m. duchamp , c. defay , and t. p. vuong , high - efficiency rectifier circuit at 2.45 ghz for low - input - power rf energy harvesting , " in _ proc . of ieee european microwave conference ( eumc ) _ , pp . 507 - 510 , nuremberg , german , oct . 2013 . s. scorcioni , l. larcher , a. bertacchini , l. vincetti , and m. maini , an integrated rf energy harvester for uhf wireless powering applications , " in _ proc . of ieee wireless power transfer ( wpt ) 92 - 95 , perugia , may 2013 . x. wang , and a. mortazawi , high sensitivity rf energy harvesting from am broadcasting stations for civilian infrastructure degradation monitoring , " in _ proc . of ieee international wireless symposium ( iws ) _ , beijing , china , april 2013 . t. thierry , f. ludivine , o. laurent , and v. valerie , cots - based modules for far - field radio frequency energy harvesting at 900mhz and 2.4ghz , " in _ proc . of ieee international new circuits and systems conference ( newcas ) _ , paris , france , june 2013 . a. nimo , d. grgic , t. ungan , and l. m. reindl , a new family of passive wireless rf harvesters based on r - c - quartz oscillators , " in _ proc . of ieee european microwave conference ( eumc ) _ , pp . 511 - 514 , nuremberg , german , oct . 2013 . s. b. alam , m. s. ullah , and s. moury , design of a low power 2.45 ghz rf energy harvesting circuit for rectenna , " in _ proc . of ieee international conference on informatics , electronics @xmath8 vision ( iciev ) _ , dhaka , bangladesh , may 2013 . s. agrawal , s.k . pandey , j. singh , and m. s. parihar , realization of efficient rf energy harvesting circuits employing different matching technique , " in _ proc . of ieee international symposium on quality electronic design ( isqed ) _ , pp . 754 - 761 , santa clara , ca , march 2014 . a. aziz , a. mutalib , and r. othman , current developments of rf energy harvesting system for wireless sensor networks , " _ advances in information sciences and service sciences ( aiss ) _ , vol . 5 , no . 11 , 328 - 338 , june 2013 . x. shao , b. li , n. shahshahan , n. goldman , t. s. salter , and g. m. metze , a planner dual - band antenna design for rf energy harvesting applications , " in _ proc . of ieee international semiconductor device research symposium ( isdrs ) _ , college park , md , dec . j. m. barcak , and h. p. partal , efficient rf energy harvesting by using multiband microstrip antenna arrays with multistage rectifiers , " in _ proc . of ieee subthreshold microelectronics conference ( subvt ) _ , pp . 1 - 3 , waltham , ma , oct . 2012 . m. arsalan , m.h . ouda , l. marnat , t. j. ahmad , a. shamim , and k. n. salama , a 5.2ghz , 0.5mw rf powered wireless sensor with dual on - chip antennas for implantable intraocular pressure monitoring , " in _ proc . of ieee international microwave symposium digest ( ims ) _ , pp . 1 - 4 , seattle , wa , june 2013 . z. zakaria , n. a. zainuddin , m. z. a. abd aziz , m. n. husain , and m. a. mutalib , a parametric study on dual - band meander line monopole antenna for rf energy harvesting , " in _ proc . of ieee international conference on rfid - technologies and applications ( rfid - ta ) _ , johor bahru , malaysia , sept . 2013 . b. li , x. shao , n. shahshahan , and n. goldsman , t. salter , and g. m. metze , an antenna co - design dual band rf energy harvester , " _ ieee transactions on circuits and systems i _ , vol . 3256 - 3266 , dec . 2013 . z. zakaria , n. a. zainuddin , m. z. a. abd aziz , m. n. husain , and m. a. mutalib , dual - band monopole antenna for energy harvesting system , " in _ proc . of ieee symposium on wireless technology and applications ( iswta ) _ , kuching , malaysia , sept . b. l. pham , and a .- v . pham , triple bands antenna and high efficiency rectifier design for rf energy harvesting at 900 , 1900 and 2400 mhz , " in _ proc . of ieee mtt - s international microwave symposium digest ( ims ) _ , seattle , wa , june 2013 . d. masotti , a. costanzo , and s. adami , design and realization of a wearable multi - frequency rf energy harvesting system , " in _ proc . of ieee european conference on antennas and propagation ( eucap ) _ , pp . 517 - 520 , rome , italy , april 2011 . s. keyrouz , h. j. visser , and a. g. tijhuis , multi - band simultaneous radio frequency energy harvesting , " in _ proc . of ieee european conference on antennas and propagation ( eucap ) _ , pp . 3058 - 3061 , gothenburg , sweden , april 2013 . d. yi , and t. arslan , broadband differential antenna for full - wave rf energy scavenging system , " in _ proc . of ieee antennas and propagation conference ( lapc ) _ , pp . 325 - 328 , loughborough , uk , nov . 2013 . a. buonanno , m. durso , and d. pavone , an ultra wide - band system for rf energy harvesting , " in _ proc . of ieee european conference on antennas and propagation ( eucap ) _ , pp . 388 - 389 , rome , italy , april 2011 . a. nimo , d. grgic , and l. m. reindl , ambient electromagnetic wireless energy harvesting using multiband planar antenna , " in _ proc . of ieee international multi - conference on systems , signals and devices ( ssd ) _ , chemnitz , german , march 2012 . j. zhang , y. huang , and p. cao , harvesting rf energy with rectenna arrays , " in _ proc . of ieee european conference on antennas and propagation ( eucap ) _ , pp . 365 - 367 , prague , czech republic , march 2012 . j. zhang , y. huang , and p. cao , a wideband cross dipole rectenna for rf wireless harvesting , " in _ proc . of ieee european conference on antennas and propagation ( eucap ) _ , pp . 3063 - 3067 , gothenburg , sweden , april 2013 . n. a. zainuddin , z. zakaria , m. n. husain , b. m. derus , m. z. a. a. aziz , m. a. mutalib , and m. a. othman , design of wideband antenna for rf energy harvesting system , " in _ proc . of ieee international conference on instrumentation , communications , information technology , and biomedical engineering ( icici - bme ) _ , pp . 162 - 166 , bandung , indonesian , nov . 2013 . l. vincetti , m. maini , e. pinotti , l. larcher , s. scorcioni , a. bertacchini , d. grossi , and a. tacchini , broadband printed antenna for radiofrequency energy harvesting , " in _ proc . of ieee international conference on electromagnetics in advanced applications ( iceaa ) _ , pp . 814 - 816 , cape town , south africa , sept . 2012 . s. agrawal , s. pandey , j. singh , and p.n . kondekar , an efficient rf energy harvester with tuned matching circuit , " _ vlsi design and test , communications in computer and information science _ , vol . 138 - 145 , 2013 . j. a. hagerty , f. b. helmbrecht , w. h. mccalpin , r. zane , and z. b. popovic , recycling ambient microwave energy with broad - band rectenna arrays , " _ ieee trans . on microwave theory and techniques _ , vol . 1014 - 1024 , march 2004 . j. a. g. akkermans , m. c. van beurden , g. j. n. doodeman , and h. j. visser , analytical models for low - power rectenna design , " _ ieee antennas and wireless propagation letters _ , vol . 187 - 190 , june 2005 . y. wu , j. p. linnartz , h. gao , m. k. matters - kammerer , and p. baltus , modeling of rf energy scavenging for batteryless wireless sensors with low input power , " in _ proc . of ieee international symposium on personal indoor and mobile radio communications ( pimrc ) 527 - 531 , london , uk , sept . 2013 . x. zhou , r. zhang , and c. k. ho , wireless information and power transfer : architecture design and rate - energy tradeoff , " _ ieee transactions on communications _ , 4757 - 4767 , november , 2013 . x. zhou , r. zhang , and c. k. ho , wireless information and power transfer : architecture design and rate - energy tradeoff , " in _ proc . of ieee global communications conference ( globecom ) _ , pp . 3982 - 3987 , 2012 . r. morsi , d. s. michalopoulos , and r. schober , multi - user scheduling schemes for simultaneous wireless information and power transfer over heterogeneous fading channels . " ( available on - line at arxiv:1401.1943 ) w. wang , l. li , q. sun , and j. jin , power allocation in multiuser mimo systems for simultaneous wireless information and power transfer , " in _ proc . of ieee vehicular technology conference ( vtc fall ) _ , las vegas , nv , usa , sept . 2013 . d. w. k. ng , e. s. lo , and r. schober , energy - efficient resource allocation in multiuser ofdm systems with wireless information and power transfer , " in _ proc . of ieee wireless communications and networking conference ( wcnc ) _ , 3823 - 3828 , shanghai , china , april 2013 . d. niyato , p. wang , and d. i. kim , admission control policy for wireless networks with rf energy transfer , " in _ proc . of ieee international conference on communications ( icc ) _ , sydney , australia , june 2014 . d. t. hoang , d. niyato , and d. i. kim , cooperative bidding of data transmission and wireless energy transfer , " in _ proc . of ieee wireless communications and networking conference ( wcnc ) _ , istanbul , turkey , april 2014 . d. t. hoang , d. niyato , p. wang , and d. i. kim , optimal decentralized control policy for wireless communication systems with wireless energy transfer , " in _ proc . of ieee international conference on communications ( icc ) _ , sydney , australia , june 2014 . d. niyato , p. wang , t. h. pink , and w. saad cooperation in delay tolerant networks with wireless energy transfer : performance analysis and optimization . " submitted to _ ieee transactions on vehicular technology_. d. w. k. ng , e. s. lo , and r. schober , spectral efficient optimization in ofdm systems with wireless information and power transfer , " in _ proc . of ieee european signal processing conference ( eusipco ) _ , pp . 1 - 5 , marrakech , morocco , sept . d. w. k. ng , e. s. lo , and r. schober , energy - efficient power allocation in ofdm systems with wireless information and power transfer , " in _ proc . of ieee international conference on communications ( icc ) _ , pp . 4125 - 4130 , budapest , hungary , june 2013 . h. ju and r. zhang , a novel model switching scheme utilizing random beamforming for opportunistic energy harvesting . " _ ieee transactions on wireless communications _ , vol . 13 , no . 2150 - 2162 , april 2014 . d. w. k. ng , e. s. lo , and r. schober , wireless information and power transfer : energy efficiency optimization in ofdma systems , " _ ieee transactions on wireless communications _ , 6352 - 6370 , dec 2013 . q. shi , l. liu , w. xu , and r. zhang , joint transmit beamforming and receive power splitting for miso swipt systems , " _ ieee transactions on wireless communications _ 600 - 603 , april 2014 . s. leng , d. w. k. ng , and r. schober , power efficient and secure multiuser communication systems with wireless information and power transfer . " in _ proc . of ieee international conference on communications ( icc ) _ , sydney , australia , june 2014 . s. zhao , q. li , q. zhang , and j. qin , antenna selection for simultaneous wireless information and power transfer in mimo systems , " _ ieee communications letters _ , vol . 789 - 792 , may 2014 . h. zhang , k. song , y. huang , and l. yang , energy harvesting balancing technique for robust beamforming in multiuser miso swipt system , " in _ proc . of ieee international conference on wireless communications @xmath8 signal processing ( wcsp ) _ , hangzhou , china , oct . 2013 . j. xu , l. liu , and r. zhang , multiuser miso beamforming for simultaneous wireless information and power transfer . " to appear in _ ieee transactions on signal processing_. m. r. a. khandaker and k. k. wong , swipt in miso multicasting systems . " _ ieee wireless communications letters _ , vol . 277 - 280 , june 2014 . s. timotheou , i. krikidis , g. zheng , and b. ottersten , beamforming for miso interference channels with qos and rf energy transfer , " _ ieee transactions on wireless communications _ , vol . 5 , pp . 2646 - 2658 , may 2014 . d. li , c. shen , and z. qiu , two - way relay beamforming for sum - rate maximization and energy harvesting , " in _ proc . of ieee international conference on communications ( icc ) _ , pp . 3115 - 3120 , budapest , hungary , june 2013 . d. li , c. shen , and z. qiu , sum rate maximization and energy harvesting for two - way af relay systems with imperfect csi , " in _ proc . of ieee international conference on acoustics , speech and signal processing ( icassp ) _ , pp . 4958 - 4962 , vancouver , bc , may 2013 . j. park , and b. clerckx , joint wireless information and energy transfer in a two - user mimo interference channel , " _ ieee transactions on wireless communications _ , vol 12 , no . 8 , pp . 4210 - 4221 , august 2013 . d. w. k. ng and r. schober , resource allocation for coordinated multipoint networks with wireless information and power transfer , " in _ proc . of ieee global communications conference ( globecom ) _ , austin , tx , usa , december 2014 . d. w. k. ng , e. s. lo , and r. schober , robust beamforming for secure communication in systems with wireless information and power transfer , " to appear in _ ieee transactions on wireless communications _ , 2014 . b. zhu , j. ge , y. huang , y. yang , and m. lin , rank - two beamformed secure multicasting for wireless information and power transfer , " _ ieee signal processing letters _ , 199 - 203 , feb . 2014 . d. w. k. ng , r. schober , and h. alnuweiri , secure layered transmission in multicast systems with wireless information and power transfer . " in _ proc . of ieee international conference on communications ( icc ) _ , sydney , australia , june 2014 . s. x. wu , a. m .- c . so , and w .- k . ma , rank - two transmit beamformed alamouti space - time coding for physical - layer multicasting , " in _ proc . ieee international conference on acoustics , speech and signal processing ( icassp ) _ , pp . 2793 - 2796 , kyoto , japan , mar . 2012 . d. w. k. ng , e. s. lo , and r. schober , multi - objective resource allocation for secure communication in cognitive radio networks with wireless information and power transfer . " ( available online at arxiv:1403.0054 ) q. li , q. zhang , and j. qin , secure relay beamforming for simultaneous wireless information and power transfer in non - regenerative relay networks , " _ ieee transactions on vehicular technology _ , vol . 5 , pp . 2462 - 2467 , jun 2014 . x. chen , c. yuen , and z. zhang , wireless energy and information transfer tradeoff for limited feedback multi - antenna systems with energy beamforming , " _ ieee transactions on vehicular technology _ 407 - 412 , jan . 2014 . k. ishibashi and h. ochiai , analysis of instantaneous power distributions for non - regenerative and regenerative relaying signals , " _ ieee transactions on wireless communications _ , vol . 258 - 265 , january 2012 . i. krikidis , s. timotheou , and s. sasaki , rf energy transfer for cooperative networks : data relaying or energy harvesting ? " _ ieee communications letters _ , vol . 11 , pp . 1772 - 1775 , november 2012 . i. krikidis , s. sasaki , s. timotheou , and z. ding , a low complexity antenna switching for joint wireless information and energy transfer in mimo relay channels , " _ ieee transactions on communications _ 5 , pp . 1577 - 1587 , april 2014 . a. a. nasir , x. zhou , s. durrani , and r. a. kennedy , relaying protocols for wireless energy harvesting and information processing , " _ ieee transactions on wireless communications _ 12 , no . 7 , pp . 3622 - 3636 , july 2013 . a. a. nasir , x. zhou , s. durrani , and r. a. kennedy , wireless energy harvesting and information relaying : adaptive time - switching protocols and throughput analysis . " ( available on - line at arxiv:1310.7648 ) d. s. michalopoulos , h. a. suraweera , and r. schober , relay selection for simultaneous information transmission and wireless energy transfer : a tradeoff perspective . " ( available online at : arxiv:1303.1647 ) i. krikidis , simultaneous information and energy transfer in large - scale networks with / without relaying , " _ ieee transactions on communications _ , _ ieee transactions on communications _ , vol . 900 - 912 , march 2014 . z. ding , s. m. perlaza , i. esnaola , and h. v. poor , power allocation strategies in energy harvesting wireless cooperative networks . " _ ieee transactions on wireless communications _ , vol . 846 - 860 , february 2014 . b. gurakan , o. ozel , j. yang , and s. ulukus , two - way and multiple - access energy harvesting systems with energy cooperation , " in _ proc . of ieee conference record of the forty sixth asilomar conference on signals , systems and computers ( asilomar ) 58 - 62 , pacific grove , ca , nov . 2012 . b. k. chalise , w .- k . ma , y. d. zhang , h. a. suraweera , and m. g. amin , optimum performance boundaries of ostbc based af - mimo relay system with energy harvesting receiver , " _ ieee transactions on signal processing _ , vol . 17 , sept . 2013 . g. l. moritz , j. l. rebelatto , r. d. souza , b. f. ucha - filho , and y. li , time - switching uplink network - coded cooperative communication with downlink energy transfer . " ( available on - line at arxiv:1312.2544 ) j. l. rebelatto , b. f. uchoa - filho , y. li , and b. vucetic , multiuser cooperative diversity through network coding based on classical coding theory , " _ ieee transactions on signal processing _ , vol . 60 , no . 916 - 926 , february 2012 . s. park , j. heo , b. kim , w. chung , h. wang , and d. hong , optimal mode selection for cognitive radio sensor networks with rf energy harvesting , " in _ proc . of ieee international symposium on personal indoor and mobile radio communications ( pimrc ) _ , pp . 2155 - 2159 , sydney , australia , sept . 2012 . n. barroca , j. m. ferro , l. m. borges , j. tavares , and f. j. velez , electromagnetic energy harvesting for wireless body area networks with cognitive radio capabilities . " ( available on - line at : http://www.anacom.pt/streaming/norbertobarroca@xmath7congressoursi2012 .pdf?contentid=1148324@xmath8field@xmath20attached@xmath7file ) j. kim and j .- w . lee , energy adaptive mac protocol for wireless sensor networks with rf energy transfer , " in _ proc . of ieee third international conference on ubiquitous and future networks ( icufn ) _ , pp . 89 - 94 , dalian , china , june 2011 . j. kim and j .- w . lee , performance analysis of the energy adaptive mac protocol for wireless sensor networks with rf energy transfer , " in _ proc . of ieee international conference on ict convergence ( ictc ) 14 - 19 , seoul , korea , sept . 2011 p. nintanavongsa , m. y. naderi , and k. r. chowdhury , medium access control protocol design for sensors powered by wireless energy transfer , " in _ proceedings of ieee infocom - mini conference _ , pp . 150 - 154 , turin , italy , april 2013 . r. doost , k. r. chowdhury , and m. di felice , routing and link layer protocol design for sensor networks with wireless energy transfer , " in _ proceedings of global telecommunications conference _ , miami , fl , december 2010 . l. zi , p. yang , w. zhang , and d. qiao , j - roc : a joint routing and charging scheme to prolong sensor network lifetime , " in _ proc . of ieee international conference on network protocols ( icnp ) 373 - 382 , vancouver , bc , oct . 2011 . b. tong , z. li , g. wang , and w. zhang , how wireless power charging technology affects sensor network deployment and routing , " in _ proc . of ieee international conference on distributed computing systems ( icdcs ) _ , pp . 438 - 447 , genoa , italy , june 2010 . a. h. coarasa , p. nintanavongsa , s. sanyal , and k. r. chowdhury , impact of mobile transmitter sources on radio frequency wireless energy harvesting , " in _ proc . of ieee international conference on computing , networking and communications ( icnc ) _ , pp . 573 - 577 , san diego , ca , jan . 2013 . p. v. mekikis , a. s. lalos , a. antonopoulos , l. alonso , and c. verikoukis , wireless energy harvesting in two - way network coded cooperative communications : a stochastic approach for large scale networks , " _ ieee communications letters _ , vol . 18 , no . 6 , pp . 1011 - 1014 , june 2014 . p. demestichas , a. georgakopoulos , d. karvounas , k. tsagkaris , v. stavroulaki , j. lu , c. xiong , and j. yao , 5 g on the horizon : key challenges for the radio - access network , " _ ieee vehicular technology magazine _ , vol . 47 - 53 , sept . 2013 . v. jungnickel , k. manolakis , w. zirwas , b. panzner , v. braun , m. lossow , m. sternad , r. apelfrojd , and t. svensson , the role of small cells , coordinated multipoint , and massive mimo in 5 g , " _ ieee communications magazine _ , vol . 44 - 51 , may 2014 . scientific committee on emerging and newly identified health risks ( scenihr ) : possible effects of electromagnetic fields on human health . docs/[email protected] ] . j. breckenkamp , e. m. g. berg - beckhoff , j. schuz , b. schlehofer , j. wahrendorft , and m. blettner , feasibility of a cohort study on health risks caused by occupational exposure to radiofrequency electromagnetic fields , " _ biomed central environmental health _ , vol . 8 , no . 23 , may 2009 . r. w. habash , j. m. elwood , d. krewski , w. g. lotz , j. p. mcnamee , and f. s. prato , recent advances in research on radiofrequency fields and health , " _ journal of toxicology and environmental health , part b _ , vol.12 , no . 4 , pp . 250288 , apr . 2009 .

two liquid state theories , the self - consistent ornstein - zernike equation ( scoza ) and the hierarchical reference theory ( hrt ) are shown , by comparison with monte carlo simulations , to perform extremely well in predicting the liquid - vapour coexistence of the hard core yukawa ( hcy ) fluid when the interaction is long range . the long range of the potential is treated in the simulations using both an ewald sum and hyperspherical boundary conditions . in addition , we present an analytical optimised mean field theory which is exact in the limit of an infinitely long range interaction . the work extends a previous one by caccamo _ et al _ [ _ phys . rev . e , _ * 60 * , 5533 ( 1999 ) ] for short range interactions . * keywords * : yukawa potential , critical phenomena , monte carlo simulations , scoza , hrt .

0704.3889

two liquid state theories are presently available which provide an accurate description of the liquid - vapour transition of simple systems including the critical region . a first one is the so - called self - consistent ornstein - zernike approximation ( scoza ) devised initially by hye and stell @xcite and later adapted for numerical calculations @xcite . it is based on the assumption that the direct correlation function behaves , to a good approximation , at long range ( attractive region of the potential ) as the potential with a density and temperature dependent prefactor which is determined by imposing consistency between the energy and compressibility routes to the thermodynamics . the second approach put forward by parola and reatto @xcite , incorporates renormalisation group ideas by gradually taking into account fluctuations of longer and longer wavelengths . it is based on the hierarchical reference theory ( hrt ) of fluids truncated at lowest order by means of an orpa - like ( optimised random phase approximation ) ansatz ( see ref . @xcite for review ) . both predict non - classical critical exponents @xcite . thermodynamic properties and phase diagrams based on the scoza and hrt approaches have been reported for a variety of potentials ( limiting ourselves to continuum systems ) including the hard core yukawa ( hcy ) potential @xcite , the square well potential @xcite , the lennard - jones ( lj ) potential @xcite , the gaussian potential @xcite , ultrasoft potentials @xcite , asakura - oosawa pair potentials @xcite , girifalco potentials @xcite , binary hard core yukawa mixtures @xcite , etc .. most of the comparisons with simulation data are so far restricted to short or medium range potentials . the aim of this article is to extend such comparisons to long range interactions . a convenient choice to do this is the hcy potential as the range can be varied from short range appropriate for modelling colloidal suspensions or protein solutions to medium range , where it mimics the familiar lj potential @xcite , to long range . for long range interactions we have considered a third approach , an optimised mean field theory ( omf ) @xcite , exact in the kac limit @xcite , which can be worked out analytically for the hcy potential ( see section 6 ) . a further advantage of the hcy potential is that the repulsive part of the potential which in scoza , hrt and omf is used as a reference potential is a hard sphere potential whose properties are accurately known @xcite . last , an approximate analytical solution ( mean spherical approximation ) is available for the hcy potential @xcite which drastically reduces the computational efforts of the solution of the scoza equations . for short inverse screening lengths @xmath0 of the yukawa potential , @xmath1 ( @xmath2 hard sphere diameter ) , a comparison between scoza , hrt and simulation results for the liquid - vapour coexistence curve is available in @xcite . in this work we extend the domain of @xmath0 from @xmath3 ( where the hcy potential approximates the lj potential ) to @xmath4 , the infinite long range potential . with respect to simulations , in this domain the range of the yukawa potential generally exceeds the dimensions of the simulation box ( for typical system sizes of a thousand particles ) and some care is necessary to properly treat the long range of the potential . this has been done both by using periodic boundary conditions ( b.c . ) and performing an ewald ( ew ) sum @xcite and by using hyperspherical b.c . @xcite . the remainder of the paper is organised as follows : in section [ mod ] we define the hcy interaction . section [ sim ] provides a description of the two monte carlo ( mc ) methods used to treat the long range of the yukawa interaction , ewald sum and hyperspherical method . after summarising the three considered theories , scoza , hrt and omf in section [ th ] we compare , in section [ results ] , the numerical results obtained with the different theories for the liquid - vapour coexistence curve and the thermodynamic properties along the coexistence boundary with the simulation results . the conclusions are presented in section [ conclusion ] . an appendix describes some properties of yukawa charge distributions relevant for elaborating the omf theory . in our system particles interact by the hard - core yukawa potential @xmath5 in the limit @xmath6 the potential gets infinitely long range and infinitely weak ( kac limit ) . we define an inverse reduced temperature @xmath7 , where @xmath8 is the boltzmann constant , @xmath9 the temperature and @xmath10 has dimension of an electric charge . the reduced density is @xmath11 . numerical simulations of fluids or plasmas involving yukawa interactions @xmath12 can be performed either in a cube of side @xmath13 with periodic b.c . ( @xmath14 ) or on the surface of a four dimensional ( 4d ) sphere of centre @xmath15 and radius @xmath16 ( equation @xmath17 ) , the hypersphere @xmath18 for short . in both geometries the helmholtz equation @xmath19 can be solved analytically . in the case of @xmath14 the solution of the helmholtz equation can be reexpressed in terms of ewald sums with good convergence properties@xcite . the use of an ewald potential is of course only necessary when the range of the yukawa potential exceeds the linear dimensions of the simulation box . molecular dynamics simulations of simple models of yukawa plasmas were performed recently with this method@xcite . for the potential model eq . ( [ potential ] ) the ewald sum for the energy is given by @xcite @xmath20 \big /2 |{\bf r}_{ij}+ l{\bf n}| \nonumber \\ & & - \frac{2\pi}{v } q^2 \alpha^{*2 } \sum_{{\bf k } = 0}^\infty \displaystyle \frac{e^{- ( { \bf k}^2 + \alpha^2 ) / 4 \eta^2}}{{\bf k}^2 + \alpha^2 } f({\bf k})f^*({\bf k } ) + u_{self } \end{aligned}\ ] ] where @xmath21 \big /2 l |{\bf n}| \nonumber \\ & & - \frac{2\pi}{v } q^2 \alpha^{*2 } \sum_{{\bf k } = 0}^\infty \displaystyle \frac{e^{- ( { \bf k}^2 + \alpha^2 ) / 4 \eta^2}}{{\bf k}^2 + \alpha^2 } \nonumber \\ & & + q^2 \alpha^{*2 } \frac{\eta}{\sqrt \pi } { e^{- \alpha^2 / 4 \eta^2 } } - q^2 \alpha^{*2 } \ ; \frac{\alpha}{2 } { \rm erfc } \big ( \displaystyle \frac{\alpha}{2 \eta } \big ) . \end{aligned}\ ] ] and @xmath22,\end{aligned}\ ] ] in eq . ( [ i2 ] ) @xmath23 , @xmath13 is the box length , @xmath24 the volume and @xmath25 denotes the complementary error function . the prime in the sum over @xmath26 , with @xmath27,@xmath28 integers , restricts it to @xmath29 for @xmath30 . the parameter @xmath31 governs the convergence of the real - space and reciprocal contributions to the energy . with @xmath32 , adopted in our calculations , only terms with @xmath30 need to be retained in eq . ( [ i2 ] ) . the sum in reciprocal space extends over all lattice vectors @xmath33 with @xmath34 . in the @xmath18 geometry the green s function of the helmholtz equation for the yukawa fluid is obtained analytically @xcite : @xmath35 where @xmath36 . the pair potential @xmath37 in @xmath18 is isotropic and depends only on the geodesic length @xmath38 . the geodesic distance between two points @xmath39 and @xmath40 of @xmath18 is given by @xmath41 the configurational energy of a system of @xmath42 yukawa hard spheres in @xmath18 is therefore given by @xmath43 in equation ( [ confi ] ) @xmath44 denotes the hard sphere potential in @xmath18 , i.e. @xmath45 the presence of the constant @xmath46 in the r.h.s . of equation([confi ] ) accounts for the fact that the energy is defined up to an additive constant . the choice of @xmath46 is a recurrent problem of simulations in @xmath18 ( see e.g. the discussion in reference@xcite ) . here we define @xmath46 by requiring that the self energies in @xmath18 and the usual euclidean space @xmath47 coincide , i.e. , @xmath48 which gives @xmath49 recent monte carlo simulations of yukawa plasmas have been performed with this method@xcite . here we apply it to a liquid of attractive yukawa spheres . the self - consistent ornstein - zernike approximation in its most general formulation is a liquid - state theory that introduces in the relation between the direct correlation function @xmath50 and the pair potential @xmath51 one or more state - dependent parameters that are determined by imposing consistency between the different routes to thermodynamics leading to a partial differential equation ( pde ) for the unknown parameter(s ) . the scoza originally grew out of the semi - analytic solution of the mean spherical approximation ( msa ) for hcy potentials @xcite and it is this scoza formulation that is considered in the present work . it is based on the ornstein - zernike equation supplemented with an msa - type closure relation @xmath52 here @xmath53 is the pair distribution function , @xmath54 is the direct correlation function of the hard - core reference system given e.g. in the waisman parameterisation @xcite and @xmath55 is the state - dependent parameter , that is not given in advance but is determined so that thermodynamic consistency is obtained between the energy and compressibility routes ( @xmath56 ) . the consistency requirement leads to a pde of the form @xmath57 where @xmath58 is the reduced ( with respect to the ideal gas ) dimensionless isothermal compressibility given by the compressibility route and @xmath59 is the excess ( over ideal gas ) internal energy per volume provided by the energy route , i.e. @xmath60 . once @xmath61 is expressed in terms of the excess internal energy the pde eq.([consist ] ) can be transformed into a pde for @xmath59 @xmath62 for the general case of an arbitrary attractive tail of the pair interaction the determination of @xmath63 must be done fully numerically @xcite and requires an enormous amount of computational cost . however , this procedure is not necessary for the fluid considered here : for certain pair potentials , like the hard - core yukawa potential or hard - core potentials with linear combinations of yukawa and exponential tails - so called sogami - ise fluids , the coefficient @xmath63 can be obtained semi - analytically due to the availability of the semi - analytic msa solution for these potentials . details of the determination of @xmath63 can be found in ref . @xcite , @xcite and @xcite . while the former formulation of scoza in ref . @xcite is based on the laplace transformation route for solving the underlying msa for hard - core yukawa fluids , we have used in the present work the reformulation described in refs . @xcite and @xcite which is based on the wertheim - baxter factorisation method for the solution of the msa . while both solution methods are equivalent for the potential considered here , the wertheim - baxter factorisation method is more flexible and allows a broader applicability of the scoza @xcite . the pde is a quasilinear diffusion equation that has been solved numerically via an implicit finite - difference algorithm in the region @xmath64\times[0,1]$ ] with suitable initial and boundary conditions ( see ref . @xcite for more details ) . we have chosen a density and temperature grid spacing of @xmath65 and @xmath66 . up to now scoza has been applied to a number of discrete @xcite and continuum systems @xcite and the results showed - when compared with computer simulations - that the theory yields very accurate predictions for the thermodynamic and structural properties throughout the temperature - density plane , even near the phase coexistence and in the critical region . indeed , scoza is one of very few liquid state theories that do not show serious problems in the critical region and even exhibit some form of scaling with non - classical , partly ising - like critical exponents @xcite . in hrt the pair potential @xmath51 is divided into a repulsive reference potential part @xmath67 and a longer - ranged attractive part @xmath68 . in the present case the reference system is the hard - sphere fluid , whose thermodynamics and correlations are accurately described by the carnahan - starling equation of state @xcite and the verlet - weis parameterisation of the two - body radial distribution function @xcite . the hrt differs from the conventional liquid state approaches in the way the attractive perturbation is dealt with : in order to take into account the effect of fluctuations , @xmath68 is introduced gradually via a sequence of @xmath69 systems whose interaction potential @xmath70 results from the sum of @xmath71 and a perturbation @xmath72 . the latter term is defined by its fourier transform @xmath73 which coincides with @xmath74 for @xmath75 , and is identically vanishing for @xmath76 . as @xmath69 evolves from @xmath77 to @xmath78 , the interaction @xmath79 goes from the reference part @xmath67 to the full potential @xmath51 . in other words the parameter @xmath69 plays the role of an infrared cutoff whose effect consists in depressing fluctuations on length scales larger than @xmath80 : the critical fluctuations are recovered only in the limit @xmath81 . the corresponding evolution of thermodynamics and correlations of increasing order can be determined via perturbation theory and is described by an exact hierarchy of integro - differential equations . as mentioned above close to a critical point and at large length scales , this hierarchy becomes equivalent to a formulation of the momentum - space renormalisation group @xcite . unlike current liquid - state theories hrt preserves the correct convexity of the free energy also below the critical temperature , producing flat isotherms in the coexistence region which correspond to infinite compressibility and constant chemical potential . the first equation of the hierarchy gives the evolution of the helmholtz free energy @xmath82 of the partially interacting system in terms of its two - body direct correlation function in momentum space @xmath83 and the full perturbation @xmath74 : @xmath84 where @xmath85 . the quantities @xmath86 , @xmath87 are linked to @xmath82 and @xmath83 by the relations @xmath88 -\frac{1}{2 } \ , \rho\!\int\!\!\frac{d^{3}{\bf k}}{(2\pi)^{3 } } \left[\phi(k)-\phi_{q}(k)\right ] \label{amod } \\ & & { \cal c}_{q}(k ) = c_{q}(k)+\phi(k)-\phi_{q}(k ) \ , , \label{cmod}\end{aligned}\ ] ] where @xmath89 is the volume and @xmath90 the number density . these modified free energy and direct correlation function have been introduced in order to remove the discontinuities which appear in @xmath82 and @xmath83 at @xmath78 and @xmath91 respectively as a consequence of the discontinuity of @xmath92 at @xmath91 . indeed , they represent the free energy and direct correlation function of the fully interacting fluid as given by a treatment such that the fourier components of the interaction with wavelengths larger than @xmath80 are not really disregarded , but approximately taken into account by mean - field theory . in the @xmath81 limit the modified quantities coincide with the physical ones , once the fluctuations have been fully included . for @xmath77 the definitions in eqs . ( [ amod ] ) and ( [ cmod ] ) reduce , instead , to the well - known mean - field expressions of the free energy and direct correlation function . a simple approximation scheme consists in truncating the hierarchy at the first equation , supplementing it with a suitable closure relation based on some ansatz for @xmath87 . as in previous applications , our form of @xmath87 has been inspired by liquid - state theories , @xmath93 where @xmath94 is the fourier transform of the direct correlation function of the hard - sphere fluid , and @xmath95 , @xmath96 are _ a priori _ unknown functions of the thermodynamic state and of q. the function @xmath96 is determined by the core condition , i.e. , the requirement that the radial distribution function @xmath97 is vanishing for every @xmath69 whenever the interparticle separation is less than the hard - sphere diameter @xmath2 . via eq . ( [ closure ] ) we can write the core condition in terms of an integral equation for @xmath98 . @xmath95 is adjusted so that @xmath87 satisfies the compressibility rule . for each q - system this constraint gives the reduced compressibility of the fluid as the structure factor evaluated at zero wavevector , and can be expressed in terms of the modified quantities @xmath86 , @xmath87 as @xmath99 this equation can be regarded as a consistency condition between the compressibility route and a route in which the thermodynamics is determined from the helmholtz free energy as obtained from eq.([hrteq ] ) . here the compressibility rule ( [ sum ] ) plays a fundamental role . in fact , when @xmath95 in eq.([closure ] ) is determined via eq.([sum ] ) and the resulting expression for @xmath87 is used in eq.([hrteq ] ) , one obtains a partial differential equation for @xmath86 which reads @xmath100 \ , , \label{hrtclos}\ ] ] where we have set @xmath101 , @xmath102 and @xmath103 \varphi(k ) \ , . \label{psi}\ ] ] eq.([hrtclos ] ) is integrated numerically from @xmath77 down to @xmath78 . at each integration step , @xmath96 is determined by the core condition @xmath104 , @xmath105 . this condition acts as an auxiliary equation for determining @xmath106 via eq.([psi ] ) . for this specific work we considered the density interval @xmath107 $ ] to numerically solve the differential equation for the free energy . the spacing of the density grid is @xmath108 , consequently the error in the determination of the coexistence densities is of the same order of magnitude . concerning the boundary conditions : for low density the system behaves as an ideal gas while at @xmath109 we assumed the validity of the standard orpa approximation . two remarks are worth mentioning : first , eq . ( [ closure ] ) assumes that the fluid direct correlation function depends linearly on the perturbation @xmath110 . this ansatz is especially appropriate when the perturbation range is much longer than that of the reference part of the interaction , i.e. longer than the particle size . the second remark relates to the implementation of the core condition : the inverse fourier transform of the function @xmath96 has been expanded in a series of legendre polynomials for @xmath111 and the series has been truncated after a finite number of terms ( typically five ) . the evolution equations for the expansion coefficients were then determined by differentiating with respect to @xmath69 the integral equation for @xmath98 , i.e. the core condition , and subsequently projecting it on the polynomials used in the expansion . however , the resulting equations are coupled to the evolution equation [ hrteq ] for the free energy , and this makes them difficult to handle . therefore , as described in detail in [ 21 , 22 ] , in the derivative of @xmath98 with respect to q the long - wavelength contributions containing the isothermal compressibility of the q - system were disregarded . physically this amounts to a decoupling of the short- and the long - range evolutions of the correlations . this approximation , as well as the previous one , are fully justified for long - range perturbations , where the short- and the long - range parts of the correlations are mainly affected by the reference and the perturbation terms respectively , but become more problematic for short - range interactions where the interplay between excluded - volume and cohesion effects is much stronger . however , this effect does not concern the present study . we consider , quite generally , a fluid of identical hard spheres of diameter @xmath2 with additional isotropic pair interaction @xmath112 . since @xmath51 is an arbitrary function of @xmath113 for @xmath114 , one can assume that @xmath51 has been regularised in the core in such a way that its fourier transform @xmath115 is a well behaved function of @xmath116 and that @xmath117 is a finite quantity . at equilibrium at inverse temperature @xmath118 and chemical potential @xmath119 ( grand canonical ( gc ) ensemble ) the pressure @xmath120 of the fluid is a convex function of @xmath121 ( at fixed @xmath118 ) @xcite even before the thermodynamic limit has been taken @xcite . as a consequence @xmath122 is continuous , its derivatives with respect to @xmath123 exist and it admits a legendre transform @xmath124 with respect to @xmath123 defined as @xmath125 the gc free energy @xmath124 is then a convex function of the density and its legendre transform with respect to @xmath90 coincides with @xmath122 . it was shown in ref.@xcite that in the case of an attractive interaction ( i.e. @xmath126 for all @xmath10 ) we have the inequality @xmath127 here the subscript mean field ( mf ) emphasises that @xmath128 is the van der waals free energy@xcite . note that @xmath129 $ ] is _ a priori _ non convex , notably in the two - phase region . > from here on we specialise to the yukawa interaction @xmath130 with @xmath131 . since @xmath132 , equation ( [ boundf ] ) does not appear to be very useful . however , it follows from appendix a that we can replace @xmath133 by the regular function @xmath134 where @xmath135 is the interaction energy of two spherical distributions @xmath136 of yukawa charges of effective charge @xmath137 . the interaction is regularised in the core but remains unchanged for @xmath138 . since @xmath139 , eq . ( [ boundf ] ) is still valid and will give a non trivial upper bound for the free energy . similar to what has been done in reference@xcite for the coulomb interaction we seek a best upper bound for @xmath140 by minimising the quantity @xmath141 to this end we consider variations of @xmath142 with respect to the charge distribution @xmath143 defined by eq . ( [ tau ] ) and work out the stationary condition ( @xmath144 ) @xmath145 first , it follows from equation ( [ q ] ) that @xmath146 second , as a consequence of the convolution relations @xmath147 and @xmath148 one has @xmath149 and thus @xmath150 finally , since @xmath151 , we have @xmath152 collecting results ( [ eq1 ] ) , ( [ eq2 ] ) and ( [ eq3 ] ) one gets @xmath153 note that we can impose @xmath154 since , if @xmath143 is solution of the stationary condition ( [ statio ] ) , then @xmath155 , where @xmath156 is an arbitrary constant , is also a solution ( @xmath137 is then multiplied by @xmath157 but @xmath158 is left unchanged ) . therefore eq.([statio ] ) can be recast in the form @xmath159 of course , eq . ( [ resuv ] ) is valid only for @xmath160 . the first contribution in the r.h.s of eq . ( [ resuv ] ) is clearly identified with the potential created by a uniform density of yukawa charges of density @xmath90 while the second contribution stems from a charge discontinuity at @xmath161 . the solution is therefore of the form @xmath162 , where @xmath163 is a constant determined by the condition @xmath154 , yielding @xmath164 ( @xmath165 packing fraction ) . furthermore , from eq . ( [ q ] ) it follows that @xmath166 so that the potential @xmath167 is seen to be the superposition of two potentials created by spherical surface and volume distributions , the expressions of which are given in appendix a by eqs . ( [ surf ] ) and ( [ vol ] ) , respectively . one finds @xmath168 one can check that @xmath167 indeed verifies eq.([resuv ] ) . finally , the optimised upper bound for the free energy can then be shown to be equal to @xmath169 an optimised mean field ( omf ) theory for the fluid can now be devised by considering the landau function@xcite @xmath170 at given @xmath118 and @xmath123 , the density @xmath90 is the one which minimises the landau function @xmath171 . in the limit @xmath172 , @xmath173 and one thus revovers the free energy of the kac model for @xmath174 . a comparison between the three theoretical approaches and simulation data for the liquid - vapour coexistence curve is made in figs . [ figure1 ] -[figure4 ] for @xmath175 and 0.1 . values for the densities , internal energies , chemical potentials and helmholtz free energies along the coexistence curve are presented in tables [ table1]- [ table4 ] . these data may be valuable for reference to future approaches . the simulation results have been obtained with the gibbs ensemble monte carlo ( gemc ) method @xcite using mostly a total of 1000 particles ( ew b.c . ) and 2000 particles ( @xmath18 b.c . ) . the number of cycles generated after equilibration varied from 7 - 40 @xmath176 according to temperature , each cycle comprising with equal probabilities , displacement of the @xmath42 particles , volume change , or exchange of 500 ( ew b.c . ) or 1000 particles ( @xmath18 b.c . ) . error bars ( two standard deviations ) were calculated by averaging densities , energies or histograms over blocks of 1 million cycles . expectedly , uncertainties are largest near the critical temperature where fluctuations of density of the coexisting phases are important and hence convergence slow . finite size effects additionally affect the critical behaviour , but also influenced to some extent the determination of the coexistence density on the liquid side at least for the smallest value @xmath177 . thus , for the @xmath18 b.c . , an increase of the number of particles from 1000 to 2000 typically lowered @xmath178 by about @xmath179 . the opposite trend was observed with the ew b.c . where @xmath178 increased by @xmath180 by an increase of @xmath42 from 1000 to 1872 ( cf . table [ table4 ] ) . it is likely that , for @xmath177 , small finite size effects are still present in both sets of results . the agreement between hrt , scoza and mc simulations turns out to be extremely good over the range of @xmath0 values considered . small differences occur only at the highest and lowest values of @xmath0 . at @xmath3 the critical temperature of hrt is slightly below that predicted by scoza . previous work @xcite showed , however , that by further increasing the value of @xmath0 the hrt critical temperature eventually gets higher than that of scoza . at @xmath177 the coexistence curves , including the critical regions , of scoza and omf are very close ( cf . [ figure4 ] ) ; as the omf theory is expected to be nearly exact at this small value of @xmath0 it is very likely that scoza becomes exact also in the kac limit , though no formal proof seems to be available . this is further confirmed by the close agreement between critical parameters of scoza and omf at @xmath181 ( see table [ table5 ] ) . the coexistence curve of hrt is found to be slightly narrower than that of scoza though in the near critical region the agreement between the two theories is quite good . as pointed out in ref . @xcite the hrt flow can not be solved conveniently in the kac limit . inspection of fig . [ figure3 ] shows that the omf theory rapidly deteriorates with increase of @xmath0 . figure [ figure5 ] compares hrt , scoza , omf and mc simulation results for the internal energy along the coexistence curve at @xmath182 . the agreement is similar to that for the densities . the variation with @xmath0 of the critical temperature and density for scoza and hrt , including data of ref . @xcite for @xmath183 , is given in table [ table5 ] . no attempt has been made to determine critical data in the simulations by fitting data away from the critical point by some power law expression . because of the finite size effects in the critical region uncertainties of the results would be large and of no use to validate the theoretical predictions . in the gemc method the order parameter @xmath184 is believed to have mf behaviour ( critical exponent @xmath185 ) in the critical region once the correlation length exceeds the linear dimension @xmath13 of the simulation box . in fact we find that for @xmath186 , @xmath187 varies linearly over the whole temperature region considered yielding an effective exponent @xmath188 ) . such an analysis was not conclusive for @xmath3 . we have compared the predictions for the liquid - vapour coexistence curve of a long - range yukawa fluid obtained from advanced theoretical approaches with gibbs ensemble monte carlo simulations . concerning the simulations care has to be taken when the range of the potential exceeds the length of the simulation box . this was done by performing an ewald sum in the case of a cubic simulation box and by using hyperspherical boundary conditions . the theoretical approaches that we applied comprised the self - consistent ornstein - zernike approximation ( scoza ) , the hierarchical reference theory ( hrt ) , as well as an optimised mean field theory ( omf ) . while the omf yields the exact result in the limit of infinite range of the potential , it deteriorates with decreasing interaction range . on the other hand , hrt and scoza turn out to be in perfect agreement with simulation results over the whole interaction range considered . this study complements a recent comparison @xcite for the case of intermediate and short range yukawa fluid . this work was supported by a grant through the programme dactions intgres amadeus under project nos . 06648pb and fr 09/2007 , by the hochschuljubilumsstiftung der stadt wien under project number 1757/2007 and by the marie curie european network mrtn - ct-2003 - 504712 . federica lo verso thanks davide pini , alberto parola and luciano reatto for helpful discussions . 99 j. s. hye , g. stell . _ j. chem . phys . , _ * 67 * , 439 ( 1977 ) . j. s. hye , g. stell . phys . , _ * 52 * , 1071 ( 1984 ) . r. dickman , g. stell . lett . , _ * 77 * , 996 ( 1996 ) . d. pini , g. stell , r. dickman . e , _ * 57 * , 2862 ( 1998 ) . d. pini , g. stell , j.s . j. thermophys . , _ * 19 * , 1029 ( 1998 ) . a. parola , l. reatto , _ phys . rev . a , _ * 31 * , 2417 ( 1985 ) . a. parola , l. reatto . phys . , _ * 44 * , 211 ( 1995 ) . j. s. hye , d. pini , g. stell _ physica a , _ * 279 * , 213 ( 2000 ) . d. pini , g. stell , n. b. wilding . phys . , _ * 95 * , 483 ( 1998 ) . c. caccamo , g. pellicane , d. costa , d. pini , g. stell _ phys e , _ * 60 * , 5533 ( 1999 ) . d. pini , ge jialin , a. parola , l. reatto . lett . , _ * 327 * , 209 ( 2000 ) . d. pini , a. parola , l. reatto . _ j. phys . : matter , _ * 18 * , s2305 ( 2006 ) . a. reiner , g. kahl . _ j. chem . _ * 117 * , 4925 ( 2002 ) . e. schll - paschinger , a. l. benavides , r. castaeda - priego . _ j. chem . _ * 123 * , 234513 ( 2005 ) . a. meroni , a. parola , l. reatto . rev . a , _ * 42 * , 6104 ( 1990 ) . m. tau , a. parola , d. pini , l. reatto . e , _ * 52 * , 2644 ( 1995 ) . e. schll - paschinger , a. reiner . _ j. chem . _ * 125 * , 164503 ( 2006 ) . j. s. hye , a. reiner . _ j. chem . , _ * 125 * , 104503 ( 2006 ) . b. m. mladek , g. kahl , m. neumann . _ j. chem . _ * 124 * , 064503 ( 2006 ) . f. lo verso , m. tau , l. reatto . _ j. phys . : condens.matter_ * 15 * , 1505 ( 2003 ) . f. lo verso , d. pini , l. reatto . _ j. phys . : condens . matter _ * 17 * , 771 ( 2005 ) . f. lo verso , r. l. c. vink , d. pini , l. reatto . e _ * 73 * , 61407 ( 2006 ) . e. schll - paschinger , g. kahl . _ europhys . letters , _ * 63 * , 538 ( 2003 ) . d. costa , g. pellicane , c. caccamo , e. schll - paschinger , g. kahl _ phys . * 68 * , 021104 ( 2003 ) . e. schll - paschinger , g. kahl . _ j. chem . phys . , _ * 118 * , 7414 ( 2003 ) . e. schll - paschinger , d. levesque , j .- j . weis , g. kahl . _ j. chem . phys . , _ * 122 * , 024507 ( 2005 ) . e. schll - paschinger , g. kahl . _ j. chem . _ * 123 * , 134508 ( 2005 ) . d. pini , m. tau , a. parola , l. reatto . e , _ * 67 * , 046116 ( 2003 ) . _ * 101 * , 1617 ( 2003 ) . m. kac , g.e . uhlenbeck , p.c . hemmer , _ j. math , phys . , _ * 4 * , 216 ( 1963 ) . n. f. carnahan , k.e . starling _ _ j . phys.,__*51 * , 635 ( 1969 ) . l. verlet , j. j. weis . rev . a,__*5 * , 939 ( 1972 ) . d. henderson e. w. grundke . _ j. chem . , _ * 63 * , 601 ( 1975 ) . e. waisman . _ mol . , _ * 25 * , 45 ( 1973 ) . l. blum , j.s . _ j. stat . _ * 19 * , 317 ( 1978 ) . g. salin , j .- m . caillol . _ j. chem . phys . , _ * 113 * , 10459 ( 2000 ) . caillol , d. gilles . _ j. stat . , _ * 100 * , 905 ( 2000 ) . caillol , d. gilles . _ j. stat . , _ * 100 * , 933 ( 2000 ) . g. salin , j .- m . caillol . , _ * 88 * , 65002 - 1 - 4 ( 2002 ) . g. salin , j .- m . caillol . _ phys . of plasmas , _ * 10 * , 1220 ( 2003 ) . caillol , d. gilles . _ j. phys . a : math . gen . , _ * 36 * , 6343 ( 2003 ) . g. kahl , e. schll - paschinger , g. stell . _ j. phys . condens . matter , _ * 14 * , 9153 ( 2002 ) . e. schll - paschinger . _ j. chem . _ * 120 * , 11698(2004 ) . hansen , i.r . _ theory of simple liquids _ academic press , london ( 1986 ) . see , for instance , k. g. wilson , j. b. kogut _ phys . c _ 12 , 75 ( 1974 ) . see , for instance , m. e. fisher _ critical phenomena _ , lecture notes in physics , vol . * 186 * , f. j. w. hahne ( ed . ) , springer , berlin , ( 1982 ) . f. nicoll , t. s. chang . lett . , _ * 62a * , 287 ( 1977 ) . n. goldenfeld . _ lectures on phase transitions and the renormalization group _ addison - wesley ( 1992 ) . _ j. phys . a : math . gen . , _ * 35 * , 4189 ( 2002 ) . _ j. stat . _ * 115 * , 1461 ( 2004 ) . a.z . panagiotopoulos . _ phys . , _ * 61 * , 813 ( 1987 ) . d. frenkel , b. smit . _ understanding molecular simulation _ academic press , london ( 1996 ) . _ * 104 * , 1931 ( 2006 ) . y. rosenfeld . 1993 , _ j. chem . phys . , _ * 98 * , 8126 ( 1993 ) . y. rosenfeld . e , _ * 47 * , 2676 ( 1993 ) . y. rosenfeld . phys . , _ * 88 * , 1357 ( 1996 ) . y. rosenfeld , g. chabrier . _ j. stat . _ * 98 * , 283 ( 1997 ) . coexistence densities , internal energy @xmath189 , excess chemical potential @xmath119 and helmholtz free energy @xmath190 at coexistence for @xmath191 . @xmath192 is the carnahan - starling hs free energy @xcite . l l l l l l l l @xmath193 & & @xmath194 & @xmath195 & @xmath196 & @xmath197 & @xmath198 & @xmath199 & @xmath200 + & scoza&0.174 & 0.464 & -1.055 & -2.650 & -2.737 & & + 0.635 & hrt & 0.181 & 0.458 & -1.106 & -2.607 & & 0.963 & 2.521 + & mc - ew&0.175(10)&0.430(15 ) & -1.06(4 ) & -2.43(3 ) & -2.74(1 ) & & + + & scoza&0.161 & 0.485 & -0.982 & -2.770 & -2.765 & & + 0.63 & hrt & 0.166 & 0.477 & -1.030 & -2.735 & & 0.890 & 2.657 + & mc - ew&0.171(8)&0.470(15 ) & -1.05(2 ) & -2.67(2 ) & -2.76(1 ) & & + + & scoza&0.138 & 0.515 & -0.862 & -2.996 & -2.825 & & + 0.62 & hrt & 0.142 & 0.509 & -0.901 & -2.967 & & 0.774 & 2.900 + & mc - ew&0.149(5)&0.501(4 ) & -0.94(2 ) & -2.92(2 ) & -2.83(1 ) & & + + & scoza&0.105 & 0.565 & -0.686 & -3.407 & -2.953 & & + 0.60 & hrt & 0.108 & 0.560 & -0.724 & -3.386 & & 0.609 & 3.333 + & mc - ew&0.109(2)&0.553(4 ) & -0.73(1 ) & -3.34(2 ) & -2.96(1 ) & & + + & scoza&0.082 & 0.606 & -0.556 & -3.803 & -3.094 & & + 0.58 & hrt & 0.083 & 0.602 & -0.578 & -3.787 & & 0.486 & 3.741 + & mc - ew&0.083(2)&0.594(4 ) & -0.58(1 ) & -3.74(2 ) & -3.10(1 ) & & + + & scoza&0.063 & 0.643 & -0.452 & -4.204 & -3.251 & & + 0.56 & hrt & 0.065 & 0.640 & -0.490 & -4.192 & & 0.396 & 4.155 + & mc - ew&0.063(2)&0.631(4 ) & -0.46(1 ) & -4.13(3 ) & -3.25(3 ) & & + + & scoza&0.049 & 0.676 & -0.367 & -4.614 & -3.424 & & + 0.54 & hrt & 0.050 & 0.674 & -0.390 & -4.606 & & 0.317 & 4.573 + & mc - ew&0.050(2)&0.670(4 ) & -0.39(1 ) & -4.57(2 ) & -3.47(8 ) & & + + & scoza&0.0284 & 0.738 & -0.235 & -5.503 & -3.834 & & + 0.50 & hrt & 0.029 & 0.736 & -0.258 & -5.500 & & 0.201 & 5.469 + & mc - ew&0.029(2)&0.733(4 ) & -0.25(1 ) & -5.47(3 ) & -3.76(10 ) & & + + @xmath201@xmath202 + @xmath203@xmath204 [ table1 ] table 2 . same as table 1 but for @xmath205 + @c l l l l l l l l @xmath193 & & @xmath194 & @xmath195 & @xmath196 & @xmath197 & @xmath198 & @xmath199 & @xmath200 + & scoza&0.174 & 0.396 & -0.987 & -2.209 & -2.773 & & + 0.905 & hrt & 0.176 & 0.393 & -1.000 & -2.195 & & 0.955 & 2.169 + & mc - ew&0.183(2)&0.395(10 ) & -1.025(50 ) & -2.19(5 ) & -2.77(1 ) & & + + & scoza&0.135 & 0.450 & -0.788 & -2.583 & -2.864 & & + 0.88 & hrt & 0.136 & 0.448 & -0.767 & -2.477 & & 0.758 & 2.557 + & mc - ew&0.135(2)&0.439(5 ) & -0.78(2 ) & -2.52(2 ) & -2.87(1 ) & & + + & scoza&0.112 & 0.484 & -0.676 & -2.852 & -2.944 & & + 0.86 & hrt & 0.114 & 0.482 & -0.687 & -2.842 & & 0.650 & 2.825 + & mc - ew&0.112(2)&0.474(4 ) & -0.67(1 ) & -2.79(2 ) & -2.95(1 ) & & + + & scoza&0.095 & 0.514 & -0.586 & -3.107 & -3.028 & & + 0.84 & hrt & 0.096 & 0.512 & -0.593 & -3.098 & & 0.561 & 3.083 + & mc - ew&0.096(2)&0.508(5 ) & -0.59(2 ) & -3.07(3 ) & -3.04(2 ) & & + + & scoza&0.0806 & 0.541 & -0.511 & -3.360 & -3.119 & & + 0.82 & hrt & 0.081 & 0.540 & -0.513 & -3.353 & & 0.485 & 3.341 + & mc - ew&0.085(3)&0.543(6 ) & -0.54(2 ) & -3.37(3 ) & -3.11(2 ) & & + + & scoza&0.0685 & 0.567 & -0.446 & -3.615 & -3.216 & & + 0.80 & hrt & 0.069 & 0.566 & -0.451 & -3.610 & & 0.424 & 3.600 + & mc - ew&0.070(2)&0.564(4 ) & -0.447(15 ) & -3.60(3 ) & -3.21(1 ) & & + + & scoza&0.058 & 0.591 & -0.389 & -3.873 & -3.321 & & + 0.78 & hrt & 0.059 & 0.590 & -0.401 & -3.868 & & 0.372 & 3.859 + & mc - ew&0.062(2)&0.594(6 ) & -0.420(15 ) & -3.89(3 ) & -3.30(2 ) & & + + & scoza&0.049 & 0.614 & -0.339 & -4.140 & -3.434 & & + 0.76 & hrt & 0.050 & 0.613 & -0.353 & -4.133 & & 0.324 & 4.124 + & mc - ew&0.050(2)&0.612(3 ) & -0.344(5 ) & -4.120(15 ) & -3.43(1 ) & & + + & scoza&0.041 & 0.636 & -0.294 & -4.413 & -3.557 & & + 0.74 & hrt & 0.042 & 0.635 & -0.304 & -4.407 & & 0.280 & 4.398 + & mc - ew&0.042(2)&0.634(4 ) & -0.30(1 ) & -4.440(25 ) & -3.55(1 ) & & + + @xmath201@xmath202 + @xmath203@xmath204 [ table2 ] table 3 . same as table 1 but for @xmath206 + @c l l l l l l l l @xmath193 & & @xmath194 & @xmath195 & @xmath196 & @xmath197 & @xmath198 & @xmath199 & @xmath200 + & scoza&0.164 & 0.367 & -0.923 & -2.063 & -2.824 & & + 1.045 & hrt & 0.163 & 0.371 & -0.916 & -2.086 & & 0.908 & 2.082 + & mc - ew&0.169(8)&0.355(8 ) & -0.93(3 ) & -1.99(2 ) & -2.82(1 ) & & + & mc - s3&0.182(5)&0.384(3 ) & -1.02(3 ) & -2.15(3 ) & & & + + & scoza&0.157 & 0.377 & -0.885 & -2.131 & -2.839 & & + 1.04 & hrt & 0.156 & 0.381 & -0.881 & -2.153 & & 0.873 & 2.149 + & mc - ew&0.160(8)&0.370(8 ) & -0.89(4 ) & -2.05(4 ) & -2.84(1 ) & & + & mc - s3&0.165(6)&0.384(6 ) & -0.93(3 ) & -2.172(22 ) & & & + + & scoza&0.133 & 0.412 & -0.764 & -2.377 & -2.899 & & + 1.02 & hrt & 0.132 & 0.415 & -0.760 & -2.394 & & 0.752 & 2.390 + & mc - ew&0.130(6)&0.400(6 ) & -0.75(2 ) & -2.31(4 ) & -2.90(1 ) & & + & mc - s3&0.133(2)&0.415(3 ) & -0.77(1 ) & -2.39(2 ) & & & + + & scoza&0.114 & 0.442 & -0.670 & -2.602 & -2.962 & & + 1.00 & hrt & 0.114 & 0.445 & -0.670 & -2.620 & & 0.662 & 2.617 + & mc - ew&0.118(4)&0.439(4 ) & -0.689(15 ) & -2.58(2 ) & -2.96(1 ) & & + & mc - s3&0.111(1)&0.441(2 ) & -0.655(50 ) & -2.60(1 ) & & & + + & scoza&0.099 & 0.469 & -0.594 & -2.817 & -3.030 & & + 0.98 & hrt & 0.099 & 0.470 & -0.594 & -2.826 & & 0.587 & 2.824 + & mc - ew&0.103(4)&0.467(4 ) & -0.62(1 ) & -2.74(2 ) & -3.02(1 ) & & + & mc - s3&0.095(1)&0.466(2 ) & -0.573(54 ) & -2.800(15 ) & & & + + & scoza&0.086 & 0.493 & -0.528 & -3.028 & -3.101 & & + 0.96 & hrt & 0.086 & 0.494 & -0.527 & -3.035 & & 0.520 & 3.032 + & mc - ew&0.085(4)&0.485(4 ) & -0.52(1 ) & -2.99(1 ) & -3.11(1 ) & & + & mc - s3&0.0846(11)&0.494(3 ) & -0.521(7 ) & -3.037(15 ) & & & + + & scoza&0.075 & 0.516 & -0.471 & -3.239 & -3.178 & & + 0.94 & hrt & 0.076 & 0.516 & -0.476 & -3.240 & & 0.469 & 3.238 + & mc - ew&0.076(2)&0.512(2 ) & -0.468(5 ) & -3.21(1 ) & -3.17(2 ) & & + & mc - s3&0.072(1)&0.509(2 ) & -0.442(4 ) & -3.20(1 ) & & & + + & scoza&0.0656 & 0.538 & -0.420 & -3.451 & -3.259 & & + 0.92 & hrt & 0.066 & 0.537 & -0.422 & -3.447 & & 0.417 & 3.446 + & mc - ew&0.068(2)&0.539(3 ) & -0.43(1 ) & -3.46(2 ) & -3.27(3 ) & & + & mc - s3&0.0678(13)&0.547(3 ) & -0.437(8 ) & -3.52(2 ) & & & + + & scoza&0.0572 & 0.559 & -0.374 & -3.667 & -3.346 & & + 0.90 & hrt & 0.058 & 0.557 & -0.380 & -3.658 & & 0.374 & 3.656 + & mc - ew&0.058(2)&0.556(3 ) & -0.375(10 ) & -3.64(2 ) & -3.34(1 ) & & + & mc - s3&0.058(1)&0.566(3 ) & -0.382(7 ) & -3.72(2 ) & & & + + @xmath201@xmath202 + @xmath203@xmath204 [ table3 ] table 4 . same as table 1 but for @xmath207 + @c l l l l l l l l @xmath193 & & @xmath194 & @xmath195 & @xmath196 & @xmath197 & @xmath198 & @xmath199 & @xmath200 + & scoza&0.158 & 0.357 & -0.900 & -2.030 & -2.863 & & + 1.10 & hrt & 0.161 & 0.353 & -0.916 & -2.010 & & -0.916 & -2.010 + & mc - ew&0.17(1)&0.340(15 ) & -0.96(2 ) & -1.96(5 ) & -2.86(2 ) & & + & mc - s3&0.167(5)&0.361(5 ) & -0.93(3 ) & -2.05(2 ) & & & + + & scoza&0.145 & 0.375 & -0.834 & -2.155 & -2.891 & & + & hrt & 0.148 & 0.371 & -0.850 & -2.132 & & -0.850 & -2.131 + 1.09 & mc - ew&0.145(10)&0.360(15 ) & -0.82(1 ) & -2.05(1 ) & -2.89(1 ) & & + & mc - ew1&0.145(4)&0.362(3 ) & -0.82(2 ) & -2.07(2 ) & -2.895(5 ) & & + & mc - s3&0.160(6)&0.392(3 ) & -0.92(2 ) & -2.25(2 ) & & & + + & scoza&0.134 & 0.391 & -0.778 & -2.270 & -2.918 & & + 1.08 & hrt & 0.138 & 0.387 & -0.800 & -2.244 & & -0.800 & -2.244 + & mc - ew&0.136(5)&0.380(5 ) & -0.79(2 ) & -2.21(2 ) & -2.92(2 ) & & + & mc - s3&0.1384(24)&0.395(3 ) & -0.8018(15 ) & -2.29(2 ) & & & + + & scoza&0.124 & 0.406 & -0.730 & -2.379 & -2.947 & & + & hrt & 0.128 & 0.401 & -0.749 & -2.347 & & -0.749 & -2.347 + 1.07 & mc - ew&0.126(2)&0.393(3 ) & -0.73(1 ) & -2.32(2 ) & -2.95(1 ) & & + & mc - ew1&0.127(2)&0.399(3 ) & -0.74(1 ) & -2.34(2 ) & -2.95(1 ) & & + & mc - s3&0.1260(15)&0.4103(21 ) & -0.737(8 ) & -2.402(13 ) & & & + + & scoza&0.116 & 0.420 & -0.686 & -2.484 & -2.977 & & + 1.06 & hrt & 0.120 & 0.415 & -0.708 & -2.452 & & -0.708 & -2.452 + & mc - ew&0.121(4)&0.416(4 ) & -0.716(15 ) & -2.46(2 ) & -2.97(1 ) & & + & mc - s3&0.118(2)&0.423(3 ) & -0.697(11 ) & -2.544(15 ) & & & + + & scoza&0.1082 & 0.434 & -0.647 & -2.587 & -3.007 & & + & hrt & 0.112 & 0.427 & -0.667 & -2.547 & & -0.667 & -2.547 + 1.05 & mc - ew&0.109(2)&0.425(3 ) & -0.65(1 ) & -2.54(2 ) & -3.000(15 ) & & + & mc - ew1&0.113(2)&0.433(3 ) & -0.68(1 ) & -2.59(2 ) & -3.00(1 ) & & + & mc - s3&0.111(2)&0.442(2 ) & -0.660(7 ) & -2.63(2 ) & & & + + & scoza&0.101 & 0.446 & -0.610 & -2.689 & -3.038 & & + 1.04 & hrt & 0.107 & 0.439 & -0.638 & -2.644 & & -0.638 & -2.644 + & mc - ew&0.102(1)&0.439(2 ) & -0.641(5 ) & -2.64(1 ) & -3.04(1 ) & & + & mc - s3&0.1047(13)&0.456(3 ) & -0.630(9 ) & -2.75(2 ) & & & + + & scoza&0.089 & 0.470 & -0.546 & -2.887 & -3.103 & & + 1.02 & hrt & 0.093 & 0.462 & -0.571 & -2.837 & & -0.570 & -2.837 + & mc - ew&0.091(1)&0.466(2 ) & -0.555(5 ) & -2.86(1 ) & -3.10(1 ) & & + & mc - s3&0.091(1)&0.479(2 ) & -0.558(6 ) & -2.94(1 ) & & & + + & scoza&0.078 & 0.493 & -0.489 & -3.085 & -3.173 & & + & hrt & 0.083 & 0.483 & -0.519 & -3.026 & & -0.519 & -3.026 + 1.00 & mc - ew&0.077(3)&0.483(5 ) & -0.48(1 ) & -3.028(25 ) & -3.17(1 ) & & + & mc - ew1&0.078(2)&0.487(5 ) & -0.484(14 ) & -3.05(3 ) & -3.185(15 ) & & + & mc - s3&0.079(1)&0.499(2 ) & -0.492(6 ) & -3.126(13 ) & & & + + & scoza&0.0572 & 0.559 & -0.374 & -3.667 & -3.346 & & + 0.98 & hrt & 0.074 & 0.503 & -0.472 & -3.215 & & -0.472 & -3.215 + & mc - ew&0.057(1)&0.539(3 ) & -0.373(6 ) & -3.55(2 ) & -3.35(3 ) & & + + @xmath201@xmath202 + @xmath203@xmath204 [ table4 ] table 5 . critical parameters . results based on the carnahan - starling hard sphere free energy @xcite for the reference system . + @c l l l @xmath208 & & @xmath209 & @xmath210 + 0.0 & omf&1.13194 & 0.24913 + + @xmath211 & scoza&1.1319 & 0.2491 + + 0.1 & scoza&1.1294 & 0.2495 + & hrt & 1.129 & 0.250 + & omf & 1.13108 & 0.24915 + + 0.5 & scoza&1.0718 & 0.2586 + & hrt & 1.073 & 0.260 + + 1.0 & scoza&0.9264 & 0.2793 + & hrt & 0.925 & 0.279 + + 1.8 & scoza&0.6527 & 0.3145 + & hrt & 0.650 & 0.314 + + @xmath212 & scoza&0.173 & 0.3895 + & hrt & 0.175 & 0.394 + + @xmath213 & scoza&0.0187 & 0.4575 + & hrt & 0.019 & 0.424 + + @xmath201 from ref . . due to a different definition , the temperatures of ref . must be multiplied by the factor @xmath214 . [ table5 ] [ figure1 ] [ figure2 ] [ figure3 ] [ figure4 ] [ figure5 ] the electrostatics of yukawa charge distributions is similar to , but not identical with the electrostatics of usual coulomb charge distributions . in this appendix we extend previous results obtained for special forms of spherical distributions of yukawa charges @xcite to a general spherical distribution @xmath215 which satisfies : @xmath216 we denote by @xmath133 the yukawa potential created by a unit point charge . we have @xmath131 and for its fourier transform @xmath217 . the yukawa potential @xmath218 created by this distribution at point @xmath219 is given by the convolution of @xmath220 and @xmath221 @xmath222 where @xmath223 denotes the infinitesimal spherical angle about vector @xmath224 . the integral @xmath225 reads @xcite @xmath226 the potential @xmath227 for @xmath228 is not particularly useful but outside the sphere ( i.e. for @xmath229 one infers from equations ( [ v ] ) and ( [ i ] ) the remarkable result as in the case of the coulomb potential , the potential outside the spherical distribution of charge @xmath133 is still a yukawa potential with the same screening parameter @xmath0 but with a different charge @xmath137 . since @xmath231 the effective charge @xmath137 is larger than the bare charge @xmath232 of the distribution . of course in the limit @xmath6 one has @xmath233 and gauss law is recovered . we now prove the theorem @xmath234 for @xmath235 it is obvious ; for @xmath236 we note that from the expression ( [ q ] ) of @xmath137 and from equation([v ] ) it follows that @xmath237 where @xmath238 . since @xmath239 is a decreasing function of @xmath113 and @xmath240 we have @xmath241 from which inequality ( [ inequav ] ) follows ( note that @xmath143 must be positive ) . for a sufficiently regular distribution @xmath143 , the potential @xmath227 is a bounded smooth function for @xmath242 , in particular @xmath243 is finite in general ( see examples at the end of the appendix ) . also in this case gauss theorem generalises easily . indeed , for the case where the two distributions do not overlap , i.e. @xmath250 , @xmath251 can be replaced by @xmath252 in equation ( [ w ] ) and thus @xmath253 note that the fourier transform @xmath254 is positive definite contrary to @xmath255 . in addition , we have the theorem @xmath256 the proof is trivial : @xmath257 the function @xmath135 is in general bounded in the core and in particular @xmath258 is finite . this charge smearing process provides an easy way to regularise the yukawa potential @xmath133 in the core which preserves the positivity of the interaction . finally , we give explicit expressions for @xmath137 , @xmath167 for @xmath259 , and the value @xmath258 in simple cases relevant for sect . thus for a surface distribution @xmath260 one finds @xmath261 for a volume distribution @xmath262 ( @xmath263 step function ) one has @xmath264

billey , jockusch , and stanley characterized-avoiding permutations by a property of their reduced decompositions . this paper generalizes that result with a detailed study of permutations via their reduced decompositions and the notion of pattern containment . these techniques are used to prove a new characterization of vexillary permutations in terms of their principal dual order ideals in a particular poset . additionally , the combined frameworks yield several new results about the commutation classes of a permutation . in particular , these describe structural aspects of the corresponding graph of the classes and the zonotopal tilings of a polygon defined by elnitsky that is associated with the permutation .

math0506242

reduced decompositions of permutations are classical objects in combinatorics that appear throughout the literature . following the work of rodica simion and frank schmidt in @xcite , the study of permutation patterns , particularly pattern avoidance , has become a frequently studied field as well . in @xcite , sara billey , william jockusch , and richard stanley relate these two concepts , possibly for the first time . there they show that @xmath0-avoiding permutations are exactly those permutations where the subsequence @xmath1 never occurs in a reduced decomposition . relatedly , victor reiner shows in @xcite that the number of @xmath2 occurrences in reduced decompositions of the longest element in the symmetric group , which has the maximal number of occurrences of @xmath0 , is equal to the number of such reduced decompositions . stanley had previously shown that this is the number of standard young tableaux of a staircase shape in @xcite . inspired by these results , and more generally by the relationship they suggest between the two aspects of permutations , this paper studies elements of the symmetric group from the combined perspectives of their reduced decompositions and their patterns . while these aspects of a permutation appear extensively in combinatorial literature , they are not often treated together . this paper strives to remedy that fact , addressing several questions where reduced decompositions and permutation patterns together lead to interesting results . after introducing basic terminology and notation in section [ section defs ] , section [ section vexillary ] generalizes the result of billey , jockusch , and stanley , via a new characterization of vexillary permutations in theorem [ vexthm ] . this characterization is based on the reduced decompositions of the permutations _ containing _ the permutation in question , and is strikingly different from all previous equivalent characterizations . in addition to requiring that each of the permutations containing the vexillary permutation has a certain kind of reduced decomposition , the proof of theorem [ vexthm ] explicitly constructs such a reduced decomposition . there are three algorithms which appear in this paper , the first of which occurs in the proof of theorem [ vexthm ] . it should be noted that these are not deterministic , and include a certain amount of choice . for instance , example [ vex ex ] describes only one possible route that the algorithm may take on a particular input . there is an equivalence relation , sometimes known as the commutation relation , on the set of reduced decompositions of a particular permutation . this and an associated graph are discussed in section [ section eq ] . theorem [ 1 lbm ] and corollary [ 1 ulbm ] characterize permutations with graphs and commutation classes having certain properties . these results are strengthened in theorem [ chainthm ] . the results in sections [ section polygon ] and [ section poset ] discuss permutation patterns with respect to a polygon defined by serge elnitsky in @xcite . the rhombic tilings of this polygon are in bijection with the commutation classes of a permutation . new results include that the number of commutation classes of a permutation is monotonically increasing with respect to pattern containment ( theorem [ monotone ] ) , and several results pertaining to a poset associated with tilings of the polygon . finally , section [ section fb ] completely describes this poset in the case of a freely braided permutation , as defined by richard green and jozsef losonczy in @xcite and @xcite . the main definitions and notation that appear throughout the paper are discussed below . for more information about these objects , including proofs of elementary facts , see @xcite and @xcite . let @xmath3 denote the symmetric group on @xmath4 elements . an element @xmath5 permutes @xmath6 by mapping @xmath7 . this permutation will be written in one - line notation @xmath8 . @xmath9 maps @xmath10 to @xmath11 , @xmath12 to itself , @xmath13 to @xmath10 , and @xmath11 to @xmath13 . for @xmath14 , the map @xmath15 transposes @xmath16 and @xmath17 , and fixes all other elements in a permutation . the symmetric group @xmath3 is the coxeter group of type @xmath18 , and it is generated by the adjacent transpositions @xmath19 . the adjacent transpositions satisfy the coxeter relations : @xmath20 equation is called the _ short braid relation _ , and equation is the _ long braid relation_. a map is written to the left of its input , so @xmath21 interchanges the positions of values @xmath16 and @xmath17 in the permutation @xmath22 , while @xmath23 interchanges the values in positions @xmath16 and @xmath17 in @xmath22 . if @xmath24 , then @xmath25 . because the symmetric group is generated by adjacent transpositions , any permutation @xmath5 can be written as @xmath26 for some @xmath27 . the least such @xmath28 is the _ length _ of @xmath22 , denoted @xmath29 . an _ inversion _ in @xmath22 is a pair @xmath30 where @xmath31 and @xmath32 . inversion set _ is @xmath33 . since @xmath34 \times [ 1,n]$ ] , the inversion set can also be viewed as an array . the number of inversions in @xmath22 is equal to @xmath29 ( see @xcite ) . for obvious reasons , the permutation @xmath35 is called the _ longest element _ in @xmath3 . for a permutation @xmath22 with @xmath36 , a string @xmath37 such that @xmath26 is a _ reduced decomposition _ of @xmath22 . ( some sources call this a _ reduced word_. ) the set @xmath38 consists of all reduced decompositions of @xmath22 . a _ factor _ is a consecutive substring of a reduced decomposition . similar to the coxeter relations , a factor @xmath39 in a reduced decomposition will be called a _ short braid move _ if @xmath40 , and a factor @xmath41 will be called a _ long braid move_. the set @xmath38 has been studied in various contexts , notably by stanley in @xcite . there , stanley computes @xmath42 for several classes of permutations in terms of the number of standard young tableaux of certain shapes . in the case of a vexillary permutation , this is equal to the number of standard young tableaux of a single shape @xmath43 ( see also exercise 7.22 of @xcite ) . the definition of vexillary permutations is postponed until section [ section vexillary ] , where they will be discussed in depth . let @xmath24 and @xmath44 for @xmath45 . the permutation @xmath22 _ contains the pattern @xmath46 _ if there exist @xmath47 such that @xmath48 is in the same relative order as @xmath49 . that is , @xmath50 if and only if @xmath51 . if @xmath22 does not contain @xmath46 , then @xmath22 _ avoids _ @xmath46 , or is _ @xmath46-avoiding_. suppose that @xmath22 contains the pattern @xmath46 , with @xmath52 as defined above . then @xmath48 is an _ occurrence _ of @xmath46 in @xmath22 . the notation @xmath53 will denote the value @xmath54 . if @xmath55 , then @xmath56 . let @xmath57 , @xmath58 , and @xmath59 . then @xmath60 is an occurrence of @xmath46 , with @xmath61 , @xmath62 , @xmath63 , and @xmath64 . also , @xmath65 . the permutation @xmath22 is @xmath66-avoiding . let @xmath22 contain the pattern @xmath46 , and let @xmath67 be a particular occurrence of @xmath46 . if @xmath68 , then @xmath69 is a _ pattern entry _ in @xmath22 . otherwise @xmath69 is a _ non - pattern entry_. if a non - pattern entry lies between two pattern entries in the one - line notation for @xmath22 , then it is _ inside _ the pattern . otherwise it is _ outside _ the pattern . `` inside '' and `` outside '' are only defined for non - pattern entries . [ obstruction ] let @xmath67 be an occurrence of @xmath70 in @xmath22 . suppose that @xmath71 is inside the pattern , that @xmath72 for some @xmath73 $ ] , and that the values @xmath74 appear in increasing order in the one - line notation for @xmath22 . let @xmath75 be maximal so that the values @xmath76 appear in increasing order in the one - line notation for @xmath22 . the entry @xmath71 is _ obstructed to the left _ if a pattern entry smaller than @xmath77 appears between @xmath77 and @xmath71 in @xmath22 . likewise , @xmath71 is _ obstructed to the right _ if a pattern entry larger than @xmath78 appears between @xmath71 and @xmath78 in @xmath22 . let @xmath79 and @xmath80 . then @xmath81 and @xmath82 are both occurrences of @xmath46 in @xmath22 . obstruction is only defined for the latter , with @xmath83 and @xmath84 . then @xmath85 , and @xmath11 is obstructed to the left and not to the right . [ nonvex obst ] let @xmath86 and @xmath87 . then @xmath88 is an occurrence of @xmath46 in @xmath22 . using @xmath89 , @xmath90 in definition [ obstruction ] shows that @xmath91 , and @xmath13 is obstructed both to the left and to the right . vexillary permutations first appeared in @xcite and subsequent publications by alain lascoux and marcel - paul schtzenberger . they were also independently found by stanley in @xcite . there have since emerged several equivalent definitions of these permutations , and a thorough discussion of these occurs in @xcite . the original definition of lascoux and schtzenberger , and the one of most relevance to this discussion , is the following . [ 2143vex ] a permutation is _ vexillary _ if it is @xmath92-avoiding . the permutation @xmath93 is vexillary , but @xmath94 is not vexillary because @xmath95 is an occurrence of @xmath92 in the latter . the following proposition is key to proving one direction of theorem [ vexthm ] . [ vex not obst ] let @xmath22 contain the pattern @xmath46 . let @xmath71 be inside the pattern , with @xmath72 and the values @xmath74 appearing in increasing order in @xmath22 . if @xmath46 is vexillary then @xmath71 can not be obstructed both to the left and to the right . such obstructions would create a @xmath92-pattern in @xmath46 . example [ nonvex obst ] illustrates a non - vexillary permutation which has an element @xmath71 that is obstructed on both sides . equivalent characterizations of vexillarity concern the inversion set @xmath96 or the following objects . the _ diagram _ of a permutation @xmath22 is @xmath97 \times [ 1,n]$ ] where @xmath98 the _ code _ of @xmath22 is the vector @xmath99 where @xmath100 is the number of elements in row @xmath16 of @xmath96 . the shape @xmath43 is the partition formed by writing the entries of the code in non - increasing order . [ vex defs ] the following are equivalent definitions of vexillarity for a permutation @xmath22 : 1 . [ vex1 ] @xmath22 is @xmath92-avoiding ; 2 . [ vex2 ] the set of rows of @xmath96 is totally ordered by inclusion ; 3 . [ vex3 ] the set of columns of @xmath96 is totally ordered by inclusion ; 4 . [ vex4 ] the set of rows of @xmath101 is totally ordered by inclusion ; 5 . [ vex5 ] the set of columns of @xmath101 is totally ordered by inclusion ; 6 . [ vex6 ] @xmath102 , where @xmath103 is the transpose of @xmath43 . see @xcite . this section proves a new characterization of vexillary permutations , quite different from those in proposition [ vex defs ] . a partial ordering can be placed on the set of all permutations @xmath104 , where @xmath105 if @xmath106 contains the pattern @xmath107 . definition [ 2143vex ] determines vexillarity by a condition on the principal order ideal of a permutation . the new characterization , theorem [ vexthm ] , depends on a particular condition holding for the principal _ dual _ order ideal . let @xmath108 be a reduced decomposition of @xmath24 . for @xmath109 , the _ shift of @xmath110 by @xmath111 _ is @xmath112 [ vexthm ] the permutation @xmath46 is vexillary if and only if , for every permutation @xmath22 containing a @xmath46-pattern , there exists a reduced decomposition @xmath113 containing some shift of an element @xmath114 as a factor . first suppose that @xmath70 is vexillary . let @xmath5 contain a @xmath46-pattern . assume for the moment that there is a @xmath115 such that 1 . @xmath116 ; 2 . @xmath117 has a @xmath46-pattern in positions @xmath118 for some @xmath119 $ ] . choose a reduced decomposition @xmath114 . let @xmath120 be the permutation obtained from @xmath117 by placing the values @xmath121 in increasing order and leaving all other entries unchanged . choose any @xmath122 . then @xmath123 it remains only to find a @xmath124 satisfying ( r@xmath10 ) and ( r@xmath12 ) . this will be done by an algorithm that takes as input a permutation @xmath5 containing a @xmath46-pattern and outputs the desired permutation @xmath124 . because the details of this algorithm can be cumbersome , a brief description precedes each of the major steps . 1 . + set @xmath125 } : = w$ ] and @xmath126 . 2 . [ alg halt ] + if @xmath127}$ ] has no entries inside the pattern , then . otherwise , choose @xmath128}$ ] inside the pattern . [ alg big ] + if @xmath128 } > \langle k \rangle$ ] , then 1 . let @xmath129})= \{y \ge x_{[i ] } : y \text { is inside the pattern}\}$ ] . 2 . consider the elements of @xmath129})$ ] in decreasing order . multiply @xmath127}$ ] on the right by adjacent transpositions ( changing _ positions _ in the one - line notation ) to move each element immediately to the right of @xmath67 . 3 . let @xmath130}$ ] be the resulting permutation . set @xmath131 and . 4 . [ alg small ] + if @xmath128 } < \langle 1 \rangle$ ] , then 1 . let @xmath132 } ) = \{y \le x_{[i ] } : y \text { is inside the pattern}\}$ ] . 2 . consider the elements of @xmath132})$ ] in increasing order . multiply @xmath127}$ ] on the right by adjacent transpositions to move each element immediately to the left of @xmath67 . 3 . let @xmath130}$ ] be the resulting permutation . set @xmath131 and . [ alg bound ] + let @xmath73 $ ] be the unique value such that @xmath133 } < \langle m+1 \rangle$ ] . 6 . [ alg between ] + if the values @xmath134 } , \langle m+1 \rangle\}$ ] appear in increasing order in @xmath127}$ ] , then define @xmath135 and @xmath136 as in definition [ obstruction ] and 1 . [ alg unob right ] if @xmath128}$ ] is unobstructed to the right , then 1 . let @xmath137})$ ] be the set of non - pattern entries at least as large as @xmath128}$ ] and lying between @xmath128}$ ] and @xmath78 in the one - line notation of @xmath127}$ ] . 2 . consider the elements of @xmath137})$ ] in decreasing order . for each @xmath138})$ ] , multiply on the right by adjacent transpositions until @xmath139 is immediately to the right of @xmath78 , or the right neighbor of @xmath139 is @xmath140 . in the latter case , @xmath141 for some @xmath142 $ ] because all larger non - pattern entries are already to the right of @xmath78 . interchange the roles of @xmath139 and @xmath143 , and move this new @xmath139 to the right in the same manner , until it is to the right of ( the redefined ) @xmath78 . 3 . let @xmath130}$ ] be the resulting permutation , with @xmath67 redefined as indicated . let @xmath144}$ ] be the non - pattern entry in the final move after any interchange of roles . this is greater than @xmath128}$ ] and the newly redefined @xmath78 , and occurs to the right of the new @xmath78 . if @xmath144}$ ] is outside of the pattern , with @xmath131 . otherwise with @xmath131 . [ alg unob left ] the entry @xmath128}$ ] is unobstructed to the left ( proposition [ vex not obst ] ) . 1 . let @xmath145})$ ] be the set of non - pattern entries at most as large as @xmath128}$ ] and lying between @xmath77 and @xmath128}$ ] in the one - line notation of @xmath127}$ ] . 2 . consider the elements of @xmath145})$ ] in increasing order . for each @xmath146})$ ] , multiply on the right by adjacent transpositions until @xmath139 is immediately to the left of @xmath77 , or the left neighbor of @xmath139 is @xmath147 . in the latter case , @xmath148 for some @xmath149 $ ] because all smaller non - pattern entries are already to the left of @xmath77 . interchange the roles of @xmath139 and @xmath150 , and move this new @xmath139 to the left in the same manner , until it is to the left of ( the redefined ) @xmath77 . 3 . let @xmath130}$ ] be the resulting permutation , with @xmath67 redefined as indicated . let @xmath144}$ ] be the non - pattern entry in the final move after any interchange of roles . this is less than @xmath128}$ ] and the newly redefined @xmath77 , and occurs to the left of the new @xmath77 . if @xmath144}$ ] is outside of the pattern , . otherwise with @xmath131 . [ alg big left ] + if @xmath127}(s ) = \langle m+1 \rangle$ ] and @xmath127}(t ) = x_{[i]}$ ] with @xmath151 , multiply @xmath127}$ ] on the left by adjacent transpositions ( changing _ values _ in the one - line notation ) to obtain @xmath130}$ ] with the values @xmath152 } , \langle m+1 \rangle]$ ] in increasing order . then @xmath130}(s)$ ] is in the half - open interval @xmath152 } , \langle m+1 \rangle)$ ] , and @xmath130}(t)$ ] is in the half - open interval @xmath153 } , \langle m+1 \rangle]$ ] . with @xmath144}:=w_{[i+1]}(t)$ ] , the pattern redefined so that @xmath154}(s)$ ] , and @xmath131 . [ alg small right ] + if @xmath127}(s ) = \langle m \rangle$ ] and @xmath127}(t ) = x_{[i]}$ ] with @xmath155 , multiply @xmath127}$ ] on the left by adjacent transpositions to obtain @xmath130}$ ] with the values @xmath156}]$ ] in increasing order . then @xmath130}(s)$ ] is in the half - open interval @xmath157}]$ ] , and @xmath130}(t)$ ] is in the half - open interval @xmath156})$ ] . with @xmath144}:=w_{[i+1]}(t)$ ] , the pattern redefined so that @xmath158}(s)$ ] , and @xmath131 . each subsequent visit to step [ alg halt ] involves a permutation with strictly fewer entries inside the pattern than on the previous visit . each multiplication by an adjacent transposition indicated in the algorithm removes an inversion , and so decreases the length of the permutation . this is crucial because of requirement ( r@xmath10 ) . consider the progression of : * step [ alg halt ] @xmath159 or begin a pass through ; * step [ alg big ] @xmath159 step [ alg halt ] ; * step [ alg small ] @xmath159 step [ alg halt ] ; * step [ alg unob right ] @xmath159 steps [ alg big ] or [ alg big left ] ; * step [ alg unob left ] @xmath159 steps [ alg small ] or [ alg small right ] ; * step [ alg big left ] @xmath159 steps [ alg big ] , [ alg between ] , or [ alg big left ] ; * step [ alg small right ] @xmath159 steps [ alg small ] , [ alg between ] , or [ alg small right ] . step [ alg unob right ] concludes with @xmath144}$ ] to the left of its lower pattern bound , and smaller pattern elements lying between @xmath144}$ ] and this bound . therefore , no matter how often step [ alg big left ] is next called , the algorithm will never subsequently go to step [ alg unob left ] before going to step [ alg halt ] . likewise , a visit to step [ alg unob left ] means that step [ alg unob right ] can never be visited until step [ alg halt ] is visited and a new entry inside the pattern is chosen . steps [ alg big ] and [ alg small ] do not change the relative positions of @xmath67 . steps [ alg unob right ] and [ alg big left ] imply @xmath144 } > x_{[i]}$ ] , while @xmath144 } < x_{[i]}$ ] after steps [ alg unob left ] and [ alg small right ] . let @xmath160 be as in step [ alg bound ] . until revisiting step [ alg halt ] , the values @xmath161 , for @xmath162 , do not increase if @xmath144 } > x_{[i]}$ ] . nor do the values @xmath161 , for @xmath163 , decrease if @xmath144 } < x_{[i]}$ ] . the other pattern values are unchanged . the definition of @xmath160 means that the reordering of values in steps [ alg big left ] and [ alg small right ] does not change the positions in which the pattern @xmath46 occurs . additionally , these steps change the value of the entry inside the pattern ( that is , @xmath144 } \neq x_{[i]}$ ] ) , but not its position . these observations indicate not only that terminates , but that it outputs @xmath124 as in equation satisfying ( r1 ) and ( r2 ) . this completes one direction of the proof . now suppose @xmath70 is not vexillary . there is an occurrence @xmath164 such that @xmath165 define @xmath166 to be the index such that @xmath167 . define @xmath168 by @xmath169 for example , if @xmath87 , then @xmath86 . if there is a reduced decomposition @xmath113 such that @xmath170 for @xmath114 and @xmath109 , then there is a @xmath171 as in equation satisfying ( r1 ) and ( r2 ) . keeping the values @xmath172 , @xmath173 , @xmath174 , and @xmath175 as defined above , the permutation @xmath22 was constructed so that @xmath176 one of the values in the consecutive subsequence @xmath177 must move to get a consecutive @xmath46-pattern in @xmath117 . however , the values @xmath178 appear in increasing order in @xmath22 , and the consecutive subsequence @xmath179 in @xmath22 is increasing . therefore , there is no way to multiply @xmath22 by adjacent transpositions , always eliminating an inversion , to obtain a consecutive @xmath46-pattern . hence , if @xmath46 is not vexillary then there exists a permutation @xmath22 containing a @xmath46-pattern such that no reduced decomposition of @xmath22 contains a shift of a reduced decomposition of @xmath46 as a factor . [ vex ex ] if @xmath180 and @xmath181 , with the chosen occurrence @xmath67 in bold , the algorithm may proceed as follows . * @xmath125 } : = { \mbox{\boldmath $ 3$}}14{\mbox{\boldmath $ 6$}}5{\mbox{\boldmath $ 2$}}$ ] . * step [ alg halt ] : @xmath182 } : = 1 $ ] . * step [ alg small ] : @xmath125 } \mapsto w_{[0 ] } s_1 = 1{\mbox{\boldmath $ 3$}}4{\mbox{\boldmath $ 6$}}5{\mbox{\boldmath $ 2 $ } } = : w_{[1]}$ ] . * step [ alg halt ] : @xmath183 } : = 5 $ ] . * step [ alg big left ] : @xmath184 } \mapsto s_5 w_{[1 ] } = 1{\mbox{\boldmath $ 3$}}4{\mbox{\boldmath $ 5$}}6{\mbox{\boldmath $ 2 $ } } = : w_{[2]}$ ] ; @xmath185 } : = 6 $ ] . * step [ alg big ] : @xmath186 } \mapsto w_{[2 ] } s_5 = 1{\mbox{\boldmath $ 3$}}4{\mbox{\boldmath $ 52$}}6 = : w_{[3]}$ ] . * step [ alg halt ] : @xmath187 } : = 4 $ ] . * step [ alg unob right ] : @xmath188 } \mapsto w_{[3 ] } = 1{\mbox{\boldmath $ 34$}}5{\mbox{\boldmath $ 2$}}6 = : w_{[4]}$ ] ; @xmath189 } : = 5 $ ] . * step [ alg big ] : @xmath190 } \mapsto w_{[4]}s_4 = 1{\mbox{\boldmath $ 342$}}56 = : w_{[5]}$ ] . * step [ alg halt ] : output @xmath191 . therefore @xmath192 , and @xmath193 . keeping the notation of equation , @xmath194 and @xmath195 . the unique reduced decomposition of @xmath196 is @xmath197 , and indeed @xmath198 let @xmath86 and @xmath87 . no element of @xmath199 contains a shift of any element of @xmath200 as a factor . suppose that @xmath113 contains a shift of @xmath114 as a factor , @xmath201 then @xmath114 can be replaced by any @xmath202 in equation . some care must be taken regarding factors in reduced decompositions . this is clarified in the following definition and lemma , the proof of which is straightforward . let @xmath5 and @xmath203 . write @xmath204 , where @xmath205 and @xmath206 . suppose that @xmath207 contains only letters in @xmath208 . if no element of @xmath209 has an element of @xmath210 as its rightmost character and no element of @xmath211 has an element of @xmath210 as its leftmost character , then @xmath207 is _ isolated _ in @xmath110 . equivalently , the values @xmath118 must appear in increasing order in @xmath106 , and the positions @xmath118 must comprise an increasing sequence in @xmath107 . if @xmath212 and a shift of @xmath207 appears as a factor in a reduced decomposition of some permutation , then @xmath207 is necessarily isolated . this is because @xmath207 has maximal reduced length in the letters @xmath213 , so any factor of length greater than @xmath214 in the letters @xmath215 is not reduced . [ isolated ] if a reduced decomposition of @xmath22 contains an isolated shift of a reduced decomposition of @xmath46 , then @xmath22 contains the pattern @xmath46 . the converse to lemma [ isolated ] holds if @xmath46 is vexillary . the characterization of vexillary in theorem [ vexthm ] differs substantially from those in proposition [ vex defs ] . there is not an obvious way to prove equivalence with any of the definitions ( v[vex2])-(v[vex6 ] ) , except via ( v[vex1 ] ) . this raises the question of whether more may be understood about vexillary permutations ( or perhaps other types , such as grassmannian or dominant permutations ) by studying their reduced decompositions or the permutations that contain those in question as patterns . theorem [ vexthm ] has a number of consequences , and will be used often in the subsequent sections of this paper . most immediately , notice that it generalizes the result of billey , jockusch , and stanley mentioned earlier : @xmath0-avoiding permutations are exactly those whose reduced decompositions contain no long braid moves , and observe that @xmath216 . recall the definition of short and long braid moves in a reduced decomposition , as well as the short and long braid relations described in equations and . it is well known that any element of @xmath38 can be transformed into any other element of @xmath38 by successive applications of the braid relations . because the short braid relation represents the commutativity of particular pairs of adjacent transpositions , the following equivalence relation is known as the _ commutation relation_. [ commutation ] for a permutation @xmath22 and @xmath217 , write @xmath218 if @xmath110 can be obtained from @xmath219 by a sequence of short braid moves . let @xmath220 be the set of commutation classes of reduced decompositions of @xmath22 , as defined by @xmath221 . the commutation classes of @xmath222 are @xmath223 , @xmath224 , and @xmath225 . [ graph ] for a permutation @xmath22 , the graph @xmath226 has vertex set equal to @xmath220 , and two vertices share an edge if there exist representatives of the two classes that differ by a long braid move . elnitsky gives a very elegant representation of this graph in @xcite , which will be discussed in depth in section [ section polygon ] . a consequence of his description , although not difficult to prove independent of his work , is the following . [ graph facts ] the graph @xmath226 is connected and bipartite . see @xcite . despite proposition [ graph facts ] , much remains to be understood about the graph @xmath226 . for example , even the size of the graph for @xmath227 ( that is , the number of commutation classes for the longest element ) is unknown . billey , jockusch , and stanley characterize all permutations with a single commutation class , and hence whose graphs are a single vertex , as @xmath0-avoiding permutations . a logical question to ask next is : for what permutations does each reduced decomposition contain at most one long braid move ? more restrictively : what if this long braid move is required to be a specific shift of @xmath228 or @xmath229 ? moreover , what are the graphs in these cases ? let @xmath230 . [ 1 lbm ] @xmath231 is the set of permutations such that every @xmath0-pattern in @xmath22 has the same maximal element and the same minimal element . assume @xmath22 has a @xmath0-pattern . suppose that every occurrence of @xmath0 in @xmath22 has @xmath232 and @xmath233 . suppose that @xmath113 has at least one long braid move . choose @xmath234 so that @xmath235 is the first such . each adjacent transposition in a reduced decomposition increases the length of the product . then by the supposition , @xmath236 is @xmath0-avoiding , so @xmath237 has no long braid moves . it remains only to consider when @xmath238 is also a long braid move . the only possible reduced configurations for such a factor @xmath239 are shifts of @xmath240 and @xmath241 . if either of these is not isolated , then it is part of a shift of @xmath242 , @xmath243 , @xmath244 , or @xmath245 . notice that * @xmath246 ; * @xmath247 ; * @xmath248 . if @xmath239 is isolated in @xmath219 then @xmath22 contains a @xmath249- or @xmath250-pattern by lemma [ isolated ] . otherwise , @xmath22 contains a @xmath251-pattern . however , every @xmath0-pattern in @xmath22 has @xmath232 and @xmath233 . therefore @xmath252 is the only long braid move in @xmath219 , so @xmath253 . now let @xmath22 be an element of @xmath231 . if @xmath22 has two @xmath0-patterns that do not have the same maximal element and the same minimal element , then they intersect at most once or they create a @xmath251- , @xmath249- , or @xmath250-pattern . these three patterns are vexillary . thus by theorem [ vexthm ] and the examples above , containing one of these patterns would imply that some element of @xmath38 has more than one long braid move . if the two @xmath0-patterns intersect at most once , their union may be a non - vexillary pattern , so theorem [ vexthm ] does not necessarily apply . however , a case analysis shows that it is possible to shorten @xmath22 by adjacent transpositions and make one @xmath0-pattern increasing ( via a long braid move ) without destroying the other @xmath0-pattern . thus an element of @xmath38 would have more than one long braid move , contradicting @xmath253 . let @xmath254 consist of permutations with some @xmath0-pattern , where every long braid move that occurs must be @xmath255 or @xmath256 . [ 1 ulbm ] @xmath257 has a unique @xmath0-pattern and @xmath258 . if @xmath22 has a unique @xmath0-pattern , then @xmath259 . a unique @xmath0-pattern implies that @xmath260 all appear to the left of @xmath173 in @xmath261 , and @xmath262 all appear to the right of @xmath173 , so the second statement follows . consider the long braid moves that may appear for elements of @xmath263 . let @xmath253 have @xmath234 distinct @xmath0-patterns . by theorem [ 1 lbm ] , these form a pattern @xmath264 in @xmath22 . the permutation @xmath46 is vexillary , so there exists @xmath109 and a reduced decomposition @xmath265 for each @xmath114 . there are elements in @xmath266 with long braid moves @xmath267 for each @xmath268 $ ] . for example , @xmath269 . therefore , if @xmath270 , then @xmath271 , so @xmath22 has a unique @xmath0-pattern . suppose that @xmath22 has a unique @xmath0-pattern . because @xmath259 , the only possible long braid moves in reduced decompositions of @xmath22 are @xmath272 or @xmath273 . [ u_n graph ] if @xmath253 and @xmath22 has @xmath234 distinct @xmath0-patterns , then @xmath274 and the graph @xmath226 is a path of @xmath275 vertices connected by @xmath234 edges . because @xmath22 contains the pattern @xmath264 , there is a subgraph of @xmath226 that is a path of @xmath275 vertices connected by @xmath234 edges . since @xmath46 accounts for all of the @xmath0-patterns in @xmath22 , this is all of @xmath226 . [ u_n(j ) graph ] if @xmath270 , then @xmath276 and the graph @xmath226 is a pair of vertices connected by an edge . in his doctoral thesis and in @xcite , elnitsky developed a bijection between commutation classes of reduced decompositions of @xmath5 and rhombic tilings of a particular @xmath277-gon @xmath278 . this bijection leads to a number of interesting questions about tilings of @xmath278 and their relations to the permutation @xmath22 itself . a number of these ideas are studied in this and the following section . [ polygon ] for @xmath5 , let @xmath278 be the @xmath277-gon with all sides of unit length such that 1 . sides of @xmath278 are labeled @xmath279 in order ; 2 . the portion labeled @xmath280 is convex ; and 3 . sides with the same label are parallel . orient the polygon so that the edge labeled @xmath10 lies to the left of the top vertex and the edge labeled @xmath281 lies to its right . polygon_. for @xmath282 , the polygon @xmath283 is a centrally symmetric @xmath277-gon . the hexagon @xmath284 can be tiled by rhombi with sides of unit length in exactly two ways . each of these tilings is the _ flip _ of the other . let @xmath285 be the set of tilings of @xmath278 by rhombi with sides of unit length . define a graph @xmath286 with vertex set @xmath285 , and connect two tilings by an edge if they differ by a flip of the tiling of a single sub - hexagon . unless otherwise indicated , the term _ tiling _ refers to an element of @xmath285 . [ elthm ] the graphs @xmath226 and @xmath286 are isomorphic . henceforth , both graphs will be denoted @xmath226 . before discussing new results related to this polygon , it is important to understand elnitsky s bijection , outlined in the following algorithm . a more thorough treatment appears in @xcite . 1 . set the polygon @xmath287 } : = x(w)$ ] , the string @xmath288}$ } } : = \emptyset$ ] , and @xmath289 . [ elnstart ] if @xmath290}$ ] has no area , then . 3 . [ elnchoice ] there is at least one tile @xmath291 that shares two edges with the right side of @xmath290}$ ] . if @xmath291 includes the @xmath292 and @xmath293 edges from the top along the right side of @xmath290}$ ] , set @xmath294}$ } } : = j{\mbox{\boldmath $ j_{[i]}$}}$ ] . 5 . let @xmath295}$ ] be @xmath290}$ ] with the tile @xmath291 removed . set @xmath131 and . yields the entire commutation class because of the choice of tile in step [ elnchoice ] . the tiling in figure [ tilingfig ] corresponds to the equivalence class consisting solely of the reduced decomposition @xmath296 . [ subgraph ] if @xmath46 is vexillary and @xmath22 contains a @xmath46-pattern , then @xmath297 is a subgraph of @xmath226 . this follows from theorems [ vexthm ] and [ elthm ] . elnitsky s correspondence , described in , combined with theorem [ 1 lbm ] and corollary [ 1 ulbm ] , indicates that any tiling of @xmath278 for @xmath253 has at most one sub - hexagon ( every tiling has exactly one sub - hexagon if @xmath22 is not @xmath0-avoiding ) . moreover , the sub - hexagon has the same vertical position for all elements of @xmath254 . under certain circumstances , the polygon @xmath278 for @xmath5 can be rotated or reflected to give a polygon @xmath298 for another @xmath299 . [ reflect ] let @xmath24 and @xmath300 . then @xmath301 and @xmath302 . [ rotate ] let @xmath24 . if @xmath303 , then @xmath304and @xmath305 where @xmath306 and all entries are modulo @xmath4 . likewise , if @xmath307 , then @xmath308 and @xmath309 where @xmath310 and all entries are modulo @xmath4 . elnitsky s result interprets the commutation classes of @xmath38 as rhombic tilings of @xmath278 , with long braid moves represented by flipping sub - hexagons . the following theorem utilizes this interpretation , and demonstrates that the number of commutation classes of a permutation is monotonically increasing with respect to pattern containment , thus generalizing one aspect of corollary [ subgraph ] . note that @xmath46 is not required to be vexillary in theorem [ monotone ] , unlike in theorem [ vexthm ] . [ monotone ] if @xmath22 contains the pattern @xmath46 , then @xmath311 . consider a tiling @xmath312 . this represents a commutation class of @xmath266 . for an ordering of the tiles in @xmath313 as defined by , label the tile @xmath314 by @xmath315 , the tile @xmath316 by @xmath317 , and so on . if the tile with label @xmath318 corresponds to the adjacent transposition @xmath319 , then @xmath320 . 1 . set @xmath125}:=w$ ] , @xmath321}:=p$ ] , @xmath322}:=t$ ] , @xmath323}:=\emptyset$ ] , and @xmath289 . [ monostart ] if @xmath324}$ ] is the identity permutation , then define @xmath325}$ ] to be the tiles of @xmath326}$ ] together with any tiling of @xmath327})$ ] . . 3 . let @xmath328}$ ] be such that the tile labeled @xmath329 includes edges @xmath324}(j_{[i]})$ ] and @xmath324}(j_{[i]}+1)$ ] . note that @xmath324}(j_{[i ] } ) > p_{[i]}(j_{[i]}+1)$ ] . 4 . define @xmath330 so that @xmath127}(r ) = \langle p_{[i]}(j_{[i ] } ) \rangle$ ] and @xmath127}(s ) = \langle p_{[i]}(j_{[i]}+1 ) \rangle$ ] . note that @xmath127}(t)$ ] is a non - pattern entry for @xmath331 . 5 . let @xmath332}$ ] be the permutation defined by @xmath333}(t ) \mapsto \left\ { \begin{array } { c@{\quad:\quad}l } w_{[i]}(t ) & t < r \text { or } t > s\\ \widetilde{w}_{[i]}(t ) & r \le t \le s \end{array } \right.\ ] ] where @xmath334}(r ) , \ldots , \widetilde{w}_{[i]}(s))$ ] is @xmath335}(r ) , \ldots , w_{[i]}(s)\}$ ] in increasing order . set @xmath130 } : = w_{[i]}v_{[i]}$ ] , and notice that @xmath336 } ) = \ell(w_{[i ] } ) - \ell(v_{[i]})$ ] . 7 . [ monochoice ] the right boundaries of @xmath337})$ ] and @xmath327})$ ] differ only in the @xmath338 edges , and the left side of this difference ( part of the boundary of @xmath337})$ ] ) is convex . therefore , this difference has a rhombic tiling @xmath339}$ ] . define @xmath325}$ ] to be the tiles in @xmath326}$ ] together with the tiles in @xmath339}$ ] . set @xmath131 and . the algorithm takes a tiling @xmath312 and outputs one of possibly several tilings @xmath340 due to the choice in steps [ monostart ] and [ monochoice ] . a tiling @xmath340 so obtained can only come from this @xmath313 , although possibly with more than one labeling of the tiles . however , this labeling of the tiles merely reflects the choice of a representative from the commutation class , so indeed @xmath341 , and @xmath311 . let @xmath342 and @xmath343 . the pattern @xmath46 occurs in @xmath22 as @xmath344 . figure [ mono ex ] depicts the output of , given the two tilings of @xmath345 . @xmath159 + @xmath159 elnitsky s bijection considers the rhombic tilings of the polygon @xmath278 . rhombi are a special case of a more general class of objects known as zonotopes . a polytope is a _ @xmath346-zonotope _ if it is the projection of a regular @xmath4-cube onto a @xmath346-dimensional subspace . centrally symmetric convex polygons are exactly the @xmath12-zonotopes . these necessarily have an even number of sides . a _ zonotopal tiling _ of a polygon is a tiling by centrally symmetric convex polygons . let @xmath347 be the set of zonotopal tilings of elnitsky s polygon . rhombi are centrally symmetric , so @xmath348 . [ 2k - gon ] there is a tiling in @xmath347 containing a @xmath349-gon with sides parallel to the sides labeled @xmath47 if and only if @xmath350 is an occurrence of @xmath351 in @xmath22 . since the tiles are convex , a @xmath349-gon in the tiling with sides as described has right side labeled @xmath352 from top to bottom and left side labeled @xmath353 from top to bottom . therefore elnitsky s bijection shows that this tile ( or rather , any decomposition of it into rhombi ) transforms the sequence @xmath354 into @xmath355 . reduced decompositions have minimal length , so no inversions can be `` undone '' by subsequent adjacent transpositions . therefore @xmath350 must be an occurrence of @xmath351 in @xmath22 . conversely , suppose that @xmath350 is an occurrence of the vexillary pattern @xmath351 in @xmath22 . for a decreasing pattern , the algorithm can be modified slightly to produce @xmath117 as in equation , where the consecutive occurrence @xmath356 is @xmath350 . let @xmath357 and @xmath358 for @xmath122 . removing the rhombi that correspond to @xmath359 yields the polygon @xmath360 , and the rhombi that correspond to @xmath361 form a sub-@xmath349-gon with sides parallel to the sides labeled @xmath52 in @xmath278 . less specifically , theorem [ 2k - gon ] states that a tiling in @xmath347 can contain a @xmath349-gon if and only if @xmath22 has a decreasing subsequence of length @xmath234 . using a group theoretic argument , pasechnik and shapiro showed in @xcite that no element of @xmath362 consists entirely of hexagons for @xmath363 . their result states that at least one rhombus must be present in a hexagonal / rhombic tiling . kelly and rottenberg had previously obtained a better bound in terms of arrangements of pseudolines in @xcite . working with reduced decompositions and elnitsky s polygons yields a different proof that no element of @xmath362 can consist of entirely hexagonal tiles for @xmath363 , and generalizes the result to other types of tiles . theorem [ 2k tiles ] is , in a sense , a counterpart to theorem [ 2k - gon ] . [ 2k tiles ] let @xmath227 be the longest element in @xmath3 . there is a tiling @xmath364 consisting entirely of @xmath349-gons if and only if one of the following is true : 1 . @xmath365 ; or 2 . @xmath366 . if @xmath367 , the result holds for all @xmath4 : every @xmath283 can be tiled by rhombi with unit side length . for the remainder of the proof , assume that @xmath368 . suppose that there exists @xmath364 consisting entirely of @xmath349-gons . then there exists @xmath369 , for some @xmath370 . because @xmath371 is the longest element in @xmath372 , the factor @xmath373 can have any of @xmath374 at either end . thus , if @xmath375 for @xmath376 , then @xmath377 . because @xmath378 $ ] , there is at most one shift by @xmath379 and at most one shift by @xmath380 . in fact , there is exactly one of each of these shifts , since all of @xmath283 must be tiled and the shifts correspond to the vertical placement of the @xmath349-gons . consider the tile corresponding to @xmath381 . this @xmath349-gon is as high vertically as possible and it is the only tile placed so high . thus it shares the top vertex and its incident sides with @xmath283 . that is , two of the tile s sides are labeled @xmath10 and @xmath4 . similarly , the tile for @xmath382 also has two sides labeled @xmath10 and @xmath4 . by theorem [ 2k - gon ] , each of these tiles corresponds to a @xmath371-pattern with @xmath383 and @xmath61 . however , once the value @xmath10 is to the right of the value @xmath4 , it remains to the right as adjacent positions are transposed to lengthen the permutation . therefore the two patterns , the tiles , and their shifts must be equal : @xmath384 . indeed , there is always a tiling @xmath364 consisting of a single @xmath277-gon . there is a poset @xmath385 that arises naturally when studying @xmath347 . for a permutation @xmath22 , let the poset @xmath385 have elements equal to the zonotopal tilings @xmath347 , partially ordered by reverse edge inclusion . in the poset @xmath386 , the tiling in figure [ tilingfig ] is less than the tiling in figure [ ztilingfig ] . [ p(w_0 ) max ] for the longest element @xmath282 , the poset @xmath387 has a maximal element equal to the tiling in @xmath388 that consists of a single @xmath277-gon . [ posetgraph ] the minimal elements of @xmath385 are the rhombic tilings , which are the vertices of the graph @xmath226 . moreover , edges in the graph @xmath226 correspond to flipping a single sub - hexagon in the tiling . therefore these edges correspond to the elements of @xmath385 that cover the minimal elements . the relationships in remark [ posetgraph ] are immediately apparent . another relationship is not as obvious . this follows from a result of boris shapiro , michael shapiro , and alek vainshtein in @xcite . [ ssvlem ] the set of all @xmath11- and @xmath389-cycles in @xmath226 form a system of generators for the first homology group @xmath390 . additionally , anders bjrner noted that gluing @xmath12-cells into those @xmath11- and @xmath389-cycles yields a simply connected complex ( @xcite ) . in @xcite , lemma [ ssvlem ] is stated only for @xmath391 . however , the proof easily generalizes to all @xmath5 . a straightforward argument demonstrates that a @xmath11-cycle in @xmath226 corresponds to @xmath392 with rhombi and two hexagons , and an @xmath389-cycle corresponds to @xmath392 with rhombi and an octagon . these are exactly the elements of @xmath385 which cover those that correspond to edges of @xmath226 . [ level2 ] the elements of @xmath385 that cover the elements ( corresponding to edges of @xmath226 ) covering the minimal elements ( corresponding to vertices of @xmath226 ) correspond to a system of generators for the first homology group @xmath390 . little is known about the structure of the graph @xmath226 for arbitrary @xmath22 . however , in some cases a description can be given via theorem [ 2k - gon ] and lemma [ ssvlem ] . [ chainthm ] the following statements are equivalent for a permutation @xmath22 : 1 . [ treecon ] @xmath226 is a tree ; 2 . [ chaincon ] @xmath226 is a path ( that is , no vertex has more than two incident edges ) ; 3 . [ posetcon ] the maximal elements of @xmath385 cover the minimal elements ; 4 . [ patterncon ] @xmath22 is @xmath251-avoiding and any two @xmath0-patterns intersect at least twice . ( [ treecon ] ) @xmath393 ( [ posetcon ] ) by corollary [ level2 ] . from theorem [ 2k - gon ] and the discussion preceding corollary [ level2 ] , an @xmath389-cycle in the graph is equivalent to having a @xmath251-pattern . similarly , a @xmath11-cycle is equivalent to two sub - hexagons whose intersection has zero area , so some reduced decomposition has two disjoint long braid moves . this implies that two @xmath0-patterns intersect in at most one position . therefore ( [ treecon ] ) @xmath393 ( [ patterncon ] ) . finally , suppose that @xmath226 is a tree and a vertex has three incident edges . the corresponding tiling has at least three sub - hexagons . however , it is impossible for every pair of these to overlap . this contradicts ( [ treecon ] ) @xmath393 ( [ posetcon ] ) @xmath393 ( [ patterncon ] ) , so ( [ treecon ] ) @xmath393 ( [ chaincon ] ) . if @xmath394 is the set of all @xmath5 for which @xmath226 is a path , then @xmath395 by corollaries [ u_n graph ] and [ u_n(j ) graph ] . following convention , the unique maximal element in a poset , if it exists , is denoted @xmath396 . remark [ p(w_0 ) max ] noted that the poset @xmath387 has a @xmath396 . in fact , there are other @xmath22 for which @xmath385 has a @xmath396 , as described below . [ max elt ] the poset @xmath385 has a @xmath396 if and only if @xmath22 is @xmath397- , @xmath249- , and @xmath250-avoiding . the definition of the poset @xmath385 and theorem [ 2k - gon ] indicate that @xmath385 has a @xmath396 if and only if the union of any two decreasing subsequences that intersect at least twice is itself a decreasing subsequence . suppose there are decreasing subsequences in @xmath22 of lengths @xmath398 that intersect @xmath399 times , for @xmath400 . let @xmath401 , and choose a @xmath275 element subsequence of @xmath402 that includes @xmath403 and one more element from each descending subsequence . let @xmath404 be the resulting pattern . no @xmath396 in @xmath385 is equivalent to there being subsequences so that @xmath405 for some @xmath406 $ ] . there are two ways to place a @xmath349-gon in a zonotopal tiling of @xmath345 , but these overlapping @xmath349-gons do not both lie in any larger centrally symmetric polygon . the permutation @xmath46 is always vexillary , so theorem [ vexthm ] implies that @xmath385 will not have a @xmath396 if @xmath22 contains such a @xmath46 . therefore , considering the permutation @xmath46 for each possible @xmath407 , the poset @xmath385 has a @xmath396 if and only if @xmath22 is @xmath397- , @xmath249- , and @xmath250-avoiding . the permutations for which @xmath385 has a @xmath396 have recently been enumerated by toufik mansour in @xcite . although the graph @xmath226 and poset @xmath385 are not known in general , there is a class of permutations for which these objects can be completely described . this paper concludes with a study of this special case . in @xcite and @xcite , green and losonczy introduce and study `` freely braided '' elements in simply laced coxeter groups . in the case of type @xmath408 , these are as follows . a permutation @xmath22 is _ freely braided _ if every pair of distinct @xmath0-patterns in @xmath22 intersects at most once . equivalently , @xmath22 is freely braided if and only if @xmath22 is @xmath251- , @xmath397- , @xmath249- , and @xmath250-avoiding . the poset of a freely braided permutation has a unique maximal element by theorem [ max elt ] . the permutation @xmath409 is not freely braided because @xmath0 and @xmath410 are both occurrences of the pattern @xmath0 , and they intersect twice . the permutation @xmath411 is freely braided . mansour enumerates freely braided permutations in @xcite . in @xcite , green and losonczy show that a freely braided @xmath22 with @xmath234 distinct @xmath0-patterns has latexmath:[\[\label{fb size } any simply laced coxeter group , here stated only for type @xmath408 . if a permutation @xmath22 is freely braided with @xmath234 distinct @xmath0-patterns , then there exists @xmath203 with @xmath234 disjoint long braid moves . [ fb tiles ] this means that there is a tiling of @xmath278 with @xmath234 sub - hexagons , none of which overlap . furthermore , equation implies that flipping any sequence of these sub - hexagons does not yield any new sub - hexagons . hence _ every _ tiling of @xmath278 has exactly @xmath234 sub - hexagons , none of which overlap , and @xmath396 in @xmath385 corresponds to the zonotopal tiling with rhombi and @xmath234 hexagons . from remark [ fb tiles ] , the structures of the graph @xmath226 and the poset @xmath385 are clear for a freely braided permutation @xmath5 . let @xmath22 be freely braided with @xmath234 distinct @xmath0-patterns . the graph @xmath226 is the graph of the @xmath234-cube , and the poset @xmath385 is isomorphic to the face lattice of the @xmath234-cube without its minimal element . the permutation @xmath413 is freely braided . its three @xmath0-patterns are @xmath414 , @xmath415 , and @xmath416 . figures [ 3cubefig ] and [ 3posetfig ] depict its graph and poset . particular thanks are due to richard stanley for his continued guidance and for the suggestion to study reduced decompositions . anders bjrner provided helpful advice and discussion , and together with richard stanley coordinated the semester on algebraic combinatorics at the institut mittag - leffler , during which much of the research for this paper occurred . thanks are also owed to john stembridge for his referral to the work of elnitsky and green and losonczy . finally , the thoughtful suggestions of two anonymous referees have been greatly appreciated . l. m. kelly and r. rottenberg , simple points in pseudoline arrangements , _ pacific j. math . _ * 40 * ( 1972 ) , 617 - 622 . a. lascoux and m. p. schtzenberger , polynmes de schubert , _ c. r. acad . paris , srie i _ * 294 * ( 1982 ) , 447 - 450 . d. v. pasechnik and b. shapiro , in search of higher permutahedra , preprint . v. reiner , note on the expected number of yang - baxter moves applicable to reduced decompositions , _ europ . j. combin . _ * 26 * ( 2005 ) , 1019 - 1021 .

we study the unitary dynamics and the thermalization properties of free - fermion - like hamiltonians after a sudden quantum quench , extending the results of s. ziraldo _ et al . _ [ phys . rev . lett . 109 , 247205 ( 2012 ) ] . with analytical and numerical arguments , we show that the existence of a stationary state and its description with a generalized gibbs ensemble ( gge ) depend crucially on the observable considered ( local versus extensive ) and on the localization properties of the final hamiltonian . we present results on two one - dimensional ( 1d ) models , the disordered 1d fermionic chain with long - range hopping and the disordered ising/ spin chain . we analytically prove that , while time averages of one - body operators are perfectly reproduced by gge ( even for finite - size systems , if time integrals are extended beyond revivals ) , time averages of many - body operators might show clear deviations from the gge prediction when disorder - induced localization of the eigenstates is at play .

1211.4465

the concept of ergodicity is at the core of classical statistical mechanics : it establishes a connection between long - time averages of observables and statistical ensemble averages @xcite . the extension of the ergodic theorem to quantum mechanics was pioneered by von neumann @xcite in a seminal paper on the unitary dynamics of closed quantum systems . the experimental possibility of studying the nonequilibrium dynamics of `` virtually '' isolated quantum systems most notably cold atomic species in optical lattices @xcite has stimulated new interest in this issue . the simplest setting for such a nonequilibrium situation is that of a sudden quench of the hamiltonian , @xmath1 , governing the dynamics of the isolated quantum system : the after - quench ( @xmath2 ) unitary evolution is simply @xmath3 but the initial state @xmath4 for instance , the ground state of @xmath5 is generally a complicated superposition of the eigenstates of @xmath6 . since the energy is conserved , @xmath7 , it is reasonable to expect that a `` generic '' ergodic evolution in the hilbert space will lead to time averages which are reproduced by the microcanonical ensemble : this is what von neumann discussed for `` macroscopic '' observables @xcite , and is generally expected to occur @xcite , independently of the initial state @xmath4 . classical physics teaches us also when to expect violations of ergodicity : on one extreme , for systems that are integrable @xcite or close enough to being integrable @xcite ; on the other , for systems with a glassy dynamics , be it due to interactions providing dynamical constraints @xcite , or to genuine disorder @xcite . quantum mechanically , dangers to ergodicity come from very similar sources : integrability , interactions , and disorder . integrability implies the existence of many constants of motion @xmath8 , and this clearly restricts the ergodic exploration of the microcanonical energy shell , leading to what one might call a `` breakdown of thermalization '' @xcite . it often results in a kind of `` generalized thermalization '' described by a statistical ensemble which maximizes entropy in the presence of the constraints @xmath9 , an ensemble introduced long ago by jaynes @xcite and known as generalized gibbs ensemble ( gge ) @xcite @xmath10 } \;,\ ] ] where @xmath11 are lagrange multipliers which constrain the mean value of each @xmath12 to its @xmath13 value : @xmath14 $ ] . on the theory side , the approach to equilibrium has been carefully investigated for one - dimensional bose - hubbard models describing quench experiments in optical lattices and superlattices @xcite . starting from nonhomogeneous initial states ( density waves ) , such as @xmath15 , and evolving the system with a bose - hubbard @xmath16 both at the integrable points @xmath17 ( free bosons ) and @xmath18 ( free hard - core bosons / spinless fermions ) , where analytic solutions are possible @xcite , and at general ( nonintegrable ) values of @xmath19 ( through time - dependent density - matrix renormalization group @xcite ) , the physical picture emerging has led to the so - called _ local relaxation conjecture_@xcite : although the system is , _ sensu stricto _ , in a pure state @xmath20 , when measured upon locally in a finite region @xmath21 , the resulting ( mixed ) reduced density matrix @xmath22 relaxes towards a stationary gibbs state of maximum entropy compatible with all the constants of motion . this relaxation is strongly tight to the `` locality '' of the observable @xmath23 one measures , and results from `` information transfer '' carried by the excitations along the system @xcite , which eventually `` thermalizes '' any finite region @xmath21 , the rest of the system acting as an effective bath for @xmath21 . remarkably , such an approach to equilibrium does not require time averages @xcite . experimentally , a fast dynamical relaxation was recently observed @xcite in a system of cold atoms , where the long - time stationary state was compatible with gge . a far more dangerous type of `` ergodicity crisis '' derives , apparently , from dynamical constraints imposed by interactions@xcite : when quenching a bose - hubbard model starting , for instance , from @xmath24 i.e , with an average density of bosons per site @xmath25 where , in equilibrium , a mott phase transition occurs for @xmath26 the ensuing dynamics leads to a fast relaxation / thermalization for quenches at small @xmath19 , while the relaxation is extremely slow ( and the more so , the more the size of the system increases ) and the dynamics appears effectively _ freezed _ for @xmath27 @xcite . concerning ergodicity breaking due to genuine disorder , both anderson localization , at the single - particle level @xcite , and many - body localization , in the presence of interactions @xcite , are well - known examples of disorder - induced phenomena occurring in equilibrium physics . quantum quenches in the presence of disorder and interactions have also been studied , in the framework of many - body localization @xcite , but the physical picture is far from being fully understood . while breaking of translational invariance and disorder in the initial state @xmath4 have apparently little or no effect on the ensuing relaxation towards a stationary state if the after - quench hamiltonian @xmath6 is translationally invariant @xcite , disorder in the final hamiltonian @xmath6 seems to play a quite crucial role : this has been seen in numerical studies of nonintegrable fermion models with disordered long - range hopping and nearest - neighbor interaction @xcite , and has been recently corroborated through analytical arguments on integrable quantum spin chains of the ising/@xmath0 class @xcite . our present study extends that of ref . , where it was shown that , in the presence of disorder in @xmath6 , the long - time after - quench dynamics _ does not relax _ towards a stationary state , and time fluctuations generally persist in the expectation values of local operators , even in the thermodynamic limit ; this is , essentially , due to the presence of a _ pure - point spectrum _ of @xmath6 associated to localized wave functions , as opposed to the smooth continuum of a system with extended states . one can view this persistence of time fluctuations in local measurements as a result of the inability of the system to carry information around @xcite , due to localization . time averages are , therefore , mandatory in comparing dynamical quantities to statistical ensemble averages . but here a further surprise emerges , which will be the main focus of this study : while time averages of one - body local observables are perfectly well reproduced by the relevant statistical ensemble the gge , being the considered models , essentially , free - fermion ones time averages of many - body local operators show clear deviations from the corresponding gge prediction , in agreement with recent numerical findings @xcite . we will exemplify these ideas on a model of disordered one - dimensional fermions with long - range hoppings , showing a transition between extended and localized eigenstates , and on disordered ising/@xmath0 models . we will start , sec . [ sec : problem ] , by stating in a more precise way the problem we want to analyze . next , in sec . [ sec : models ] , we describe the models we have investigated . in sec . [ sec : gge - one - body ] we will analyze the essential reason why gge works perfectly well in predicting infinite - time averages of _ one - body operators _ , even for finite systems ( as long as one integrates over revivals ) . in sec . [ sec : gge - many - body ] we will discuss why this is not generally the case for many - body operators , unless time fluctuations of one - body green s functions vanish for large @xmath28 . next , in secs . [ sec : spinless ] and [ sec : ising ] we will present our results for the two models we have considered . section [ discussion : sec ] contains a final discussion , and our conclusions and perspectives . let us start by defining the problem . consider a standard quantum quench : an initial state @xmath4 , ground state of some hamiltonian @xmath5 , evolves under a different time - independent hamiltonian @xmath6 . given an observable @xmath29 , its quantum average can be separated in two terms ( we take @xmath30 ) : @xmath31 where the time - independent part @xmath32 results from an infinite - time average of @xmath33 @xmath34 while @xmath35 represents a fluctuating part ( with a vanishing time average ) . an alternative standard decomposition of @xmath33 proceeds by introducing the ( many - body ) eigenstates @xmath36 of @xmath6 , with energy @xmath37 , and inserting resolutions of the identity in eq . . defining @xmath38 and @xmath39 , one gets @xmath40 where the first term , @xmath41 , is the so - called diagonal average @xcite , while the second , time - dependent , part has been recast as a fourier transform of a weighted joint ( many - body ) density of states @xmath42 . with the quite usual assumption of no energy degeneracy , @xmath43 if @xmath44 , one can quickly prove that the diagonal average indeed coincides with the long - time average @xcite : @xmath45 while time fluctuations are given by the fourier transform of @xmath46 : @xmath47 as discussed in ref . , the behavior of the fluctuating part @xmath35 , relaxing and decaying to 0 or remaining finite ( with persistent oscillations ) for @xmath48 , is strongly tied to the `` smoothness '' of @xmath46 in the thermodynamic limit ( for finite systems , @xmath46 is always a series of discrete dirac @xmath49 s , hence @xmath35 will never go to zero for @xmath48 , and revivals will appear ) . indeed , @xmath35 will decay to zero for large @xmath28 if @xmath46 is smooth enough , due to the destructive interference induced in the @xmath50 integral by the strongly oscillating phase @xmath51 ( riemann - lebesgue lemma ) ; on the contrary , disorder and an important _ pure - point spectrum _ part , i.e. , @xmath49 functions associated to localized eigenstates which do not merge smoothly into a continuum , will lead to persistent time fluctuations @xmath35 for local operators @xcite . in the rest of the paper we will concentrate on two `` solvable '' models with disorder which will be presented in sec . [ sec : models ] : disordered one - dimensional fermions with long - range hoppings ( which can show either power - law localized states or extended ones , depending on a parameter controlling the long - range hopping variance ) , and disordered ising/@xmath0 model . both , being quadratic fermionic models , can be quite effectively numerically diagonalized , i.e. , one can find the one - body spectrum @xmath52 and the corresponding quasiparticle creation operator @xmath53 to express @xmath54 . this , in turn , allows us to calculate the fermionic local one - body green s functions : @xmath55 where @xmath56 creates a fermion at site @xmath57 . we will show that , in view of wick s theorem , a key issue in understanding the validity of gge for general many - body observables in such free - fermion - like hamiltonians has to do with the nature of the long - time fluctuations of the one - body green s functions . that might seem a simple matter to explore , but unfortunately , in the general disordered case , the one - body green s functions are essentially impossible to obtain analytically , and their numerical study is often elusive : you can only study a finite system - size @xmath58 for a finite time @xmath28 , and whether the fluctuations will eventually vanish or not for @xmath59 ( first ) and @xmath48 ( after , otherwise you always get revivals ) is often hard to tell . it turns out that a useful tool to distinguish the presence or absence of time fluctuations for large @xmath28 is given by the time - averaged fluctuations : @xmath60 which is zero when the fluctuations vanish for large @xmath28 , and is finite otherwise . we will show how to make progress analytically , for this quantity , to pin point the behavior of the large-@xmath28 fluctuations of one - body green s functions . this analytical progress will require an assumption of _ absence of gap degeneracies _ , i.e. , @xmath61 only when @xmath62 and @xmath63 or @xmath64 and @xmath65 , restricted , however , to the _ one - body spectrum _ , an assumption that can be argued to be a reasonable one , especially in the presence of disorder . its general validity for many - body energy eigenvalues , on the contrary , is rather tricky ( see discussion in sec . [ discussion : sec ] ) . in order to get smoother results we will perform averages of @xmath66 over different realizations of disorder . we stress , however , that the disorder average will always be performed _ after _ the computation of @xmath66 for each realization : performing the disorder average before taking the squared time integral , i.e. , on @xmath35 , would effectively kill the time fluctuations @xmath35 , by a kind of `` self - averaging '' @xcite . in this work we concentrate on two `` solvable '' models possessing a simple fermionic description . the first model describes spinless fermions hopping on a chain @xcite : @xmath67 where @xmath68 ( @xmath69 ) creates ( destroys ) a fermion at site @xmath57 and @xmath70 is a ( real ) hopping integral between sites @xmath71 and @xmath72 . we will in general take the @xmath73 s to be random and long ranged , with a gaussian distribution of zero mean , @xmath74 , and variance given by : @xmath75 here @xmath76 is a real positive parameter setting how fast the hoppings variance decays with distance . the peculiarity of this long - range - hopping model is that , regardless of the value of @xmath77 ( which hereafter is fixed to @xmath78 ) , it has an anderson transition from ( metallic ) extended eigenstates , for @xmath79 , to ( insulating ) power - law localized eigenstates for @xmath80 @xcite . physically , this is due to the fact that , for small @xmath76 , long - range hoppings are capable of overcoming the localization due to disorder . the clean nearest - neighbor hopping model is recovered by taking @xmath81 , where @xmath82 is the kronecker delta : we will always use this choice for the initial hamiltonian @xmath5 , with the corresponding ground state @xmath4 being the filled fermi sea . ( the reason behind this simple choice for @xmath5 is that the long - time fluctuation properties do not depend , qualitatively , on the initial hamiltonian being ordered or not , see ref . ) . being quadratic in the fermion operators , @xmath83 can be diagonalized for any chain of size @xmath58 in terms of new fermionic operators @xmath84 where @xmath85 are the wave functions of the eigenmodes of energy @xmath52 : @xmath86 . the energies @xmath52 and the associated wave functions @xmath85 are obtained , for any given realization of the hoppings @xmath70 in a chain of size @xmath58 with open boundary conditions , by numerically diagonalizing the @xmath87 one - body hopping matrix . the second hamiltonian we considered describes a disordered ising/@xmath0 chain in a transverse field @xcite : @xmath88 where @xmath58 is the size of the chain , @xmath89 ( @xmath90 ) are spin-@xmath91 pauli matrices for the @xmath57 site , and periodic boundary conditions are assumed , @xmath92 . @xmath93 , @xmath94 and @xmath95 are real and describe , respectively , the nearest - neighbor spin couplings and the transverse magnetic field . a quadratic fermionic hamiltonian is obtained here by applying a jordan - wigner transformation @xcite @xmath96 , where @xmath97 : @xmath98 where @xmath99 , and the anisotropy parameter @xmath100 is such that @xmath101 and @xmath102 . in the relevant sub - sector with an even number of fermions , we have to apply anti - periodic boundary conditions @xmath103 @xcite . at variance with the fermion - hopping case , eq . , there are now , for @xmath104 , bcs terms @xmath105 which create ( and destroy ) pairs of fermions . using a bogoliubov rotation we therefore define new fermions @xcite @xmath106 which diagonalize the hamiltonian @xmath107 . the ( positive ) eigenvalues @xmath108 and the associated eigenfunctions @xmath109 are obtained , once again , by diagonalizing a @xmath110 one - body matrix @xcite . notice also the strict particle - hole symmetry present even in the general disordered case @xcite : for every positive eigenvalue @xmath111 , with associated @xmath109 , there is a negative eigenvalue @xmath112 associated to @xmath113 . when considering quenches for this hamiltonian , we always start from a clean @xmath5 , with @xmath114 , @xmath115 ( ising case ) , and @xmath116 , while , for the final disordered hamiltonian , we take @xmath117 , @xmath115 , and @xmath118 , where @xmath119 sets the disorder strength and @xmath120 , @xmath121 are uncorrelated uniform random numbers in @xmath122 . the aim of our analysis is to study the behavior of different disorder realizations for very large @xmath58 ( ideally , in the thermodynamic limit ) . practically , the largest size we will consider is @xmath123 . given a disorder realization for @xmath123 , we will generate corresponding realizations at smaller @xmath58 by cutting away the same amount of sites from the two edges . in this way , we obtain a smoother behavior for all quantities versus @xmath58 . for every quantity @xmath124 considered , we have checked its probability distribution @xmath125 for different realizations of disorder . sometimes @xmath125 deviates strongly from a gaussian distribution , and is nearly ( although not precisely ) lognormal , i.e. , it is @xmath126 which is approximately gaussian distributed . in such a situation , we will calculate and plot the median ( i.e. , the geometric mean ) @xmath127_{\mathrm{av}}^\mathrm{g } = \exp ( \left [ \log x \right]_{\mathrm{av } } ) $ ] and the geometric standard deviation @xmath128 ) $ ] , rather than the usual ( arithmetic ) mean @xmath127_{\mathrm{av}}$ ] , and its standard deviation @xmath129 $ ] ( in the plots , the error bars will then go from @xmath127_{\mathrm{av}}^\mathrm{g } \exp ( -\sigma[\log x ] ) $ ] to @xmath127_{\mathrm{av}}^\mathrm{g } \exp ( \sigma[\log x ] ) $ ] ) . for both the models considered above , after diagonalization , the hamiltonian is expressed as : @xmath130 where @xmath52 is the positive excitation energy of the state @xmath131 and @xmath132 is the energy of the state @xmath133 annihilated by all the @xmath134 . notice that , in diagonalizing @xmath83 , eq . , one generally obtains some negative @xmath52 : in such a case , it is enough to perform a particle - hole transformation @xmath135 to change the sign of @xmath52 . physically , that implies that all negative energies are occupied in the ground state @xmath133 , and the resulting excitations describe particles or holes . in the following two sections we will use expression for @xmath6 : all the observations made are valid for both the models we have just described . notice that the number operators @xmath136 commute with @xmath6 in eq . and are therefore obvious constants of motion in the gge averages , @xmath137 . in this section we are going to show that , for a free fermion hamiltonian of the form given in eq . , the long - time average ( and the diagonal average ) of any one - body operator coincides with the corresponding gge average , for any system size @xmath58 , and any possible quench : @xmath138 with some important remarks to be made when there are degeneracies in the one - body spectrum ( see below ) . as we shall see , this can be traced back to the constraints that the gge sets through the constants of motion @xmath12 . we stress that , remarkably , this equality holds even for a finite - size chain @xmath58 , while , usually , statistical ensembles need a thermodynamic limit . examples of one - body operators are , in real space , @xmath139 or @xmath140 , the local density @xmath141 , the density @xmath142 , and , in momentum space , @xmath143 ( where @xmath144 ) , etc . more generally , a one - body fermionic operator can always be written , neglecting irrelevant constants and rewriting the @xmath145 s in terms of the @xmath146 ( inverting eqs . or ) , as : @xmath147 where @xmath148 , @xmath149 , and @xmath150 are @xmath87 matrices . let us start showing that @xmath151 . if @xmath152 denotes a general many - body eigenstate of @xmath6 , then clearly only the diagonal elements of @xmath148 enter in the diagonal matrix element : @xmath153 where @xmath154 is the occupation number of eigenmode @xmath155 in the eigenstate @xmath156 . in terms of @xmath157 , the diagonal average of @xmath158 is readily expressed as : @xmath159 where we have added and extra sum over @xmath160 , using @xmath161 , and then recognized two resolutions of the identity @xmath162 . notice , therefore , that the initial state enters only through @xmath163 , i.e. , exactly the constants of motion which are constrained and reproduced by the gge averages : @xmath164 . the conclusion is therefore simple , as the gge average of @xmath158 is : @xmath165 where we used that , by construction of the gge , @xmath166 . concerning the equality @xmath167 , we should pay attention to the cases in which @xmath6 has degenerate single - particle eigenvalues , @xmath168 ( for instance , when @xmath6 is disorder - free ) . in these cases , the time average of @xmath169 suppresses all the oscillatory factors @xmath170 occurring in the @xmath149 and @xmath150 terms of eq . , but all the factors @xmath171 corresponding to degenerate eigenvalues appearing in the @xmath148 terms survive . therefore : @xmath172 where we have singled out the diagonal elements , and the second sum runs over all the degenerate pairs @xmath173 such that @xmath174 . because of degeneracies , however , there is more freedom in the choice of the fermionic operators @xmath175 : we can always perform a unitary rotation in each degenerate subspace in such a way that @xmath176 for @xmath177 . with such a choice of the @xmath175 s , the extra terms in eq . vanish , and we recover the initial statement in eq . , i.e. , for any size and any quench , the long - time average of any one - body operator is equal to the gge one . we stress the fact that , for any finite system , @xmath178 will have recurrent fluctuations @xmath179 , the so - called _ returns _ or _ revivals _ , due to the discreteness of the finite - size spectrum : nevertheless , integrating over all times ( across revivals ) is guaranteed to reproduce the gge average : @xmath180 this statement , however , does not imply that the fluctuating part @xmath179 decreases to @xmath181 for @xmath48 , as indeed evident from the presence of finite - size revivals . as we shall see , @xmath179 might indeed persist for all times even in the thermodynamic limit , when @xmath6 is disordered : this in turn implies that @xmath182 might not exist in some cases , preventing a straightforward application of wick s theorem to extend the equalities of averages in eq . to many - body operators in the previous section we have shown that the gge average of a one - body operator coincides exactly with its long - time average . here we will see that , for a general many - body observable @xmath23 , the situation is more complicated , and gge can be proven to correctly predict long - time averages under two additional requirements : ( 1 ) @xmath23 is a finite sum of powers of some fermionic operators , and ( 2 ) the time fluctuations of the one - body green s functions associated to such fermionic operators are vanishing . whenever either of the two conditions is not realized , @xmath183 is not guaranteed to coincide with @xmath32 : we will indeed discuss cases of a definite disagreement between the two averages because of a violation of condition ( 2 ) above , i.e. , the persistence of time fluctuations of one - body green s functions . the key to the story is wick s theorem @xcite , which clearly applies to the free - fermion hamiltonians we are discussing @xcite . any many - body observable @xmath23 can be expressed as a linear combination of powers of the original real - space fermions @xmath145 and @xmath184 . since @xmath185 is a bcs - slater determinant , we can expand @xmath186 , using wick s theorem , as a sum of products of one - body green s functions @xmath187 and @xmath188 . to make things more clear , let us consider , for instance , the density - density correlations @xmath189 with @xmath190 , a two - body operator whose wick s expansion reads : @xmath191 this expansion clearly involves a finite number of terms ( condition ( 1 ) ) . now suppose ( condition ( 2 ) ) that the time fluctuations of the green s functions vanish for large @xmath28 , hence the limits @xmath192 and @xmath193 exist . from the analysis of the previous section , it is obvious that such limits must coincide with the corresponding gge averages : @xmath194 , and @xmath195 . it follows therefore that @xmath196 exists ( i.e. , its fluctuating part @xmath197 vanishes for large @xmath28 ) and is given by : @xmath198 where the final step uses the fact that wick s theorem also applies to gge averages of free - fermion hamiltonians @xcite . notice that , since long - time fluctuations of @xmath199 vanish , the infinite - time limit dominates the time average , and this implies : @xmath200 for a similar discussion when @xmath6 is a translationally invariant disorder - free ising-@xmath201 chain , see ref . . a similar proof works quite generally for all observables @xmath23 provided the two stipulated conditions are satisfied . a few important remarks are in order : ( _ i _ ) the existence of _ time limits _ of green s functions , as opposed to time averages , are crucial in applying wick s theorem , because it is generally false that the `` time average of a sum of products '' coincides with the `` sum of products of time averages '' ; ( _ ii _ ) for definiteness , we have chosen , above , the real - space fermions @xmath184 to expand @xmath23 , but similar arguments can be made in any one - body fermionic basis @xmath202 , for instance , a momentum space basis . notice , in this respect , that @xmath23 might involve an infinite expansion in terms of the @xmath184 s and a finite one in terms of the @xmath203 s ( condition ( 1 ) ) . in this case , if the time fluctuations of @xmath204 and @xmath205 vanish ( condition ( 2 ) ) one can still conclude that @xmath206 ; ( _ iii _ ) whenever the time fluctuations of the one - body green s functions do not vanish for large @xmath28 , and/or the expansion of the operator @xmath23 involves an infinite number of wick s contractions there is no guarantee that gge will not reproduce long - time averages : we simply can not prove it by using wick s theorem . nevertheless , we will later discuss ( see sec . [ sec : many - body - hop ] ) explicit cases where the persistence of one - body time fluctuations , due to disorder and to the presence of localized eigenstates , indeed leads to a definite discrepancy between @xmath32 and @xmath183 . let us consider quantum quenches with a final hamiltonian given by @xmath83 , eq . , describing disordered one - dimensional spinless fermions with long - range hopping . we will consider observables @xmath23 in real space and in momentum space , and hence we will need to ascertain the time dependence of both real - space and momentum - space green s functions . to unify the treatment of both cases , we will consider a general fermionic operator @xmath207 obtained by applying a unitary transformation @xmath208 ( of matrix elements @xmath209 , with @xmath210 the identity ) to the @xmath211 s which diagonalize @xmath83 : for the original real - space fermions @xmath145 , @xmath85 is the real - space wave function of the @xmath155 eigenstate , while for the momentum space fermions @xmath212 , @xmath213 . as discussed in secs . [ sec : gge - one - body ] and [ sec : gge - many - body ] , if the green s functions associated to the @xmath214 s have vanishing long - time fluctuations , then also the long - time fluctuations of @xmath23 disappear and @xmath215 . we will show the crucial role played by the localization of the eigenfunctions , which we will characterize through the standard inverse participation ratio ( ipr ) @xcite . when quenching to a final hamiltonian with @xmath76 in the localized phase ( @xmath80 ) , the real - space green s functions @xmath216 will be shown to have persistent time fluctuations , i.e. , @xmath217 in the thermodynamic limit @xmath59 , while quenching with @xmath76 in the extended phase ( @xmath79 ) leads to vanishing time fluctuations , @xmath218 . in both cases , however , the eigenfunctions appear to be _ extended _ when analyzed in momentum space , which results in momentum space green s functions with vanishing time fluctuations , @xmath219 . we will then explicitly discuss ( see sec . [ sec : many - body - hop ] ) the discrepancy between time averages and gge averages for real - space many - body operators , such as density - density spatial correlations , when quenching to the localized phase . as discussed in sec . [ sec : gge - one - body ] , gge correctly predicts the infinite - time average of any one - body operator , and hence , in particular , of the one - body green s functions . , with @xmath220 ( solid line ) , and its running - time average @xmath221 ( dashed line ) , for two different values of @xmath76 and two different sizes . the horizontal dash - dotted line is the gge average of @xmath222 . the data are obtained using a single realization , with @xmath223 for the extended case , and @xmath224 for the localized one ( which shows no size effect ) . disorder realizations for the smaller @xmath58 chain are obtained , here and in the following , by removing sites from the right and left edges of the larger chain . , scaledwidth=42.0% ] figure [ fig : te ] exemplifies this , showing the time evolution of @xmath225 , the expectation value of the fermion density at site @xmath57 , for two sizes and two values of @xmath76 , one in the extended phase ( @xmath223 ) and one in the localized phase ( @xmath224 ) . we now address in a more general way the question of the time fluctuations of the one - body green s functions @xcite . since @xmath83 conserves the total number of fermions , the anomalous green s functions @xmath226 are always zero . concerning @xmath227 , using the expansion @xmath207 we can express it as : @xmath228 where @xmath229 is the @xmath13 green s function of the normal modes . assuming the model has no single - particle energy degeneracy ( i.e. , if @xmath230 then @xmath62 ) , only the diagonal terms with @xmath231 will contribute to the infinite - time average of @xmath227 : @xmath232 hence , the time fluctuations of @xmath227 will be given by : @xmath233 now we calculate @xmath234 by squaring the previous expression and taking the infinite - time average . if we assume there is no gap degeneracy ( i.e. , if @xmath235 then @xmath64 and @xmath65 ) we arrive at@xcite : @xmath236 which turns out to be a very useful and sharp tool in the analysis of the time fluctuations of the green s functions . we will study it in various situations ( different quenches and different choices of the fermionic operators @xmath214 ) to understand when and why @xmath234 , which is always finite for any finite @xmath58 ( due to revivals ) , goes to zero in the thermodynamic limit @xmath59 . we will see , in this respect , the crucial role played by the weights @xmath237 and by localization . let us consider first real - space green s functions @xmath216 , which for @xmath238 , have a simple physical meaning : the expectation value at time @xmath28 of the local density at the site @xmath57 . from eq . we get : @xmath239 where @xmath240 is the real - space wave functions of the eigenstate @xmath155 . depending on @xmath76 , the eigenfunctions @xmath85 are either localized ( for @xmath80 ) or extended ( for @xmath79 ) @xcite . to pin - point this , we could monitor the ipr of the @xmath155 eigenstate , @xmath241 . as it turns out , all eigenstates behave in the same way for the present model : either all localized , @xmath242 for @xmath243 , or all extended , @xmath244 for @xmath243 , without any mobility edge . for that reason , we can just monitor the average ipr defined as : @xmath245 shown in the inset of fig . [ fig : delta2 ] ( top ) for @xmath223 ( `` extended '' points ) and @xmath224 ( localized " points ) . correspondingly , the main panel of fig . [ fig : delta2 ] ( top ) shows the average value of @xmath246 as a function of the chain size @xmath58 for both choices of @xmath76 . notice that when eigenstates are extended , the weights @xmath247 in eq . can be essentially taken out of the sum . but , without weights , it is a simple matter to show that : @xmath248 where @xmath249 is the total number of fermions in the initial state . hence , @xmath250 is expected to go to zero as @xmath251 when quenching towards a phase with extended eigenstates ( @xmath79 ) , as indeed found numerically . on the contrary , weights are of paramount importance when quenching to a phase with localized eigenstates ( @xmath80 ) , because they move the important contributions to @xmath250 from the average @xmath252 , which is of order @xmath251 , to rare large values @xcite , leading to a finite @xmath250 which is rather insensitive to the size @xmath58 . ( top ) and @xmath253 ( bottom ) for the disordered long - range hopping model @xmath83 as a function of the chain size @xmath58 for different values of @xmath76 . here @xmath254 with @xmath220 , and @xmath255 with @xmath256 . the data are obtained starting from the ground state of a clean fermionic chain with nearest - neighbor hopping and quenching to @xmath83 with different values of @xmath76 . the localized " points are for @xmath257 , while the extended " points are for @xmath258 . we used @xmath259 realizations of disorder . in all cases we report the usual ( arithmetic ) mean ( the error bar , when visible , is the standard deviation ) , except for @xmath246 in the localized phase , where we plot the median ( the geometric mean , see sec . [ sec : models ] for details ) . in the inset , the ipr ( see eqs . and ) as a function of size . notice how for all cases ( localized " and extended " quenches ) the eigenstates of the final hamiltonian are `` extended '' in reciprocal space and consequently @xmath260 goes to zero for @xmath59 . for a smoother size scaling , each disorder realization of the largest @xmath58 generated is employed , by removing the same amount of sites from the two edges , to generate realizations for smaller @xmath58 . , title="fig:",scaledwidth=42.0% ] ( top ) and @xmath253 ( bottom ) for the disordered long - range hopping model @xmath83 as a function of the chain size @xmath58 for different values of @xmath76 . here @xmath254 with @xmath220 , and @xmath255 with @xmath256 . the data are obtained starting from the ground state of a clean fermionic chain with nearest - neighbor hopping and quenching to @xmath83 with different values of @xmath76 . localized " points are for @xmath257 , while the extended " points are for @xmath258 . we used @xmath259 realizations of disorder . in all cases we report the usual ( arithmetic ) mean ( the error bar , when visible , is the standard deviation ) , except for @xmath246 in the localized phase , where we plot the median ( the geometric mean , see sec . [ sec : models ] for details ) . in the inset , the ipr ( see eqs . and ) as a function of size . notice how for all cases ( localized " and extended " quenches ) the eigenstates of the final hamiltonian are `` extended '' in reciprocal space and consequently @xmath260 goes to zero for @xmath59 . for a smoother size scaling , each disorder realization of the largest @xmath58 generated is employed , by removing the same amount of sites from the two edges , to generate realizations for smaller @xmath58 . , title="fig:",scaledwidth=42.0% ] consider now the green s functions in momentum space @xmath261 , representing , for @xmath262 , the expectation value at time @xmath28 of the momentum distribution . since @xmath263 , the @xmath264 s are straightforwardly related to the @xmath216 s through a double summation on @xmath71 and @xmath72 with oscillating phase factors @xmath265 . however , for @xmath59 these are infinite sums , and this might change the behavior of the time fluctuations . we now show that , even when @xmath216 has persistent time fluctuations , the corresponding @xmath264 averages them out , and @xmath266 for @xmath59 . indeed , using eq . , @xmath267 where @xmath213 are the fourier transforms of the real - space wave functions @xmath85 . figure [ fig : delta2 ] ( bottom ) shows that @xmath268 for @xmath59 , regardless of the value of @xmath76 . notice that , perhaps counter - intuitively , the average ipr in @xmath269-space @xmath270 always decreases as @xmath251 for both @xmath79 ( extended real - space wave functions ) and @xmath80 ( localized real - space wave functions ) . versus the site index @xmath57 , where @xmath85 is a typical extended eigenstate of @xmath83 for @xmath223 and @xmath123 , with energy @xmath52 in the middle of the band . ( b ) the corresponding momentum space @xmath271 , with @xmath272 , quite clearly extended . ( c ) plot of @xmath273 for a toy extended wave function @xmath274 , where @xmath275 is a random sign . , scaledwidth=42.0% ] the extended case @xmath79 is particularly intriguing , because one would expect that an extended real - space wave function should look `` localized '' in momentum space , i.e. , composed of a small number of @xmath269 waves . this expectation , quite reasonable for ordinary extended states of nondisordered systems , is not correct , in general , in the presence of disorder , as quite evident from figs . [ fig : eigenstate](a ) and [ fig : eigenstate](b ) . a simple example demonstrates the crucial role played by disorder . consider a toy real - space extended wave function with @xmath276 where @xmath275 is a random sign on every site . without @xmath277 , the momentum space function @xmath278 would be localized , with a peak at @xmath256 . when @xmath277 is accounted for , @xmath279 becomes extremely irregular but extended over all @xmath269 points , see fig . [ fig : eigenstate](c ) , indeed with a strong resemblance to the actual momentum space wave function of fig . [ fig : eigenstate](b ) . in the presence of disorder , therefore , being `` extended in real - space '' does not imply a sharply defined momentum . effectively , therefore , going to momentum space averages out persistent time fluctuations which are seen in real space when eigenstates are localized , an effect akin to `` self - averaging '' of extensive quantities in disordered systems @xcite . from the general analysis of secs . [ sec : gge - one - body ] and [ sec : gge - many - body ] and the study of the one - body green s functions of sec . [ sec : g - hop ] , we can conclude that many - body operators involving a finite expansion in terms of momentum space operators @xmath212 , such as correlations @xmath280 , will have time averages which coincide with gge averages , regardless the value of @xmath76 . the same is true in the delocalized phase ( @xmath79 ) for many - body operators with a finite expansion in real space , because the time fluctuations of @xmath216 vanish . when @xmath80 ( localized phase ) the gge ability in describing time averages of many - body operators is not guaranteed , because @xmath216 have persistent time fluctuations and wick s theorem is of no help . here we will show that , when @xmath80 , gge fails in predicting the spatial density - density correlations @xmath189 . to see this , we compare , see eqs . and , the time average of @xmath199 ( i.e. , a time average of a sum of products of @xmath281 s ) with the corresponding gge average ( a sum of products of time averages ) . in fig . [ fig : te - rho ] we plot @xmath199 , together with its running - time average ( i.e. , @xmath282 ) and the gge average for two chain sizes and two values of @xmath76 . first we notice that , as in the case of @xmath216 , increasing the size @xmath58 in the delocalized phase ( @xmath79 ) strongly decreases the time fluctuations , which are , on the contrary , unaffected by @xmath58 in the localized phase ( @xmath80 ) . the second feature emerging from fig . [ fig : te - rho ] is that , while in the delocalized phase the time average tends to the gge value , there is a marked and clear discrepancy between the two in the localized phase . , with @xmath283 and @xmath284 ( solid line ) , and its running - time average @xmath285 ( dashed line ) , for two different values of @xmath76 and two different sizes . the horizontal dash - dotted line is the gge average for @xmath286 . the data are obtained using a single realization ( see caption of fig . [ fig : te ] for details ) . , scaledwidth=42.0% ] the difference between the gge average and the time average can be explicitly computed using the same strategy ( and assumptions ) of sec . [ sec : g - hop ] : @xmath287 \nonumber \\ & = & \lim_{t \to \infty } \frac{1}{t } \int_0^t { \mathrm{d } } t \left [ |\delta g_{j_1j_2}(t)|^2 - \delta g_{j_1j_1 } ( t ) \delta g_{j_2j_2}(t ) \right ] \nonumber \\ & = & \sum_{\mu_1 \mu_2 } \left ( | u_{j_1\mu_1 } |^2 | u_{j_2\mu_2 } |^2 - u_{j_1\mu_1}^\ast u_{j_1\mu_2 } u_{j_2\mu_1 } u_{j_2\mu_2}^\ast \right ) \left| g_{\mu_1\mu_2}^0 \right|^2 \;,\end{aligned}\ ] ] where we first used the wick s expansions of @xmath288 and @xmath289 , then used the relationships @xmath290 and @xmath291 ( since gge works for one - body averages ) , and finally made use of eq . and of the no - gap - degeneracy assumption . the result is closely reminiscent of eq . for @xmath292 , except that now the weights have two contributions . using the relation @xmath293 we can finally reexpress @xmath294 as an explicitly positive quantity as follows : @xmath295 \left| g_{\mu_1\mu_2}^0 \right|^2 \nonumber \\ & = & \frac{1}{2 } \sum_{\mu_1\mu_2 } \left| u_{j_1\mu_1 } u_{j_2\mu_2 } - u_{j_1\mu_2 } u_{j_2\mu_1 } \right|^2 \left| g_{\mu_1\mu_2}^0 \right|^2 \geq 0 \;.\end{aligned}\ ] ] this explicitly shows that , apart from the trivial cases in which all terms of the summation are zero ( e.g. , @xmath296 , when @xmath297 becomes a one - body operator for which gge works and @xmath298 ) , the gge average overestimates the time average of @xmath299 , as exemplified in fig . [ fig : te - rho ] for a single realization . the question now is whether or not @xmath294 goes to zero for @xmath243 . in fig . [ fig : error ] we plot the average value of @xmath294 as a function of @xmath58 for the same set of quenches presented before . @xmath300_{\rm av}$ ] goes to zero in the thermodynamic limit in the extended " phase , and this confirms our general analysis : the real - space green s functions have vanishing fluctuations in the extended phase , and gge works for many - body observables with a finite expansion . on the contrary , in the localized " phase , @xmath300_{\rm av}^{\rm g}= \exp ( \left [ \log \delta_{j_1j_2 } \right]_{\mathrm{av } } ) $ ] remains finite even in thermodynamic limit , and since @xmath300_{\rm av } \geq [ \delta_{j_1j_2}]_{\rm av}^{\rm g}$ ] ( by jensen s inequality ) , this ensures that also @xmath300_{\rm av}$ ] is finite for @xmath243 . clearly , the persistent time fluctuations of the green s functions lead to time _ correlations _ between the @xmath216 appearing in the expansion of @xmath199 , see eq . , which reduce the time average with respect to the corresponding `` sum of products '' of time averages . , the discrepancy between the gge average and the time average , for the density - density correlation @xmath189 , as a function of the chain size , for with @xmath283 and two values of @xmath301 . the data are obtained using the same quenches of fig . [ fig : delta2 ] ( see its caption for details ) . in the localized phase ( @xmath224 ) we plot the median @xmath300_{\rm av}^{\rm g}= \exp ( \left [ \log \delta_{j_1j_2 } \right]_{\mathrm{av } } ) $ ] because @xmath294 is there roughly log - normal distributed . , scaledwidth=42.0% ] consider now the case of an ising chain in transverse field , described in sec . [ sec : models ] . the most important difference with respect to the case of spinless fermions with long - range hoppings is the presence of particle non - conserving ( bcs - like ) terms , which mix particles and holes . because of that , it is convenient to work with nambu vectors@xcite , which make the algebra very similar to the previous one , with very similar results . in particular , we will show that , due to disorder and localization of eigenstates , the time fluctuations of the local magnetization @xmath302 remain finite ; however , considering for instance , the total magnetization @xmath303 which is extended over the whole chain , introduces an infinite summation which effectively leads to a self - averaging of time fluctuations . consider , for simplicity , only real - space green s functions . in terms of the real - space fermion operators , @xmath145 defines the nambu vector @xmath304 . similarly , in terms of the fermions @xmath134 which diagonalize @xmath305 , we define a second nambu vector @xmath306 . these two vectors are connected through the relation @xmath307 where @xmath19 is a @xmath308 unitary matrix which describes the bogoliubov rotation@xcite performed to diagonalize @xmath305 . notice that the index @xmath155 in the nambu vector @xmath309 runs from @xmath78 to @xmath310 : for every index @xmath311 $ ] ( associated to @xmath134 , and with energy @xmath312 ) there is a corresponding index @xmath313 $ ] ( associated to @xmath314 , and with energy @xmath315 ) ; similarly , for every @xmath316 $ ] we define @xmath317 $ ] . the time evolution of @xmath309 is simple : @xmath318 , where the energy @xmath319 when @xmath320 and @xmath321 when @xmath322 . the pairs @xmath323 and @xmath324 entering in the one - body standard and anomalous green s functions can both be obtained from the nambu pairs @xmath325 ( where now @xmath71 and @xmath72 run from @xmath78 to @xmath310 ) : @xmath326 , and the associated ( nambu ) green s function closely resembles eq . : @xmath327 where @xmath328 . assuming , again , no energy degeneracy , the time fluctuations of @xmath329 read : @xmath330 in computing the time - averaged squared fluctuations of @xmath227 we have to take care of gap degeneracies due to the particle - hole symmetry of the spectrum @xmath331 . taking due care of that , the value of @xmath332 turns out to be : @xmath333 where the over - line denotes the infinite - time average , and we used the relation @xmath334 . the first term is due to the cases in which @xmath335 and @xmath336 , similarly to what is found in eq . . the second term originates from particle - hole symmetry ( present even when the system is disordered ) and occurs when @xmath337 and @xmath338 . as we did for the spinless fermionic hamiltonian , one can show that : @xmath339 moreover , using the fact that @xmath340 we can repeat the same observations presented in sec . [ sec : g - hop ] : if the eigenstates are delocalized ( clean chain case ) @xmath332 goes to zero for @xmath59 , while @xmath332 remains finite when the eigenstates are localized . for any finite disorder amplitude @xmath119 , the hamiltonian @xmath305 has always localized states . indeed , as shown in the inset of fig . [ fig : ising ] , the average ipr defined similarly to eq . , @xmath341 is finite for @xmath243 . this localization leads to persistent time fluctuations : @xmath342 in the thermodynamic limit . this is shown in fig . [ fig : ising ] where we plot the average value of @xmath343 for a quench from the ground state of a clean ising chain at the critical point ( @xmath344 , @xmath115 , @xmath114 , and @xmath345 ) to a disordered ising chain at the infinite randomness critical point ( @xmath346 , @xmath115 , @xmath347 $ ] , and @xmath348 $ ] ) @xcite . , with @xmath23 the local transverse magnetization @xmath349 or the total magnetization @xmath350 , as a function of the chain size . the initial state is the ground state of a clean ising chain at the critical point ( @xmath344 , @xmath115 , @xmath114 , and @xmath345 ) and the final hamiltonian is a disordered ising chain at the infinite - randomness critical point ( @xmath346 , @xmath115 , @xmath347 $ ] , and @xmath348 $ ] ) . data obtained with @xmath259 disorder realizations . for a smoother size scaling , each disorder realization of the largest @xmath58 generated is employed , by removing the same amount of sites from the two edges , to generate realizations for smaller @xmath58 . for @xmath351 we plot the median @xmath352_{\rm av}^{\rm g}= \exp ( \left [ \log \delta^2_{\sigma_j^z } \right]_{\mathrm{av } } ) $ ] because @xmath353 is roughly log - normal distributed.,scaledwidth=42.0% ] we now show that , while each @xmath354 has nonvanishing time fluctuations , the average magnetization per site @xmath355 has vanishing time fluctuations for large @xmath58 , due to cancellations reminiscent of self - averaging in extensive observables @xcite . indeed , @xmath356 , which implies that : @xmath357 i.e. , an expression entirely similar to eq . for @xmath358 except for the weight @xmath359 which is now replaced by the averaged weight @xmath360 . with the same steps done to obtain eq . , we finally get : @xmath361 which is definitely different from eq . , the site average having been performed _ before _ taking the squared time fluctuations . in fig . [ fig : ising ] we plot @xmath362 , averaged over disorder realizations , as a function of the chain size @xmath58 : we clearly see that , even if the eigenstates of the @xmath305 are localized , the time fluctuations of @xmath350 decay , and @xmath363 for large @xmath58 . this behavior for @xmath350 is similar to that of the green s functions @xmath264 for the disordered long - range hopping fermions analyzed in sec . [ sec : g - hop ] , where the infinite site summations lead to a cancellation of the time fluctuations of the various terms . let us discuss some of the most relevant recent papers appearing in the literature , in the light of what we have presented in our paper . a detailed analysis of the validity of gge averages for quantum quenches where the final hamiltonian was integrable , disorder - free , and translationally invariant ( the 1d quantum ising/@xmath201 spin chains and the luttinger model ) has been made by cazalilla _ et al . _ @xcite , showing that , for a general class of initial states @xmath185 , the time fluctuations of the one - body green s functions vanish and the gge averages are correct , in the thermodynamic limit , for both local and nonlocal observables . these results are in complete agreement with what we have shown here , since homogeneous hamiltonians have extended eigenstates and the time fluctuations of the one - body green s functions decay for @xmath48 . in our study , we have extended the analysis of ref . to quantum quenches with a final hamiltonian @xmath6 which is _ disordered _ : we have shown that the localization properties of @xmath6 are crucial for the relaxation of time fluctuations and , ultimately , also for the validity of the gge . in particular , we showed that , for one - body operators , infinite - time averages are always reproduced by gge , while for many - body operators , the localization of eigenstates of @xmath6 and the ensuing absence of relaxation of one - body real - space green s functions are , in principle , dangerous for the validity of gge . quantum quenches with integrable hamiltonians having a transition between extended and localized states have been analyzed in a recent work by gramsch _ et al . _ they have studied the aubry - andr model @xcite for hard - core bosons in a one - dimensional quasiperiodic potential , @xmath364 , where @xmath365 ( @xmath366 ) creates ( annihilates ) a hard - core boson at site @xmath57 , @xmath367 , @xmath368 is an irrational number , and @xmath369 is the strength of the quasiperiodic potential . this model can be diagonalized by mapping it , through jordan - wigner @xcite , onto a noninteracting spinless fermion chain with the same potential . the quasiperiodic on - site potential , in the absence of a true disorder , is able to induce a transition to a phase with localized eigenstates @xcite at a finite strength of @xmath370 . reference considered , among others , two operators that are particularly relevant for our discussion : the local density of bosons @xmath371 , a one - body operator in terms of jordan - wigner fermions , and the boson momentum distribution @xmath372 , which is , on the contrary , a many - body operator when written in terms of fermions , because of a jordan - wigner string @xcite . the numerical results of ref . show that when the eigenstates of the final hamiltonian are extended ( @xmath373 ) , the time fluctuations of both @xmath371 and @xmath374 vanish , and the gge predicts the time averages quite well , consistently with our analysis ; when the eigenstates of the final hamiltonian are localized ( @xmath375 ) the situation is more complex : the time fluctuations of @xmath371 do not relax but gge predicts well the time average ( again consistently with our analysis of sec . [ sec : gge - one - body ] ) , while the time fluctuations of @xmath374 appear to vanish , but gge seems to fail . the failure of gge in predicting time averages of a many - body observable like @xmath374 when persistent time fluctuations of the one - body green s functions are at play ( localized phase , @xmath375 ) is perfectly in line with our results ( see sec . [ sec : many - body - hop ] ) . what is definitely beyond our analysis , but not in contradiction with it , is the fact that the time fluctuations of @xmath374 relax for large @xmath28 : this is likely an effect of cancellation of fluctuations due to the summation of many terms , similar to what we have found for extensive operators ( see sec . [ sec : ising ] ) or for momentum space green s functions ( see sec . [ sec : g - hop ] ) . another paper quite relevant for our study is that of khatami _ et al . _ @xcite , where they analyze quenches with a final hamiltonian similar to our @xmath83 , eq . , supplemented by an interaction term @xmath376 which definitely breaks integrability . two comments are in order here . first , as shown in ref . , interactions do not change the picture dramatically : numerically , a metal - insulator transition occurs around @xmath377 , with a quite clear metallic phase for @xmath378 , and an insulating one for @xmath379 . second , by comparing after - quench time averages with the microcanonical average for the momentum distribution function @xmath380 and the density - density structure factor @xmath381 , ref . shows that quenches in the metallic phase ( @xmath378 ) are well described by the microcanonical ensemble , while thermalization appears to break down when quenching to the insulating phase ( @xmath379 ) . these results are definitely in line with what we have found , and suggest that , independently of the integrability of the hamiltonian , the localization properties of @xmath6 are crucial for the after - quench thermalization . finally , let us mention a technical point related to the relaxation of time fluctuations of general many - body operators of nonintegrable models @xcite with the technique we have used for one - body green s functions of free - particle models . in principle , one could compute @xmath382 for a general operator @xmath23 , starting from eq . . however , to make progress , one would need to stipulate something about gap degeneracies in the _ many - body _ spectrum , i.e. , @xmath383 with @xmath384 and @xmath385 ( apart from the trivial case @xmath386 and @xmath387 ) . ( the assumption of the absence of gap degeneracies is often used in the literature , and dates back to the original paper of von neumann @xcite , who , however , carefully stipulates it to hold only within each microcanonical energy shell , and not for the many - body spectrum at large . ) the condition of absence of gap degeneracies @xcite is clearly untenable for models with noninteracting quasiparticles : you can produce an exponentially large number of many - body states @xmath388 and @xmath389 whose spectral gap @xmath390 coincides exactly with @xmath391 : simply operate on @xmath36 and @xmath392 by applying , in identical fashion , an arbitrary number of particles and/or holes , @xmath393 , and @xmath394 . then @xmath395 because quasiparticles do not interact , and therefore @xmath396 . one might argue that this proliferation of exactly degenerate gaps is a peculiarity of models with noninteracting quasiparticles @xcite : interaction effects might change the picture completely . this is definitely an interesting point , which deserves further studies , but certainly also a very hard one , because the combination of disorder and interactions makes the analysis highly nontrivial . nevertheless , let us mention the following simple argument . suppose that quasiparticles interact , but there is still an exponentially large ( in the number of particles @xmath397 ) number of states with very small spectral gap differences @xmath398 : @xmath399 . then , all these spectral gap quasi - degeneracies will appear , effectively , as true degeneracies until a time @xmath400 is reached , and that time might indeed be very large . in conclusion , we have analyzed after - quench relaxation and thermalization issues for `` solvable '' models with disordered fermions ( or spins ) in one dimension . several points are still open and deserve further studies . let us just mention two of them : what is the role of a mobility edge , separating localized from extended states , in the after - quench relaxation ? what is the role of interactions in combination with disorder in the out - of - equilibrium dynamics of closed quantum systems ? a special thanks goes to alessandro silva , our former collaborator in a recent publication on the same subject , for many discussions and a careful reading of the manuscript . we also thank f. becca , m. fabrizio , j. marino , p. smacchia , g. menegoz , and a. russomanno for discussions .

we describe two ground based observing campaigns aimed at building a grid of approximately 200 spectrophotometric standard stars ( spss ) , with an internal% precision and tied to vega within% , for the absolute flux calibration of data gathered by gaia , the esa astrometric mission . the criteria for the selection and a list of candidates are presented , together with a description of the survey strategy and the adopted data analysis methods . we also discuss a short list of notable rejected spss candidates and difficult cases , based on identification problems , literature discordant data , visual companions , and variability . in fact , all candidates are also monitored for constancy ( within mmag , approximately ) . in particular , we report on a calspec standard , 1740346 , that we found to be a scuti variable during our short - term monitoring ( 12 h ) campaign . [ firstpage ] catalogs techniques : spectroscopic techniques : photometric stars : variables .

1207.6042

gaia is an esa ( european space agency ) all sky astrometric , photometric , and spectroscopic survey mission aimed at measuring parallaxes , proper motions , radial velocities , and astrophysical parameters of @xmath010@xmath3 stars ( @xmath01% of the galactic stellar population ) down to magnitude g@xmath020 , extracted from @xcite . ] , corresponding to v@xmath02025 mag , depending on spectral type . the astrometric accuracy is expected to be 514 @xmath4as for bright stars ( v@xmath512 mag ) , and to reach @xmath0300 @xmath4as down to v@xmath020 mag . radial velocities will be measured for stars brighter than v@xmath017 mag , depending on spectral type , and their precision will range from 1 km s@xmath6 for the bright stars down to 1520 km s@xmath6 for the faintest stars , bluer stars having higher uncertainties . the updated science performances of gaia can be found on the gaia esa webpage . the expected launch will be in august 2013 , from the esa launch site at kourou in french guiana . gaia will operate for approximately 5 years , with a possible 1 year extension , and the final catalogue is expected to be published 3 years after mission completion , while a set of intermediate releases is presently being defined . , height=264 ] although the primary scientific goal of gaia is the characterization of the milky way , its scientific impact will range from solar system studies to distant quasars , from unresolved galaxies to binaries , from supernovae to microlensing events , from fundamental physics to stellar variability . the wide variety of scientific topics is illustrated by almost 900 papers in ads ( astrophysics data system , of which more than 200 refereed ) to date , on a diversity of subjects , from the description of various mission components ( including software , pipelines , data treatment philosophy ) to simulations of the expected scientific harvest in many diverse areas . some papers summarize the expected science results ( see , e.g. , * ? ? ? * ) , but no single paper can be complete in this respect , given the huge range of possibilities opened by gaia . three main instruments can be found on board gaia , the af ( astrometric field ) , consisting of 62 ccds illuminated with white light , which will provide astrometric measurements and integrated gaia g - band magnitudes ( hereafter g ) ; the bp ( blue photometer ) and rp ( red photometer ) , consisting of two strips of 7 ccds each and providing prism dispersed , slitless spectra at a resolution of r=@xmath7@xmath020100 , covering the passbands shown in figure [ fig_bands ] and also providing integrated bp and rp magnitudes ( hereafter g@xmath8 and g@xmath9 ) and the g@xmath8g@xmath9 colour , which will be fundamental for chromaticity corrections of the astrometric measurements ; and the rvs ( radial velocity spectrograph ) , providing r@xmath011000 spectra in the calcium triplet region ( 84708740 ) projected onto 12 ccds . the mission output will thus be accurate positions , proper motions and parallaxes , low resolution bp / rp spectra , integrated g , g@xmath8 , and g@xmath9 magnitudes and the g@xmath8g@xmath9 colour , plus medium resolution rvs spectra and radial velocities for stars brighter than v@xmath017 mag . a classification of all observed objects will be performed on the basis of bp / rp and rvs spectra and when possible their parametrization will be performed as well , which for stars will provide t@xmath10 , log@xmath11 , e(b v ) , [ fe / h ] , and [ @xmath12/fe ] . although gaia is in principle a self - calibrating mission , some gaia measurements need to be tied to existing absolute reference systems , and many gaia algorithms need to be trained . thus extensive theoretical computations and observing campaigns are being carried out . to make a few examples : radial velocity standards that are stable to 1 km s@xmath6 are being obtained @xcite ; extended libraries of observed and theoretical spectra @xcite are being established ; the ecliptic poles that will be repeatedly observed in the initial calibration phase of gaia observations are being observed to produce catalogues of magnitudes and high - resolution spectra @xcite . also , the selection and analysis of reference stars ( and galaxies , quasars , asteroids , solar system objects and so on ) for the training of gaia algorithms is being carried on by different groups . this paper is the first of a series , which will present different aspects of the survey and of its data products . the series will include technical papers on the instrumental characterization , data papers presenting flux tables , photometric measurements , and lightcurves of our spss candidates , and scientific follow - up papers based on survey data and , when needed , on additional data . this paper presents the ongoing observational survey aimed at building the grid of spectrophotometric standard stars ( spss ) for the absolute flux calibration of gaia spectra and integrated magnitudes . the structure of the paper is the following : the gaia external calibration model is briefly illustrated in section [ sec - model ] ; the selection criteria and a list of candidate spss are presented in section [ sec - spss ] ; the observing campaigns and facilities are described in section [ sec - survey ] ; a description of the data treatment principles and methods can be found in section [ sec - reds ] ; and a set of preliminary results is presented in section [ sec - res ] . calibrating ( spectro)photometry obtained from the usual type of ground based observations ( broadband imaging , spectroscopy ) is not a trivial task , but the procedures are well known ( see , e.g. , * ? ? ? * ) and many groups have developed sets of appropriate standard stars for the more than 200 photometric known systems , and for spectroscopic observations . in the case of gaia , several instrumental effects much more complex than those usually encountered redistribute light along the sed ( spectral energy distribution ) of the observed objects . the most difficult gaia data to calibrate are the bp and rp slitless spectra , requiring a new approach to the derivation of the calibration model and to the spss grid needed to perform the actual calibration . some important complicating effects are : * the large focal plane with its large number of ccds makes it so that different observations of the same star will be generally on different ccds , with different quantum efficiencies , optical distorsions , transmissivity and so on . therefore , each wavelength and each position across the focal plane has its ( sometimes very different ) psf ( point spread function ) ; * tdi ( time delayed integration ) continuous reading mode , combined with the need of compressing most of the data before on - ground transmission , make it necessary to translate the full psf into a linear ( compressed into 1d ) lsf ( line spread function ) , which of course adds complication into the picture ; * in - flight instrument monitoring is foreseen , but never comparable to the full characterization that will be performed before launch , so the real instrument at a certain observation time will be slightly different from the theoretical one assumed initially , and this difference will change with time ; * finally , radiation damage ( or cti , charge transfer inefficiencies ) deserves special mention , for it is one of the most important factors in the time variation of the instrument model @xcite . it has particular impact onto the bp and rp dispersed images because the objects travel along the bp and rp ccd strips in a direction that is parallel to the spectral dispersion ( wavelength coordinate ) and therefore the net effect of radiation damage can be to alter the sed of some spectra . several solutions are being implemented to mitigate cti effects , but the global instrument complexity calls for a new approach to spectra flux calibrations . a flux calibration model is currently implemented in the photometric pipeline , which splits the calibration into an _ internal _ and an _ external _ part . the internal calibration model @xcite uses a large number of well behaved stars ( internal standards ) , observed by gaia , to report all observations to a _ reference _ instrument , on the same instrumental relative flux and wavelength scales . once each observation for each object is reported to the internal reference scales , the absolute or _ external _ calibration @xcite will use an appropriate spss set to report the relative flux scale to an absolute flux scale in physical units , tied to the calibration of vega ( see also section [ sec - spss ] ) . alternative approaches where the internal and external calibration steps are more inter - connected are being tested to maximise the precision and the accuracy of the gaia calibration @xcite . the gaia calibration model was also described by @xcite , @xcite , and @xcite . the final flux calibrated products will be : averaged ( on all transits or observations ) white light magnitudes , g ; integrated bp / rp magnitudes , g@xmath8 and g@xmath9 ; flux calibrated bp / rp spectra ; rvs spectra and integrated g@xmath13 magnitudes , possibly also flux calibrated @xcite . the g@xmath8g@xmath9 colour will be used to correct for chromaticity effects in the global astrometric solution . only for specific classes of objects , epoch spectra and magnitudes will be released , with variable stars as an obvious example . the external calibration model contains as discussed a large number of parameters , requiring a large number ( about 200 ) of calibrators . with the standard calibration techniques @xcite , the best possible calibrators are hot , almost featureless stars such as wd or hot subdwarfs . unfortunately , these stars are all similar to each other , forming an intrinsically degenerate set . the gaia calibration model instead requires to differentiate as much as possible the calibrators , by including smooth spectra , but also spectra with absorption features , both narrow ( atomic lines ) or wide ( molecular bands ) , appearing both on the blue and the red side of the spectrum . an experiment described by @xcite shows that the inclusion of just a few m stars with large molecular absorptions in the gaia spss set can improve the calibration of similarly red stars by a factor of more than ten ( from a formal error of 0.15 mag to an error smaller than 0.01 mag ) . 45 deg from the ecliptic poles ( roughly where gaia is observing more often ) are shaded in grey . our _ pillars _ are shown as three four - pointed stars , the _ primary spss candidates _ as six - pointed stars , and the _ secondary spss candidates _ as five - pointed stars . the stars size is proportional to the spss brightness , ranging from v@xmath08 ( largest symbols ) to 15 mag ( smallest symbols ) , approximately . ] in conclusion , the complexity of the instrument reflects in a complex calibration model , that requires a large set of homogeneously calibrated spss , covering a range of spectral types . no such database exists in the literature , and new observations are necessary to build it . the gaia photometric calibration model implies specific needs as it comes to _ ( i ) _ the selection criteria of the spss candidates and _ ( ii ) _ the characteristics of their flux tables ( i.e. , their calibrated spectra ) . the derived formal requirements @xcite define both the spss grid and the observing needs and can be summarized as follows : * spectral resolution r=@xmath14 , i.e. , they should oversample the gaia bp / rp resolution by a factor of 45 at least ; * wavelength coverage : 330010500 , corresponding to the full coverage of the bp and rp spectrophotometers ; * large sample ( approximately 200300 stars ) , covering different spectral types , although a large fraction should consist in hot stars , as featureless as possible ; * magnitude range 9@xmath5v@xmath515 mag : when observed by gaia they should ensure an end - of - mission s / n@xmath0100 over most of the wavelength range , without saturating ; * typical uncertainty on the absolute flux , with respect to the assumed calibration of vega @xcite of @xmath03% , excluding small troubled areas in the spectral range ( telluric bands residuals , extreme red and blue edges ) , where it can be somewhat worse ; * very homogeneous data treatment and quality , i.e. , the spss flux tables should have @xmath01% internal precision ; * photometric stability within @xmath15 mmag , necessary to ensure the above accuracy and precision . the calspec @xcite and the @xcite databases are very good starting points ( see also * ? ? ? * for further references ) , but new observations are needed . it is clear that if we add the requirements deriving from a ground - based campaign to the above ones , it becomes very difficult to assemble the grid in a relatively short time . therefore we decided to proceed in steps . the link between vega and our spss will be ensured by three _ pillars _ ( section [ sec - pillars ] ) ; these will enable to calibrate the _ primary spss _ ( section [ sec - primaries ] ) , our ground - based calibrators spread over the whole sky . the primary spss will in turn enable to calibrate our _ secondary spss _ ( section [ sec - secondaries ] ) , which constitute the actual gaia grid , together with the eligible _ primaries_. the basic principles of our calibration strategy were first outlined by @xcite . the sky distribution of our candidates is shown in figure [ fig_sky ] , while the magnitude and spectral type distributions are shown in figure [ fig_hist ] . more details on the selection criteria , sources , and candidate lists can be found in @xcite and @xcite . our three pillars are the calspec pillars and were selected from @xcite and @xcite . they are the da ( pure hydrogen atmosphere ) white dwarfs ( wd ) named g 191-b2b , gd 71 , and gd 153 , three well known and widely used standards . a fourth star from @xcite , hz 43 , was excluded from our list because it is member of a binary system . its companion , a dme star @xcite , at a distance of @xmath03@xmath15 , is brighter longward of @xmath07000 @xcite , and therefore not usable in our ground - based campaign , where the actual seeing ranges from @xmath00.5 `` up to @xmath162 '' in some cases and the slit width is 10``-12 '' for our spectra . the flux calibrated spectra of the _ pillars _ , available in the calspec database , are tied to the revised vega flux , and their flux calibrations are based on the comparison of wd model atmosphereserg @xmath17 s@xmath6 @xmath6 at 5556 . ] and spectra obtained with the faint object spectrograph ( fos ) aboard hst . the _ pillars _ are in the temperature range @xmath18 k and the fos spectrophotometry agrees with the model fluxes to within 2% over the whole uv - visible range . in addition , the simulated b and v magnitudes of the data agree to better than 1% with the landolt photometry @xcite . some of the most recent literature measurements for the three _ pillars _ are listed in table [ tab_pillars ] . .pillars[tab_pillars ] [ cols="<,^,^,^,^ , < " , ] [ sec - rej ] we discuss in the following sections some preliminary results of our survey : a few interesting cases of problematic candidates are described , and a list of notable rejected spss candidates can be found in table [ tab_rej ] ; two stars showed variability larger than @xmath15 mmag in our short - term constancy monitoring . identification problems are common , especially when large databases are automatically matched ( as done within simbad , for example ) , and when stars have large proper motions . we found our first case when a discrepancy became evident between the ltt 377 literature spectrum @xcite and our observed spectrum , which was more consistent with an f type rather than the expected k spectral type . we contacted the simbad and eso staff , because their sites reported the information from @xcite as well , and we concluded that the eso standard was not ltt 377 , but another star named cd -34 241 , of spectral type f. this was confirmed by older literature papers like @xcite , where ltt 377 was identified as cd -34 239 , and by literature proper motions and coordinates . we could trace back the error to @xcite , where the wrong association was probably done for the first time , and then propagated down to simbad and eso . the correct identification of both stars is shown in figure [ fig_ids1 ] ( left panel ) and [ fig_wd1148 ] were created with the eso skycat tool and images from the digitized sky survey . skycat was developed by eso s data management and very large telescope ( vlt ) project divisions with contributions from the canadian astronomical data center ( cadc ) . ] . we decided to keep both stars in our candidates lists ( see tables [ tab_primaries ] and [ tab_secondaries ] ) . a similar case was wd 0204 - 306 for which we obtained an unexpectedly red spectrum . we traced literature identifications back to @xcite , who correctly identified wd 0204 - 306 as associated with lp 885 - 23 ( an m star ) in a binary system , with a separation of 73 " . at some point , the two stars got confused and in simbad wd 0204 - 306 ( a white dwarf ) was cross - identified with lp 885 - 23 ( an m star ) . given the reported distance between the two stars , we identified wd 0204 - 306 as lp 885 - 22 , as shown in figure [ fig_ids1 ] . also in this case , having observations of both stars , we kept both in our _ secondary spss candidates_. the mistake was reported to the simbad staff and now the database is corrected . a more critical example was wd 1148 - 230 ( figure [ fig_wd1148 ] ) , having very different coordinates in the @xcite catalogue ( coming from * ? ? ? * and reporting r.a.=11:50:38.8 h and dec=23:20:34 deg ) and in simbad . the simbad coordinates were from the 2mass catalogue ( * ? ? ? * reporting r.a.=11:50:06.09 h and dec=23:16:14.0 deg ) . magnitudes were also significantly different . unlike in the previous cases , we had insufficient literature information to confirm one or the other identification , so we decided to reject this spss candidate , although we suspect that the mistake resides in the simbad automatic association between wd 1148 - 230 by @xcite and the 2mass catalogue . finally , we report on the case of wd 2256 + 313 , which was reported to have v=13.96 mag @xcite , but when observed in san pedro mrtir appeared to be much fainter than that ( and of uncertain spectral type , see also * ? ? ? * ) , possibly with v@xmath1615 mag , so was removed from our candidates list . in a few cases candidates that appeared relatively isolated on the available finding charts turned out to be in a crowded area where no aperture photometry or reliable wide slit spectroscopy could be performed from the ground , or showed previously unseen companions . generally , stars with high proper motion could appear isolated in some past finding chart , but later moved too close to another star to be safely observed from the ground . one example of candidate which appeared relatively isolated judging from the @xcite finding charts , but turned out to be in a crowded field when observed at san pedro mrtir was wd 1911 + 135 , that was promptly rejected , together with wd 1943 + 163 and wd 2046 + 396 . examples of candidates showing the presence of previously unknown and relatively bright companions were wd 0406 + 592 and wd 2058 + 181 ( figure [ fig_crowd ] ) . these stars do not have a particularly high proper motion , and appeared easy to identify on the corresponding finding charts , so we did not expect them to show close visual companions , when observed from the tng and san pedro mrtir , respectively . both stars were rejected . our auxiliary campaign started giving results as far as the short - term constancy monitoring ( 12 h ) is concerned . the ability of one lightcurve to detect magnitude variations is measured using the spread of reference star s magnitude differences . these appear as 1 , 2 , and 3 @xmath19 limits in figure [ fig_var ] , where we present the differential lightcurves our only _ confirmed _ variable star . 1 , 2 , and 3 @xmath19 variations are marked with dotted lines . ] star 1740346 , one of the currently used calspec standards and one of our _ primary spss _ candidates , showed variability with an amplitude of 10 @xmath1 0.8 mmag in b band when observed with bfosc@cassini in loiano , on 1 september 2010 ; with dolores@tng , on 31 september 2009 ; and with bfosc@cassini , on 26 may 2009 . the variability period is 50 min , approximately , thus showing properties typical of @xmath2 scuti variables . a preliminary determination of 1740346 parameters can be found in @xcite , using literature data and stellar models , resulting in a mass of @xmath01.3 m@xmath20 , an effective temperature of @xmath08300 k , and a distance of @xmath0750 pc . these parameters are also compatible with a @xmath2 scuti type star . we are gathering detailed follow - up observations and a complete characterization of star 1740346 will be the subject of a forthcoming paper ( marinoni et al . , in preparation ) . the differential lightcurve is presented in figure [ fig_var ] ( top panel ) . we have described a large ( more than 400 nights ) ground - based survey which started in 2007 and is expected to end in 20132014 , aimed at building a grid of spss for the flux calibration of gaia spectra and magnitudes . the technical complexity of gaia requires a large ( @xmath0200 ) set of spss flux tables , calibrated in flux with high precision ( @xmath01% ) and accuracy ( @xmath03% with respect to vega ) , and covering a range of spectral types . spss candidates need to be monitored for constancy ( within @xmath15 mmag ) to ensure the quoted precision in the final calibration . we discussed the adopted calibration strategy , the selection requirements and a list of candidate spss . a brief overview of the adopted data reduction and analysis procedures was also presented , and more details will be discussed in a series of future papers dealing with all technical aspects , data products , photometric catalogues , flux tables , and lightcurves . some preliminary results were presented , showing the data quality , a few problematic cases of candidate spss that were rejected because of identification problems , close companions , and variability . in particular , we detected a new variable star , a calspec standard which is most probably a @xmath2 scuti variable ; follow - up observations for its characterization are ongoing . all data products will be eventually made public together with each gaia data release , within the framework of the dpac ( data processing and analysis consortium ) publication policies . at the moment the accumulated data and literature information are stored locally and can be accessed upon request . we would like to acknowledge the support of the inaf ( istituto nazionale di astrofisica ) and specifically of the bologna observatory ; of the asi ( agenzia spaziale italiana ) under contracts to inaf i/037/08/0 and i/058/10/0 , dedicated to the gaia mission , and the italian participation to dpac ( data analysis and processing consortium ) . this work was supported by the /micinn/ ( spanish ministry of science and innovation ) feder through grant aya2009 - 14648-c02 - 01 and consolider csd2007 - 00050 . ep acknowledges the hospitality os the asdc ( asi sciece data center ) , where part of this work was carried out . we warmly thank the technical staff of the san pedro mrtir , calar alto , loiano , la silla ntt and rem , and roque de los muchachos tng observatories . we made use of the following softwares and online databases ( in alphabetical order ) : 2mass , calspec , cataxcorr , eso - dss , eso skycat tool , iraf , kawka webpage , miles , ngsl , saoimage ds9 , sdss and segue , sextractor , simbad , supermongo , ucac3 , usno catalogues , villanova white dwarf catalogue . we thank g. s. aldering , m. .a . barstow , m. e. kayser , and a. korn for sharing their information with us . we also thank m. bessell , who was the referee of this paper and provided extremely useful comments not only to improve the paper , but for the whole project . the survey presented in this paper relies on data obtained at eso ( proposals 182.d-0287 , 086.d-0176 , 087.d-0213 , and 089.d-0077 ) , calar alto ( proposals h07 - 2.2 - 024 , f08 - 2.2 - 043 , h08 - 2.2 - 041 , f10 - 2.2 - 027 , h10 - 2.2 - 042 , h10 - 2.2 - 042 , and f12 - 2.2 - 034 ) , tng ( proposals aot16_37 , aot17_3 , aot18_14 , aot19_14 , aot20_41 , and aot21_1 ) , loiano ( 10 accepted proposals starting from june 2007 ) , san pedro mrtir ( 7 accepted proposals starting from october 2007 ) , and rem ( proposals aot16_16012 , aot17_17012 , aot18_18002 , aot19_19010 , aot20_78 , aot21_2 , aot22_18 , aot23_7 , aot24_21 ) . federici l. , bragaglia a. , diolaiti e. , bellazzini m. , cacciari c. , pancino e. , altavilla g. , montegriffo p. , rossetti e. , jordi c. , figueras f. , trager s. , 2006 , gaia technical report no . gaia - c5-tn - oabo - lf-001 + star & ra ( j2000 ) & dec ( j2000 ) & b & v & type & star & ra ( j2000 ) & dec ( j2000 ) & b & v & type + & ( hh : mm : ss ) & ( dd : pp : ss ) & ( mag ) & ( mag ) & & & ( hh : mm : ss ) & ( dd : pp : ss ) & ( mag ) & ( mag ) & + wd 2359 - 434 & 00:02:10.77@xcite compilation and online updates ; @xmath21sdss seventh data release @xcite ; @xmath22sdss , derived with the segue pipeline @xcite and the transformations by @xcite ; @xmath23sdss , derived with the segue pipeline @xcite ; @xmath24@xcite ; @xmath25@xcite ; @xmath26@xcite ; @xmath27tycho-2 catalogue of bright sources @xcite ; @xmath28@xcite ; @xmath29from t@xmath30 by @xcite ; @xmath31@xcite ; @xmath32@xcite ; @xmath33@xcite , for approximate johnson magnitudes the formulae v = vt0.090*(bt vt ) and b v=0.850*(bt vt ) where used ; @xmath34@xcite ; @xmath35@xcite ; @xmath36@xcite ; @xmath37subdwarf database " @xcite ; @xmath38misclassified as a wd by @xcite according to @xcite ; @xmath39from t@xmath30 by @xcite ; @xmath40@xcite ; @xmath41henry draper catalogue @xcite ; @xmath42@xcite ; @xmath43@xcite ; @xmath44ucac3 @xcite ; @xmath45hipparcos input catalogue @xcite ; @xmath46@xcite ; @xmath47@xcite ; @xmath48@xcite ; @xmath49from t@xmath30 by @xcite ; @xmath50@xcite ; @xmath51@xcite ; @xmath52@xcite ; @xmath53@xcite ; @xmath54@xcite ; @xmath55@xcite ; @xmath56@xcite ; @xmath57@xcite ; @xmath58@xcite ; @xmath59@xcite ; @xmath60@xcite ; @xmath61@xcite ; @xmath627th sdss photometric data release @xcite ; @xmath63@xcite ; @xmath64@xcite ; @xmath65@xcite ; @xmath66@xcite ; @xmath67@xcite ; @xmath68@xcite ; @xmath69@xcite ; @xmath70@xcite ; @xmath71@xcite ; @xmath72@xcite ; @xmath73@xcite ; @xmath74@xcite ; @xmath75@xcite ; @xmath76stetson standard in m 5 @xcite ; data available at http://cadcwww.dao.nrc.ca/community/stetson/standards ; @xmath77@xcite ; @xmath78possible identification problem , see also section [ sec - id ] . ] & 43:09:56.02@xmath79 & 13.12@xmath80 & 13.05@xmath80 & da5@xmath80 & hd 271759 & 06:00:41.34@xmath33 & 66:03:14.03@xmath33 & 11.00@xmath27 & 11.20@xmath27 & a2@xmath41 + wd 0004 + 330 & 00:07:32.26@xmath79 & + 33:17:27.60@xmath79 & 13.57@xmath80 & 13.85@xmath80 & da1@xmath80 & hd 271783 & 06:02:11.36@xmath27 & 66:34:59.13@xmath27 & 12.63@xmath27 & 12.23@xmath27 & f5@xmath41 + sdss 03932 & 00:07:52.22@xmath21 & + 14:30:24.72@xmath21 & 15.37@xmath22 & 15.07@xmath22 & a0@xmath23 & hip 28618 & 06:02:27.88@xmath24 & 66:47:28.68@xmath24 & 12.20@xmath27 & 12.30@xmath27 & b8@xmath41 + wd 0009 + 501 & 00:12:14.80@xmath79 & + 50:25:21.40@xmath79 & 14.78@xmath80 & 14.36@xmath80 & da8@xmath80 & wd0604 - 203 & 06:06:13.39@xmath79 & 20:21:07.20@xmath79 & 11.75@xmath42 & 11.80@xmath42 & da@xmath42 + wd 0018 - 267 & 00:21:30.73@xmath79 & 26:26:11.46@xmath79 & & 13.80@xmath80 & da9@xmath80 & wd0621 - 376 & 06:23:12.63@xmath79 & 37:41:28.01@xmath79 & 11.76@xmath80 & 12.09@xmath80 & da1@xmath80 + sdss 03532 & 00:24:38.62@xmath21 & 01:11:39.75@xmath21 & 15.14@xmath22 & 15.04@xmath22 & a0@xmath23 & wd0644 + 375 & 06:47:37.99@xmath24 & + 37:30:57.07@xmath24 & 11.99@xmath80 & 12.08@xmath80 & da2@xmath80 + wd 0038 + 555 & 00:41:21.99@xmath79 & + 55:50:08.40@xmath79 & 14.10@xmath80 & 14.08@xmath80 & dq5@xmath80 & wd0646 - 253 & 06:48:56.09@xmath79 & 25:23:47.00@xmath79 & 13.30@xmath80 & 13.40@xmath80 & da2@xmath80 + ltt 377 & 00:41:30.47@xmath24 & 33:37:32.03@xmath24 & 11.97@xmath25 & 10.53@xmath25 & k9@xmath26 & g193 - 26 & 07:03:26.29@xmath79 & + 54:52:06.00@xmath79 & 13.59@xmath43 & 13.02@xmath43 & g@xmath39 + wd 0046 + 051 & 00:49:09.90@xmath24 & + 05:23:19.01@xmath24 & 12.93@xmath80 & 12.39@xmath80 & dz7@xmath80 & wd0713 + 584 & 07:17:36.26@xmath24 & + 58:24:20.51@xmath24 & 12.06@xmath27 & 12.02@xmath27 & da4@xmath80 + wd 0047 - 524 & 00:50:03.68@xmath79 & 52:08:15.60@xmath79 & 14.19@xmath80 & 14.20@xmath80 & da2@xmath80 & wd0721 - 276 & 07:23:20.10@xmath79 & 27:47:21.60@xmath79 & 13.50@xmath80 & 13.40@xmath80 & da1@xmath80 + wd 0050 - 332 & 00:53:17.44@xmath79 & 32:59:56.60@xmath79 & 13.11@xmath80 & 13.36@xmath80 & da1@xmath80 & wd0749 - 383 & 07:51:32.58@xmath44 & 38:28:36.41@xmath44 & 13.53@xmath80 & 13.66@xmath80 & da@xmath80 + wd 0104 - 331 & 01:06:46.86@xmath79 & 32:53:12.45@xmath79 & 13.28@xmath80 & 13.57@xmath80 & daz3@xmath80 & g251 - 54 & 08:11:06.24@xmath24 & + 79:54:29.57@xmath24 & 10.58@xmath45 & 10.01@xmath45 & g0@xmath45 + wd 0106 - 358 & 01:08:20.80@xmath80 & 35:34:43.00@xmath80 & 14.54@xmath80 & 14.72@xmath80 & da2@xmath80 & gj2066 & 08:16:07.98@xmath24 & + 01:18:09.26@xmath24 & 11.63@xmath25 & 10.09@xmath25 & m2@xmath31 + wd 0109 - 264 & 01:12:11.65@xmath27 & 26:13:27.69@xmath27 & 12.91@xmath80 & 13.15@xmath80 & da1@xmath80 & g114 - 25 & 08:59:03.37@xmath24 & 06:23:46.19@xmath24 & 12.52@xmath46 & 11.97@xmath46 & f7@xmath47 + wd 0123 - 262 & 01:25:24.45@xmath79 & 26:00:43.90@xmath79 & 15.35@xmath80 & 14.95@xmath80 & dc@xmath80 & wd0859 - 039 & 09:02:17.30@xmath79 & 04:06:55.45@xmath79 & 13.02@xmath80 & 13.19@xmath80 & da2@xmath80 + g245 - 31 & 01:38:39.39@xmath79 & + 69:38:01.50@xmath79 & 15.26@xmath28 & 14.50@xmath28 & k@xmath29 & wd0912 + 536 & 09:15:56.23@xmath79 & + 53:25:24.90@xmath79 & 14.19@xmath80 & 13.85@xmath80 & db / dc@xmath80 + wd 0134 + 833 & 01:41:28.74@xmath79 & + 83:34:58.90@xmath79 & 12.88@xmath80 & 13.11@xmath80 & da2@xmath80 & wd0943 + 441 & 09:46:39.08@xmath79 & + 43:54:52.37@xmath79 & 13.19@xmath80 & 13.12@xmath80 & da4@xmath80 + gj70 & 01:43:20.18@xmath24 & + 04:19:17.97@xmath24 & 12.45@xmath25 & 10.92@xmath25 & m2@xmath31 & g43 - 5 & 09:49:51.59@xmath27 & + 06:36:35.64@xmath27 & 12.90@xmath48 & 12.48@xmath48 & k@xmath49 + g72 - 34 & 01:46:03.66@xmath79 & + 35:54:49.40@xmath79 & 13.84@xmath28 & 12.98@xmath28 & k@xmath29 & wd0954 - 710 & 09:55:22.89@xmath79 & 71:18:08.31@xmath79 & 13.60@xmath80 & 13.48@xmath80 & da4@xmath80 + wd 0147 + 674 & 01:51:10.29@xmath79 & + 67:39:31.30@xmath79 & 14.17@xmath80 & 14.42@xmath80 & da2@xmath80 & g236 - 30 & 10:28:48.37@xmath79 & + 62:59:45.00@xmath79 & 13.62@xmath34 & 12.87@xmath34 & g5@xmath34 + wd 0148 + 467 & 01:52:02.96@xmath24 & + 47:00:06.65@xmath24 & 12.50@xmath80 & 12.44@xmath80 & da3@xmath80 & wd1029 + 537 & 10:32:10.26@xmath50 & + 53:29:36.40@xmath50 & 14.18@xmath80 & 14.46@xmath80 & da1@xmath80 + wd 0204 - 306@xmath78 & 02:07:02.28@xmath79 & 30:23:32.20@xmath79 & & 16.18@xmath32 & da@xmath80 & wd1031 - 114 & 10:33:42.76@xmath44 & 11:41:38.35@xmath44 & 12.85@xmath80 & 13.03@xmath80 & da2@xmath80 + lp885 - 23@xmath78 & 02:07:06.33@xmath79 & 30:24:22.90@xmath79 & & 13.06@xmath32 & m0@xmath32 & wd1034 + 001 & 10:37:03.81@xmath79 & 00:08:19.30@xmath79 & 12.86@xmath51 & 13.23@xmath51 & doz1@xmath80 + wd 0214 + 568 & 02:17:33.52@xmath79 & + 57:06:47.50@xmath79 & 13.56@xmath80 & 13.68@xmath80 & da2@xmath80 & wd1041 + 580 & 10:44:46.10@xmath52 & + 57:44:35.00@xmath52 & 14.37@xmath80 & 14.60@xmath80 & da1@xmath80 + wd 0227 + 050 & 02:30:16.62@xmath24 & + 05:15:50.68@xmath24 & 12.75@xmath25 & 12.80@xmath25 & da3@xmath80 & wd1053 - 550 & 10:55:13.54@xmath79 & 55:19:05.20@xmath79 & 14.42@xmath80 & 14.32@xmath80 & da4@xmath80 + wd 0302 + 621 & 03:06:16.69@xmath79 & + 62:22:22.68@xmath79 & 15.17@xmath80 & 14.95@xmath80 & da4/6@xmath80 & wd1056 - 384 & 10:58:20.11@xmath79 & 38:44:25.10@xmath79 & 13.86@xmath53 & 14.05@xmath53 & da2@xmath80 + wd 0316 - 849 & 03:09:59.89@xmath33 & 84:43:21.14@xmath33 & 11.62@xmath33 & 10.55@xmath33 & dah@xmath80 & g146 - 76 & 10:59:57.48@xmath27 & + 44:46:43.75@xmath27 & 11.15@xmath27 & 10.47@xmath27 & g / k @xmath39 + g174 - 44 & 03:17:23.31@xmath79 & + 52:17:42.40@xmath79 & 14.49@xmath34 & 13.75@xmath34 & k0@xmath35 & wd1104 + 602 & 11:07:42.80@xmath79 & + 59:58:29.90@xmath79 & 13.78@xmath80 & 13.80@xmath80 & da3@xmath80 + hg7 - 15 & 03:48:11.86@xmath24 & + 07:08:46.47@xmath24 & 12.11@xmath48 & 10.65@xmath48 & m1@xmath77 & wd1105 - 048 & 11:07:59.95@xmath79 & 05:09:25.90@xmath79 & 13.09@xmath51 & 13.06@xmath51 & da3@xmath80 + wd 0435 - 088 & 04:37:47.42@xmath36 & 08:49:10.70@xmath36 & 14.10@xmath80 & 13.77@xmath80 & dq7@xmath80 & g10 - 4 & 11:10:60.00@xmath24 & + 06:25:11.51@xmath24 & 12.13@xmath54 & 11.41@xmath54 & k@xmath39 + wd 0446 - 789 & 04:43:46.47@xmath79 & 78:51:50.40@xmath79 & 13.36@xmath80 & 13.47@xmath80 & da3@xmath80 & g254 - 24 & 11:32:23.31@xmath24 & + 76:39:18.03@xmath24 & 12.18@xmath55 & 11.53@xmath55 & g0@xmath35 + wd 0447 + 176 & 04:50:13.52@xmath24 & + 17:42:06.21@xmath24 & 12.63@xmath36 & 12.65@xmath37 & sdo@xmath38 & wd1134 + 300 & 11:37:05.10@xmath24 & + 29:47:58.34@xmath24 & 12.41@xmath53 & 12.49@xmath53 & da2@xmath80 + wd 0455 - 282 & 04:57:13.90@xmath80 & 28:07:54.00@xmath80 & 13.63@xmath80 & 13.95@xmath80 & da1@xmath80 & sdss 09310 & 11:38:02.62@xmath21 & + 57:29:23.89@xmath21 & 15.24@xmath22 & 14.99@xmath22 & a0/f3@xmath23 + wd 0501 - 289 & 05:03:55.51@xmath80 & 28:54:34.57@xmath80 & 13.55@xmath80 & 13.90@xmath80 & do@xmath80 & g10 - 54 & 11:49:48.20@xmath79 & + 06:08:52.14@xmath79 & 13.17@xmath56 & 12.57@xmath56 & g@xmath39 + g191 - 52 & 05:44:43.55@xmath79 & + 56:15:30.80@xmath79 & 14.02@xmath34 & 13.26@xmath34 & g@xmath39 & wd1153 - 484 & 11:56:11.43@xmath79 & 48:40:03.18@xmath79 & 12.65@xmath80 & 12.85@xmath80 & da2@xmath80 + u1050 - 027792 & 05:52:18.18@xmath79 & + 15:51:52.70@xmath79 & 14.42@xmath40 & 13.70@xmath40 & & wd1210 + 533 & 12:13:24.64@xmath79 & + 53:03:57.36@xmath79 & 13.78@xmath80 & 14.12@xmath80 & dao1@xmath80 + wd 0552 - 041 & 05:55:09.53@xmath36 & 04:10:07.10@xmath36 & 15.50@xmath80 & 14.45@xmath80 & dc / dz@xmath80 & wd1211 - 169 & 12:14:10.53@xmath33 & 17:14:20.19@xmath33 & 11.04@xmath34 & 10.13@xmath34 & dah@xmath80 + hd 270422 & 05:56:47.74@xmath33 & 66:39:05.27@xmath33 & 10.92@xmath27 & 10.05@xmath27 & g0@xmath41 & gj459.3 & 12:19:24.09@xmath24 & + 28:22:56.52@xmath24 & 12.06@xmath45 & 10.62@xmath45 & m2@xmath45 + hd 270477 & 05:59:33.36@xmath33 & 67:01:13.72@xmath33 & 10.73@xmath27 & 10.28@xmath27 & f8@xmath41 & sdss 12720 & 12:22:41.66@xmath21 & + 42:24:43.66@xmath21 & 15.18@xmath22 & 15.04@xmath22 & a0/f2@xmath23 + hd 271747 & 05:59:58.62@xmath27 & 66:06:08.91@xmath27 & 11.82@xmath27 & 11.29@xmath27 & g0@xmath41 & wd1223 - 659 & 12:26:42.02@xmath79 & 66:12:18.70@xmath79 & 14.37@xmath80 & 13.97@xmath80 & da7@xmath80 + wd1234 + 481 & 12:36:45.18@xmath79 & + 47:55:22.34@xmath79 & 14.09@xmath80 & 14.42@xmath80 & da1@xmath80 & wd2047 + 372 & 20:49:06.69@xmath79 & + 37:28:13.90@xmath79 & 13.07@xmath80 & 12.93@xmath80 & da3@xmath80 + sa 104 - 428 & 12:41:41.28@xmath79 & 00:26:26.20@xmath79 & 13.58@xmath51 & 12.63@xmath51 & g8@xmath57 & wd2111 + 498 & 21:12:44.05@xmath79 & + 50:06:17.80@xmath79 & 12.84@xmath80 & 13.08@xmath80 & da1@xmath80 + sa 104 - 490 & 12:44:33.46@xmath79 & 00:25:51.70@xmath79 & 13.07@xmath51 & 12.57@xmath51 & g3@xmath58 & wd2105 - 820 & 21:13:13.90@xmath80 & 81:49:04.00@xmath80 & 13.82@xmath80 & 13.61@xmath80 & da5@xmath80 + g14 - 24 & 13:02:01.58@xmath79 & 02:05:21.42@xmath79 & 13.52@xmath46 & 12.81@xmath46 & k0@xmath39 & wd2111 + 261 & 21:13:45.93@xmath79 & + 26:21:33.20@xmath79 & 14.92@xmath80 & 14.68@xmath80 & da6@xmath80 + gj2097 & 13:07:04.31@xmath44 & + 20:48:38.54@xmath44 & 14.10@xmath59 & 12.54@xmath59 & m1@xmath31 & wd2117 + 539 & 21:18:56.27@xmath27 & + 54:12:41.25@xmath27 & 12.40@xmath80 & 12.33@xmath80 & da3@xmath80 + sdss 08393 & 13:10:32.07@xmath21 & + 54:18:33.66@xmath21 & 15.30@xmath22 & 15.08@xmath22 & a0/f3@xmath23 & wd2115 - 560 & 21:19:36.52@xmath79 & 55:50:14.20@xmath79 & 14.43@xmath80 & 14.28@xmath80 & daz5@xmath80 + gj507.1 & 13:19:40.13@xmath24 & + 33:20:47.49@xmath24 & 12.10@xmath48 & 10.57@xmath48 & m2@xmath31 & wd2122 + 282 & 21:24:58.30@xmath80 & + 28:26:05.00@xmath80 & 13.80@xmath80 & 14.00@xmath80 & da0@xmath74 + wd1319 + 466 & 13:21:15.08@xmath79 & + 46:23:23.68@xmath79 & 14.55@xmath80 & 14.55@xmath80 & da3@xmath80 & wd2136 + 828 & 21:33:43.25@xmath79 & + 83:03:32.40@xmath79 & 13.01@xmath80 & 13.02@xmath79 & da3@xmath80 + wd1323 - 514 & 13:26:09.65@xmath79 & 51:41:35.78@xmath79 & 14.60@xmath80 & 14.60@xmath80 & da2@xmath80 & wd2134 + 218 & 21:36:36.30@xmath80 & + 22:04:33.00@xmath80 & 14.41@xmath80 & 14.45@xmath80 & da3@xmath80 + wd1327 - 083 & 13:30:13.64@xmath24 & 08:34:29.49@xmath24 & 12.40@xmath25 & 12.34@xmath25 & da4@xmath80 & wd2140 + 207 & 21:42:42.00@xmath79 & + 20:59:58.24@xmath79 & 13.40@xmath80 & 13.24@xmath80 & dq6@xmath80 + gj521 & 13:39:24.10@xmath24 & + 46:11:11.37@xmath24 & 11.50@xmath27 & 10.26@xmath27 & m2@xmath31 & wd2147 + 280 & 21:49:54.53@xmath79 & + 28:16:59.80@xmath79 & 14.66@xmath80 & 14.68@xmath80 & db4@xmath80 + wd1408 + 323 & 14:10:26.95@xmath79 & + 32:08:36.10@xmath79 & 13.96@xmath80 & 13.97@xmath80 & da3@xmath80 & wd2152 - 548 & 21:56:21.27@xmath79 & 54:38:23.00@xmath79 & 13.80@xmath80 & 14.30@xmath80 & da1@xmath80 + sdss 09626 & 14:29:51.06@xmath21 & + 39:28:25.43@xmath21 & 15.23@xmath22 & 14.99@xmath22 & a0@xmath23 & gj851 & 22:11:30.09@xmath24 & + 18:25:34.29@xmath24 & 11.37@xmath25 & 10.23@xmath25 & m2@xmath31 + gj570.2 & 14:57:32.30@xmath24 & + 31:23:44.61@xmath24 & 12.68@xmath48 & 11.54@xmath48 & m2@xmath31 & wd2211 - 495 & 22:14:11.91@xmath27 & 49:19:27.26@xmath27 & 11.37@xmath80 & 11.71@xmath80 & da1@xmath80 + g15 - 10 & 15:09:46.02@xmath24 & 04:45:06.61@xmath24 & 12.67@xmath45 & 12.01@xmath45 & g2@xmath60 & wd2216 - 657 & 22:19:48.35@xmath79 & 65:29:18.11@xmath79 & 14.57@xmath80 & 14.43@xmath80 & dz5@xmath80 + wd1509 + 322 & 15:11:27.66@xmath79 & + 32:04:17.80@xmath79 & 14.20@xmath80 & 14.11@xmath80 & da3@xmath80 & gj863 & 22:33:02.23@xmath24 & + 09:22:40.70@xmath24 & 11.91@xmath25 & 10.74@xmath25 & m0@xmath31 + m5-s1490 & 15:17:38.64@xmath76 & + 02:02:25.60@xmath76 & 15.08@xmath76 & 14.10@xmath76 & & sdss 14276 & 22:42:04.17@xmath21 & + 13:20:28.61@xmath21 & 14.48@xmath22 & 14.32@xmath22 & a0@xmath23 + g167 - 50 & 15:35:31.55@xmath79 & + 27:51:02.20@xmath79 & 14.25@xmath34 & 13.50@xmath34 & g@xmath61 & wd2251 - 634 & 22:55:10.00@xmath80 & 63:10:27.00@xmath80 & & 14.28@xmath80 & da@xmath80 + g179 - 54 & 15:46:08.25@xmath79 & + 39:14:16.40@xmath79 & 13.90@xmath61 & 13.41@xmath61 & f@xmath61 & wd2309 + 105 & 23:12:21.62@xmath44 & + 10:47:04.25@xmath44 & 12.78@xmath51 & 13.09@xmath51 & da1@xmath80 + g224 - 83 & 15:46:14.68@xmath79 & + 62:26:39.60@xmath79 & 13.86@xmath34 & 12.67@xmath34 & k@xmath61 & g190 - 15 & 23:13:38.82@xmath24 & + 39:25:02.59@xmath24 & 11.57@xmath48 & 10.98@xmath48 & f6@xmath47 + wd1553 + 353 & 15:55:01.99@xmath79 & + 35:13:28.70@xmath79 & 14.64@xmath80 & 14.75@xmath80 & da2@xmath80 & sdss 00832 & 23:30:24.90@xmath21 & 00:09:34.90@xmath21 & 15.15@xmath22 & 14.99@xmath22 & a0@xmath23 + g16 - 20 & 15:58:18.62@xmath27 & + 02:03:06.11@xmath27 & 11.34@xmath61 & 10.75@xmath61 & k@xmath39 & wd2329 + 407 & 23:31:35.65@xmath79 & + 41:01:30.70@xmath79 & 13.85@xmath80 & 13.82@xmath80 & da3@xmath80 + wd1606 + 422 & 16:08:22.20@xmath79 & + 42:05:43.20@xmath79 & 13.93@xmath80 & 13.82@xmath80 & da4@xmath80 & wd2331 - 475 & 23:34:02.20@xmath79 & 47:14:26.50@xmath79 & 13.15@xmath21 & 13.44@xmath80 & da1@xmath80 + wd1615 - 154 & 16:17:55.26@xmath79 & 15:35:51.90@xmath79 & 13.22@xmath80 & 13.42@xmath80 & da2@xmath80 & g241 - 64 & 23:41:24.49@xmath79 & + 59:24:34.90@xmath79 & 13.45@xmath34 & 12.70@xmath34 & k@xmath39 + gj625 & 16:25:24.62@xmath24 & + 54:18:14.77@xmath24 & 11.80@xmath27 & 10.17@xmath27 & m2@xmath31 & g171 - 15 & 23:45:02.71@xmath27 & + 44:40:03.60@xmath27 & 12.00@xmath27 & 11.75@xmath27 & g0@xmath75 + g180 - 58 & 16:28:16.87@xmath24 & + 44:40:38.28@xmath24 & 11.87@xmath48 & 11.12@xmath48 & g / k@xmath39 & wd2352 + 401 & 23:54:56.25@xmath79 & + 40:27:30.10@xmath79 & 15.13@xmath80 & 14.94@xmath80 & dq6@xmath80 + wd1626 + 368 & 16:28:25.03@xmath79 & + 36:46:15.40@xmath79 & 14.02@xmath80 & 13.83@xmath80 & dza6@xmath80 + wd1637 + 335 & 16:39:27.83@xmath44 & + 33:25:22.30@xmath44 & 14.85@xmath80 & 14.65@xmath80 & da5@xmath80 + sdss 13028 & 16:40:24.18@xmath21 & + 24:02:14.91@xmath21 & 15.45@xmath22 & 15.26@xmath22 & a0@xmath23 + wd1659 - 531 & 17:02:56.33@xmath62 & 53:14:36.63@xmath62 & 13.57@xmath80 & 13.47@xmath80 & da4@xmath80 + g139 - 16 & 17:09:47.38@xmath79 & + 08:04:25.50@xmath79 & 13.31@xmath43 & 12.61@xmath43 & k@xmath39 + g170 - 47 & 17:32:41.63@xmath24 & + 23:44:11.64@xmath24 & 9.54@xmath27 & 8.94@xmath27 & g0@xmath47 + 2mass j175713 & 17:57:13.25@xmath79 & + 67:03:40.90@xmath79 & 11.91@xmath79 & 12.01@xmath79 & a3@xmath63 + tyc4213 - 617 & 18:00:02.14@xmath33 & + 66:45:54.96@xmath33 & 11.24@xmath27 & 10.68@xmath27 & + bd+661071 & 18:02:10.92@xmath33 & + 66:12:26.39@xmath33 & 10.93@xmath27 & 10.52@xmath27 & f5@xmath61 + g184 - 17 & 18:40:29.27@xmath79 & + 19:36:06.65@xmath79 & 14.90@xmath46 & 14.08@xmath46 & k@xmath39 + wd1837 - 619 & 18:42:29.73@xmath64 & 61:51:45.10@xmath64 & 15.01@xmath80 & 14.90@xmath80 & dc5@xmath80 + g184 - 20 & 18:43:52.50@xmath79 & + 16:00:34.20@xmath79 & 13.37@xmath65 & 12.61@xmath66 & g@xmath39 + wd1845 + 019 & 18:47:39.08@xmath79 & + 01:57:35.62@xmath79 & 12.73@xmath80 & 12.95@xmath80 & da2@xmath80 + wd1900 + 705 & 19:00:10.25@xmath79 & + 70:39:51.24@xmath79 & 13.24@xmath80 & 13.19@xmath80 & dap4@xmath27 + gj745a & 19:07:05.56@xmath24 & + 20:53:16.97@xmath24 & 12.40@xmath25 & 10.77@xmath25 & m2@xmath31 + gj745b & 19:07:13.20@xmath24 & + 20:52:37.24@xmath24 & 12.38@xmath25 & 10.77@xmath25 & m2@xmath31 + wd1918 + 725 & 19:18:10.5@xmath80 & + 72:37:24.00@xmath80 & 14.70@xmath67 & 15.12@xmath67 & da2@xmath80 + wd1914 - 598 & 19:18:44.84@xmath79 & 59:46:33.80@xmath79 & 14.34@xmath80 & 14.39@xmath80 & da@xmath80 + wd1919 + 145 & 19:21:40.40@xmath80 & + 14:40:43.00@xmath80 & 13.07@xmath80 & 13.01@xmath80 & da3@xmath80 + wd1936 + 327 & 19:38:28.21@xmath79 & + 32:53:19.90@xmath79 & 13.46@xmath80 & 13.58@xmath80 & da2@xmath80 + g23 - 14 & 19:51:49.61@xmath24 & + 05:36:45.84@xmath24 & 11.42@xmath68 & 11.02@xmath68 & g5@xmath45 + wd2000 - 561.1 & 20:04:18.00@xmath80 & 56:02:47.00@xmath80 & & 15.20@xmath69 & da1@xmath80 + wd2004 - 605 & 20:09:05.24@xmath70 & 60:25:41.60@xmath70 & 13.10@xmath80 & 13.40@xmath80 & da1@xmath80 + wd2014 - 575 & 20:18:54.90@xmath71 & 57:21:34.00@xmath71 & 13.40@xmath80 & 13.70@xmath80 & da2@xmath80 + wd2028 + 390 & 20:29:56.16@xmath79 & + 39:13:32.00@xmath79 & 13.22@xmath80 & 13.37@xmath80 & da2@xmath80 + wd2032 + 248 & 20:34:21.88@xmath24 & + 25:03:49.72@xmath24 & 11.47@xmath25 & 11.55@xmath25 & da2@xmath80 + wd2034 - 532 & 20:38:16.84@xmath79 & 53:04:25.40@xmath79 & 14.41@xmath80 & 14.46@xmath80 & db4@xmath80 + g24 - 25 & 20:40:16.10@xmath27 & + 00:33:19.74@xmath27 & 11.23@xmath27 & 10.61@xmath27 & g0@xmath72 + wd2039 - 202 & 20:42:34.75@xmath24 & 20:04:35.95@xmath24 & 12.32@xmath25 & 12.40@xmath25 & da3@xmath80 + sdss 14511 & 20:42:42.40@xmath21 & 00:34:03.71@xmath21 & 15.34@xmath22 & 15.11@xmath22 & a0/f0@xmath23 + wd2039 - 682 & 20:44:21.47@xmath73 & 68:05:21.30@xmath73 & 13.19@xmath80 & 13.25@xmath80 & da3@xmath80 + sdss 15724 & 20:47:38.19@xmath21 & 06:32:13.11@xmath21 & 15.06@xmath22 & 14.87@xmath22 & a0/f2@xmath23 +

adaptive networks rely on in - network and collaborative processing among distributed agents to deliver enhanced performance in estimation and inference tasks . information is exchanged among the nodes , usually over noisy links . the combination weights that are used by the nodes to fuse information from their neighbors play a critical role in influencing the adaptation and tracking abilities of the network . this paper first investigates the mean - square performance of general adaptive diffusion algorithms in the presence of various sources of imperfect information exchanges , quantization errors , and model non - stationarities . among other results , the analysis reveals that link noise over the regression data modifies the dynamics of the network evolution in a distinct way , and leads to biased estimates in steady - state . the analysis also reveals how the network mean - square performance is dependent on the combination weights . we use these observations to show how the combination weights can be optimized and adapted . simulation results illustrate the theoretical findings and match well with theory . diffusion adaptation , adaptive networks , imperfect information exchange , tracking behavior , diffusion lms , combination weights , energy conservation .

1112.6212

adaptive network consists of a collection of agents that are interconnected to each other and solve distributed estimation and inference problems in a collaborative manner . two useful strategies that enable adaptation and learning over such networks in real - time are the incremental strategy @xcite and the diffusion strategy @xcite . incremental strategies rely on the use of a hamiltonian cycle , i.e. , a cyclic path that covers all nodes in the network , which is generally difficult to enforce since determining a hamiltonian cycle is an np - hard problem . in addition , cyclic trajectories are not robust to node or link failure . in comparison , diffusion strategies are scalable , robust , and able to match well the performance of incremental networks . in adaptive diffusion implementations , information is processed locally at the nodes and then diffused in real - time across the network . diffusion strategies were originally proposed in @xcite and further extended and studied in @xcite . they have been applied to model self - organized and complex behavior encountered in biological networks , such as fish schooling @xcite , bird flight formations @xcite , and bee swarming @xcite . diffusion strategies have also been applied to online learning of gaussian mixture models @xcite and to general distributed optimization problems @xcite . there have also been several useful works in the literature on distributed consensus - type strategies , with application to multi - agent formations and distributed processing @xcite . the main difference between these works and the diffusion approach of @xcite is the latter s emphasis on the role of adaptation and learning over networks . in the original diffusion least - mean - squares ( lms ) strategy @xcite , the weight estimates that are exchanged among the nodes can be subject to quantization errors and additive noise over the communication links . studying the degradation in mean - square performance that results from these particular perturbations can be pursued , for both incremental and diffusion strategies , by extending the mean - square analysis already presented in @xcite , in the same manner that the tracking analysis of conventional stand - alone adaptive filters was obtained from the counterpart results in the stationary case ( as explained in ( * ? ? ? * ch . 21 ) ) . useful results along these lines , which study the effect of link noise during the exchange of the weight estimates , already appear for the traditional diffusion algorithm in the works @xcite and for consensus - based algorithms in @xcite . in this paper , our objective is to go beyond these earlier studies by taking into account additional effects , and by considering a more general algorithmic structure . the reason for this level of generality is because the analytical results will help reveal which noise sources influence the network performance more seriously , in what manner , and at what stage of the adaptation process . the results will suggest important remedies and mechanisms to adapt the combination weights in real - time . some of these insights are hard to get if one focuses solely on noise during the exchange of the weight estimates . the analysis will further show that noise during the exchange of the regression data plays a more critical role than other sources of imperfection : this particular noise alters the learning dynamics and modes of the network , and biases the weight estimates . noises related to the exchange of other pieces of information do not alter the dynamics of the network but contribute to the deterioration of the network performance . to arrive at these results , in this paper , we first consider a generalized analysis that applies to a broad class of diffusion adaptation strategies ( see further ahead ; this class includes the original diffusion strategies and as two special cases ) . the analysis allows us to account for various sources of information noise over the communication links . we allow for noisy exchanges during _ each _ of the three processing steps of the adaptive diffusion algorithm ( the two combination steps and and the adaptation step ) . in this way , we are able to examine how the three sets of combination coefficients @xmath0 in influence the propagation of the noise signals through the network dynamics . our results further reveal how the network mean - square - error performance is dependent on these combination weights . following this line of reasoning , the analysis leads to algorithms and further ahead for choosing the combination coefficients to improve the steady - state network performance . it should be noted that several combination rules , such as the metropolis rule @xcite and the maximum degree rule @xcite , were proposed previously in the literature especially in the context of consensus - based iterations @xcite . these schemes , however , usually suffer performance degradation in the presence of noisy information exchange since they ignore the network noise profile @xcite . when the noise variance differs across the nodes , it becomes necessary to design combination rules that are aware of this variation as outlined further ahead in section vi - b . moreover , in a mobile network @xcite where nodes are on the move and where neighborhoods evolve over time , it is even more critical to employ adaptive combination strategies that are able to track the variations in the noise profile in order to cope with such dynamic environments . this issue is taken up in section vi - c . we use lowercase letters to denote vectors , uppercase letters for matrices , plain letters for deterministic variables , and boldface letters for random variables . we also use @xmath1 to denote conjugate transposition , @xmath2 for the trace of its matrix argument , @xmath3 for the spectral radius of its matrix argument , @xmath4 for the kronecker product , and @xmath5 for a vector formed by stacking the columns of its matrix argument . we further use @xmath6 to denote a ( block ) diagonal matrix formed from its arguments , and @xmath7 to denote a column vector formed by stacking its arguments on top of each other . all vectors in our treatment are column vectors , with the exception of the regression vectors , @xmath8 , and the associated noise signals , @xmath9 , which are taken to be row vectors for convenience of presentation . we consider a connected network consisting of @xmath10 nodes . each node @xmath11 collects scalar measurements @xmath12 and @xmath13 regression data vectors @xmath8 over successive time instants @xmath14 . note that we use parenthesis to refer to the time - dependence of scalar variables , as in @xmath12 , and subscripts to refer to the time - dependence of vector variables , as in @xmath8 . the measurements across all nodes are assumed to be related to an unknown @xmath15 vector @xmath16 via a linear regression model of the form @xcite : @xmath17 where @xmath18 denotes the measurement or model noise with zero mean and variance @xmath19 . the vector @xmath16 in denotes the parameter of interest , such as the parameters of some underlying physical phenomenon , the taps of a communication channel , or the location of food sources or predators . such data models are also useful in studies on hybrid combinations of adaptive filters @xcite . the nodes in the network would like to estimate @xmath16 by solving the following minimization problem : @xmath20 in previous works @xcite , we introduced and studied several distributed strategies of the diffusion type that allow nodes to cooperate with each other in order to solve problems of the form in an adaptive manner . these diffusion strategies endow networks with adaptation and learning abilities , and enable information to diffuse through the network in real - time . we review the adaptive diffusion strategies below . in @xcite , two classes of diffusion algorithms were proposed . one class is the so - called combine - then - adapt ( cta ) strategy : @xmath21\\ \end{aligned}\right.\end{aligned}\ ] ] and the second class is the so - called adapt - then - combine ( atc ) strategy : @xmath22\\ { \bs{w}}_{k , i}&=\sum_{l\in{\mc{n}}_k}a_{2,lk}{\bs{\psi}}_{l , i}\\ \end{aligned}\right.\end{aligned}\ ] ] where the @xmath23 are small positive step - size parameters and the @xmath0 are nonnegative entries of the @xmath24 matrices @xmath25 , respectively . the coefficients @xmath0 are zero whenever node @xmath26 is not connected to node @xmath11 , i.e. , @xmath27 , where @xmath28 denotes the neighborhood of node @xmath11 . the two strategies and can be integrated into one broad class of diffusion adaptation @xcite : @xmath29\\ \label{eqn : idealdiffusionpostdiff } { \bs{w}}_{k , i}&=\sum_{l\in{\mc{n}}_k}a_{2,lk}{\bs{\psi}}_{l , i}\end{aligned}\ ] ] several diffusion strategies can be obtained as special cases of through proper selection of the coefficients @xmath0 . for example , to recover the cta strategy , we set @xmath30 , and to recover the atc strategy , we set @xmath31 , where @xmath32 denotes the @xmath24 identity matrix . in the general diffusion strategy , each node @xmath11 evaluates its estimate @xmath33 at time @xmath34 by relying solely on the data collected from its neighbors through steps and and on its local measurements through step . the matrices @xmath35 , @xmath36 , and @xmath37 are required to be left or right - stochastic , i.e. , @xmath38 where @xmath39 denotes the @xmath40 vector whose entries are all one . this means that each node performs a convex combination of the estimates received from its neighbors at every iteration @xmath34 . the mean - square performance and convergence properties of the diffusion algorithm have already been studied in detail in @xcite . for the benefit of the analysis in the subsequent sections , we present below in the recursion describing the evolution of the weight error vectors across the network to do so , we introduce the error vectors : @xmath41 and substitute the linear model into the adaptation step to find that @xmath42 where the @xmath43 matrix @xmath44 and the @xmath15 vector @xmath45 are defined as : @xmath46 we further collect the various quantities across all nodes in the network into the following block vectors and matrices : @xmath47 then , from , , and , the recursion for the network error vector @xmath48 is given by @xmath49 where @xmath50 each of the steps in involves the sharing of information between node @xmath11 and its neighbors . for example , in the first step , all neighbors of node @xmath11 send their estimates @xmath51 to node @xmath11 . this transmission is generally subject to additive noise and possibly quantization errors . likewise , steps and involve the sharing of other pieces of information with node @xmath11 . these exchange steps can all be subject to perturbations ( such as additive noise and quantization errors ) . one of the objectives of this work is to analyze the _ aggregate _ effect of these perturbations on general diffusion strategies of the type and to propose choices for the combination weights in order to enhance the mean - square performance of the network in the presence of these disturbances . to node @xmath11 . ] so let us examine what happens when information is exchanged over links with additive noise . we model the data received by node @xmath11 from its neighbor @xmath26 as @xmath52 where @xmath53 and @xmath54 are @xmath15 noise signals , @xmath9 is a @xmath13 noise signal , and @xmath55 is a scalar noise signal ( see fig . [ fig : noise ] ) . observe further that in , we are including several sources of information exchange noise . in comparison , references @xcite only considered the noise source @xmath53 in and one set of combination coefficients @xmath56 ; the other coefficients were set to @xmath57 for @xmath58 and @xmath59 . in other words , these references only considered and the following traditional cta strategy without exchange of the data @xmath60 compare with ; note that the second step in only uses @xmath61 : @xmath62\\ \end{aligned}\right.\end{aligned}\ ] ] the analysis that follows examines the aggregate effect of all four noise sources appearing in , in addition to the three sets of combination coefficients appearing in . we introduce the following assumption on the statistical properties of the measurement data and noise signals . [ asm : all ] 1 . the regression data @xmath8 are temporally white and spatially independent random variables with zero mean and covariance matrix @xmath63 . 2 . the noise signals @xmath64 , @xmath53 , @xmath55 , @xmath9 , and @xmath54 are temporally white and spatially independent random variables with zero mean and ( co)variances @xmath19 , @xmath65 , @xmath66 , @xmath67 , and @xmath68 , respectively . in addition , @xmath65 , @xmath66 , @xmath67 , and @xmath68 are all zero if @xmath27 or @xmath69 . 3 . the regression data @xmath70 , the model noise signals @xmath71 , and the link noise signals @xmath72 , @xmath73 , @xmath74 , and @xmath75 are mutually - independent random variables for all @xmath76 and @xmath77 . using the perturbed data , the diffusion algorithm becomes @xmath78\\ \label{eqn : noisydiffusionpostdiffold } { \bs{w}}_{k , i}&\!=\!\sum_{l\in{\mc{n}}_k}a_{2,lk}{\bs{\psi}}_{lk , i}\end{aligned}\ ] ] where we continue to use the symbols @xmath79 to avoid an explosion of notation . from and , expressions can be rewritten as @xmath80\\ \label{eqn : noisydiffusionpostdiff } { \bs{w}}_{k , i}&\!=\!\sum_{l\in{\mc{n}}_k}a_{2,lk}{\bs{\psi}}_{l , i}\!+\!{\bs{v}}_{k , i}^{(\psi)}\end{aligned}\ ] ] where we are introducing the symbols @xmath81 and @xmath82 to denote the aggregate @xmath15 zero - mean noise signals defined over the neighborhood of node @xmath11 : @xmath83 with covariance matrices @xmath84 it is worth noting that @xmath85 and @xmath86 depend on the combination coefficients @xmath56 and @xmath87 , respectively . this property will be taken into account when optimizing over @xmath56 and @xmath87 in a later section . we further introduce the following scalar zero - mean noise signal : @xmath88 for @xmath89 , whose variance is @xmath90 to unify the notation , we define @xmath91 . then , from , , and , it is easy to verify that the noisy data @xmath92 are related via @xmath93 for @xmath94 . continuing with the adaptation step and substituting , we get @xmath95\end{aligned}\ ] ] then , we can derive the following error recursion for node @xmath11 ( compare with ): @xmath96 where the @xmath43 matrix @xmath97 and the @xmath98 vector @xmath99 are defined as ( compare with and ): @xmath100 we further introduce the block vectors and matrices : @xmath101 and the corresponding covariance matrices for @xmath102 and @xmath103 : @xmath104 then , from , , and , we arrive at the following recursion for the network weight error vector in the presence of noisy information exchange : @xmath105\!-\!{\bs{v}}_i^{(\psi)}\nonumber\\ { } & = { \mc{a}}_2^{\t}\left[(i_{nm}\!-\!{\mc{m}}{\bs{\mc{r}}}_i')({\mc{a}}_1^{\t}{\wt{\bs{w}}}_{i-1}\!-\!{\bs{v}}_{i-1}^{(w ) } ) \!-\!{\mc{m}}{\bs{z}}_i\right]\!-\!{\bs{v}}_i^{(\psi)}\end{aligned}\ ] ] that is , @xmath106 compared to the previous error recursion , the noise terms in consist of three parts : * @xmath107 is contributed by the noise introduced at the information - exchange step _ before _ adaptation . * @xmath108 is contributed by the noise introduced at the adaptation step . * @xmath103 is contributed by the noise introduced at the information - exchange step _ after _ adaptation . given the weight error recursion , we are now ready to study the mean convergence condition for the diffusion strategy in the presence of disturbances during information exchange under assumption [ asm : all ] . taking expectations of both sides of , we get @xmath109 where @xmath110 from , , , and , it can be verified that @xmath111 whereas , from and assumption [ asm : all ] , we get @xmath112\nonumber\\ { } & = -\left(\sum_{l\in{\mc{n}}_k}c_{lk}r_{v , lk}^{(u)}\right)w^o\end{aligned}\ ] ] let us define an @xmath113 matrix @xmath114 that collects all covariance matrices @xmath115 , @xmath116 , weighted by the corresponding combination coefficients @xmath117 , such that its @xmath118th @xmath43 submatrix is @xmath119 . note that @xmath114 itself is _ not _ a covariance matrix because @xmath120 for all @xmath11 . then , from and , we arrive at @xmath121 therefore , using and , expression becomes @xmath122 with a driving term due to the presence of @xmath123 . this driving term would disappear from if there were no noise during the exchange of the regression data . to guarantee convergence of , the coefficient matrix @xmath124 must be stable , i.e. , @xmath125 . since @xmath126 and @xmath127 are right - stochastic matrices , it can be shown that the matrix @xmath124 is stable whenever @xmath128 itself is stable ( see appendix [ app : meanconvergence ] ) . this fact leads to an upper bound on the step - sizes @xmath23 to guarantee the convergence of @xmath129 to a steady - state value , namely , we must have @xmath130 for @xmath131 , where @xmath132 denotes the largest eigenvalue of its matrix argument . note that the neighborhood covariance matrix @xmath133 in is related to the combination weights @xmath117 . if we further assume that @xmath37 is doubly - stochastic , i.e. , @xmath134 then , by jensen s inequality @xcite , @xmath135 since ( i ) @xmath132 coincides with the induced @xmath136-norm for any positive semi - definite hermitian matrix ; ( ii ) matrix norms are convex functions of their arguments @xcite ; and ( iii ) by , @xmath117 are convex combination coefficients . thus , we obtain a sufficient condition for the convergence of in lieu of : @xmath137 } } \ ] ] for @xmath131 , where the upper bound for the step - size @xmath138 becomes independent of the combination weights @xmath117 . this bound can be determined solely from knowledge of the covariances of the regression data and the associated noise signals that are accessible to node @xmath11 . it is worth noting that for traditional diffusion algorithms where information is perfectly exchanged , condition reduces to @xmath139}\end{aligned}\ ] ] for @xmath131 . comparing with , we see that the link noise @xmath9 over regression data reduces the dynamic range of the step - sizes for mean stability . now , under , and taking the limit of as @xmath140 , we find that the mean error vector will converge to a steady - state value @xmath141 : @xmath142 it is well - known that studying the mean - square convergence of a single adaptive filter is a challenging task , since adaptive filters are nonlinear , time - variant , and stochastic systems . when a network of adaptive nodes is considered , the complexity of the analysis is compounded because the nodes now influence each other s behavior . in order to make the performance analysis more tractable , we rely on the energy conservation approach @xcite , which was used successfully in @xcite to study the mean - square performance of diffusion strategies under perfect information exchange conditions . that argument allows us to derive expressions for the mean - square - deviation ( msd ) and the excess - mean - square - error ( emse ) of the network by analyzing how energy ( measured in terms of error variances ) flows through the nodes . from recursion and under assumption [ asm : all ] , we can obtain the following weighted variance relation for the global error vector @xmath48 : @xmath143\}\\ { } & \quad+\e\|{\mc{a}}_2^{\t}(i_{nm}\!-\!{\mc{m}}{\bs{\mc{r}}}_i'){\bs{v}}_{i-1}^{(w)}\|_{\sigma}^2+\e\|{\bs{v}}_i^{(\psi)}\|_{\sigma}^2 \end{aligned } } \ ] ] where @xmath144 is an arbitrary @xmath113 positive semi - definite hermitian matrix that we are free to choose . moreover , the notation @xmath145 stands for the quadratic term @xmath146 . the weighting matrix @xmath147 in can be expressed as @xmath148 where @xmath124 is given by and @xmath149 denotes a term on the order of @xmath150 . evaluating the term @xmath149 requires knowledge of higher - order statistics of the regression data and link noises , which are not available under current assumptions . however , this term becomes negligible if we introduce a small step - size assumption . [ asm : smallstepsize ] the step - sizes are sufficiently small , i.e. , @xmath151 , such that terms depending on higher - order powers of the step - sizes can be ignored . hence , in the sequel we use the approximation : @xmath152 observe that on the right - hand side ( rhs ) of relation , only the first and third terms relate to the error vector @xmath153 . by assumption [ asm : all ] , the error vector @xmath153 is independent of @xmath154 and @xmath155 . thus , from , the third term on rhs of can be expressed as @xmath156\cdot\e{\wt{\bs{w}}}_{i-1}\}\nonumber\\ { } & = -2\,{\mathfrak{re}}(z^*{\mc{m}}{\mc{a}}_2\sigma{\mc{a}}_2^{\t}{\mc{a}}_1^{\t}\cdot\e{\wt{\bs{w}}}_{i-1})+o({\mc{m}}^2)\end{aligned}\ ] ] since we already showed in the previous section that @xmath157 converges to a fixed bias @xmath141 , quantity will converge to a fixed value as well when @xmath140 . moreover , under assumption [ asm : all ] , the second , fourth , and fifth terms on rhs of relation are all fixed values . therefore , the convergence of relation depends on the behavior of the first term @xmath158 . although the weighting matrix @xmath147 of @xmath153 is different from the weighting matrix @xmath144 of @xmath159 , it turns out that the entries of these two matrices are approximately related by a linear equation shown ahead in . introduce the vector notation @xcite : @xmath160 then , by using the identity @xmath161 , it can be verified from that @xmath162 where the @xmath163 matrix @xmath164 is given by @xmath165 to guarantee mean - square convergence of the algorithm , the step - sizes should be sufficiently small and selected to ensure that the matrix @xmath164 is stable @xcite , i.e. , @xmath166 , which is equivalent to the earlier condition @xmath125 . although more specific conditions for mean - square stability can be determined without assumption [ asm : smallstepsize ] @xcite , it is sufficient for our purposes here to conclude that the diffusion strategy is stable in the mean and mean - square senses if the step - sizes @xmath23 satisfy or and are sufficiently small . the conclusion so far is that sufficiently small step - sizes ensure convergence of the diffusion strategy in the mean and mean - square senses , even in the presence of exchange noises over the communication links . let us now determine expressions for the error variances in steady - state . we start from the weighted variance relation . in view of , it shows that the error variance @xmath167 depends on the mean error @xmath129 . we already determined the value of @xmath168 in . we continue to use the vector notation and proceed to evaluate all the terms , except the first one , on rhs of in the following . for the _ second _ term , it can be expressed as @xmath169^*\sigma\end{aligned}\ ] ] where we used the identity @xmath170^*\sigma$ ] for any hermitian matrix @xmath171 , and @xmath172 denotes the autocorrelation matrix of @xmath173 . it is shown in appendix [ app : rz ] that @xmath172 is given by @xmath174 where @xmath175 is defined in , @xmath123 is in , and @xmath176 are two @xmath113 positive semi - definite block diagonal matrices : @xmath177\end{aligned}\ ] ] from expression and assumption [ asm : smallstepsize ] , the _ third _ term on rhs of is given by @xmath178\sigma\}\nonumber\\ { } & = -\!\left[\vec\left({\mc{a}}_2^{\t}{\mc{a}}_1^{\t}(\e{\wt{\bs{w}}}_{i-1})z^*{\mc{m}}{\mc{a}}_2 + { \mc{a}}_2^{\t}{\mc{m}}z(\e{\wt{\bs{w}}}_{i-1})^*{\mc{a}}_1{\mc{a}}_2\right)\right]^*\sigma\end{aligned}\ ] ] likewise , the _ fourth _ term on rhs of is approximated by @xmath179\right)\right]^*\sigma\nonumber\\ { } & \approx\left[\vec({\mc{a}}_2^{\t}{\mc{r}}_v^{(w)}{\mc{a}}_2)\right]^*\sigma\end{aligned}\ ] ] where we are now ignoring terms on the order of @xmath180 and @xmath150 . the _ fifth _ term on rhs of is given by @xmath181^*\sigma\end{aligned}\ ] ] let us introduce @xmath182 at steady - state , as @xmath140 , by and , the weighted variance relation becomes @xmath183^*\sigma\end{aligned}\ ] ] where we are using the compact notation @xmath184 to refer to @xmath185 doing so allows us to represent @xmath147 by the more compact relation @xmath186 on rhs of ; we shall be using the weighting matrix @xmath144 and its vector representation @xmath187 interchangeably for ease of notation ( likewise , for @xmath147 and @xmath188 ) . the steady - state weighted variance relation can be rewritten as @xmath189^*\sigma\end{aligned}\ ] ] where the term @xmath190 is contributed by the model noise @xmath191 while the remaining terms @xmath192 are contributed by the link noises @xmath193 . recall that we are free to choose @xmath144 and , hence , @xmath187 . let @xmath194 , where @xmath195 is another arbitrary positive semi - definite hermitian matrix . then , we arrive at the following theorem . [ lemma : steadystatevariancerelation ] under assumptions [ asm : all ] and [ asm : smallstepsize ] , for any positive semi - definite hermitian matrix @xmath195 , the steady - state weighted error variance relation of the diffusion strategy is approximately given by @xmath196^*(i_{n^2m^2}-{\mc{f}})^{-1}\vec(\omega ) \end{aligned } } \ ] ] where @xmath197 is given in , @xmath198 in , @xmath199 in , and @xmath164 in . each subvector of @xmath200 corresponds to the estimation error at a particular node , say , @xmath201 for node @xmath11 . the network msd is defined as @xcite : @xmath202 since we are free to choose @xmath195 , we select it as @xmath203 . then , expression gives @xmath204^*(i_{n^2m^2}-{\mc{f}})^{-1}\vec(i_{nm } ) \end{aligned } } \ ] ] similarly , if we instead select @xmath205 , where @xmath206 then expression would allow us to evaluate the network emse as : @xmath207^*(i_{n^2m^2}-{\mc{f}})^{-1}\vec({\mc{r}}_u ) \end{aligned } } \ ] ] where , under assumption [ asm : all ] , the network emse is given by @xmath208 we showed in the earlier sections that the link noise over regression data biases the weight estimators . in this section we examine how the results simplify when there is no sharing of regression data among the nodes . [ asm : nodatasharing ] nodes do not share regression data within neighborhoods , i.e. , assume @xmath209 . by assumptions [ asm : smallstepsize ] and [ asm : nodatasharing ] , matrices @xmath210 in , , and become @xmath211 where @xmath212 is given in . then , the network msd and emse expressions and simplify to : @xmath213^ * ( i_{n^2m^2}-{\mc{f}})^{-1}\vec(i_{nm } ) \end{aligned } } \ ] ] and @xmath214^ * ( i_{n^2m^2}-{\mc{f}})^{-1}\vec({\mc{r}}_u ) \end{aligned } } \ ] ] recalling that @xmath198 and @xmath164 are related to the combination matrices @xmath215 , or , equivalently , @xmath216 , results and express the network msd and emse in terms of @xmath216 . however , it is generally difficult to use these expressions to optimize over @xmath216 to reduce the impact of link noise . instead , by substituting into and using the fact that @xmath164 is stable , we can arrive at another useful expression for the network msd : @xmath217^ * \sum_{j=0}^{\infty}{\mc{f}}^j\vec(i_{nm})\nonumber\\ { } & = \frac{1}{n}\!\left[\vec({\mc{a}}_2^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a}}_2\!+\!{\mc{r}}_v)\right]^ * \sum_{j=0}^{\infty}({\mc{b}}^{\t}\otimes{\mc{b}}^*)^j\vec(i_{nm})\nonumber\\ { } & = \frac{1}{n}\!\left[\vec({\mc{a}}_2^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a}}_2\!+\!{\mc{r}}_v)\right]^ * \sum_{j=0}^{\infty}\vec({\mc{b}}^{*j}{\mc{b}}^j)\end{aligned}\ ] ] that is , @xmath218 } \ ] ] where @xmath124 is given in . similarly , the network emse can be expressed as @xmath219 } \ ] ] expressions and reveal in an interesting way how the noise sources originating from any particular node end up influencing the overall network performance . let us denote @xmath220 the error recursion can be rewritten as @xmath221 where @xmath222 then , @xmath223 under assumption [ asm : nodatasharing ] , @xmath224 in and can be simplified as @xmath225 where @xmath226 are given in and . by assumption [ asm : all ] , @xmath224 are temporally independent for different @xmath34 and @xmath227 where @xmath124 is given by . as @xmath140 , the first term on rhs of becomes @xmath228\right\}\nonumber\\ { } & \stackrel{(a)}{\approx}\lim_{i\rightarrow\infty}\tr\left[\left(\e{\bm{\phi}}_{0,i}\right ) ( \e{\wt{\bs{w}}}_{-1}{\wt{\bs{w}}}_{-1}^*)\left(\e{\bm{\phi}}_{0,i}\right)^*\right]\nonumber\\ { } & = \lim_{i\rightarrow\infty}\tr\left[{\mc{b}}^{i+1}(\e\,{\wt{\bs{w}}}_{-1}{\wt{\bs{w}}}_{-1}^*){\mc{b}}^{(i+1)*}\right]\nonumber\\ { } & \stackrel{(b)}{=}0\end{aligned}\ ] ] where ( a ) is obtained by approximating the expectation of the product by the product of expectations and ( b ) is due to the stability of @xmath124 . therefore , the steady - state value of gives @xmath229\nonumber\\ { } & \stackrel{(a)}{\approx}\lim_{i\rightarrow\infty}\sum_{m=0}^{i}\tr\left[{\mc{b}}^{i - m}({\mc{a}}_2^{\t}{\mc{m}}{\mc{s } } { \mc{m}}{\mc{a}}_2+{\mc{r}}_v){\mc{b}}^{(i - m)*}\right]\nonumber\\ { } & \stackrel{(b)}{=}\lim_{i\rightarrow\infty}\sum_{j=0}^{i}\tr\left[{\mc{b}}^{j}({\mc{a}}_2^{\t}{\mc{m}}{\mc{s } } { \mc{m}}{\mc{a}}_2+{\mc{r}}_v){\mc{b}}^{j*}\right]\nonumber\\ { } & = \sum_{j=0}^{\infty}\tr\left[{\mc{b}}^{j}({\mc{a}}_2^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a}}_2+{\mc{r}}_v){\mc{b}}^{*j}\right]\end{aligned}\ ] ] where , by and , ( a ) is due to @xmath230 and ( b ) is simply a change of variable : @xmath231 . since the @xmath232th term of the summation in or is contributed by the term @xmath233 , which consists of all the noise sources at time @xmath234 , expression shows how various sources of noises are involved and how they contribute to the network msd . before we optimize the combination matrices @xmath216 , we first specialize the msd expression and the emse expression for the atc and cta algorithms . for the atc algorithm , we set @xmath31 and @xmath235 , and for the cta algorithm , we set @xmath236 and @xmath30 . let us denote @xmath237 then , we get @xmath238\!\!\\ \label{eqn : noisyatcemse } \!\!{\textrm{emse}}_{\textrm{atc}}&\!\approx\!\frac{1}{n}\!\sum_{j=0}^{\infty}{\mathrm{tr}}\!\left[{\mc{b}}_{\textrm{atc}}^j ( { \mc{a}}^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a}}\!+\!{\mc{r}}_v^{(\psi ) } ) { \mc{b}}_{\textrm{atc}}^{*j}{\mc{r}}_u\right]\!\!\end{aligned}\ ] ] and @xmath239\\ \label{eqn : noisyctaemse } { \textrm{emse}}_{\textrm{cta}}&\approx\frac{1}{n}\sum_{j=0}^{\infty}{\mathrm{tr}}\left[{\mc{b}}_{\textrm{cta}}^j ( { \mc{m}}{\mc{s}}{\mc{m}}+{\mc{r}}_v^{(w)}){\mc{b}}_{\textrm{cta}}^{*j}{\mc{r}}_u\right]\end{aligned}\ ] ] minimizing the msd expression or the emse expression for the atc algorithm over left - stochastic matrices @xmath240 is generally nontrivial . we pursue an approximate solution that relies on optimizing an upper bound and performs well in practice . let us use @xmath241 to denote the nuclear norm ( also known as the trace norm , or the ky fan @xmath242-norm ) of matrix @xmath243 @xcite , which is defined as the sum of the singular values of @xmath243 . therefore , @xmath244 for any @xmath243 and @xmath245 when @xmath243 is hermitian and positive semi - definite . let us also denote @xmath246 as the block maximum norm of matrix @xmath243 ( see appendix [ app : meanconvergence ] ) . then , @xmath247 & = \big\|{\mc{b}}_{\textrm{atc}}^j({\mc{a}}^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a } } + { \mc{r}}_v^{(\psi)}){\mc{b}}_{\textrm{atc}}^{*j}\big\|_*\nonumber\\ { } & \le\|{\mc{b}}_{\textrm{atc}}^j\|_*\cdot\|{\mc{a}}^{\t}{\mc{m}}{\mc{s } } { \mc{m}}{\mc{a}}+{\mc{r}}_v^{(\psi)}\|_*\cdot\|{\mc{b}}_{\textrm{atc}}^{*j}\|_*\nonumber\\ { } & \le c^2\cdot\|{\mc{b}}_{\textrm{atc}}^j\|_{b,\infty}^2\cdot { \mathrm{tr}}({\mc{a}}^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a}}+{\mc{r}}_v^{(\psi)})\nonumber\\ { } & \le c^2\cdot\|{\mc{b}}_{\textrm{atc}}\|_{b,\infty}^{2j}\cdot { \mathrm{tr}}({\mc{a}}^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a}}+{\mc{r}}_v^{(\psi)})\nonumber\\ { } & \le c^2\cdot(\|{\mc{a}}\|_{b,\infty}\cdot\|i_{nm}-{\mc{m}}{\mc{r}}_u\|_{b,\infty})^{2j } { \mathrm{tr}}({\mc{a}}^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a}}+{\mc{r}}_v^{(\psi)})\nonumber\\ { } & = c^2\cdot\rho(i_{nm}-{\mc{m}}{\mc{r}}_u)^{2j}\cdot{\mathrm{tr } } ( { \mc{a}}^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a}}+{\mc{r}}_v^{(\psi)})\end{aligned}\ ] ] where @xmath248 is some positive scalar such that @xmath249 because @xmath241 and @xmath246 are submultiplicative norms and all such norms are equivalent @xcite . in the last step of we used lemmas [ lemma : rghtstocmat ] and [ lemma : blockdiagonal ] from appendix [ app : meanconvergence ] . thus , we can upper bound the network msd by @xmath250 ^ 2}\end{aligned}\ ] ] where the combination matrix @xmath251 appears only in the numerator . the result motivates us to consider instead the problem of minimizing the upper bound , namely , @xmath252 using and , the cost function in can be expressed as @xmath253\end{aligned}\ ] ] problem can therefore be decoupled into @xmath10 separate optimization problems of the form : @xmath254\\ \st & \;\ ; \sum_{l\in{\mc{n}}_k}a_{lk}=1,\;\;a_{lk}\ge0,\;\;a_{lk}=0\;\mbox{if}\;l\notin{\mc{n}}_k \\ \end{aligned } } \ ] ] for @xmath255 . with each node @xmath94 , we associate the following nonnegative _ variance product _ measure : @xmath256 this measure incorporates information about the link noise covariances @xmath257 . the solution of is then given by @xmath258 we refer to this combination rule as the relative variance combination rule ; it is an extension of the rule devised in @xcite to the case of noisy information exchanges . in particular , the definition of the scalars @xmath259 in is different and now depends on both subscripts @xmath26 and @xmath11 . minimizing the emse expression for the atc algorithm over left - stochastic matrices @xmath240 can be pursued in a similar manner by noting that @xmath260\le c^2[\rho(i_{nm}\!-\!{\mc{m}}{\mc{r}}_u)]^{2j}\ , { \mathrm{tr}}({\mc{a}}^{\t}{\mc{m}}{\mc{s}}{\mc{m}}{\mc{a}}\!+\!{\mc{r}}_v^{(\psi)})\,{\mathrm{tr}}({\mc{r}}_u)\end{aligned}\ ] ] thus , minimizing the upper bound of the network emse leads to the same solution . using the same argument , we can also show that the same result minimizes the upper bound of the network msd or emse for the cta algorithm . to apply the relative variance combination rule , each node @xmath11 needs to know the variance products , @xmath259 , of their neighbors , which in general are not available since they require knowledge of the quantities @xmath261 . therefore , we now propose an adaptive combination rule by using data that are available to the individual nodes . for the atc algorithm , we first note from and that @xmath262 for @xmath89 . since the algorithm converges in the mean and mean - square senses under assumption [ asm : smallstepsize ] , all the estimates @xmath263 tend close to @xmath16 as @xmath140 . this allows us to estimate @xmath264 for node @xmath11 by using instantaneous realizations of @xmath265 , where we replace @xmath51 by @xmath266 . similarly , for node @xmath11 itself , we can use realizations of @xmath267 to estimate @xmath268 . to unify the notation , we define @xmath269 . let @xmath270 denote an estimator for @xmath264 that is computed by node @xmath11 at time @xmath34 . then , one way to evaluate @xmath270 is through the recursion : @xmath271 for @xmath272 , where @xmath273 is a forgetting factor that is usually close to one . in this way , we arrive at the adaptive combination rule : @xmath274^{-1}}{\sum_{m\in{\mc{n}}_k}[{\widehat{\bm{\gamma}}}_{mk}^{2}(i)]^{-1 } } , & { \textrm{if $ l\in{\mc{n}}_k$ } } \\ 0 , & { \textrm{otherwise } } \\ \end{cases } } \ ] ] the diffusion strategy is adaptive in nature . one of the main benefits of adaptation ( by using constant step - sizes ) is that it endows networks with tracking abilities when the underlying weight vector @xmath16 varies with time . in this section we analyze how well an adaptive network is able to track variations in @xmath16 . to do so , we adopt a random - walk model for @xmath16 that is commonly used in the literature to describe the non - stationarity of the weight vector @xcite . [ asm : randomwalk ] the weight vector @xmath16 changes according to the model : @xmath275 where @xmath276 has a constant mean @xmath16 for all @xmath34 , @xmath277 is an i.i.d . random sequence with zero mean and covariance matrix @xmath278 ; the sequence @xmath277 is independent of the initial conditions @xmath279 and of all regression data and noise signals across the network for all time instants . we now define the error vector at node @xmath11 as @xmath280 so that the global error recursion ( [ eqn : noisyerrorrecursion1 ] ) for the network is replaced by @xmath281 where the @xmath282 vector @xmath283 is defined as @xmath284 by assumptions [ asm : all ] and [ asm : randomwalk ] , it can be verified that the condition for mean convergence continues to be @xmath285 , where @xmath124 is defined in . in addition , it can also be verified that the error recursion converges in the mean sense to the same non - zero bias vector @xmath141 as in . from and under assumption [ asm : smallstepsize ] , we can derive the weighted variance relation : @xmath286\}\nonumber\\ { } & \qquad+{\mathbb{e}}\|{\mc{a}}_2^{\t}(i_{nm}-{\mc{m}}{\bs{\mc{r}}}_i'){\bs{v}}_{i-1}^{(w)}\|_{\sigma}^2\nonumber\\ { } & \qquad+{\mathbb{e}}\|{\mc{a}}_2^{\t}{\mc{m}}{\bs{z}}_i\|_{\sigma}^2+{\mathbb{e}}\|{\bs{v}}_i^{(\psi)}\|_{\sigma}^2\end{aligned}\ ] ] where @xmath164 is given in . if the step - sizes are sufficiently small , then we can assume that the network continues to be mean - square stable . the steady - state performance is affected by the non - stationarity of @xmath16 . from assumption [ asm : smallstepsize ] , at steady - state , expression becomes @xmath287^*(i_{n^2m^2}\!-\!{\mc{f}})^{-1}{\mathrm{vec}}(\omega)\end{aligned}\ ] ] where @xmath197 is given in , @xmath288 in , @xmath199 in , @xmath164 in , and @xmath289 is the covariance matrix of @xmath283 : @xmath290 by , , and , we get @xmath291 then , following the same argument that led to , we find that the network msd is now given by : @xmath292^*(i_{n^2m^2}-{\mc{f}})^{-1}\vec(i_{nm } ) \end{aligned } } \ ] ] similarly , the network emse is given by : @xmath293^*(i_{n^2m^2}-{\mc{f}})^{-1}\vec({\mc{r}}_u ) \end{aligned } } \ ] ] where @xmath212 is defined in . observe that the main difference relative to and is the addition of the term @xmath289 . therefore , all the results that were derived in the earlier section , such as and , continue to hold by adding @xmath289 . in particular , if assumptions [ asm : smallstepsize ] and [ asm : nodatasharing ] are adopted , expressions and can be approximated as @xmath294^*(i_{n^2m^2}-{\mc{f}})^{-1}\vec(i_{nm } ) \end{aligned } } \ ] ] and @xmath295^*(i_{n^2m^2}-{\mc{f}})^{-1}\vec({\mc{r}}_u ) \end{aligned } } \ ] ] where @xmath198 is now given in . we simulate two scenarios : noisy information exchanges and non - stationary environments . we consider a connected network with @xmath296 nodes . the network topology is shown in fig . [ fig : topology ] . the unknown complex parameter @xmath16 of length @xmath297 is randomly generated ; its value is @xmath298 $ ] . we adopt uniform step - sizes , @xmath299 , and uniformly white gaussian regression data with covariance matrices @xmath300 , where @xmath301 are shown in fig . [ fig : regressor ] . the variances of the model noises , @xmath302 , are randomly generated and shown in fig . [ fig : modelnoise ] . we also use white gaussian link noise signals such that @xmath303 , @xmath304 , and @xmath305 . all link noise variances , @xmath306 , are randomly generated and illustrated in fig . [ fig : linknoise ] from top to bottom . we assign the link number by the following procedure . we denote the link from node @xmath26 to node @xmath11 as @xmath307 , where @xmath308 . then , we collect the links @xmath309 in an ascending order of @xmath26 in the list @xmath310 ( which is a set with _ ordered _ elements ) for each node @xmath11 . for example , for node @xmath311 in fig . [ fig : topology ] , it has @xmath312 links ; the ordered links are then collected in @xmath313 . we concatenate @xmath314 in an ascending order of @xmath11 to get the overall list @xmath315 . eventually , the @xmath316th link in the network is given by the @xmath316th element in the list @xmath317 . nodes.,height=172 ] we examine the simplified cta and atc algorithms in and , namely , no sharing of data among nodes ( i.e. , @xmath209 ) , under various combination rules : ( i ) the relative variance rule in , ( ii ) the metropolis rule in @xcite : @xmath318 where @xmath319 denotes the degree of node @xmath11 ( including the node itself ) , ( iii ) the uniform weighting rule : @xmath320 and ( iv ) the adaptive rule in with @xmath321 . we plot the network msd and emse learning curves for atc algorithms in figs . [ fig : msd_atc ] and [ fig : emse_atc ] by averaging over 50 experiments . for cta algorithms , we plot their network msd and emse learning curves in figs . [ fig : msd_cta ] and [ fig : emse_cta ] also by averaging over 50 experiments . moreover , we also plot their theoretical results and in the same figures . from fig . [ fig : sim2 ] we see that the relative variance rule makes diffusion algorithms achieve the lowest msd and emse levels at steady - state , compared to the metropolis and uniform rules as well as the algorithm from @xcite ( which also requires knowledge of the noise variances ) . in addition , the adaptive rule attains msd and emse levels that are only slightly larger than those of the relative variance rule , although , as expected , it converges slower due to the additional learning step . the value for each entry of the complex parameter @xmath322 is assumed to be changing over time along a circular trajectory in the complex plane , as shown in fig . [ fig : track ] . the dynamic model for @xmath323 is expressed as @xmath324 , where @xmath325 , @xmath326 , and @xmath327 . the covariance matrices @xmath328 are randomly generated such that @xmath329 when @xmath330 , but their traces are normalized to be one , i.e. , @xmath331 , for all nodes . the variances for the model noises , @xmath332 , are also randomly generated . we examine two different scenarios : the low noise - level case where the average noise variance across the network is @xmath333 db and the noise variances are shown in fig . [ fig : hsnr ] ; and the high noise - level case where the average variance is @xmath334 db and the variances are shown in fig . [ fig : lsnr ] . we simulate 3000 iterations and average over 20 experiments in figs . [ fig : track_hsnr ] and [ fig : track_lsnr ] for each case . the step - size is 0.01 and uniform across the network . for simplicity , we adopt the simplified atc algorithm where @xmath209 , and only use the uniform weighting rule . the tracking behavior of the network , denoted as @xmath335 , is obtained by averaging over all the estimates , @xmath336 , across the network . figs . [ fig : track_hsnr ] and [ fig : track_lsnr ] depict the complex plane ; the horizontal axis is the real axis and the vertical axis is the imaginary axis . therefore , for every time @xmath34 , each entry of @xmath323 or @xmath337 represents a point in the plane . when @xmath34 is increasing , @xmath338 moves along the red trajectory ( in @xmath339 ) , @xmath340 along the blue trajectory ( in @xmath341 ) , @xmath342 along the green trajectory ( in @xmath343 ) , and @xmath344 along the magenta trajectory ( in @xmath345 ) . from fig . [ fig : track ] , it can be seen that diffusion algorithms exhibit the tracking ability in both high and low noise - level environments . in this work we investigated the performance of diffusion algorithms under several sources of noise during information exchange and under non - stationary environments . we first showed that , on one hand , the link noise over the regression data biases the estimators and deteriorates the conditions for mean and mean - square convergence . on the other hand , diffusion strategies can still stabilize the mean and mean - square convergence of the network with noisy information exchange . we derived analytical expressions for the network msd and emse and used these expressions to motivate the choice of combination weights that help ameliorate the effect of information - exchange noise and improve network performance . we also extended the results to the non - stationary scenario where the unknown parameter @xmath16 is changing over time . simulation results illustrate the theoretical findings and how well they match with theory . following @xcite , we first define the block maximum norm of a vector . [ def : vecblkmaxnorm ] given a vector @xmath347 consisting of @xmath10 blocks @xmath348 , the block maximum norm is the real function @xmath349 , defined as @xmath350 where @xmath351 denotes the standard @xmath136-norm on @xmath352 . similarly , we define the matrix norm that is induced by the block maximum norm as follows : [ def : blkmaxnorm ] given a block matrix @xmath353 with block size @xmath43 , then @xmath354 denotes the induced block maximum ( matrix ) norm on @xmath355 . [ lemma : blkunitaryinvariant ] the block maximum matrix norm is block unitary invariant , i.e. , given a block diagonal unitary matrix @xmath356 consisting of @xmath10 unitary blocks @xmath357 , where @xmath358 , for any matrix @xmath353 , then @xmath359 where @xmath360 denotes the block maximum matrix norm on @xmath355 with block size @xmath43 . [ lemma : rghtstocmat ] let @xmath361 be a right - stochastic matrix . then , for block size @xmath43 , @xmath362 from definition [ def : blkmaxnorm ] , we get @xmath363_{lk}x_k\|_2}{\max_{k}\|x_k\|_2}\nonumber\\ { } & \le\max_{x\in{\mathbb{c}}^{mn}\backslash\{0\ } } \frac{\max_{l}\sum_{k=1}^{n}[a]_{lk}\|x_k\|_2}{\max_{k}\|x_k\|_2}\nonumber\\ { } & \le\max_{x\in{\mathbb{c}}^{mn}\backslash\{0\ } } \frac{\max_{l}(\sum_{k=1}^{n}[a]_{lk})\cdot\max_k\|x_k\|_2}{\max_{k}\|x_k\|_2}\nonumber\\ { } & \le\max_{x\in{\mathbb{c}}^{mn}\backslash\{0\}}\frac{\max_{l}1\cdot\max_{k}\|x_k\|_2}{\max_{k}\|x_k\|_2}\nonumber\\ { } & = 1\end{aligned}\ ] ] where @xmath364 consists of @xmath10 blocks @xmath348 , and @xmath365_{lk}$ ] denotes the @xmath366th entry of @xmath240 . on the other hand , for any induced matrix norm , say , the block maximum norm , it is always lower bounded by the spectral radius of the matrix @xcite : @xmath367 combining and completes the proof . [ lemma : blockdiagonal ] let @xmath368 be a block diagonal hermitian matrix with block size @xmath43 . then the block maximum norm of the matrix @xmath251 is equal to its spectral radius , i.e. , @xmath369 denote the @xmath11th @xmath43 submatrix on the diagonal of @xmath251 by @xmath370 . let @xmath371 be the eigen - decomposition of @xmath370 , where @xmath372 is unitary and @xmath373 is diagonal . define the block unitary matrix @xmath374 and the diagonal matrix @xmath375 . then , @xmath376 . by lemma [ lemma : blkunitaryinvariant ] , the block maximum norm of @xmath251 with block size @xmath43 is @xmath377 where we used the fact that the induced @xmath136-norm is identical to the spectral radius for hermitian matrices @xcite . on the other hand , any matrix norm is lower bounded by the spectral radius @xcite , i.e. , @xmath378 combining and completes the proof . now we show that the matrix @xmath379 is stable if @xmath128 is stable . for any induced matrix norm , say , the block maximum norm with block size @xmath43 , we have @xcite @xmath380 where , from and , @xmath126 and @xmath127 satisfy lemma [ lemma : rghtstocmat ] . by and , it is straightforward to see that @xmath128 is block diagonal with block size @xmath43 . then , by lemma [ lemma : blockdiagonal ] , expression can be further expressed as @xmath381 which completes the proof . in appendix i of @xcite and the matrix @xmath180 in lemma 2 of @xcite are block diagonal , the @xmath382 norm used in these references should simply be replaced by the @xmath360 norm used here and as already done in @xcite . ] let us denote the @xmath366th submatrix of @xmath383 by @xmath384 . by assumptions [ asm : all ] and expression , @xmath385 can be evaluated as @xmath386 where , by expressions and , @xmath387 when @xmath388 , expression reduces to @xmath389 when @xmath390 , expression becomes @xmath391 where @xmath392 denotes the kronecker delta function . evaluating the last term on rhs of requires knowledge of the excess kurtosis of @xmath393 , which is generally not available . in order to proceed , we invoke a separation principle to approximate it as @xmath394 substituting into leads to @xmath395\end{aligned}\ ] ] from and , we get @xmath396\end{aligned}\ ] ] substituting into , we obtain @xmath397\end{aligned}\ ] ] from and , we arrive at expression . l. li and j. a. chambers , `` distributed adaptive estimation based on the apa algorithm over diffusion netowrks with changing topology , '' in _ proc . ieee workshop stat . signal process . ( ssp ) _ , cardiff , uk , aug . 2009 , pp . 757760 . n. takahashi and i. yamada , `` link probability control for probabilistic diffusion least - mean squares over resource - constrained networks , '' in _ proc . speech , signal process . ( icassp ) _ , dallas , tx , mar . 2010 , pp . 35183521 . z. towfic , j. chen , and a. h. sayed , `` collaborative learning of mixture models using diffusion adaptation , '' in _ proc . workshop machine learn . signal process . ( mlsp ) _ , beijing , china , sept . 2011 , pp . j. chen and a. h. sayed , `` diffusion adaptation strategies for distributed optimization and learning over networks , '' to appear in _ ieee trans . signal process . [ also available online at http://arxiv.org/abs/1111.0034 as _ arxiv:1111.0034v2 [ math.oc]_ , oct . 2011 . ] r. abdolee and b. champagne , `` diffusion lms algorithms for sensor networks over non - ideal inter - sensor wireless channels , '' in _ proc . sensor systems ( dcoss ) _ , barcelona , spain , june 2011 , pp . 16 . l. xiao , s. boyd , and s. lall , `` a scheme for robust distributed sensor fusion based on average consensus , '' in _ proc . acm / ieee int . sensor networks ( ipsn ) _ , los angeles , ca , apr . 2005 , pp . 6370 . d. mandic , p. vayanos , c. boukis , b. jelfs , s. l. goh , t. gautama , and t. rutkowski , `` collaborative adaptive learning using hybrid filters , '' in _ proc . speech , signal process . ( icassp ) _ , honolulu , hi , apr . 2007 , pp . 921924 . tu and a. h. sayed , `` optimal combination rules for adaptation and learning over netowrks , '' in _ proc . workshop comput . advances multi - sensor adapt . process . ( camsap ) _ , san juan , puerto rico , dec . 2011 , pp . 317320 .

borisov , mamaev and kilin have recently found certain poisson structures with respect to which the reduced and rescaled systems of certain non - holonomic problems , involving rolling bodies without slipping , become hamiltonian , the hamiltonian function being the reduced energy . we study further the algebraic origin of these poisson structures , showing that they are of rank two and therefore the mentioned rescaling is not necessary . we show that they are determined , up to a non - vanishing factor function , by the existence of a system of first - order differential equations providing two integrals of motion . we generalize the form of that poisson structures and extend their domain of definition . we apply the theory to the rolling disk , the routh s sphere , the ball rolling on a surface of revolution , and its special case of a ball rolling inside a cylinder . # 1#2 ( # 1 # 2 ) # 1#2

math-ph0401054

in recent years there has been an increasing interest in the geometric treatment of non - holonomic mechanical systems , see , e.g. , @xcite . in particular , it has been recognised that the hamiltonian formulation of such systems can be stated in terms of an _ almost - poisson _ bracket , that is , a biderivation of functions of phase space , antisymmetric in its arguments but which does not necessarily fulfil the jacobi identity ( see , e.g. , @xcite ) . therefore , for researchers in this field , it seems to be usual the conceptual association of the hamiltonian formulation of non - holonomic mechanical systems with almost - poisson structures . on the other hand , there exist non - holonomic systems which , after certain reductions are performed , admit a hamiltonian formulation after a rescaling of time is carried out , by means of rescaling factors ( sometimes called _ invariant measures _ ) of the reduced vector field of the system . this is the case for the so - called lr systems , which are systems formulated on compact lie groups endowed with a left - invariant metric and right - invariant non - holonomic constraints . after a rescaling of time , their corresponding reduced systems become integrable hamiltonian systems describing geodesic flows on unit spheres @xcite . in @xcite , a necessary and sufficient condition for the existence of an invariant measure for the reduced dynamics of generalized chaplygin systems of mechanical type another recent work on this line is @xcite . for a classic treatment of the theory of chaplygin s reducing multiplier , see section iii-12 of @xcite . thus , it could be conceptually associated as well the existence of specific rescaling factors for these reduced systems with the possibility of formulating them in a hamiltonian way . in addition , borisov , mamaev and kilin @xcite have recently found a poisson structure for each studied case of reduced non - holonomic systems , such that the reduced system becomes hamiltonian , with respect to such a structure , after a rescaling , the hamiltonian function being the reduced energy . the examples treated by them are classical in the literature , consisting mainly of rolling bodies without slipping , namely a rigid body of revolution rolling on a plane , in particular the _ routh s sphere _ ( see section [ sect_routh_sphere ] ) , the rolling disk ( to be treated in section [ sect_rolling_disk ] ) , the motion of a homogeneous ball on a surface of revolution ( called sometimes _ routh s problem _ , see , e.g. , @xcite ) , and other cases . there is a strong emphasis in these references in the sense that the poisson structure for each case can be found _ after a rescaling of time _ of the reduced vector field . our primary motivation for this work was to understand the origin of the two integrals of motion appearing in the mentioned problem of a ball rolling without slipping inside a surface of revolution , which are not given , in general , in an explicit form but being related to the solutions of a system of first order non - autonomous differential equations @xcite . this also happens in the other mentioned cases . the results of @xcite suggest that such systems can be interpreted as the equations providing a set of functionally independent casimir functions of the poisson structure they find for each specific case . therefore , it seemed to be worth investigating further such poisson structures , in particular to clarify their domain of definition and basic properties . let us note that another recent approach , devoted to the study of poisson structures which can be associated to never vanishing vector fields on manifolds of arbitrary dimension @xmath0 , with _ fibrating periodic flows _ , is given in @xcite . it follows that the previously mentioned poisson structures have a rather peculiar form . in particular , the associated characteristic distributions have rank two in the open sets of the reduced spaces considered in @xcite . this property implies that such poisson structures , when multiplied by a never vanishing function , are again poisson structures of the same type . the immediate consequence is that the above mentioned reduced non - holonomic systems are already hamiltonian with respect to one of these poisson structures without any need of rescaling . other interesting result is that , in the cases studied , the poisson structures obtained can be extended from their original domains of definition , namely ( open sets of ) semialgebraic subvarieties of @xmath1 , to an open set of the ambient space . such extended poisson structures become zero only at the so - called _ singular equilibria _ of the reduced systems . moreover , the existence of these ( extended ) poison structures , from an algebraic point of view , is only caused by the existence of integrals of motion of the reduced vector field related to the solutions of the mentioned systems of first order differential equations . this paper is organized as follows . in section [ pois_str_rank2 ] we briefly review some notions of poisson geometry and in particular , of poisson structures of rank two . in section [ pois_str_rank2 ] we show the explicit expressions of certain bivectors in @xmath2 and @xmath1 , determined up to a non - vanishing factor function , by choosing the 1-forms in their kernels to have a specific form , and we prove that they are in fact poisson bivectors of rank two . section [ sect_of_examples ] is devoted to show the application of the previous results in specific examples , namely , the rolling disk , the routh s sphere , and the ball rolling on a surface of revolution . we will use the formulation of @xcite , @xcite and @xcite , respectively , of these problems , rather than that of @xcite . however , we point out the equivalence of both treatments in the last case . we also treat the special case of a ball rolling inside a cylinder . finally , we end with some conclusions and an outlook for further research . for the sake of completeness and in order to fix some notations , we will recall some well - known notions on poisson manifolds , and in particular , we will focus on poisson structures of rank two . for more details see , e.g. , @xcite . given a differentiable manifold @xmath3 , a _ poisson structure _ on @xmath3 is defined by an antisymmetric bilinear map @xmath4 which is a derivation on both of its arguments , satisfying moreover the jacobi identity . a manifold @xmath3 endowed with a poisson structure is called a _ poisson manifold_. thus , it is possible to associate to each function @xmath5 a unique vector field @xmath6 such that , for any other function @xmath7 , we have @xmath8 . the vector field @xmath6 is called _ hamiltonian vector field _ associated to the _ hamiltonian function _ this association defines an homomorphism of the lie algebra @xmath9 onto the lie algebra of vector fields in @xmath3 . a _ casimir function _ or _ casimir _ for short , is a function @xmath10 such that @xmath11 . the poisson structure is called _ non degenerate _ if only the constant functions are casimir functions . moreover , on every poisson manifold , there exists a unique twice contravariant antisymmetric tensor field ( called _ bivector field _ for short ) @xmath12 such that @xmath13 for every pair of functions @xmath14 . this tensor field is called the _ poisson tensor _ of the structure , and the manifold @xmath3 , endowed with its poisson structure , will be denoted @xmath15 . the existence of such a tensor field is due only to the antisymmetry and derivation properties of the poisson bracket . the fulfillment of the jacobi identity for the poisson bracket is equivalent @xcite to the vanishing of the _ schouten nijenhuis bracket _ of @xmath12 with itself , @xmath16=0 $ ] . the schouten nijenhuis bracket @xcite is the unique extension of the lie bracket of vector fields to the exterior algebra of multivector fields . some of its properties are @xmath17=-(-1)^{(p-1)(q-1)}[q , p ] \nonumber\\ & & [ p , q\wedge r]=[p , q]\wedge r+(-1)^{(p-1)q } q\wedge[p , r ] \label{props_scho}\\ & & [ p\wedge r , q]=p\wedge[r , q]+(-1)^{(q-1)r } [ p , q]\wedge r \nonumber\end{aligned}\ ] ] where @xmath18 are completely antisymmetric contravariant tensors of degree @xmath19 , respectively . for more details and properties on the schouten nijenhuis bracket see , e.g. , @xcite and references therein . take a local chart of @xmath3 , with domain @xmath20 and associated local coordinates @xmath21 , where @xmath22 . we will denote by @xmath23 , ( @xmath24 ) the components of the poisson tensor @xmath12 in the previous chart . the expression of the poisson bracket of the restriction of the two functions @xmath25 to @xmath20 , also denoted by @xmath25 , reads @xmath26 where summation in the repeated indices is understood . in particular we have @xmath27 . the poisson tensor admits the local expression @xmath28 in these coordinates . given a poisson manifold @xmath15 , it can be defined the fibered morphism @xmath29 such that for any pair of 1-forms @xmath30 , @xmath31 , @xmath32 . the image of the morphism @xmath33 , @xmath34 , is called _ the characteristic distribution _ of the poisson structure , and the _ characteristic space _ on @xmath35 is the vectorial subspace @xmath36 of @xmath37 . the _ rank _ of the structure on the point @xmath38 is the rank of @xmath39 , i.e. , the dimension of @xmath40 . note that the annihilator of the characteristic distribution , i.e. , @xmath41 , is @xmath42 , and we have @xmath43 , for all @xmath35 . in general , the rank of the structure varies with @xmath38 and thus @xmath44 is not in general a subbundle of @xmath45 . consider now a poisson manifold @xmath15 , @xmath46 , such that in the domain of a local chart @xmath47 the structure has constant rank equal to two . the theorem 11.5 of chapter iii in @xcite ( or corollary 2.3 . in @xcite ) assures us that the associated local coordinates , denoted @xmath48 , can be chosen such that for @xmath49 , @xmath50 we are now in a position to prove a simple result , but important for our purposes here : let @xmath15 be a poisson manifold of ( locally ) constant rank equal to two . then , for each never vanishing smooth function @xmath51 , @xmath52 is a poisson manifold of ( locally ) constant rank equal to two , with the same characteristic distribution . [ prop_conf ] _ proof _ we have to prove that the schouten nijenhuis bracket @xmath53 $ ] vanishes , the other needed properties being obvious . from the paragraph 18.8 of chapter v of @xcite , we have that @xmath54=2 a\ , \l^\sharp(da)\wedge \l\,.\ ] ] it suffices to compute the previous expression on a coordinate neighbourhood like that described in the previous paragraph , with respect to the poisson tensor @xmath12 @xcite . we have @xmath54(dx , dy , dz_k ) = 2 a ( \l^\sharp(da)\wedge \l)(dx , dy , dz_k)=0\,,\quad 1\leq k\leq n-2\,,\ ] ] because @xmath55 are casimir functions of @xmath12 , and @xmath56 enter at least once as argument of @xmath12 in all terms of the previous expression . for other possible arguments the expression vanishes by the same reason . let @xmath3 be a @xmath57-dimensional manifold and @xmath58 , @xmath59 two vector fields such that for all @xmath35 , the lie bracket @xmath60_x$ ] belongs to the subspace of @xmath37 generated by @xmath61 and @xmath62 . then , @xmath63 is a poisson tensor of rank two except where @xmath58 and @xmath59 are linearly dependent . this is easily seen by deducing from the properties of the schouten nijenhuis bracket ( [ props_scho ] ) the relation @xmath64=2 x\wedge y\wedge [ x , y]$ ] , see also @xcite . note that it is essential in proposition [ prop_conf ] the assumption that the initial bivector is poisson , which assures the existence of local coordinates satisfying ( [ pois_prop ] ) . the existence of a bivector whose rank is always two is not enough to conclude that it is a poisson bivector . a simple counter - example is the following . take @xmath65 , with coordinates @xmath66 . let @xmath58 , @xmath59 be vector fields given by @xmath67 then , @xmath68 is an everywhere rank two bivector but is not poisson , since @xmath60=2\pd{}{z}$ ] and @xmath69=[x\wedge y , x\wedge y ] = 4 \pd{}{x}\wedge\pd{}{y}\wedge\pd{}{z}\,.\ ] ] the vector fields @xmath58 , @xmath59 and @xmath60 $ ] in this example close on a lie algebra isomorphic to the heisenberg weyl lie algebra @xmath70 , see , e.g. , @xcite . we will construct in this section some poisson structures of rank two in @xmath2 and @xmath1 by imposing that the kernel of the corresponding bivectors consists of a set of two and three specific 1-forms , respectively . such 1-forms will determine codistributions which are integrable in the sense of frobenius . we will prove that the resulting bivectors are in fact poisson . consider the euclidean space @xmath2 , with coordinates @xmath71 . the equations of motion of the reduced non - holonomic systems encountered in the examples are observed to have integrals of motion which are related to the solutions of a system of differential equations of the type @xmath72 where @xmath73 , @xmath74 , are two given ( smooth ) functions of their arguments , which do not include @xmath75 . we consider the system ( [ syst_first_ints ] ) as the pfaffian system @xmath76 , where the 1-forms @xmath77 in @xmath2 , are given by @xmath78 these two 1-forms determine a codistribution integrable in the sense of frobenius @xcite , since there exist a set of four 1-forms @xmath79 such that @xmath80 for @xmath81 . for example , we can take @xmath82 in order to satisfy the integrability condition . thus , there will exist ( locally ) functions @xmath83 such that @xmath84 , @xmath85 . the subvarieties solution of the pfaffian system @xmath76 are defined by the equations @xmath86 , where @xmath87 are constants , @xmath85 . more specifically , in the actual examples , the system ( [ syst_first_ints ] ) takes the form of a non - autonomous first order system of linear differential equations @xmath88 or , written in matrix form , @xmath89 where @xmath90 the previous functions @xmath91 can be identified with the initial conditions of the solution of ( [ matr_equation ] ) . in fact , such a solution can be expressed as @xmath92 , where @xmath93 , @xmath94 , and @xmath95 is a @xmath96-valued curve ( @xmath97-valued curve if @xmath98 for all @xmath99 ) , solution of the right - invariant matrix system ( see , e.g. , @xcite ) @xmath100 then , @xmath101 gives the desired functions : with a slight abuse of notation , we have @xmath102 where we have used that @xmath103 and @xmath104 . however , note that the solution of ( [ right_inv_matrix_syst_x4 ] ) can not be expressed in an explicit way in the general case , and therefore , the functions @xmath105 , @xmath106 can not be explicitly written in general . now , we impose that the 1-forms ( [ one_forms_first_ints ] ) generate the kernel of the bivector in @xmath2 @xmath107 the resulting bivectors will clearly have rank two . moreover , they are poisson , according to the following result consider in @xmath2 a bivector of type ( [ bivector_x4 ] ) , such that @xmath108 , @xmath109 , where @xmath110 , @xmath111 are given by ( [ one_forms_first_ints ] ) . then the bivector is of the form @xmath112 where @xmath113 and @xmath114 . each of these bivectors is poisson , and of rank two on points where @xmath115 . [ propo_pois_x4 ] _ proof _ the case of @xmath116 is trivial . we will assume @xmath115 in the domain of interest . take @xmath12 and @xmath110 , @xmath111 , as stated . the conditions @xmath108 , @xmath109 give rise to an algebraic system for the six independent functions @xmath23 , which can be easily solved for five of them , in terms of the remaining one and the functions entering into the 1-forms . we choose @xmath117 to be the undetermined function . then the solution reads @xmath118 thus the resulting bivectors are as claimed . to see that each of them is poisson , consider the bivector of the family with @xmath119 , i.e. , @xmath120 . this bivector is of the form given in example 1 , and @xmath121=0 $ ] , thus @xmath120 is poisson . it is moreover of rank two , therefore by proposition [ prop_conf ] , the claim follows . note that the vector fields @xmath122 of the previous theorem satisfy @xmath123 , @xmath85 , which in principle might seem a stronger condition than that the bivector ( [ bivector_x4 ] ) annihilates the 1-forms @xmath77 . now , given a ( hamiltonian ) function @xmath124 , the hamiltonian vector field @xmath125 with respect to a poisson structure of the family described on theorem [ propo_pois_x4 ] takes the form @xmath126\ , , \label{ham_vf_x4}\end{aligned}\ ] ] where @xmath20 and @xmath127 are given by ( [ def_u_v_x4 ] ) . obviously , @xmath128 is a first integral of @xmath125 , since @xmath129 . other two first integrals are the functions @xmath91 such that @xmath130 , since by construction @xmath131 , @xmath85 . these two first integrals are common to all hamiltonian vector fields of type ( [ ham_vf_x4 ] ) . on the other hand , given a specific vector field @xmath58 in @xmath2 , which is recognized to be of the form ( [ ham_vf_x4 ] ) , it could be regarded as a hamiltonian vector field with respect to _ one specific _ poisson structure of the family described in theorem [ propo_pois_x4 ] . we will treat in this section analogous questions to that of the previous section , but now in the euclidean space @xmath1 , with coordinates @xmath132 . the motivation is that typically , the reduced orbit spaces for the non - holonomic problems of interest , are semialgebraic varieties of @xmath1 , essentially determined by the zero level set of a function @xmath133 , quadratic in its arguments , which are moreover subject to certain constraints . more specifically , in the examples it will have the form @xmath134 , with @xmath135 , @xmath136 , and @xmath137 , or with @xmath138 , @xmath139 , and @xmath137 . however , for what follows @xmath140 can be in principle any differentiable function in @xmath1 . we will consider then the pfaffian system @xmath141 , where @xmath142 and @xmath77 are 1-forms in @xmath1 whose coordinate expression is again ( [ one_forms_first_ints ] ) . these three 1-forms also determine determine a codistribution integrable in the sense of frobenius in @xmath1 , because we have again @xmath80 with ( [ one_forms_frob ] ) , @xmath143 and @xmath144 . we impose now that @xmath145 , where @xmath12 is the bivector in ( some open set of ) @xmath1 @xmath146 the resulting bivectors are again generically of rank two and poisson , as follows consider in @xmath1 a bivector of type ( [ bivector_x5 ] ) , such that @xmath147 , @xmath108 and @xmath109 , where @xmath142 , and @xmath110 , @xmath111 are given by ( [ one_forms_first_ints ] ) . then the bivector is of the form @xmath148\ , , \label{pois_biv_x5}\ ] ] where @xmath149 and @xmath150 . each of these bivectors is poisson , and of rank two on points where @xmath151 . [ propo_pois_x5 ] _ proof _ once more , the case of @xmath152 is trivial , thus we will assume again that @xmath151 in the domain of interest . take @xmath12 , @xmath153 , @xmath110 and @xmath111 as stated . the idea of the proof is similar to that of theorem [ propo_pois_x4 ] . first of all , since the kernel of @xmath33 has generically dimension three , then the rank of @xmath33 is two . the conditions @xmath147 , @xmath108 and @xmath109 give rise again to an algebraic system for the functions @xmath23 , out of which all can be solved for except one of them , namely @xmath117 , which we will write as @xmath154 . the solution then reads @xmath155 thus the resulting bivectors take the stated form . to see that each of them is poisson , consider the bivector of the family with @xmath156 , i.e. , @xmath157 , where @xmath158 . we have to show that the schouten nijenhuis bracket of @xmath159 with itself vanish , i.e. , @xmath160=0 $ ] . by linearity and using the first property of ( [ props_scho ] ) we have @xmath161&=&[\overline u\wedge v,\overline u\wedge v]+2\,[\overline u\wedge v , y\wedge z ] + [ y\wedge z , y\wedge z ] \nonumber % \label{scho_l_0}\end{aligned}\ ] ] by example 1 we know that @xmath162 = 2\,\overline u\wedge v\wedge[\overline u , v]$ ] and analogously , @xmath163=2\,y\wedge z\wedge[y , z]$ ] . now , using again the second and third properties of ( [ props_scho ] ) we can write @xmath164&=&v\wedge z\wedge[\overline u , y ] -\overline u\wedge z\wedge[v , y ] \nonumber\\ & & + y\wedge v\wedge[\overline u , z]-y\wedge \overline u\wedge[v , z ] \nonumber\end{aligned}\ ] ] we have to calculate now some lie brackets . we have @xmath121=[v , z]=[u , z]=0 $ ] but @xmath165 = -[v(z\phi)]u\,,\quad [ y , z]=-[z(u\phi)]v+[z(v\phi)]u \nonumber\\ & & \fl\quad\quad [ { \overline u},y]=(z\phi)[u(u\phi)]v -\{(z\phi)[u(v\phi)]+(u\phi)[v(z\phi)]-(v\phi)[u(z\phi)]\}u \nonumber\\ & & \fl\quad\quad [ v , y]=[v(u\phi)]v-[v(v\phi)]u\ , , \quad[{\overline u},z]=-[z(z\phi)]u \nonumber\end{aligned}\ ] ] then , summing up , we have @xmath166=2\ , u\wedge v\wedge z \{(z\phi)([v , u]\phi)+(u\phi)([z , v]\phi)+(v\phi)([u , z]\phi)\}=0\,.\nonumber\end{aligned}\ ] ] since the rank of any of the @xmath12 , and in particular @xmath159 , is two , applying proposition [ prop_conf ] ends the proof . note that the vector fields @xmath167 and @xmath168 of theorem [ propo_pois_x5 ] satisfy @xmath169 , @xmath85 , @xmath170 and @xmath171 . these requirements might seem _ a priori _ to be stronger conditions to that imposed in the theorem . if we are given now a ( hamiltonian ) function @xmath172 , the hamiltonian vector field @xmath125 with respect to a poisson structure of the family described in theorem [ propo_pois_x5 ] reads , using ( [ def_y_x5 ] ) , @xmath173\,u \label{ham_vf_x5}\\ & & \quad+[(z\phi)(uh)-(zh)(u\phi)]\,v + [ ( u\phi)(vh)-(v\phi)(u h)]\,z \big\}\ , , \nonumber\end{aligned}\ ] ] where @xmath20 , @xmath127 and @xmath168 are given by ( [ def_u_v_z_x5 ] ) . by construction @xmath128 is a first integral of @xmath125 . other first integrals are the functions @xmath91 such that @xmath130 , as in the previous section . these two first integrals are common to all hamiltonian vector fields of type ( [ ham_vf_x5 ] ) . however , given a specific vector field @xmath125 of type ( [ ham_vf_x5 ] ) , it fixes the specific function @xmath5 and therefore the specific poisson bivector of the family ( [ pois_biv_x5 ] ) with respect to which @xmath125 is hamiltonian . in this section we will show how the preceding results can be directly applied in the cases of reduced systems corresponding to specific examples of non - holonomic systems , i.e. , the rolling disk , the routh s sphere , the ball rolling on a surface of revolution and its special case of a ball rolling inside a cylinder . for this example we will follow the treatment and use some of the results of @xcite , see details therein . this problem has been treated as well , e.g. , in @xcite . consider a homogeneous disk , which rolls without slipping on a horizontal plane under the influence of a vertical gravitational field of strenght @xmath7 . the resulting non - holonomic system has two evident symmetry groups . one is the symmetry group @xmath174 consisting of translations in the horizontal plane and rotations about the vertical axis , and the second is the @xmath175 symmetry consisting of rotations about the principal axis perpendicular to the plane of the disk . after these two symmetries have been reduced out , in particular by using invariant theory for the reduction of the @xmath175 symmetry , it is obtained a system giving the evolution on the reduced orbit space , which is a semialgebraic variety of @xmath1 . in particular , the system can be restricted to a smooth open subset as it has been done in @xcite . thus , consider a reference homogeneous disk of radius @xmath176 and mass @xmath177 , lying flat in a fixed reference frame with center of mass at the origin . the position of the moving disk is given by transforming the position of the reference disk by means of a translation @xmath178 ( e.g. , of the center of mass ) and a rotation @xmath179 . the tensor of inertia @xmath180 with respect to the principal axes of the disk is diagonal , @xmath181 . let us call @xmath182 the vertical unitary vector in the fixed frame of reference . we define the unitary vector @xmath183 with respect that frame as the pre - image of @xmath184 under the rotation @xmath179 , @xmath185 . the vector @xmath186 in the fixed disk , rotated by @xmath179 gives the vector in the moving disk pointing from the center of mass to the point of contact of the moving disk with the horizontal plane . if we denote @xmath187 , the relation between @xmath186 and @xmath183 is @xmath188 . we denote by @xmath189 the components of the angular velocity vector @xmath190 of the disk , following @xcite , after the mentioned symmetry group @xmath174 is reduced out , the equations of motion read @xmath191 which have a first integral given by the total energy of the disk @xmath192 the second of eqs . ( [ eqs_mot_roll_disk ] ) expresses the non - holonomic constraint of rolling without slipping , i.e. , instantaneous velocity of the point of contact equal to zero . we recall briefly now how the further reduction of the @xmath175 symmetry is performed . let us denote by @xmath193 the components of @xmath183 . the @xmath175 symmetry action consists of rotating both vectors @xmath183 and @xmath190 simultaneously as mentioned , and it is not a free action since the isotropy subgroup of pairs @xmath194 is @xmath175 . thus , we will use invariant theory in order to perform the reduction . a set of invariants for this action is easily constructed @xcite : @xmath195 with the relations @xmath196 since @xmath183 is a unitary vector , we have that @xmath197 and @xmath198 , thus the completely reduced orbit space @xmath3 is the semialgebraic variety of @xmath1 @xmath199 where @xmath133 is the polynomial function @xmath200 . however , @xmath3 is _ not _ a smooth submanifold of @xmath1 . the singular points of @xmath3 are @xmath201 the non - smoothness of @xmath3 is due to the fact that the @xmath175 action is not free , see @xcite . the somehow redundant variables @xmath202 therefore parametrize the reduced orbit space @xmath3 . the induced system from ( [ eqs_mot_roll_disk ] ) will be written in terms of the orbit variables : simply calculating their time - derivatives , using the equations of motion ( [ eqs_mot_roll_disk ] ) and that @xmath203 and @xmath204 , we arrive to the following system @xmath205 where @xmath206 and the dot means derivative with respect to time . the reduced energy , obtained from ( [ ener_mot_roll_disk ] ) , reads @xmath207 although in principle the expressions ( [ sist_sig_5_roll_disk ] ) and ( [ ener_sig_5_roll_disk ] ) are only defined on @xmath3 , their right hand sides make sense for @xmath208 , so we will consider this extended domain for the vector field @xmath58 whose integral curves are given by ( [ sist_sig_5_roll_disk ] ) and the reduced energy function @xmath209 . however , if we restrict ourselves to the original domain @xmath3 , and morever to points with @xmath210 , we can define a smooth open dense subset @xmath211 given by @xmath212 diffeomorphic to @xmath2 @xcite . the induced vector field @xmath213 and energy @xmath214 on @xmath215 can be easily found from ( [ sist_sig_5_roll_disk ] ) and ( [ ener_sig_5_roll_disk ] ) by solving for @xmath216 . the integral curves of @xmath213 are the solutions of the system @xmath217 meanwhile @xmath218 these expressions are eqs . ( 18 ) and ( 19 ) of @xcite , respectively . the reduced vector field @xmath58 satisfies @xmath219 as well as @xmath220 in @xmath221 , meanwhile @xmath222 in @xmath215 . in addition , @xmath58 has a family of equilibrium points belonging to the singular set @xmath223 , called _ singular equilibria _ , given by @xmath224 , and a family of _ regular equilibria _ given by the set of constants @xmath225 these regular equilibria , in the original system , correspond to periodic motions of the disk in which the point of contact describes a circle and the center of mass stands at constant height . these motions are contained in the set of _ steady motions _ of the rolling disk , according to routh s terminology @xcite . they have received an extensive treatment in @xcite , although by using the system ( [ sist_rd_sig4 ] ) . now , both of the systems ( [ sist_sig_5_roll_disk ] ) and ( [ sist_rd_sig4 ] ) admit two first integrals related to the solutions ( in the sense explained in section [ pois_struc_r4 ] ) of the non - autonomous linear system @xmath226 which can be written in matrix form as @xmath227 this equation is the same as eq . ( 69 ) of @xcite , where its solutions have been studied in great detail , including their asymptotic behaviour . however , the important point for us is that the systems ( [ sist_sig_5_roll_disk ] ) and ( [ sist_rd_sig4 ] ) are good candidates to be formulated as hamiltonian systems with respect to poisson structures of the type described in theorems [ propo_pois_x5 ] and [ propo_pois_x4 ] , respectively . let @xmath142 and @xmath110 , @xmath111 be the 1-forms , defined in @xmath215 ( resp . @xmath221 ) by @xmath228 applying the results of sections [ pois_struc_r4 ] and [ pois_struc_r5 ] to these 1-forms , we have the bivectors of the form @xmath229 , defined in @xmath215 , where @xmath230 and @xmath231 is a non - vanishing function , are poisson tensors of rank two in @xmath215 . the vector field @xmath213 in @xmath215 , whose integral curves are the solutions of ( [ sist_rd_sig4 ] ) , is a hamiltonian vector field with respect to the poisson bivector @xmath232 with the specific function @xmath233 and hamiltonian function @xmath214 given by ( [ ener_rd_sig4 ] ) , i.e. , @xmath234 in @xmath215 . [ prop_lam_s_roll_disk_s4 ] the bivectors @xmath235 $ ] , defined in @xmath236 , where @xmath20 and @xmath127 are given by ( [ def_u_v_z_x5_roll_disk ] ) , @xmath237 , @xmath238 , and @xmath239 is a non - vanishing function , are poisson tensors of rank two in @xmath221 , except in the set of singular equilibria , where they vanish . the vector field @xmath58 in @xmath221 , whose integral curves are the solutions of ( [ sist_sig_5_roll_disk ] ) , is a hamiltonian vector field with respect to the poisson bivector @xmath12 with the specific function @xmath156 and hamiltonian function @xmath209 given by ( [ ener_sig_5_roll_disk ] ) , i.e. , @xmath240 in @xmath221 . both propositions can be proved by direct computations . the poisson hamiltonian structure of the systems ( [ sist_sig_5_roll_disk ] ) and ( [ sist_rd_sig4 ] ) could be used to have an interpretation of their geometry . for example , the invariant submanifolds mentioned in the analysis of the reduced vector field ( [ sist_rd_sig4 ] ) in @xcite , could be understood as the symplectic leaves of the rank - two poisson structure(s ) @xmath232 of proposition [ prop_lam_s_roll_disk_s4 ] . for this example we will follow the treatment and use some of the results of @xcite , see details therein . this problem has been treated as well , e.g. , in @xcite . consider a sphere of mass @xmath177 and radius @xmath176 with its center of mass at a distance @xmath30 ( @xmath241 ) from its geometric center . the line joining both centers is a principal axis of inertia , with associated moment of inertia @xmath242 . any axis orthogonal to the previous , passing though the geometric center , has an associated moment of inertia @xmath243 . this sphere is supposed to roll on a horizontal plane under the influence of a vertical gravitational field of strenght @xmath7 . the resulting non - holonomic system has as well two symmetry groups . one is again the group @xmath174 consisting of translations in the horizontal plane and rotations about the vertical axis . the other is the @xmath175 symmetry consisting of rotations about the principal axis of inertia which joins the center of mass and the geometric center of the ball . again , after these symmetries have been reduced out by a similar procedure to that of the rolling disk , it is obtained a system giving the evolution on the reduced orbit space , which coincides with that of the rolling disk . therefore , let us consider a reference ball as the one described , with the geometric center at the origin , and the center of mass at the point @xmath244 , where @xmath182 denotes the vertical unitary vector in this fixed frame . the position and attitude of the moving ball is given by transforming the position of the reference ball by means of a translation @xmath178 ( e.g. , of the center of mass ) and a rotation @xmath179 . we denote by @xmath186 the vector in the fixed sphere such that rotated by @xmath179 gives the vector in the moving sphere pointing from the center of mass to the point of contact . the unitary vector @xmath183 in the fixed frame is the pre - image of @xmath184 under the rotation @xmath179 , @xmath185 . the relation between @xmath183 and @xmath186 is @xmath245 . the components of the angular velocity @xmath190 of the ball will be denoted by @xmath189 . following @xcite , after the reduction of the mentioned @xmath174 symmetry , the equations of motion read @xmath246 which have a first integral given by the total energy of the ball @xmath247 the second of eqs . ( [ eqs_mot_routh_esf ] ) expresses again the non - holonomic constraint of rolling without slipping . now , the reduction of the @xmath175 symmetry is performed in an analogous way as in the case of the rolling disk , see section [ sect_rolling_disk ] , where @xmath193 denote as well the components of @xmath183 . the @xmath175 action consists of rotating both vectors @xmath183 , @xmath190 simultaneously , with respect to the principal axis joining the geometric and mass centers . this action is not free , since @xmath175 leaves invariant pairs of points of the form @xmath194 . the corresponding set of invariants is again ( [ inv_sigma ] ) with the relations ( [ rels_inv_sigma ] ) . thus , the reduced orbit space @xmath3 is the semialgebraic variety of @xmath1 described in the previous example of the rolling disk , with the same notations . however , the reduced system reads now , using ( [ eqs_mot_routh_esf ] ) , @xmath248 where @xmath249 and @xmath250 . the reduced energy is @xmath251 these are eqs . ( 23 ) , ( 24 ) and ( 25 ) in @xcite . in this case the expressions ( [ sist_sig_5_routh_esf ] ) and ( [ ener_sig_5_routh_esf ] ) make sense for all @xmath252 . we will consider this extended domain for the vector field @xmath58 whose integral curves are given by ( [ sist_sig_5_routh_esf ] ) and for the reduced energy function @xmath209 . restricting ourselves to points in @xmath3 with @xmath210 , we find again the smooth submanifold @xmath211 given by ( [ smooth_subman ] ) . the integral curves of the projected vector field @xmath213 are the solutions of the system ( @xcite , eq . ( 38 ) ) @xmath253 and the restricted reduced energy @xmath214 is @xmath254 the reduced vector field @xmath58 satisfies @xmath219 and @xmath220 in @xmath221 , and @xmath222 in @xmath215 . moreover , @xmath58 has a family of singular equilibrium points , belonging to the singular set @xmath223 , given by @xmath224 , which physically correspond to the spinning of the ball about its symmetry axis when it is vertical ( then the reduced energy becomes @xmath255 ) . it has as well a family of regular equilibria given by the set of constants @xmath256 where @xmath257 . these regular equilibria , in the original system , correspond to periodic motions of the ball in which the point of contact describes a circle and the center of mass stands at constant height . in this case , both of the systems ( [ sist_sig_5_routh_esf ] ) and ( [ sist_resf_sig4 ] ) admit two first integrals related to the solutions ( in the sense of section [ pois_struc_r4 ] ) of the non - autonomous linear system @xmath258 thus , the mentioned systems are other good candidates on which to apply the poisson approach of section [ pois_struc_r4_r5 ] . let @xmath142 and @xmath110 , @xmath111 be the 1-forms , defined in @xmath215 ( resp . @xmath221 ) by @xmath259 we have the following results : the bivectors of the form @xmath229 , defined in @xmath215 , where @xmath260 , @xmath261 and @xmath231 is a non - vanishing function , are poisson tensors of rank two in @xmath215 . the vector field @xmath213 in @xmath215 , whose integral curves are the solutions of ( [ sist_resf_sig4 ] ) , is a hamiltonian vector field with respect to the poisson bivector @xmath232 with the specific function @xmath262 and hamiltonian function @xmath214 given by ( [ ener_resf_sig4 ] ) , i.e. , @xmath234 in @xmath215 . [ prop_lam_s_routh_esf_s4 ] the bivectors @xmath235 $ ] , defined in @xmath252 , where @xmath20 and @xmath127 are given as in proposition [ prop_lam_s_routh_esf_s4 ] , @xmath237 , @xmath238 , and @xmath239 is a non - vanishing function , are poisson tensors of rank two in @xmath221 , except in the set of singular equilibria , where they vanish . the vector field @xmath58 in @xmath221 , whose integral curves are the solutions of ( [ sist_sig_5_routh_esf ] ) , is a hamiltonian vector field with respect to the poisson bivector @xmath12 with the specific function @xmath263 and hamiltonian function @xmath209 given by ( [ ener_sig_5_routh_esf ] ) , i.e. , @xmath240 in @xmath221 . both propositions can be proved as well by direct computations . in this case , the equations ( [ sys_chap_rs ] ) can be explicitly integrated in an easy way . from the second of these equations we have the relation @xmath264 . substituting into the first , we can also integrate to obtain the relation @xmath265 . the constants @xmath266 are integration constants ( essentially the initial conditions of the system ( [ sys_chap_rs ] ) ) . these two expressions are the desired first integrals ( casimir functions of the preceding poisson structures ) . the second of them is known as _ jellet s integral _ , see @xcite and references therein , see also p. 184 of @xcite . the invariant submanifolds thoroughly studied in @xcite , could be interpreted in this framework as the symplectic leaves of the rank - two poisson structure(s ) @xmath232 of proposition [ prop_lam_s_routh_esf_s4 ] , determined by the level sets of the first integrals @xmath267 and @xmath268 . for this example we will follow the treatment and use some results of @xcite , see therein for more details . this problem has been treated as well , e.g. , in @xcite . in particular routh , in the last of these references , noticed the existence of two integrals of motion given by a system of two linear differential equations , solved them in special cases , and described a family of stationary periodic motions together with a necessary condition for their stability . later , in @xcite , it has been shown that the condition is also sufficient . both of @xcite and @xcite prove that the corresponding reduced system has integral curves consisting of either periodic orbits or equilibrium points . consider a homogeneous ball of mass @xmath177 , radius @xmath176 and moment of inertia @xmath3 with respect to any principal axis . the ball rolls without slipping on a surface of revolution , under the influence of a vertical gravitational field of strenght @xmath7 . we take the origin of coordinates at a point of the axis of symmetry of the surface ( the intersection of this axis with the surface at its vertex ) , and we consider a horizontal plane passing through it . we parametrize the position of the center of mass of the ball by its coordinates @xmath269 on this horizontal plane , and its height will be parametrized via the smooth _ profile function _ @xmath270 of the surface , @xmath271 . note that not all surfaces of revolution can be parametrized well in this way , e.g. , the cylinder , which requires a separate treatment , see section [ ball_in_a_cylinder ] below . we will assume that @xmath272 is a smooth even function , thus we will have that @xmath273 , @xmath274 we denote by @xmath189 the components of the angular velocity vector @xmath190 of the ball , and @xmath275 the components of a unit vector @xmath276 normal to the surface at the point of contact ( directed towards the center of the ball ) . the unit vector in the vertical coordinate axis is @xmath182 . the equations of motion can be easily computed by the classical equations of the variation of the angular momentum , and implementing the non - holonomic constraint of non - slipping of the point of contact , i.e. , that its instantaneous velocity vanishes . they read ( with respect to the center of mass of the ball , compare with eqs . ( 5 ) , ( 7 ) of @xcite and section 2 of @xcite ) @xmath277 the total mechanical energy of this system is then @xmath278 and is a first integral for the system ( [ eqs_rolling_ball ] ) . the system ( [ eqs_rolling_ball ] ) and the energy ( [ ener_ball_surf ] ) admit a further reduction of the @xmath175 symmetry consisting of rotations of the system about the vertical axis , and thus rotating both of @xmath190 and @xmath276 simultaneously . this action , as in the previous cases , is not free , since the isotropy subgroup of pairs @xmath279 is @xmath175 ( these pairs correspond to motions of the ball spinning around the vertical axis when being at the vertex of the surface ) , and we will use again invariant theory in order to perform the reduction , but now as it has been done in @xcite . first of all , we define the vector @xmath280 and the scalar @xmath281 as follows : @xmath282 , @xmath283 . then , a full set of invariant polynomials , which parametrize the orbit space of the @xmath175 action , is @xmath284 with the relations @xmath285 therefore , the completely reduced orbit space @xmath286 is now the semialgebraic variety of @xmath1 @xmath287 where now @xmath133 is the polynomial function @xmath288 . @xmath286 is _ not _ a smooth submanifold of @xmath1 , because the previous @xmath175 action is not free . instead , @xmath286 is homeomorphic to a cone in @xmath2 times @xmath289 @xcite , which can be easily seen from the relation @xmath290 when it is written as @xmath291 . the vertex of the cone is determined by @xmath292 , therefore the singular points of @xmath286 are @xmath293 calculating the time derivatives of the invariants , using ( [ eqs_rolling_ball ] ) , and the relations @xmath294 ( we will use the notation @xmath295 , @xmath296 and @xmath297 in what follows ) we arrive to the system in the reduced orbit space @xmath286 @xmath298 and the reduced energy @xmath299 where @xmath300 . these are the equations found in lemmas 2.2 and 2.3 ( i ) of @xcite . we observe that the right hand sides of ( [ eqs_herm ] ) and ( [ ered_herm ] ) make sense in an open set @xmath221 of @xmath1 larger than @xmath286 , namely @xmath301 . this is due to the fact that they are defined in the limit @xmath302 , because of the assumption that the odd - order derivatives at 0 of @xmath272 vanish . ( for points strictly in @xmath286 with @xmath303 this assumption would not be necessary , since these points also have @xmath304 , @xmath305 ) . we will consider the enlarged domain @xmath221 for the vector field @xmath58 whose integral curves are the solutions of ( [ eqs_herm ] ) , and also for the reduced energy ( [ ered_herm ] ) , compare with p. 500 of @xcite . the regular stratum of @xmath286 , i.e. , @xmath306 , can be covered by two charts @xcite , whose corresponding neighbourhoods can be chosen to be the smooth open dense subsets @xmath307 given by @xmath308 however , for our purposes here , it will be enough to consider just @xmath309 , in order to endow it with poisson structures of the type described in theorem [ propo_pois_x4 ] , which afterwards could be compared with the poisson structure given originally in @xcite . the procedure for @xmath310 is analogous . thus , we will consider the induced vector field @xmath213 and energy @xmath214 on @xmath309 , which can be found from ( [ eqs_herm ] ) and ( [ ered_herm ] ) by solving for @xmath311 . the integral curves of @xmath213 are the solutions of the system @xmath312 meanwhile @xmath313 now , the reduced vector field @xmath58 satisfies @xmath219 and @xmath220 in @xmath221 , and @xmath222 in @xmath309 . the vector field @xmath58 has a family of singular equilibrium points consisting of the singular set @xmath314 , that is , @xmath315 , which as already mentioned , correspond to the spinning of the ball about the vertical when being at the vertex of the surface ( then the reduced energy becomes @xmath316 ) . @xmath58 has as well a family of regular equilibria given by the set of constants @xmath317 these regular equilibria correspond in the original system to rotations of the ball along a parallel of the surface of revolution at constant height . in addition , both of the systems ( [ eqs_herm ] ) and ( [ eqs_herm_rd_p4 ] ) admit two first integrals of motion related to the solutions ( in the sense explained in section [ pois_struc_r4 ] ) of the non - autonomous linear system ( see also @xcite , lemma 2.3 . ( ii ) ) @xmath318 let @xmath142 and @xmath110 , @xmath111 , be the 1-forms , defined in @xmath309 ( resp . @xmath221 ) by @xmath319 we have the following results , applying the theorems of section [ pois_struc_r4_r5 ] , which can be proved by direct computations : the bivectors of the form @xmath229 , defined in @xmath309 , where @xmath320 and @xmath321 is a non - vanishing function , are poisson tensors of rank two in @xmath309 . the vector field @xmath213 in @xmath309 , whose integral curves are the solutions of ( [ eqs_herm_rd_p4 ] ) , is a hamiltonian vector field with respect to the poisson bivector @xmath232 with the specific function @xmath322 and hamiltonian function @xmath214 given by ( [ ered_herm_rd_p4 ] ) , i.e. , @xmath234 in @xmath309 . [ hamilton_eqs_roll_ball_surf_p4 ] the bivectors @xmath235 $ ] , defined in @xmath236 , where @xmath20 and @xmath127 are given by ( [ def_u_v_z_x5_ball_surf ] ) , @xmath323 , @xmath238 , and @xmath239 is a non - vanishing function , are poisson tensors of rank two in @xmath221 , except in the set of singular equilibria , where they vanish . the vector field @xmath58 in @xmath221 , whose integral curves are the solutions of ( [ eqs_herm ] ) , is a hamiltonian vector field with respect to the poisson bivector @xmath12 with the specific function @xmath324 and hamiltonian function @xmath209 given by ( [ ered_herm ] ) , i.e. , @xmath240 in @xmath221 . for the present case , a poisson structure analogous to one of the structures @xmath232 of proposition [ hamilton_eqs_roll_ball_surf_p4 ] has been found , to the best of our knowledge , by the first time in @xcite , see their equation ( 3.11 ) for @xmath325 . in fact , up to a rescaling , they are the same , by using the identifications @xmath326 the ( local ) poisson bivector found in @xcite for this case reads in their coordinates @xmath71 as @xmath327 which , in particular , is also of the type described in example 1 . therefore , multiples of this bivector are again poisson bivectors and hence , the rescaling introduced in @xcite , by means of an _ invariant measure _ , in order to render the reduced system hamiltonian , is unnecessary . on the other hand , hermans in @xcite has not noticed the existence of any of these poisson structures of rank two but he constructed a closed 2-form , with domain contained in @xmath309 , which vanish in a set containing the set of regular equilibria , but has rank four otherwise . for this construction , which uses _ non - holonomic reduction _ @xcite , it is indeed _ necessary to rescale _ the original reduced vector field , see section 4.1 of @xcite . in this section we will treat the special case of a ball rolling inside of a cylinder , which can not be parametrized as in section [ ball_surf_revol ] . in contrast with the general case , this case is completely and explicitly solvable , as it is well known , see , e.g. , @xcite . however , we will give an independent treatment . for this specific system , we will easily find a family of poisson structures of rank two , generated by two of them , with respect to which the reduced system is hamiltonian with the reduced energy as hamiltonian function . consider therefore the case of the ball rolling inside a surface of revolution , with the following variations : the center of mass of the ball will be parametrized by the vector @xmath178 , with cylindrical coordinates @xmath328 , where @xmath329 is the radius of the cylinder on which the center of mass of the ball moves , and @xmath330 is the height with respect to the gravitational energy reference point . the normal vector @xmath276 reads then @xmath331 . the system ( [ eqs_rolling_ball ] ) becomes in the coordinates @xmath189 and @xmath332 @xmath333 likewise , the energy ( [ ener_ball_surf ] ) reads now @xmath334 which is conserved by the system ( [ syst_cyl ] ) . obviously , @xmath335 is a first integral of the system as well . let us consider now the system obtained after the reduction of the @xmath175 symmetry of rotations of the whole system about the vertical axis , as in the general case . although now the @xmath175 action is free , we will use again invariant theory in order to perform the reduction . consider the invariants similar ( but not equal ) to ( [ inv_sigma ] ) : @xmath336 which in this case can be regarded as coordinates on @xmath2 . then , the reduced system for @xmath337 reads @xmath338 which preserves the reduced energy @xmath339 the reduced vector field @xmath58 in the reduced space reads then @xmath340 and we have @xmath219 in all points of @xmath2 . the general solution of ( [ sist_red_sigma_ball_cyl ] ) can be given explicitly . it reads @xmath341 where we have defined the constants @xmath342 , @xmath343 and @xmath344 . it is clear that the reduced system , if @xmath345 , has integral curves consisting of either periodic orbits or equilibrium points , belonging to the set @xmath346 . these equilibrium points correspond to rotations of the ball inside the cylinder at constant height , as in the general case . on this occasion , the reduced system can be reconstructed easily to the complete system , thus the general solution of ( [ syst_cyl ] ) is @xmath347 where @xmath348 and @xmath349 , @xmath350 are given by ( [ sol_sist_red_sigma_ball_cyl ] ) . if we denote @xmath351 , @xmath352 , we have the relations for the initial conditions @xmath353 the complete system , when @xmath354 , is then isochronus with two frequencies , the motions being periodic ( relative equilibria , projecting to equilibrium points in the reduced space ) or quasi - periodic , otherwise . the solutions with @xmath355 correspond to _ falling motions _ of the ball , rolling along a vertical generatrix of the cylinder . the explicit expression of these solutions is @xmath356 we will treat now the question of writting the vector field @xmath58 , given by ( [ red_vf_ball_cyl ] ) , as a hamiltonian vector field with respect to a poisson structure of rank two , with hamiltonian function @xmath209 . we first observe that the reduced vector field @xmath58 is annihilated by the 1-forms @xmath357 and then , it is easy to apply theorem [ propo_pois_x4 ] , to obtain the following results : the bivector @xmath358 is a poisson bivector on @xmath2 of rank two such that @xmath359 . likewise , the bivector @xmath360 $ ] is a poisson bivector on @xmath2 of rank two such that @xmath361 . in addition , we have @xmath362 , where @xmath58 is given by ( [ red_vf_ball_cyl ] ) and @xmath209 by ( [ e_red_ball_cyl ] ) . [ prop_pois_1_cyl ] now , the pfaff systems @xmath363 and @xmath364 can be easily integrated , giving non - trivial casimir functions of @xmath365 , and first integrals of @xmath58 : we have @xmath366 , and @xmath367 , where @xmath368 , @xmath369 and @xmath370 . [ prop_cas_pois_1_cyl ] as a consequence , we have that the reduced vector field @xmath58 has in principle four first integrals , namely , @xmath209 , @xmath105 , @xmath106 and @xmath371 , but clearly , they form a functionally dependent set . however , for example , we have that @xmath372 is generically an independent set of first integrals , although in the equilibrium points one becomes dependent of the other two . in the falling motions , @xmath373 , therefore @xmath374 and @xmath375 become proportional . incidentally , we also observe that @xmath376 and @xmath377 . in addition , the poisson bivectors @xmath374 , @xmath375 are compatible in the sense that their schouten nijenhuis bracket vanishes , @xmath378=0 $ ] , which can be checked , e.g. , using the properties ( [ props_scho ] ) . thus we have the following result : the pencil of bivectors @xmath379 consists of poisson bivectors of rank two such that @xmath380 for all @xmath381 . moreover , the functions @xmath105 , @xmath382 and @xmath383 are ( functionally dependent ) casimir functions of @xmath384 ( and therefore , first integrals of @xmath58 ) for all @xmath381 . [ prop_cas_pois_lambda_cyl ] _ proof _ that the rank of @xmath384 , for all @xmath381 , is two , is obvious when one realizes that it does not contain terms on @xmath385 and therefore the rank must be an even number between zero and four . the other statements are a matter of computation using the above observations . we have shown the form of certain poisson structures of rank two with respect to which certain reduced problems of non - holonomic mechanics become hamiltonian . we have shown that in @xmath2 and @xmath1 , from an algebraic point of view , these poisson structures are defined , up to a factor function , by the choice of the kernel of bivectors on these spaces to be generated by 1-forms of a specific type . such 1-forms define integrable codistributions in the sense of frobenius , and are chosen in order to accommodate and generalize the systems of first order non - autonomous differential equations which appear after reduction in certain non - holonomic mechanical systems , whose solutions are related to first integrals of such reduced systems . we have applied the theory to the cases of the rolling disk , the routh s sphere , and the ball rolling on a surface of revolution , explicitly recovering as a particular case some results of @xcite . thus , we have shown that the framework suggested by borisov , mamaev and kilin @xcite can be improved along the lines discussed , namely , that those reduced systems need no rescaling to become hamiltonian with respect to a poisson structure of rank two , and that the domain of definition of the poisson structures introduced therein can be extended , including even the set of singular equilibria of the reduced systems . a natural question is whether a similar approach could be used in other non - holonomic systems , maybe of higher dimension . however , there are more fundamental points still to be better understood . for example , to what extent the mentioned poisson structures can be useful to investigate the intimate nature of these and maybe other non - holonomic systems , for example in order to characterize their integrability properties @xcite ; see also the recent work @xcite . another question could be to clarify the origin of the system of differential equations giving the conservation laws for the mentioned reduced non - holonomic systems , see also @xcite and references therein . this work is part of the research contract hprn - ct-2000 - 00113 , supported by the european commission funding for the human potential research network mechanics and symmetry in europe ( masie ) . the author is specially indebted to f. fass for valuable comments , questions and ideas , also concerning previous versions of this article . likewise , the author acknowledges useful comments and warm hospitality from j.f . cariena , r. cushman and t. ratiu at their respective institutions ( universidad de zaragoza , university of utrecht and epfl ) . aaaaa cariena j.f . and ramos a. , _ applications of lie systems in quantum mechanics and control theory _ , in classical and quantum integrability , grabowski j. , marmo g. and urbaski p. eds . , banach center publications * 59 * , polish academy of sciences , warszawa , 2003 . marle ch .- , _ on symmetries and constants of motion in hamiltonian systems with nonholonomic constraints _ , in classical and quantum integrability , grabowski j. , marmo g. and urbaski p. eds . , banach center publications * 59 * , polish academy of sciences , warszawa , 2003 . schouten j.a . , _ on the differential operators of first order in tensor calculus _ , in convegno internazionale di geometria differenziale . italia , 20 - 26 settembre 1953 , edizioni cremonese della casa editrice perrella , roma , 1954 .

we present timing and spectral analysis of 2.2 ms of rossi x - ray time explorer ( _ rxte _ ) archival data from cyg x-1 . using a generic comptonization model we reveal that the spectrum of cyg x-1 consists of three components : a thermal seed photon spectrum , a comptonized part of the seed photon spectrum and the iron line . we find a strong correlation between the 0.1 - 20 hz frequencies of quasiperiodic oscillations ( qpos ) and the spectral index . presence of two spectral phases ( states ) are clearly seen in the data when the spectral indices saturate at low and high values of qpo frequencies . this saturation effect was discovered earlier in a number of black hole candidate ( bhc ) sources and now we strongly confirm this phenomenon in cyg x-1 . in the soft state this index - qpo frequency correlation shows a saturation of the photon index at high values of the low frequency . the saturation level of is the lowest value found yet in bhcs . the bolometric luminosity does not show clear correlation with the index . we also show that fe k emission line strength ( equivalent width , ew ) correlates with the qpo frequency . the ew increases from 200 ev in the low / hard state to 1.5 kev in the high / soft state . the observational correlations revealed compel us to propose a scenario for the spectral transition and iron line formation which occur in bhc sources . we also present the spectral state ( power - law index ) evolution for eight years of cyg x-1 observations by _

astro-ph0602091

cyg x-1 is one of the brightest high - energy sources in the sky , with an average 1 - 200 kev energy flux of @xmath4 ergs @xmath5s@xmath6 . its optical companion is an o9.7 iab supergiant hde 226868 . estimates of the mass , m , of the x - ray star , @xmath7 [ e.g. , @xcite ] strongly suggest the presence of a black hole . observed spectral and temporal x - ray characteristics are extensively studied based on the large amount of data collected in the _ rxte _ archive ( see 2 for the data description ) . our analysis includes @xmath0 2.2 ms of rossi x - ray time explorer ( _ rxte _ ) archival data from cyg x-1 to study the spectral and timing properties of this classical bhc source . one of the basic questions addressed in many observational and theoretical studies concerning relativistic compact objects is how to observationally distinguish between a neutron star ( ns ) and a black hole ( bh ) . cyg x-1 , being extensively studied , has often been used as the prototypical example of a bh . the different patterns , for example correlations between spectral and timing characteristics of bh and ns sources , has been proposed as a criteria for determination of the nature of the compact object . recently belloni ( 2005 ) and @xcite published a concise review of the observational features of the spectral states in bh sources where they also point out a link between timing and spectral properties of x - ray radiation and plasma ejection leading to radio jets . @xcite , hereafter tf04 , and @xcite , hereafter tsh05 , present theoretical and observational arguments how to distinguish between bh and ns binaries . tf04 present observational evidence that in bhs two distinct phases occur : one of them , the steep power - law phase ( so `` called '' high / soft state ) , is the signature of a bh . in the soft state of bh the spectral index - quasiperiodic oscillation ( qpo ) frequency correlation shows a flattening , or `` saturation '' of the photon index @xmath8 at high values of the low frequency @xmath9 . this saturation effect was identified as a bh signature . tsh05 demonstrate that this saturation is not present in at least one ns source . they show that for 4u 1728 - 34 the index @xmath10 increases monotonically with @xmath9 . we show here that cyg x-1 is a perfect example of a bh source as the suggested bh index - qpo frequency correlation is observed with clear features of the saturation at high and low frequencies . long - term monitoring of cyg x-1 has revealed two distinct spectral states and transitions between them [ see a review of cyg x-1 early observations in @xcite ] . simultaneous , low- and high energy x - ray observations during interstate transitions have been obtained by several groups in 1996 [ see references in @xcite and @xcite ] . in the low / hard state , the power - law portion of the spectrum is relatively flat with a photon index @xmath10 of about 1.5 . they found that a majority of the time cyg x-1 stayed in the low / hard with an occasional transition ( during their observations , duty cycle of this state was about 90 % ) to the soft state where the power - law spectrum became significantly steeper ( with @xmath11 ) . also , one or two times per year cyg x-1 exhibited so called `` failed state transitons '' , when it started to transition but did not reach a soft state , stopping at some intermediate state and their falling back . thus one can claim that the source was predominantly in a hard state in 90s . however wilms et al . ( 2005 ) have recently shown that since 2000 the source spent @xmath12 of the time in the intermediate and the soft state . we came to the same conclusion as a result of our analysis of the source spectral transition ( see more discussion of this phenomenon in 4 , the last paragraph ) . as @xcite pointed out there is strong evidence that the observed qpo characteristics are related to spectral properties : the qpo amplitude increases as the energy spectrum becomes harder . they also discovered qpo low - frequencies varying in the range of 4 - 10 hz during the spectral transition . in fact , using these observations di matteo & psaltis ( 1999 ) , hereafter dp99 , found that the photon index can be correlated with the qpo frequency . one can see a few points of this correlation in the frequency range from 1 to 10 hz in their figure 1 . this behavior was later confirmed by @xcite , hereafter p03 , using observations of the cyg x-1 spectral transition in 2000 - 2001 . it is worth noting that dp99 also suggested that the index - qpo frequency correlation can be a common phenomenon for black hole sources . in this paper we present a detailed study of spectral transitions in cyg x-1 and demonstrate how the energy spectra are related to the power density spectra ( pdss ) , in particular the qpo features . we find that the index - qpo correlation is similar to previous findings for bh sources , e.g. dp99 , @xcite and tf04 , where the qpo frequency - index correlation is presented for large samples of bh sources . in pdss observed by _ rxte _ for cyg x-1 , we show that these qpo low frequencies tightly correlate with the break frequency @xmath13 . @xcite presented a model for the radial oscillations and diffusion in the transition layer ( tl ) surrounding the bh and ns . using dimensional analysis , they identified the corresponding radial oscillation and diffusion frequencies in the tl with the low - lorentzian @xmath2 and break frequencies @xmath13 for 4u 1728 - 34 . they predicted values for @xmath13 related to the diffusion in the transition layer , that are consistent with the observed @xmath13 . both the keplerian and radial oscillations , along with diffusion in the transition layer , are controlled by the same parameter : the reynolds number @xmath14 ( inverse of @xmath15-viscosity parameter ) , which in turn is related to the accretion rate [ see also @xcite , hereafter tlm98 ] . it is worth noting that the identification of the break frequency as a diffusion effect ( the inverse of time of the diffusion propagation in the bounded configuration ) was later corroborated by both @xcite and @xcite . particularly , wood et al . demonstrated that the black hole candidate ( bhc ) xte 1118 + 480 x - ray light curves with fast rise / exponential decay profile are a consequence of the diffusion matter propagation in the disk . on the other hand , gilfanov & arefiev ( 2005 ) studied x - ray variability of persistent lmxbs in the @xmath16 hz frequency range aiming to detect pds features associated with the diffusion time scale of the accretion disk @xmath17 . as this is the longest intrinsic time scale of the disk , the power spectrum is expected to be independent of the frequency below @xmath18 . they found that the break frequency correlates very well with the binary orbital frequency in a broad range of binary periods from @xmath19 min to 33.5 days , in accord with theoretical expectations for the diffusion time scale of the disk . @xcite found while the low - energy x - ray ( 1.3 - 12 kev ) and high - energy x - ray ( 20 - 200 kev ) fluxes strongly anticorrelate during the spectral transition , the bolometric luminosity in the soft states may only be 50%-70% greater than the hard state luminosity . on the other hand , @xcite found that the increase of the bolometric flux in the high / soft state with respect to that in the low / hard state is about s factor of @xmath20 . in this paper we further explore this issue of the bolometric luminosity using the data collected from the pca and hexte detectors of _ rxte_. in fact , we confirm zhang s et al . finding that the bolometric luminosity slightly increases when cyg x-1 undergoes transition from the low / hard to the soft states . we also comment on the issue of how the wind in cyg x-1 affects the bolometric luminosity . @xcite and @xcite argue that the x - ray source in cyg x-1 is powered mainly by accretion from the strong stellar wind of the supergiant star . cyg x-1 probably represents a situation intermediate between pure , spherical wind accretion and accretion by roche lobe overflow . as @xcite pointed out the density of the wind determines the size of x - ray ionization zone surrounding the black hole . this in turn controls the acceleration of the wind in the direction of the black hole . during the low / hard state , the strong wind is fast and accretion rate is relatively low , while during the soft state , the weaker , ionized wind attains only a moderate velocity and the accretion rate increases . it is evident that the thomson optical depth of the wind increases in spectral transition because of decrease of the wind velocity ( even if outflow mass rate is constant through the state transition ) . we further investigate the effects of the wind in cyg x-1 in terms of power and energy spectra , bolometric luminosities and the strength of k@xmath3 iron line emission . the iron line observed in cyg x-1 is the strongest among the galactic black holes . @xcite discovered the broad fe k@xmath3 line , with equivalent width @xmath21 ev and fwhm=1.2 kev in an _ exosat _ spectrum of cyg x-1 . because of these features are broad one should be concerned that the profiles are artifacts of inadequate continuum models or instrumental effects . miller et al . ( 2002 ) argue that an fe k@xmath3 line is required to obtain statistically acceptable fits to spectra observed from cyg x-1 with a number of instruments , for a variety of continuum models and source luminosities [ see @xcite , and @xcite for _ asca _ results , and see @xcite and @xcite for _ bepposax _ results ] . it is important to emphasize that most of the observations indicate that the relatively strong broad fe k@xmath3 line emission which ew is within a range of 100 - 300 ev while miller et al . claim their best - fit model for chandra spectrum includes a broad line ( @xmath22 kev , fwhm@xmath23 kev , @xmath24 ev ) component along with a narrow gaussian emission line ( @xmath25 kev , fwhm@xmath26 ev , @xmath27 ev ) component . it is clear that the inferred values of line energy e , fwhm and ew are affected by the energy resolution of a given instrument and by the continuum model applied [ see e.g. @xcite ] . recently @xcite revealed a strong and broad fe k@xmath3 line in another black hole source , grs 1915 + 105 ( @xmath28 ev ) . they suggest that there should be a link between ews and qpo frequencies in bh sources . here we show that indeed there are correlations between the strength of the iron line , qpo frequency and the spectral index . the observational signature of the mass accretion rate @xmath29 is the qpo low frequency as it has been shown for bh sources by tf04 . the qpo frequency is related not only to @xmath29 but also to the size of the comptonizing region , @xmath30 , i.e. @xmath31 . the behavior of @xmath32 with respect to spectral index @xmath10 connects the characteristics of the comptonization and spectral state with @xmath29 . this is graphically represented for bhs in the observations of @xcite . we similarly employ this type of analysis to compare cyg x-1 spectral states with other bhs to show their qualitative differences . in 2 we present the details of the our spectral and timing data analysis of archival rxte data from the bh source cyg x-1 . in 3 we present and discuss the results of the data analysis and we compare them to that presented by tf04 for other bh sources . in 3 we also offer an explanation various correlations found in cyg x-1 . discussion and conclusions follow in 4 . for our analysis we used data from proportional counter array ( pca ) and high - energy x - ray timing experiment ( hexte ) onboard _ rxte_. the data is available through the gsfc public archive . cyg x-1 is one of the sources most extensively observed by _ rxte_. we searched the entire archive for public data . the summary of the _ rxte _ observation proposals and data used in the present analysis , and a reference to the corresponding proposal ids , are given in table 1 . each proposal consists of a set observations that can be divided into intervals of continuous on - source exposure ( usually about 3 ks ) corresponding to one _ rxte _ orbit . for each proposal we provide its archival identification number ( proposal i d ) , the dates between which the data were collected , the total on - source exposure , the number of continuous data intervals @xmath33 , the average number of operational pcus in the proposal @xmath34 . the data spans @xmath35 years of data with almost 2.2 ms of the total on - source exposure . we calculated an energy spectrum and an averaged pds for each continuous interval of data , which correspond to one orbital _ rxte _ revolution . data reduction and analysis was conducted with ftools 5.3 software according to recipes in the `` _ rxte _ cook book '' . before being transmitted to ground - based station the pca data is preprocessed by six event analyzers , two of which are always operating in standard1 and standard2 modes . to avoid telemetry overload for very bright sources such as cyg x-1 the counts from several energy ranges are processed by separate on - board event analyzers and stored in separate data files . during most observations , binned mode with several millisecond time resolution combining counts from 0 to 35 pca energy channels is available . depending on rxte epoch , this channel range corresponds to energy range changing from 1.5 - 10 kev to 2.0 - 15 kev . we were mostly interested in low frequency range ( @xmath36 100 hz ) and the temporal resolution of this mode is sufficient for our analysis . otherwise , we used sub - millisecond resolution event or single bit mode data . the data was rebinned to a @xmath37 second time resolution to obtain a nyquist frequency of 1024 hz . pdss are normalized to give rms fractional variability per hz . for the pds modeling we used the broken power law ( see * ? ? ? * for definition ) component to fit broad band frequency noise and lorentzians to describe qpo profiles . spectral data reduction and modeling was performed using the xspec astrophysical fitting package . first we performed the data screening to calculate good time intervals for fourier analysis . we excluded the data collected for elevation angles less than @xmath38 and during south atlantic anomaly passage . to avoid the electron contamination we also applied the condition for electron rate in the pcu 2 ( which is operational during all observations ) to be less than 0.1 counts / sec . we extract energy spectra from standard2 data files using counts from upper xenon layer of all operational detectors . then we applied a deadtime correction to account for detector dwell time after each event detection . current response matrices for pca and hexte give 10%-20% offset for relative cross - normalization . henceforth , when we fit pca and hexte spectra simultaneously , we multiply the physical model applied to the data by a constant factor to account for this instrumental discrepancy . we fix the factor value at unity for pca data set while allowing it to change for the hexte cluster a and b spectra . in table [ statestab ] we give the values of the best fit model parameters along with the offset values obtained for representative pca / hexte spectrum for each source state . to describe the continuum spectrum we use the bulk motion comptonization ( _ bmc _ ) model which is a generic comptonization model . this model can be used if the photon energy is less than the mean electron energy of the compton cloud @xmath39 . the choice of a particular theoretical model is provided by robust nature of the bmc model for different spectral states and independence of a specific type of comptonization scenario involved . the bmc model spectrum is a sum of the blackbody component ( which is the disk radiation directly seen by the observer ) and the fraction of the blackbody component comptonized in the corona with the variable comptonization fraction . the model has four parameters : @xmath40 is a color temperature of thermal photon spectrum , @xmath15 is the energy spectral index ( @xmath41 , where @xmath10 is the photon index ) , the parameter @xmath42 is related to the weight of the comptonized component , @xmath43 , and a normalization of the blackbody component . the bmc model is valid for the general case of comptonization when both bulk and thermal motion are included . for the thermal comptonization @xmath39 is related to electron temperature @xmath44 . when the bulk motion comptonization is dominant @xmath39 is related to the bulk kinetic energy of the electrons @xmath45 . for the thermal comptonization for energies less than @xmath39 comptt and bmc models are identical . the thermal comptonization and dynamical ( bulk motion ) comptonization are presumably responsible for the spectral formation in the hard state and the soft state respectively . therefore , one needs a generic model such as bmc that describes the spectral shape regardless of the specific type of comptonization . although the thermal comptonization model can properly fit the spectral shape for the observed spectra for all spectral states but it would give physically unreasonable values of the best - fit parameters for the soft state spectra notably , wilms et al . ( 2005 ) show optical depth of compton cloud inferred from comptt model significantly decreases towards high / soft state . it is very difficult ( in the framework of any reasonable physical model ) to explain tendency when the mass accretion rate increases during the hard - to - soft state transition . we remind a reader that tsh05 used two bmc components to fit the spectral data from ns source 4u 1728 - 34 . the radiation from the central object and inner parts of accretion disk is comptonized by a surrounding cloud . as long as the spectral index for each bmc component is a physical characteristic of the comptonizing region , tsh05 used this parameter for each of bmc components describing disk and ns comptonized radiation . in fact , the bmc xspec model spectrum ( titarchuk et al . 1997 ) is a sum of the ( disk or ns ) black - body component and comptonized black - body component . in the bh case one can conclude that only one the bmc component is needed in order to describe the comptonization spectrum . the soft photons generated by the solid surface are absent in this case . another important issue in modeling the bhc spectra is a proper account for iron fluorescence line at @xmath46 kev . as it was already mentioned , the iron line observed in cyg x-1 is one of the strongest among bhc sources and is confirmed by various instruments . the site of the iron line origin is not yet identified and the process of its formation is under debate . currently it is fashionable to explain the iron line in the framework of the reflection model by @xcite ( hereafter mz95 , see pexrav and pexriv models in xspec ) . the geometry required by the model implies that the source of comptonized component ( exponentially cut - off power law ) is located above a flat reflecting surface ( accretion disk ) . the appearance of the iron line and so called a reflection hump at 15 - 20 kev is then produced as a result of compton reflection . despite the fact that reflection model provides a good spectral fits to the data , the best - fit parameters obtained for a soft state are hard to explain ( we provide more details in the discussion section ) . in figure [ bmc_vs_pexrav ] we compare the fits given by bmc and pexrav models for the observation taken when the source was in the soft state . the reflection model fit requires an additional gaussian to account for the iron line . the line profile produced by pexrav alone can not account for the total iron line strength even for high values of a reflection factor @xmath30 . the large `` reflection '' component , in turn , leads to overestimated spectral indices for the comptonized component . an alternative mechanism for the line formation is proposed by @xcite , lat04 hereafter . in their scenario the iron line is formed in outflowing wind material . it is important to note that an absorption edge is _ physically _ required to compensate for absorption of photons above the k - threshold energy [ see @xcite ] . our results favor this model as it is consistent with equivalent width ( ew ) - spectral index correlations and is capable of accounting for high ew values . to illustrate our arguments and to investigate the mutual dependence of predicted spectral properties on the specific model we fit the subset of cyg x-1 observations which includes low / hard , soft states and transitions between them . the models and the results of fits are presented on figure [ models ] . the power law plus blackbody empirical model and bmc model provide qualitatively similar results when the gaussian is used with the edge and without it . considering the restricted rxte energy resolution at low energies the exact values of gaussian ews have to be handled with care . however , statistically the edge is highly significant . in figure [ counts ] we demonstrate the consistency of the our spectral model using a representative spectrum from cyg x-1 . for all fits we fix and relate the line energy at 6.4 kev to k threshold energy at 7.1 kev [ see details of this relation in @xcite ] . to account for interstellar absorption we use a fixed hydrogen column of @xmath47 @xmath5 . finally , according to the above arguments , the xspec model for the spectral fitting reads as _ wabs(gaussian+bmc)*edge_. we use @xmath15 , the disk color temperatures , disk blackbody normalizations as free parameters of the spectral continuum model . the spectral fits was obtained using 3.5 - 30.0 kev energy range . we add 0.5% error to the data to account for systematic uncertainty in the pca calibration . the typical quality of fit is good with @xmath48 in the range of 0.5 - 1.5 . the evolution of spectral properties of the source during the transition from the low / hard to the soft states is shown on figure [ spevol ] and in table [ statestab ] . the temperature profile of the thermal ( blackbody ) component is plotted versus the index @xmath10 in figure [ kt ] . this temperature is presumably related to the disk . it changes only slightly in the narrow range of 0.5 -0.6 kev and is nearly independent of the spectral state . when the source is in the low / hard state the emerging radiation spectrum is presumably formed as a result of comptonization of soft photons generated in the disk . we present a @xmath49- diagram for the low / hard state in upper left hand panel of figure [ spevol ] ( @xmath50 is the energy flux ) . as a source progresses to higher luminosity states the spectrum becomes softer , the power - law part of the spectrum becomes steeper and the contribution of the blackbody ( thermal ) component increases . the strength of the iron line also increases ( see the right upper panel in figure [ spevol ] ) . one can explain this evolution of the spectrum by an increasingly efficient deposition of the gravitational energy in the disk which becomes stronger towards the soft states ( see e.g. * ? ? ? * ; * ? ? ? the luminosity reaches the highest level in the soft state when the spectral index @xmath10 is about 2 . the power - law plateau is clearly seen in @xmath49- diagram presented in the lower left hand panel of figure [ spevol ] . then , as the spectral index increases we observe a slight decrease in flux . we identify this phase as a very soft state ( the so called thermal dominated state " ) , in addition to the canonical low / hard , intermediate and soft states . the energy spectrum is dominated by thermal component ( more than 70% of total flux ) and the maximum observed spectral index is @xmath51 . it is important to emphasize that , typically , the high / soft state in the bh sources is observed when an extended power - law component has index @xmath52 [ see e.g. @xcite , @xcite , hereafter brt99 , tf04 ] . this relatively low index for cyg x-1 soft state can be explained by a higher temperature of the converging flow than that for the high / soft state of other bh sources ( see * ? ? ? * hereafter lt99 , and also tf04 ) . the main reason for this may be the fact that in the soft state of cyg x-1 , the energy release in the disk and in the compton cloud are comparable while in the typical lmxb bh sources , like grs 1915 + 105 , xte j1550 - 564 , gro 1655 - 40 , in the high / soft state the energy release in the disk is much greater . when spectral index progresses to values higher than @xmath53 the luminosity decreases ( see the lower right hand panel of figure [ spevol ] ) . it is worth noting that we use a terminology for the spectral states in cyg x-1 based on our physical scenario of the spectral evolution there ( see more of the details in the discussion section ) . in our classification the difference between soft and very soft states corresponds to the difference in the spectral indices . in soft state when the saturation of the spectral index @xmath10 vs qpo frequency @xmath9 occurs , @xmath10 is about 2 ( see 3.4 ) . whereas in the very soft state the spectra become softer and @xmath10 increases to the values of 2.7 . no qpo frequencies are observed in this state . in the scheme of mcclintock and remillard ( 2004 ) ( and every other `` high '' or `` very high state '' bhc sources ) this classification can look different . one should be careful in applying the definition of spectral states ( particularly soft states ) for some particular source . the next important question is how this spectral evolution is related to the timing characteristics of the source . in our study of the power density spectrum ( pds ) , we reveal that the pds features , break frequency @xmath13 , and lorentzian low - frequency @xmath9 and q - value of the qpo frequency evolve and increase while the source progresses toward the soft state . but the qpo frequencies are completely washed out in the very soft state . @xcite , hereafter tcw02 , predicted that when the source is embedded in the optically thick medium the qpo features must be absent in the pds of the source because of photon scattering . when the radiation from the central source passes through the surrounding cloud of optical depth @xmath54 and of radius @xmath30 the direct ( unscattered ) and scatterred fractions of the radiation are @xmath55 and [ @xmath56 respectively . consequently the rms amplitude of the direct component decreases exponentially with @xmath54 . tcw02 show that the rms amplitude of the scattered component @xmath57 for a given rotational frequency @xmath58 decreases as @xmath59 where @xmath60 ^ 4 + ( 3r\tau_0\omega/4c)^2\}^{1/2}$ ] . as seen from equation ( [ bomega ] ) , the qpo amplitude of the radiation scatterred in the wind ( cloud ) decreases very rapidly with @xmath54 and with @xmath61 ( @xmath62 ) , namely @xmath63 a radius r of @xmath64 cm was chosen as a typical radius of the wind inferred from the observations [ @xcite ] and theoretically inferred using luminosity and spectrum as observable characteristics of the source [ @xcite ] . thus we can conclude that the variability of the scatterred part of radiation is completely washed out in the wind of radius of order @xmath64 cm even for @xmath65 . the variability of the direct ( unscattered ) component is preserved but its rms amplitude decreases as @xmath55 with @xmath54 . the power spectrum analysis of cyg x-1 data confirms this expectation ( see pottschmidt et 2003 ; axelsson , borgonovo & larsson 2005 and our results in this paper ) . the power spectrum in the very soft state is featureless ( see fig . [ power ] ) . the emission of the central source is presumably obscured by the optically thick wind and consequently all photons emanating from the central source are scattered . the direct component of the central source radiation that carries information about the variability is suppressed by scattering . @xcite argue that there is a particular state of cyg x-1 when the wind velocity is very low and thus one can expect high accumulation material in the wind and high optical depth of the wind . the wind downscattering of the photons emanating from the inner compton cloud leads to the softening of spectrum @xcite and consequently to a decrease in x - ray luminosity of the source . the softening of the spectrum can also be a result of effective cooling the compton cloud by the disk soft photons . if the power law component of the soft state is formed in the converging flow , then the high indices are a result of the low flow temperature ( lt99 ) . the index increases and saturates to the critical value about 2.8 with the mass accretion rate for the low temperatures of the flow ( see lt99 and titarchuk & zannias 1998 , hereafter tz98 ) . to calculate the power spectrum of the central source emission affected by scattering one should present the quantitative model of the resulting pulse affected by scattering @xmath66 which can be written as follows : @xmath67 where @xmath68 is the input pulse of the central source oscillations and @xmath69 is the response pulse of the scattering medium . the power spectrum of @xmath66 is a product of the power spectrum of @xmath68 and @xmath69 , i.e. @xmath70 where @xmath71 @xmath72 @xmath73 are fourier transforms of @xmath74 respectively . the time response of the innermost part of the source @xmath68 can be considered in terms of diffusion propagation of perturbation in the disk transition layer ( tl ) . any local perturbation in tl ( bounded medium ) would propagate diffusively outward over time scale ( see tl99 ) @xmath75 where @xmath76 is the characteristic thickness of the tl , @xmath77 is the mean free perturbation path related to the turbulent mhd viscosity @xmath78 , the matter density @xmath79 , the number density @xmath80 and the perturbation interaction cross - section @xmath81 in the tl and @xmath82 . even though the specific mechanism providing this viscosity needs to be understood , this time scale @xmath83 apparently `` controls '' the diffusion supply of the matter into the innermost region of the accretion disk ( tl ) . the response function @xmath68 is a solution the time - dependent problem for perturbation diffusion that mathematically is formulated as an initial value problem for the diffusion equation in the bounded medium ( see more details of the diffusion theory in st80 , st85 , t94 ) . this solution is a linear combination @xmath84 where @xmath85 is a free path crossing time , @xmath86 is kth eigen diffusion time , @xmath87 , @xmath88 is expansion coefficient of the source perturbation function @xmath89 @xmath90 that is related to eigenvalues @xmath91 and eigen functions @xmath92 of the appropriate space diffusion operator and @xmath93 is the operator weight function . if @xmath94 then the response function as a solution of the diffusion problem can be presented by a single exponent , namely @xmath95 because in this case @xmath96 and @xmath97 , for @xmath98 , where @xmath99 ( st80 , st85 , t94 ) . the scattering response function @xmath69 is a solution of the diffusion problem for the photon scattering in the extended envelope ( wind ) that surrounds the central source . the formula for the photon diffusion time @xmath100 is very similar to formula ( [ cstime ] ) , i.e. @xmath101 where @xmath102 is the envelope thickness and the thomson optical depth of the envelope @xmath103 . when @xmath104 ( similarly that for the perturbation diffusion ) the scattering diffusion function @xmath69 can be presented by the exponent @xmath105 it is worth noting that @xmath106 , because @xmath107 by definition of @xmath108 and @xmath109 . for these particular response functions ( see eqs . [ xtdiff ] , [ gtdiff0 ] ) the power spectra are @xmath110 @xmath111 and @xmath112 because @xmath106 we have the following asymptotics for the power spectrum @xmath113 as a function of @xmath114 ) @xmath115 one can clear see the change of the power - law index of the @xmath116power diagram from @xmath117 to @xmath118 about 0.08 , 0.5 , 3 hz in the low / hard , intermediate and the soft states respectively . thus one conclude that the perturbation diffusion times in the tl ( compton cloud ) @xmath119 hz ) are about @xmath120 s in these states . the change of the power - law index of the diagram from ( 1 ) to ( -1 ) [ that corresponds to 0 -(-2 ) changes of that in the power spectrum ( eqs . [ pwasym1 ] , [ pwasym2 ] ) ] takes place in the low - frequency parts of the diagrams for the low / hard , intermediate and soft states ( black , blue , red respectively ) . the power at high frequencies decays very fast and the corresponding power - law index of pds is about 4 ( see eq . [ pwasym3 ] ) . however in the @xmath116power diagram for the very soft state the index changes from 0 to ( -1 ) at low - frequency ( about 0.2 hz ) but this change is not described by aforementioned formulas ( [ power0 ] , see also [ pwasym1 ] , [ pwasym2 ] ) . we argue this particular power - law index transition occurs in the source when the optical depth of the wind @xmath121 . in order to demonstrate this effect one must calculate the power spectrum of the scattering response function @xmath69 for @xmath121 . the derivation of these formulas are out of scope of the paper and we shall present these results elsewhere . it is worth commenting that we have already shown here , the resulting power spectrum is a product of two power spectra related to the disk pds and another one to compton cloud ( scattering ) pds ( see eq . [ pwsp ] ) . in the general case , each of these spectra is a power spectrum of the series of eigen exponential shots [ see for example , formula [ xtdiff ] for @xmath68 ] . for frequencies much less than the characteristic scattering frequency , @xmath123 the resulting spectrum is represented by a power - law red noise component . in our simple treatment of this problem considered here and applicable to the hard and intermediate states only ( see eq . [ power0 ] ) the low - frequency power - law index of pds is zero . in fact , the index and power - law cuttoff frequency increases with mass accretion rate when the source undergoes transition to a soft state . we shall provide all details of this correlation in a future publication . another observational appearance of the wind is the strong broad feature of k@xmath3 line in the spectrum that is present in all spectral states of cyg x-1 ( see fig . [ spevol ] ) . in figure [ line_index ] ( upper panel ) we demonstrate how equivalent width ew of the k@xmath3 line increases with the photon index @xmath10 from about 150 ev in the low / hard state to about 1.3 kev in the high / soft and very soft states . one can see signs of saturation of the ew at about 1.3 kev for indices above 2.1 when no qpo is present in the power spectrum . wilms et al . ( 2005 ) have also found that one needs a strong k@xmath3 line for fitting data with a comptonization model ( comptt ) . the strength of the line increases almost linearly with a accretion disk flux until it satures at the values of @xmath124 kev . the x - ray photons presumably originated in the innermost part of source , in an area less than 40 schwarzschild radii , illuminate the wind . in this case the wind gas is heated by compton scattering and photoionizations from the central object . it is cooled by radiation , ionization , and adiabatic expansion losses ( lat04 ) . the photons above the k - edge energy are absorbed and ionize iron atoms that leads to the formation of the strong k@xmath3 line . lat04 calculated ionization , temperature structure and the equivalent widths of fe k@xmath3 line formed in the wind . for the wide set of parameters of the wind ( velocity , the thomson optical depth @xmath125 ) and the incident comptonization spectrum ( the index and the compton cloud electron temperature ) they established that ew of the line should be about 1 kev and less for the line to be observed . lat04 also predicted that for this case the inner radius of the wind should be situated at @xmath126 schwarzschild radii away from the central object . in the framework of the wind model we can inferred the optical depth of the outer shield ( wind ) as a function of the index using the ew of k@xmath3 line [ see @xcite , ct95 ] : @xmath127\}de . \label{ew_tau}\ ] ] where @xmath128 is the fluorescence yield , @xmath129 is the k shell ionization threshold energy , @xmath130 and @xmath131 are the abundances of elements ( in units of the cosmic abundances ) with a charge @xmath132 and the iron abundance , respectively . this integral can easy be calculated using the following formula @xmath133^m}{(3m+\alpha-1)m!}\left(\frac{7.8~{\rm kev } } { e_{th}}\right)^{3m-1}. \label{ew_tau_form}\ ] ] we apply the value of @xmath134 kev [ see @xcite ] , and @xmath135 [ see @xcite ] . in figure [ line_index ] ( lower panel ) we present the inferred dependence of @xmath136 on the photon index . to obtain a description of the x - ray photon spectrum of cyg x-1 many authors ( see e.g. gilfanov , churazov & revnivtsev 1999 , hereafter gcr99 ; p03 ) used an empirical model in which each source spectrum is a sum of power law spectrum with photon index @xmath10 and a multi - temperature disk blackbody ( makishima et al . 1986 ) . to this continuum , a reflection spectrum after mz95 was added . gcr99 emphasized that this empirical model is obviously oversimplified and therefore the best - fit parameter values do not necessarily represent physically meaningful quantities . particular problems arise with the values of the reflection factor @xmath137 and the equivalent width of the iron line . the best - fit values of @xmath137 for the soft state exceed the unity considerably @xmath138 , which is physically meaningless in the geometry of the reflection . in fact , lapidus , sunyaev & titarchuk ( 1985 ) , showed that the maximum reflection factor for geometrically thin infinite disk illuminated by isotropic radiation from the central object is 0.25 . the values of the ew of k@xmath3 inferred by gcr99 vary from 80 to 300 ev . these ews are related to the line component included in the spectrum in addition to the refelection component . because the reflection model includes its own iron line component ( mz95 ) thus the actual strength of the line is much higher in cyg x-1 than that presented in gcr99 . one can determine that the total iron line strength strongly increases with photon index because of the power of the so called `` reflection component '' r , which also strongly increases with the index ( see figs 5 , 8 in cgr99 ) . in this sense our inferred strong lines in the soft states of cyg x-1 agree with those in gcr99 . in figure [ ind_qpo ] we present the correlation of photon index @xmath10 versus low qpo frequency @xmath9 ( orange points in the upper panel ) and the correlation of @xmath10 versus break frequency @xmath13 ( the lower panel ) observed in cyg x-1 . we compare the index - qpo correlation with those observed in grs 1915 + 105 , xte j1550 - 564 . one notices similar properties for all of these sources : i. when qpo is detected the photon index does not go above 3 , ii . the index saturates . at low and high values of the qpo frequencies . in cyg x-1 the saturation level of the index for high values of low qpo frequency is remarkably lower than for the other two sources . tf04 argues that the index saturation level is determined by the temperature of the converging flow where the soft ( disk ) photons are upscattered by electrons to the energy of falling electrons ( lt99 ) . in principle , one can evaluate the mass of the central bh using the index - qpo relation because qpo frequencies are inversely proportional mass ( tf04 ) . the simple slide of the index - qpo correlation for xte j1550 - 564 ( pink line ) over the frequency axis gives us the index - qpo correlation for grs 1915 + 105 ( blue line ) . the shifting factor is 10/12 which gives the relative bh mass in xte j1550 - 564 with respect to that in grs 1915 + 105 . but one caveat should be taken into account : this sliding method works if the index - qpo relations are self - similar with respect to each other as occurs for grs 1915 + 105 and xte j1550 - 564 . in figure [ flux - index ] we show how the 1 - 30 kev flux ( black crosses ) varies during the spectral transition from the low / hard state ( l / h ) to very soft ( vs ) state . we use joint pca / hexte spectral fits for bolometric corrections for high energies . we extracted hexte spectra from cluster a and cluster b using the same screening criteria that we obtained for pca spectra . to fit pca / hexte data we apply the model obtained for pca multiplied by high energy cutoff ( _ highecut _ ) component to account for high energy turnover in the hard tail of the x - ray spectrum . the flux is then calculated by integrating the best - fit model spectrum over the energy interval from 1 kev to 300 kev . absolute normalization of hexte data for both clusters were allowed to change free with respect to pca normalization . the resulting fits show the hexte normalization consistently less than the normalization for pca by 15 - 20% as expected . resulting values of bolometric flux are shown on figure [ flux - index ] ( red circles ) . we confirmed zhang s et al . ( 1997 ) claims that the bolometric flux remains almost unchanged ( within 50% , almost ) during l / h to vs transition . this phenomenon can be explained as a combined effect of the mass accretion rate increase and the comptonization enhancement decrease when the system undergoes the spectral transition . in the low / hard state the relative small energy release ( low mass accretion rate ) in the disk is compensated by high comptonization efficiency in the corona while in the very soft ( thermal dominated ) state the situation is opposite . sunyaev & titarchuk ( 1980 ) , hereafter st80 , and sunyaev & titarchuk ( 1985 ) , hereafter st85 , derive the asymptotic form of the comptonization enhancement factor @xmath139 for both regimes of the energy spectral index ( @xmath140 and @xmath141 ) . chakrabarti & titarchuk ( 1995 ) , hereafter ct95 , provides the general formula for @xmath139 ( ct95 , formula 14 ) which combines these two asymptotic . in order to infer the bolometric flux dependence @xmath142 on the photon index using the observable thermal ( disk ) flux @xmath143 one should multiply @xmath143 by @xmath144 $ ] and apply formula ( 14 ) in ct95 for @xmath139 , namely @xmath145 , \label{flux_bol}\ ] ] where @xmath146 , @xmath40 is a color temperature of disk radiation ( see fig [ kt ] ) and @xmath147 is electron temperature of compton cloud . the enhancement factor @xmath139 depends on @xmath147 only for @xmath140 ( @xmath148 ) ( see formula 14 in ct95 ) . in order to infer @xmath147 we apply use the same pca / hexte data . to calculate the electron temperature @xmath147 we use the _ highecut _ parameter , @xmath149 and well - known relation between the electron temperature and the high energy cutoff of the comptonization spectrum , namely @xmath150 ( see e.g. st80 and t94 ) . we present the results of calulations of @xmath151 as a function of @xmath152 for different values of @xmath153 ( see fig . [ ecomp ] ) . in figure [ flux - index ] we demonstrate a dependence of @xmath142 on the photon index @xmath152 ( blue triangles ) . the theoretically predicted and observed flux values are in good agreement along . thus _ the inferred luminosty - index relation supports the idea that the variation of luminosity with index is due to the combined effect of the disk mass accretion rate and comptonization of the disk photons in the corona_. the soft photon radiation is completely comptonized in the low / hard state while the relative contribution of the comptonized radiation decreases toward the very soft ( thermal dominated ) state . in figure [ comp_ratio ] we present the observed correlation between the ratio of the comptonized flux to the bolometric flux and the index . the comptonized spectrum is a convolution of the soft photon spectrum with the upscattering comptonization green s function . this component of the observable spectrum and consequently the related flux can be obtained using the bmc model . the inferred comptonized fraction of the spectrum helps us to reveal the relative size of the corona ( compton cloud ) with respect to the disk emission region . one can see that in the soft states the coronal region becomes more compact . this effect is also confirmed by the observed index - qpo correlations ( see fig . [ ind_qpo ] ) . the qpo frequencies increases with the index as a result of the mass accretion rate increase . in fact , the coronal region is pushed close to bh when mass accretion rate goes up ( see tlm98 and tf04 ) . in the soft states ( when the compton cloud is relatively cold ) the converging flow site ( @xmath154 ) is only the place where the soft ( disk ) photons get scattered due to the dynamical comptonization . in figure [ br_lowfr ] we present the observed correlation between the break frequency @xmath13 and the low frequency @xmath9 . we fit this correlation by the broken power law of the form @xmath155 the best - fit parameters : the normalization @xmath156 , the `` low''-frequency index @xmath157 and the `` high '' frequency index @xmath158 and the `` boundary '' index @xmath159 hz . titarchuk , osherovich & kuznetsov ( 1999 ) found that the `` high '' frequency index @xmath160 is a canonical index of break - low frequency correlation for a quite a few bh and ns sources . in fact , titarchuk & osherovich ( 1999 ) identified using dimensional analysis the corresponding radial oscillation and diffusion frequencies in the transition layer ( tl ) with the low - lorentzian @xmath2 and break frequencies @xmath13 for 4u 1728 - 34 . they predicted values for @xmath13 related to the diffusion in the transition layer , that are consistent with the observed @xmath13 . to99 argue that @xmath2 is the inverse of the oscillation time of the tl radial mode , @xmath161 and @xmath162 is the inverse of the diffusion time of the perturbation propagation in the tl @xmath163 where @xmath109 is the tl radial size ( see definition of @xmath164 in the text after eq . [ cstime ] ) . @xmath165 is the coefficient which depends on the specific outer and inner boundary conditions imposed in the tl ( see , titarchuk , bradshaw & wood 2001 , for details of the boundary problem of the diffusion and oscillations in tl ) . diffusion time @xmath83 is related to the diffusion length @xmath166 where @xmath167 is a free propagation path of the perturbation in the tl , @xmath168 is a number of matter interactions , related to the effective viscosity in the tl . factor @xmath169 is related to the source distribution of the perturbation in the tl ( see sunyaev & titarchuk 1985 for details ) . thus the relation between @xmath13 and @xmath9 reads as follows @xmath170 because the perturbation dimensionless depth @xmath171 is related to @xmath109 then @xmath172 thus one can find taking into account eqs.([lowfr ] , [ break - low2 ] ) that @xmath13 should not be a linear function of @xmath9 if @xmath173 . to99 found that @xmath174 when @xmath175 . the number of interactions in the tl hydrodynamical flow @xmath33 and the perturbation depth @xmath176 depends strongly on the mass accretion rate . for relatively high rates one should expect that @xmath177 and @xmath178 while these values are about one for relatively small mass accretion rates that occur in the low / hard state . in the latter case the dependence of @xmath13 on @xmath9 is almost linear . this diffusion effect is confirmed by the correlation of @xmath13 and @xmath9 observed in cyg x-1 ( see fig [ br_lowfr ] ) . for higher values of frequencies @xmath13 ans @xmath9 ( which correspond to higher values of mass accretion rate ) @xmath174 while for lower values @xmath179 . the low frequency @xmath9 is inversely proportional to the tl size which its turn is proportional to the mass of the central object @xmath180 ( bh or ns ) . thus one can conclude that @xmath9 and @xmath13 are inversely proportional to m when mass accretion rate in a source is relatively small . this inverse proportionality of @xmath13 vs @xmath181 can be used for the mass determination of the objects that mass differs by order of magnitude from galactic bhs for example , of the supermassive or intermediate bhs provided the break frequency is detected there . recently , fiorito & titarchuk ( 2004 ) applied rescaling of qpo frequency @xmath9 to evaluate a bh central mass in ultraluminous source m82 x-1 while mchardy et al . ( 2005 ) and dewangan , titarchuk & griffiths ( 2006 ) applied rescaling of @xmath13 for the bh mass determination in agn and ulx respectively . we have presented a detailed spectral and timing analysis of x - ray data for cyg x-1 collected with the _ rxte_. we find observational evidence for the correlation of spectral index with low - frequency features : @xmath13 and @xmath2 . the photon index @xmath10 steadily increases from 1.5 in the low / hard state to values exceeding 2.1 in the soft state . the low frequency @xmath9 is detected throughout the low / hard and intermediate states , while it disappears when the source undergoes transition to the very soft ( thermal dominated ) state . like in other bh sources , there is an indication of saturation of the index in the soft state for cyg x-1 ( see fig.[ind_qpo ] ) . this saturation effect , which is presumably due to photon trapping in the converging flow , can be considered to be a bh signature . we want to stress that this saturation is a model independent phenomenon found in the present analysis of x - ray _ rxte _ data . on the other hand , _ the saturation of the index with mass accretion rate increase ( which is strongly related to the qpo frequency increase ) has to apply to any bh , because the photons are unavoidably trapped in the central accretion flow . it is a necessary condition of bh presense in the accreting systems_. thus the index saturation with qpo frequency seen in the source ( rather than the presense of the tail in the soft state ) is a signature of the horizon . in fact , one can see high energy tails with indices at about the bh saturation value 2.8 in the high / soft state observations of weakly magnetized accreting ns binaries , for example gx 17 + 2 ( farinelli et al . 2005 ) and 4u 1728 + 34 ( tsh05 ) . farinelli et al . ( 2005 ) presented two spectra of gx 17 + 2 observed in 1997 by bepposax . using these spectra alone one can not establish the evolution of the indices with the mass accretion rate . tsh05 analyzing @xmath182 ms of _ rxte _ archival data for 4u 1728 - 34 reveal the spectral evolution of the comptonized blackbody spectra and qpo frequencies when the source transitions between hard to soft states . contrary to the bh sources , the indices of 4u 1728 - 34 spectra do not saturate as qpo frequency increases . they increase from @xmath183 ( in the hard state ) to @xmath184 ( in the soft state ) with no signature of saturation versus qpo frequency ( or mass accretion rate ) . the ns soft state spectrum consists of two blackbody components that are only slightly comptonized ( inferred photon indices of the comptonization green s function are @xmath185 ) . thus one can claim ( as expected from theory ) that in ns sources thermal equilibrium is established for high mass accretion rate ( soft ) state . in bhs the equilibrium is never established because of the presence of the event horizon . the emergent bh spectrum , even in the soft state , has a power - law component which index saturates with mass accretion rate . it is worth noting that there is a particular state in bh source when it can show signs of the thermal equilibrium : the emergent spectrum consists of one or two thermal components . but no qpo is observed then . in this ( very soft , thermal dominated ) state the source is presumably covered by a powerful wind that thermalizes the radiation of the central source and prevents to see any qpo generated in the source ( see fig . [ power ] ) . one can argue that the bulk motion comptonization ( bmc ) is ruled out as a main radiative process in the soft spectral states of black - hole binaries because of the inefficiency of producing photons with energies @xmath186 kev and the lower relative normalization of the bmc component ( see a recent paper by niedzwiecki & zdziarski 2005 on monte carlo simulation of the bulk motion comptonization , hereafter nz05 ) . the production of high energy photons in monte carlo simulation with steep power - laws ( @xmath187 ) is a technically difficult problem because of poor statistics . it requires a long simulations and special methods for the treatment of the poor statistics at high energies . lt99 implemented this technique and found in their simulations that the spectra extend up to 200 - 300 kev . in fact , nz05 do not show histograms of their simulated spectra but rather show the best - fit curves describing results of their simulations . it is impossible to quantify uncertainties and the quality of counting statistics as a function of energy in the nz05 simulations . the absolute normalization of the comptonized component is always determined by the seed photon normalization ( illumination pattern ) . lt99 and then turolla , zane & titarchuk ( 2001 ) and nz05 confirm that the illumination pattern does not affect the shape of the bmc spectrum . thus , the relative normalization of the comptonized ( bmc ) with respect to that of blackbody is a model dependent parameter . lt99 show that the bmc normalization can be very high and comparable with the disk bb component when the innermost part the accretion flow ( bmc area of 1 - 3 @xmath188 ) is illuminated by the soft photons coming from the geometrical thick disk situated very close . the normalization is quite low when the seed photons illuminated the converging flow region come from the geometrically thin disk ( see nz05 ) . in contrast to the claim by nz05 , which is based on their spectral modeling , we find that cyg x-1 observations show a very strong bulk motion comptonization signature in the soft state as a photon index saturation with qpo frequency ( mass accretion rate ) . this is a well - known signature of the photon trapping in the converging flow discovered by tz98 and then confirmed in monte carlo simulations by lt99 . we also demonstrate that the fe k@xmath3 line equivalent width correlates with spectral index and correspondingly with qpo frequencies when they are present in the data ( see figs . [ line_index ] , [ ind_qpo ] ) . this leads us to conclude that the compactness of the x - ray emission area ( taking the qpo frequency value as a compactness indicator ) is higher for softer spectra ( related to higher mass accretion rate ) . on the other hand the fe k@xmath3 emission - line strength ( ew ) is about one kev when the power spectrum is featureless . it happens in the very soft ( thermal dominated ) state . thus the observations may be suggesting that the photospheric radius of the fe k@xmath3 emission is orders of magnitude larger than that for the x - ray continuum . we propose that fe k@xmath3 line emission originates in the wind where the photons emanated from the central part of the source are downscattered by electrons and absorbed by partly ionized iron atoms . in figure [ geometry ] we present a scenario that is infered using our spectral and timing data analysis of cyg x-1 spectral state transitions . in the low / hard state x - ray radiation comes from a relatively extended area . a compton cloud covers the large portion of the accretion disk that generates soft photons . most of these photons are upscattered in the hot compton cloud . on its way to the observer some fraction of the comptonized radiation is reprocessed in the relatively transparent cloud . here fe k@xmath3 line is formed because the photons of energy close to k shell ionization edge and higher are effectively absorbed and then remitted ( with a certain probability ) as k@xmath3-photons . qpo features are not washed out by photon scattering in the wind shell in this state because the optical depth of the wind is less then 0.5 ( see fig . [ line_index ] , lower panel ) . the wind becomes thicker in the intermediate and soft states and it is optically thick in the very soft ( thermal dominated ) state . this picture is supported by the fact that the oscillation amplitude and the qpo strength steadily decreases when the source proceeds to soft state and the qpo features are completely washed out in the very soft state ( see fig . [ power ] ) . also the strength of k@xmath3 ( that is presumably generated in the wind ) increases with the index ( see fig . [ line_index ] , upper panel ) . on the other hand , the compton cloud becomes very compact and cooler in softer states with respect to that in the low / hard state . the compton cloud size approaches that of the converging inflow which is a few schwarzschild radii . also , in perfect agreement with our scenario , the inferred fraction of the comptonized component in the emergent spectrum ( see fig . [ comp_ratio ] ) decreases with the index ( i.e. with mass accretion rate ) . there several other studies of the x - ray spectral and timing properties of cyg x-1 and other bhs which directly relate to the results of our paper . particularly , pottschmidt et al . ( 2003 , p03 ) also found correlations between the photon index and qpo frequencies with a clear sign of the index saturation at high values of the ( low frequency ) qpo(see fig . 7 , panel b in p03 ) . we found that our qpo frequencies @xmath9 and qpo frequencies @xmath189 in p03 are almost identical and they show the similar correlations with the photon index @xmath10 . p03 s frequencies @xmath190 are presumably very close to our break frequency @xmath13 . to obtain a description of the x - ray photon spectrum , p03 like many other authors ( see e.g. gcr99 ) used an empirical model in which each source spectrum is a sum of power law spectrum with photon index @xmath10 and a multi - temperature disk blackbody . to this continuum , a reflection spectrum after mz95 was added . gcr99 emphasized that this empirical model is an oversimplification and therefore the best - fit parameter values do not necessarily represent the physically meaningful quantities . gcr99 also showed that the reflection model can properly mimic the reprocessed component of the spectrum . particularly for intermediate and soft states ( when @xmath191 in the reflection modeling ) one can clearly see the high strength in the fluorescent k@xmath3 iron line at 6 - 7 kev and the deep broad smeared iron k - edge at @xmath192 kev ( see figs . 6 - 7 in gcr99 ) . we confirm gcr99 s conclusions that these line and k - edge edge features are more pronounced towards soft states . however , we argue that these spectral features are results of reprocessing of the central source hard radiation in the outflow warm absorber ( wind ) that surrounds the central bh . this our claim is supported by observed qpo power decay towards the softer states that is presumably a result of reprocessing of time signal in the extended relatively warm wind ( see fig [ power ] ) . recently proga ( 2005 ) and proga & kallman ( 2004 ) demonstrated using the results of hydro simulations , that the disk illumination by the radiation of the central source or just local disk luminosity can launch a wind off the disk photosphere . they argue that radiation pressure due to uv ( line resonances ) couples the x - ray and uv radiation processes by driving the disk material above the disk where the most part of the x - ray emission of the central source propagates . a strong disk wind develops when the local disk flux @xmath193 is more than 0.3 @xmath194 . for a less luminous disk the line force can still force material off the disk but it fails to accelerate the flow to escape velocity . one should stress that the timing and spectral data do reveal signatures of reprocessed component of the hard radiation in terms of high iron line equivalent width ( about one kev in the soft states ) and broad iron k - edge features . the strengths of these features seen in the observations strongly exclude their formation by disk reflection . the highest value of ew due to reflection must be less than 100 ev ( see basko 1978 and george & fabian 1991 ) . in fact , the best - fit values of the reflection scaling factor @xmath195 inferred from the cyg x-1 observations contradict to the basic assumption of the reflection model for which @xmath30 can not be more than 0.25 . as a result of our data analysis we have also found using the reflection model by mz95 that @xmath30 correlates with spectral index and values of @xmath30 are as high as @xmath196 at @xmath197 ( see fig [ rel_refl ] ) . in principle , the reflection effect from outer relatively cold parts of the disk would be detected in the low state of the source when the spectrum is relatively hard ( photon index @xmath198 ) . in this case one can see a reflection hump in the spectrum around 10 - 15 kev ( st80 ) . the main problem with the detection of this feature along with low ew of the line is that the fraction of reflected emission @xmath169 in the resulting spectrum is about 8% and less , i.e. @xmath199 where @xmath200 is albedo of the cold material ( basko , sunyaev & titarchuk 1974 , titarchuk 1987 ) . such tiny reflection features will be readily washed out by intervening wind clouds ( see above ) . the disk reflection effects become much weaker for softer spectra ( states ) , for which photon index more than 2 . thus the strong iron and k - edge features detected in the soft state are likely not due to reflection from the disk at all . rather they should be formed in the extended wind clouds surrounding the central black hole . the relative large time lag observed in the soft state ( see gcr99 ) really corroborate this scenario of reprocessing of the spectrum of the central object in the very extended environment ( wind ) of the source . mcconnell et al . ( 2002 ) using gro / comptel observations of cyg x-1 claim that the high energy tail in the soft state shows no cutoff up to at least 0.5 mev and 10 mev respectively . they demonstrate that their analysis from batse , osse and comptel show that the combined spectrum can be described by a single power law with a best - fit photon index of @xmath201 . for these particular states for which @xmath202 _ rxte _ power spectra are featureless ( see fig . [ power ] ) , no qpos are seen in these states . in fact , grove ( 2005 , private communications ) confirms that osse power spectra of cyg x-1 also do not show any signatures of qpo but only red noise . it means that in these observations the central bh is not seen directly but through a very extended cloud that ultimately washes out the timing and spectral information of the central object . it is also worth pointing out that the comptel exposure time is about @xmath203 s that is much longer than the spectral formation time near the central black hole(which is of order of crossing time of the innermost part of bh , a few times of @xmath204 s ) . the hydrodynamical time scale can be longer by two orders of magnitude than the spectral formation time but it is still 7 orders of magnitude shorter than the comptel exposure time . one can not exclude that the high energy emission detected by comptel is probably not related to central bh , but it can be a result of some other process , for example in the outflow or jets ( see the recent discovery of very high energy gamma rays associated with an x - ray binary ls 5039 in a paper by aharonian et al . 2005 ) . in figure [ index_vs_time ] we present timescale photon index variation for different states in cyg x-1 during @xmath0 8 years of _ rxte _ observations . in particular , cyg x-1 can be rather stable in the low / hard state for the several months , while it is very dynamic in the soft and very soft states where it changes its spectral index on timescale of a day . thus cyg x-1 stays in the low / hard with an occasional transition ( once per several years ) to the soft state where the power - law spectrum becomes significantly steeper ( with @xmath187 ) . also , one or two times per year cyg x-1 exhibits so called `` failed state transions '' , when it starts to transition but does not reach a soft state , stops at some intermediate state and falls back to a hard one . changes from 1.5 to 2.6 respectively . , title="fig:",width=3,height=307 ] changes from 1.5 to 2.6 respectively . , title="fig:",width=3,height=307 ] changes from 1.5 to 2.6 respectively . , title="fig:",width=3,height=307 ] changes from 1.5 to 2.6 respectively . , title="fig:",width=3,height=307 ] l low / hard state + bmc + @xmath10 , @xmath205 + @xmath153 , kev @xmath206 + a @xmath207 + gaussian ( 6.4 kev fixed ) + @xmath208 , @xmath209 + ew , ev @xmath210 + highecut + @xmath211 , kev @xmath212 + @xmath149 , kev @xmath213 + constant ( cross - normalization ) + pca / hexte a @xmath214 + pca / hexte b @xmath214 + @xmath215 0.84 ( 159 ) + + intermediate state + bmc + @xmath10 , @xmath216 + @xmath153 , kev @xmath217 + a @xmath218 + gaussian ( 6.4 kev fixed ) + @xmath208 , @xmath219 + ew , ev @xmath220 + highecut + @xmath211 , kev @xmath221 + @xmath149 , kev @xmath222 + constant ( cross - normalization ) + pca / hexte a @xmath223 + pca / hexte b @xmath224 + @xmath215 0.90 ( 159 ) + + very soft state + bmc + @xmath10 , @xmath225 + @xmath153 , kev @xmath226 + a @xmath227 + gaussian ( 6.4 kev fixed ) + @xmath208 , @xmath228 + ew , ev @xmath229 + highecut + @xmath211 , kev @xmath230 + @xmath149 , kev @xmath231 + constant ( cross - normalization ) + pca / hexte a @xmath232 + pca / hexte b @xmath233 + @xmath215 1.16 ( 159 ) + [ statestab ]

we have performed a series of three - dimensional simulations of the interaction of a supersonic wind with a non - spherical radiative cloud . these simulations are motivated by our recent three - dimensional model of a starburst - driven galactic wind interacting with an inhomogeneous disk , which show that an optically emitting filament can be formed by the break - up and acceleration of a cloud into a supersonic wind . in this study we consider the evolution of a cloud with two different geometries ( fractal and spherical ) and investigate the importance of radiative cooling on the cloud s survival . we have also undertaken a comprehensive resolution study in order to ascertain the effect of the assumed numerical resolution on the results . we find that the ability of the cloud to radiate heat is crucial for its survival , with a radiative cloud experiencing a lower degree of acceleration and having a higher relative mach number to the flow than in the adiabatic case . this diminishes the destructive effect of the kelvin - helmholtz instability on the cloud . while an adiabatic cloud is destroyed over a short period of time , a radiative cloud is broken up via the kelvin - helmholtz instability into numerous small , dense cloudlets , which are drawn into the flow to form a filamentary structure . the degree of fragmentation is highly dependent on the resolution of the simulation , with the number of cloudlets formed increasing as the kelvin - helmholtz instability is better resolved . nevertheless , there is a clear qualitative trend , with the filamentary structure still persistent at high resolution . the geometry of the cloud effects the speed at which the cloud fragments ; a wind more rapidly breaks - up the cloud in regions of least density . a cloud with a more inhomogeneous density distribution fragments faster than a cloud with a more uniform structure ( e.g. a sphere ) . we confirm the mechanism behind the formation of the h emitting filaments found in our global simulations of a starburst - driven wind . based on our resolution study , we conclude that bow shocks around accelerated gas clouds , and their interaction , are the main source of the soft x - ray emission observed in these galactic - scale winds .

0907.4004

the interstellar medium is known to be inhomogeneous , consisting of various gaseous phases , from cool molecular clouds to tenuous million degree gas @xcite . a study of the interaction of a supersonic wind with a dense cloud is a problem with many astronomical applications , such as galactic winds , supernova remnants , and broad absorption line quasars , and has received much attention . in the past this interaction has been studied both analytically ( e.g. * ? ? ? * ; * ? ? ? * ) and numerically ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? over the last two decades , numerous two- and three- dimensional simulations have been performed . many early attempts assumed an adiabatic interaction ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , while later attempts have included radiative cooling ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . among the plethora of simulations reported in the literature , the effects of thermal conduction ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , magnetic fields ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , photoevaporation ( e.g * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and the presence of multiple clouds ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have been considered . the wind / cloud interaction has also been investigated in the laboratory via laser experiments ( e.g. * ? ? ? * ; * ? ? ? * ) . in this paper , we report on our high resolution three - dimensional simulations of the interaction of a supersonic wind with a non - uniform radiative cloud . this work is motivated by our recent simulations of a starburst - driven galactic wind ( * ? ? ? * ; hereafter paper i ) , which showed that a optically emitting filament could be formed by the break - up and ram - pressure acceleration of a cloud into a supersonic wind . these simulations were three - dimensional , radiative , and incorporated an inhomogeneous disk that allowed us to study the interaction of a galactic scale wind with fractal clouds of disk gas . the interaction of a spherical cloud with a shock wave is often characterized by four evolutionary phases ( e.g. * ? ? ? * ; and references therein ) : 1 . an initial phase where the blast wave interacts with the cloud . a shock passes into the cloud , with another shock reflected into the surrounding medium . 2 . a phase of shock compression where the flow around the cloud converges on the axis at the rear . during this phase , the cloud begins to flatten , with its transverse size greatly reduced . a shock is also driven into the back of the cloud . the re - expansion phase where the shock transmitted into the cloud has reached the back surface and produces a strong rarefaction back into the cloud . this leads to expansion of the shocked cloud downstream . 4 . a final phase where the cloud fragments and is destroyed by hydrodynamical instabilities ( e.g. kelvin - helmholtz and rayleigh - taylor instabilities ) . @xcite performed an extensive series of two - dimensional simulations of an spherical adiabatic cloud interacting with a shock wave . they found that irrespective of the initial parameters , the cloud was destroyed within several cloud crushing times . @xcite confirm this result in their three - dimensional simulations . however , @xcite in a short letter , reported their two - dimensional study of the evolution a cloud in a radio galaxy cocoon . their simulations included the effects of radiative cooling and showed that merging of the front and back shocks leads to the formation of an elongated structure . this structure breaks - up into fragments which are not immediately destroyed . this increased ability for the cloud to survive has been reproduced by all studies that implement radiative cooling in their simulations . in addition , it has been shown that strong cooling in the cloud can cause a thin dense shell that acts to protect the cloud from ablation @xcite . thermal conduction can suppress the hydrodynamical instabilities that act to fragment a cloud @xcite , while the presence of a magnetic field has been shown to both hasten ( e.g * ? ? ? * ) and delay ( e.g. * ? ? ? * ; * ? ? ? * ) the cloud s destruction . more recently , @xcite performed three - dimensional simulations of the wind / cloud interaction that included the effects of radiative cooling , thermal conduction , and magnetic fields . they showed that in the presence of an ambient magnetic field , the effect of thermal conduction in stabilizing the cloud may be diminished depending on the alignment of the field . clearly the survival of a cloud interacting with a strong wind is a problem of significantly more complexity than indicated by the purely adiabatic scenario investigated by @xcite and others . in order to compare the results of our current study to those of our global model we exclude the thermal conduction , magnetic fields , or photoevaporation . however , we consider the possible effect of these phenomena on our results in [ missing ] . in this work , we further investigate the effects of radiative cooling on the survival of the cloud , by performing high @xmath1 pc resolution simulations of both a radiative and an adiabatic cloud , allowing us to perform a direct comparison between the two scenarios . symmetry is _ not _ assumed , with the entire cloud being modeled , in contrast to the strategy employed in many previous three - dimensional models . this allows us to fully investigate of the turbulent flow on the evolution of the cloud . in order to understand the effects of the cloud s structure on the wind / cloud interaction , we consider two different cloud geometries : a more realistic fractal cloud , similar to those that comprised the inhomogeneous disk in paper i , and the idealized case of a spherical cloud . the effect of the cloud s geometry on its evolution has been investigated before . @xcite considered the interaction of a shock wave with a spherical and two different prolate cloud geometries , each with a different alignment of the cloud s major axis . in addition , @xcite considered both spherical and elliptical cloud geometries . both studies show that the initial geometry of the cloud can alter the evolution of the cloud significantly . one of the most important results from paper i was our suggestion of a mechanism for the formation of the filaments seen in starburst - driven winds at optical wavelengths ( see * ? ? ? * ; for review ) . according to paper i the filaments are formed from clouds of disk gas that have been accelerated into the outflow by the ram - pressure of the wind . an important question that arises is : can the clouds survive being immersed in a hot supersonic wind long enough to form a filament and remain sufficiently cool to emit at optical temperatures ? or will they be heated and destroyed by hydrodynamical instabilities ? here we set out to answer this question . another significant result to arise from paper i is our proposal of four different mechanisms that would give rise to soft x - ray emission that is spatially correlated with the filamentary optical emission ; a major finding of recent _ chandra _ observations @xcite . our global simulations found that soft x - rays can arise from ( i ) mass - loading from ablated clouds , ( ii ) the intermediate temperature interface between the hot wind and cool filaments , ( iii ) bow shocks upstream of clouds accelerated into the outflow , and ( iv ) interactions between these bow shocks . the first two mechanisms involve the mixing of hot and cold gas , and are possibly caused by numerical diffusion in the simulations , and thus may not be physical . to investigate this possibility , we have performed a detailed resolution study of the wind / cloud interaction . the study also allows us to test the impact of the numerical resolution on the evolution of the cloud . the simulations were performed using the ppmlr ( piecewise parabolic method with a lagrangian remap ) code utilized in paper i. ppmlr is a multidimensional hydrodynamics code based on the method described by @xcite and has been extensively modified ( e.g. * ? ? ? * ; * ? ? ? * ) from the original vh-1 code @xcite . thermal cooling , based on output from the mappings iii code @xcite has been implemented . this allows for a realistic evolution of a radiatively cooling gas @xcite . the simulations discussed in this paper are three - dimensional and utilize cartesian ( x , y , z ) coordinates . they were performed on the sgi altix computer operated by the the australian partnership for advanced computing . in order to study the wind / cloud interaction in sufficient detail , whilst still retaining the ability to follow the evolution and survival of the cloud over a period of approximately 1 million years , the simulations cover a physical spatial range of 50 @xmath2 50 @xmath2 150 pc , with the cloud centered on the origin . as our intent is to compare these simulations to the formation and survival of the clouds found in paper i , we choose initial conditions for the wind based upon the results of those simulations ( table [ init_cond ] ) . in order to understand our choice , we briefly recount the formation and evolution of the starburst - driven winds which we simulated in paper i : 1 . a series of small bubbles of hot ( @xmath3 ) gas form in the starburst region . as these bubbles expand , they merge and follow the path of least resistance out of the disk of the galaxy , i.e. the tenuous hot gas surrounding the denser disk clouds . as the bubble breaks out of the disk , it begins sweeping up the the surrounding halo gas entering the `` snow - plow '' phase of its evolution . the structure of the wind in the phase is characterized by 5 different zones : ( i ) the injection zone , ( ii ) a supersonic free wind , ( iii ) a region of hot , shocked turbulent gas , ( iv ) a cooler `` shell '' of swept - up halo gas , and ( v ) the undisturbed ambient gas . clouds of disk gas inside and surrounding the central injection zone are broken - up by the freely expanding wind ; fragments of disk gas are accelerated into the outflow by the ram - pressure of the wind . since it is the interaction of the clouds with the freely expanding wind that results in their fragmentation , we select the temperature , density and velocity of our supersonic wind to be that of the inner free - wind region , namely @xmath4 , @xmath5 , and @xmath6 respectively . lcc wind temperature ( k ) & @xmath7 & @xmath8 + wind density ( @xmath9 ) & @xmath10 & 0.1 + wind velocity ( @xmath11 ) & @xmath12 & @xmath13 + wind mach number & @xmath14 & 4.6 + average cloud temperature ( k ) & @xmath15 & @xmath16 + cloud velocity ( @xmath11 ) & @xmath17 & 0 + cloud radius ( pc ) & @xmath18 & 5 + fractal cloud volume ( @xmath19 ) & @xmath20 & 1491 + spherical cloud volume ( @xmath19 ) & @xmath21 & 523 + our simulations make use of two cloud geometries : a fractal shaped and a spherical shaped cloud . the fractal cloud was chosen in order to allow us to compare the break - up of the clouds in this study to the results of our global model in paper i and has the same form as the clouds in the inhomogeneous disk used in the global study . the use of the fractal cloud also allows us to investigate the effects of inhomogeneities in the clouds s density distribution on its evolution . a spherical cloud was also modeled in order to allow us to better understand the importance of the assumed cloud geometry . to create the fractal cloud we first created a 1024 @xmath2 1024 @xmath2 1024 sized fractal cube using the method described in @xcite and paper i. a single cloud was isolated and extracted from this cube using a blob - coloring technique where each cell is examined , and any discontinuous group of non - zero cells given a unique label . an appropriate cloud was then selected and placed at cell number ( 256,256,256 ) of a 512 @xmath2 512 @xmath2 1536 sized grid , i.e. the origin of the simulation . the smaller resolution grids used in our fractal simulations were created by downsizing this larger grid ( see table [ sim_param ] ) . these arrays represent the density of the cloud . the initial grid was setup by first setting the density ( @xmath10 ) and pressure ( @xmath22 ) to be that of the hot wind . the fractal cloud was then created by adding the density array representing the cloud to the density of the halo gas . to create the spherical cloud , a high density spherical region of radius @xmath23 was centered on the origin of the computational grid . the radial profile of the density of the spherical cloud is described by an exponential with a scaling radius of 3 , in order to mimic the tapered density distribution in the fractal cloud s core . a similar tapered density distribution was considered by @xcite for an adiabatic cloud . the boundary condition on the inner z axis was set to have a fixed inflow with the same properties as that of the hot wind ( e.g. @xmath24 , @xmath5 , and @xmath25 ) . all other boundaries were set to be inflowing / outflowing . figure [ fig : dens_profile ] shows the initial density distribution of both the fractal and spherical clouds . the average number density of the fractal cloud is set to be @xmath26 and has a total mass of @xmath27 , occupying a volume of @xmath28 . the spherical cloud is setup to occupy a similar volume to the dense core of the fractal cloud , having a radius of @xmath23 and occupying a volume of @xmath29 . the average density of the spherical cloud is @xmath30 and has a total mass of @xmath31 . the lower average density of the fractal cloud is due to the large volume of less dense ( @xmath32 ) gas that surrounds the cloud core ( fig . [ fig : dens_profile ] ; left panel ) . in order to understand the effect of the cloud s assumed initial density on its evolution , a simulation in which the density of the fractal cloud was doubled ( @xmath33 ) was also performed . in all simulations , the temperature and velocity of each cloud was set to be @xmath34 and @xmath35 respectively . while these simulations were designed to be applicable to starburst - driven winds and therefore have densities and temperatures typically found in a such an environment , they are also applicable to other astronomical phenomena that involve the interaction of a supersonic wind with a cloud of gas ( e.g supernova remnants ) . when cooling is included , the ppmlr code has a one parameter scaling , which is discussed in @xcite . in general , the density scale of the simulations is inversely proportional to the spatial scale , so that within reason , these simulations can be adapted to problems on both larger and small scales . the interaction of a supersonic wind with a cloud of dense gas is a problem of some complexity . whilst there are many factors that could affect a clouds evolution and survival , such as thermal conduction @xcite , magnetic fields @xcite and photoevaporation @xcite ( see [ missing ] ) , here we focus on the importance of radiative cooling and the effect on the clouds initial structure . we have also performed a comprehensive resolution study in order to ascertain the degree as to which the assumed resolution effected our global simulations in paper i. we adopt the following naming convention for our simulations : an @xmath36 or an @xmath37 indicates whether the simulation includes radiative cooling or is adiabatic , while an @xmath38 or an @xmath39 indicates if the geometry of the cloud is fractal or spherical respectively . the numerical value indicates the number of cells in the x - plane of the computational grid . for example , the simulation rf384 includes radiative cooling , incorporates a fractal cloud and utilizes a computational grid of size @xmath40 cells . an exception to the naming convention is rfd384 , which is identical to rf384 , but whose cloud is twice as dense . ccccccc rf064 & @xmath41 & 0.78 & 63 & 1387 & yes & f + rf096 & @xmath42 & 0.52 & 63 & 1387 & yes & f + rf128 & @xmath43 & 0.39 & 63 & 1387 & yes & f + rf192 & @xmath44 & 0.26 & 63 & 1387 & yes & f + rf256 & @xmath45 & 0.20 & 63 & 1387 & yes & f + rf384 & @xmath40 & 0.13 & 63 & 1387 & yes & f + rf512 & @xmath46 & 0.10 & 63 & 1387 & yes & f + af384 & @xmath40 & 0.13 & 63 & 1387 & no & f + rs384 & @xmath40 & 0.13 & 91 & 703 & yes & s + as384 & @xmath40 & 0.13 & 91 & 703 & no & s + rfd384 & @xmath40 & 0.13 & 126 & 2770 & yes & f + in total , eleven simulations were performed , with the purpose of each falling within the four different categories outlined below . 1 . _ resolution study _ - models * rf064 * , * rf096 * , * rf128 * , * rf192 * , * rf256 * , * rf384 * , and * rf512 * form our resolution study . these seven simulations all include radiative cooling and utilize a fractal cloud . the resolution of each simulation is given in table [ sim_param ] and ranges from 0.79 to 0.10 pc per cell width . 2 . _ cloud structure _ - model * rs384 * has a resolution of 0.13 pc per cell width , includes radiative cooling and utilizes a spherical cloud . by comparison to rf384 , this model is designed to investigate the effect of the shape of the cloud on its break - up and survival . _ radiative cooling _ - models * af384 * and * as384 * have the same properties as rf384 and rs384 respectively , but are adiabatic in nature . both models are designed to help understand the importance of radiative cooling on the evolution and survival of a cloud . _ cloud density _ - model * rfd384 * is identical to rf384 , but has a larger cloud density and mass of @xmath33 and @xmath47 respectively . this model is designed to investigate the effect of the clouds initial density on its evolution . a summary of the parameters used in each simulation is given in table [ sim_param ] . for each simulation we record the density , temperature , pressure , velocity , emissivity , and a cloud gas tracer in each cell at intervals of 0.01 myr . with the exception of rf512 , each simulation is followed until the time at which the cloud flows off computational grid . in the case of rf512 , a computational error occurred while the simulation was in progress . unfortunately this simulation is too computationally expensive to re - run , and we are therefore only able to follow the evolution of rf512 to a time t = 0.7 myr . as such , our resolution study only considers the first 0.7 myr of the evolution . based on their two - dimensional adiabatic simulations , @xcite suggest that a resolution of 120 cells per cloud radius is necessary in order to sufficiently capture the hydrodynamics of the wind / cloud interaction . while this proposed benchmark is easily achieved for a two - dimensional study , it becomes computationally demanding for a fully three - dimensional simulation that include more complicated physics ( e.g. radiative cooling , thermal conduction ) . in order to overcome this difficulty , other authors have performed two - dimensional axisymmetric simulations @xcite , assumed symmetry in the solution @xcite , and neglected the more complex physics in their three - dimensional calculations @xcite . we _ do not _ assume symmetry as we show in [ evolution ] that , even in the idealized case of a spherical cloud , the solution becomes asymmetric as the cloud is broken - up and accelerated into the turbulent flow . as the ppmlr code used in this work and paper i utilizes a uniform grid , the computational resources required to model the entire cloud at the 120 cells per cloud radius resolution suggested by @xcite would be excessive . we are forced to limit the main simulations in this study to a resolution of 38 cells per cloud radius ( we consider a resolution of 50 cells per cloud radius in our resolution study ) . nevertheless , it has been shown that the global properties of the interaction , as well as the averaged characteristics of cloud ablation process , can be well described at resolutions below this criterion ( see , for example , * ? ? ? * ; * ? ? ? we start with the simple case of the interaction of a spherical radiative cloud ( rs384 ) with a supersonic wind , before describing the evolution of the fractal cloud . in order to illustrate the hydrodynamics of the wind / cloud interaction , throughout this section we will use slices of the density , temperature , pressure and velocity through the central y=0 plane . figure [ fig : dens_sph ] shows the evolution of the spherical cloud a 6 different epochs from 0.1 to 1.1 myr at 0.2 million year intervals . each panel represents the logarithm of the number density ( @xmath9 ) . the top panels of figure [ fig : var_sph ] show the logarithm of the temperature ( @xmath48 ) , the middle panels show the logarithm of the pressure ( @xmath49 ) , and the bottom panels show the magnitude of the velocity ( @xmath50 ) . the evolution at two different epochs is given : 0.35 myr ( left panels ) and 0.75 myr ( right panels ) . the spherical cloud is initially at rest and is immersed in a hot ( @xmath51 ) , supersonic ( @xmath52 ) wind . a bow shock is immediately formed upstream of the cloud . at 0.10 myr ( fig . [ fig : dens_sph ] ; upper left panel ) , the `` front '' of the cloud has been exposed to the high pressure of the shock , while gas is ablated from the `` back '' of the cloud a result of the kelvin - helmholtz instability and the strong rarefaction that is formed . a high density shock ( @xmath53 ) begins to propagate through the cloud , reflecting of the back wall at approximately 0.30 myr ( fig . [ fig : dens_sph ] ; middle left panel ) . the kelvin - helmholtz instability continues to work , stripping material from the edge of the cloud ( e.g. 0.5 - 1.1 myr ) . this material is funneled to approximately 5 pc behind the cloud where it combines and condenses forming a tail of dense ( @xmath54 ) , cool ( @xmath55 ) cloudlets , which are entrained into the hot turbulent flow downstream from the main cloud ( fig . [ fig : var_sph ] ; upper left panel ) . as the cloud evolves ( figs . [ fig : dens_sph ] ; right panels ) , the tail of cool gas continues to grow , becoming thicker as the cloud elongates . the cloud material appears as a filament of cool , low velocity ( @xmath56 ) gas , immersed inside a region of hot ( @xmath57 ) , turbulent gas with velocity @xmath58 . small cloudlets continue to be broken off the main cloud as the kelvin - helmholtz instability acts to shed its outer later . however , at 0.75 myr only 12% of the mass of material that remains on the computational grid is found mixed into the hot wind . the bulk of the mass is still found in the cloud s elongated core and tail , and despite the driving of radiative shocks into the cloud(s ) there is little increase in the temperature of the cloud material as it radiates heat , remaining cool and cohesive as it leaves the computational grid . figure [ fig : dens_cld ] shows the evolution of the radiative fractal cloud in model rf384 from 0.1 to 1.1 myr through density slices at intervals of 0.2 myr . figure [ fig : dens_hd ] is identical to figure [ fig : dens_cld ] , but shows the evolution of a cloud ( rfd384 ) with twice the density and mass than rf384 . as with the radiative spherical cloud , a bow shock immediately forms upstream of the cloud . however , a significant effect of the inhomogeneous structure of the cloud is the formation of a shock off each ridge on the cloud s surface that is exposed to the wind . this has the effect of creating a `` web '' of interacting shocks . by 0.3 myr ( fig . [ fig : dens_cld ] ; middle left panel ) , the density at the front of the cloud has jumped to @xmath59 , but fallen to approximately @xmath60 at the rear of the cloud . despite the high pressure exerted on the cloud , very little heating occurs , with the cloud maintaining temperatures of @xmath61 ( fig . [ fig : var_cld ] ; left upper two panels ) . as the cloud evolves ( fig . [ fig : dens_cld ] ; right panels ) , small cloudlets are broken off the main cloud by the kelvin - helmholtz instability . these cloudlets form a filamentary structure downstream of the main cloud and have velocities in the range of @xmath62 , somewhat slower than the velocity of the surrounding stream ( @xmath63 ) ( fig . [ fig : var_cld ] ; bottom panels ) . the cloud is exposed to the high temperature and velocity of the wind at all points along the front surface of the cloud . the cloud breaks up fastest in regions where the density is lowest and the radius of the cloud is at its smallest . the cloud is continually eroded by the kelvin - helmholtz instability , the fragments of which are immersed in a low pressure , turbulent gas . at 0.75 myr , the percentage of the mass of cloud material remaining on the computational grid that has mixed into the hot wind is @xmath64 25% . the bulk of the original cloud mass is found in the remaining core remnant and the stream of small , dense @xmath64 1 pc cloudlets . if these cloudlets are exposed to the wind they produce their own high pressure bow shock upstream of their position . however , the majority of the cloudlets are sheltered from the wind by other fragments broken off from the main cloud . these cloudlets survive to leave the computational grid . the higher density of the cloud in model rfd384 ( fig . [ fig : dens_hd ] ) results in the cloud retaining is structural integrity far longer than the lower density cloud in rf384 . while the evolution of the higher density cloud is overall very similar to that in rf384 , the break - up of the cloud via the kelvin - helmholtz instability is slower : at 1.1 myr ( bottom right panel ) , a significant bulk of the cloud material still remains on the computational grid and displays a similar morphology to the cloud in rf384 at 0.7 myr ( fig . [ fig : dens_cld ] ; top right panel ) . in contrast to rf384 , at 0.75 myr only 11% of the clouds mass remaining on the computational grid is found mixed into the hot wind . however , it is likely this amount will increase as the cloud is further eroded by the kelvin - helmholtz instability . as observed by @xcite , the initial shape of the cloud has an effect on its subsequent evolution , with our fractal cloud fragmenting faster than the spherical cloud . while this is due in part to the lower average density of the fractal cloud , even the high density fractal cloud , whose average density and total mass is greater than that of the spherical cloud ( see table [ sim_param ] ) , has a greater degree of fragmentation and is less cohesive when it leaves the computational grid . this is a result of the inhomogeneous nature of the fractal cloud s initial density distribution and the larger cross - section is presents to the incident wind ( fig . [ fig : dens_profile ] ; left panel ) . the cloud first fragments along regions where the wind finds paths of least resistance , i.e. regions of low density . as a result the fractal cloud breaks - up into multiple core fragments . as the wind finds no regions of least resistance in the spherical cloud , it is able to retain a single cohesive structure for a longer period of time . thus , not only is the initial geometry of the cloud important in determining its evolution , the distribution of the cloud s density determines how quickly the cloud begins to fragment . more homogeneous density distributions would result in less initial fragmentation . in order to understand the degree as to which the inclusion of radiative cooling affects the survival of a cloud , we performed 2 simulations in which cooling was neglected : as384 and af384 for the spherical and fractal case respectively . figure [ fig : dens_adi ] shows the density of the adiabatic clouds in as384 ( left ) and af384 ( right ) at 0.3 , 0.6 , 0.9 myr epochs . it can be immediately seen that in the absence of radiative cooling there is a far greater degree of mixing of cloud material with the surrounding stream , with 40% and 59% of the fractal and spherical clouds masses respectively on the computational grid found mixed into the hot wind at 0.75 myr . the destruction of the adiabatic fractal cloud occurs faster and is more complete than the adiabatic spherical cloud . at 0.9 myr the cloud has been almost completely destroyed ( fig . [ fig : dens_adi ] ; bottom right panel ) , with the cloud material mixed into the hot gas . the initial interaction with the wind is the same as in the radiative case , with a bow shock forming upstream of the cloud . however , the cloud gas quickly begins to heat to temperatures of the order of @xmath65 ( fig . [ fig : var_af ] ; upper left panel ) and the cloud expands . again the cloud breaks up first in regions where the density is low and the cloud radius is at its smallest . the high pressure exerted on the cloud ( fig . [ fig : var_af ] ; middle left panel ) and the kelvin - helmholtz instability act to strip material from the cloud . while this material is able to survive in the radiative model , here the cloudlets broken off the main cloud are quickly heated and destroyed . the bulk velocity of the gas downstream of the main cloud is lower than that found in the radiative case ( fig . [ fig : var_af ] ; lower panels ) . in the case of the adiabatic spherical cloud , the shedding of the cloud s outer layer by the kelvin - helmholtz instability , also seen in the radiative model , is greatly enhanced . by 0.9 myr ( fig . [ fig : dens_adi ] ; bottom left panel ) , only the cloud core remains and is subsequently destroyed by the kelvin - helmholtz instability as the simulation progresses . as with the fractal cloud , the bulk velocity of the gas downstream of the main cloud is lower than that of the radiative spherical cloud at the same time ( fig . [ fig : var_as ] ; lower right panel ) . the evolution of a spherical adiabatic cloud is discussed in more detail below . one of the significant effects of the inclusion of radiative cooling is the longer life time of the impacted cloud . while this effect has been observed in the past by other authors ( e.g. * ? ? ? * ; * ? ? ? * ) , our study is of higher resolution and does not assume any symmetry , making it is a useful exercise to directly compare the simple case of the evolution of the spherical cloud in both our adiabatic ( as384 ) and radiative ( rs384 ) models in order to determine the mechanism behind the radiative cloud s survival . the initial interaction of the wind and cloud is shown via density slices in figure [ fig : dens_initial ] at 0.05 , 0.20 and 0.35 myr epochs in both adiabatic ( left ) and radiative ( right ) models . it can be clearly seen that while the evolution begins almost identically with a bow shock forming upstream of the cloud , in the adiabatic case cloud material immediately starts being ablated from the back of the cloud . as in the radiative model , a shock propagates through the adiabatic cloud . however , the initial density increase observed in the cloud is not as extreme ( e.g. @xmath66 ) . the cloud material is heated to temperatures of the order of @xmath67 and the cloud expands . the shock travels though the cloud , reflecting off the back surface at approximately 0.27 myr , somewhat faster than in the radiative cloud . the adiabatic cloud expands transversely as it is accelerated downstream . however , this transverse expansion is suppressed in the radiative cloud as a result of the lower degree of heating of the cloud gas ( fig . [ fig : var_sph ] ; upper panels ) . in both cases , the kelvin - helmholtz instability acts to strip material from the edges of the cloud forming a tail of material downstream of the cloud position ( fig . [ fig : dens_initial ] ; middle and lower panels ) . in the adiabatic model , this tail is geometrically thick with density and temperature @xmath68 and @xmath69 respectively . in contrast , the tail formed in the radiative model is geometrically thin with density @xmath54 and temperature @xmath70 . in the adiabatic model , the internal cloud shock reflects again off the front the cloud at approximately 0.35 myr . at this time , the transverse expansion of the cloud persists and the kelvin - helmholtz instability continues to strip material from the clouds exterior into the downstream tail of gas . [ fig : dens_initial ] ; lower left panel ) . the transverse expansion ( fig . [ fig : var_as ] ; middle panels ) results in a higher rate of acceleration in the adiabatic model . as a consequence , the cloud has a lower relative mach number relative to the stream ( fig . [ fig : var_as ] ; lower left panel ) than the radiative cloud ( fig . [ fig : var_sph ] ; lower left panel ) . the growth rate of the kelvin - helmholtz instability is lower for higher mach numbers and its effect is strongly diminished for the radiative cloud . the adiabatic cloud is more easily disrupted and destroyed . this can be dramatically seen in the middle left panel of figure [ fig : dens_adi ] where the kelvin - helmholtz instability has stripped the entire outer layer of the adiabatic spherical cloud ( see also fig . 3 of * ? ? ? * ) . since a radiative cloud is broken - up via the kelvin - helmholtz instability into a filamentary structure of small @xmath64 1 pc sized clouds , the survival of these small clouds is of interest . we now compare the cloud crushing time ( @xmath71 ) and the kelvin - helmholtz timescale ( @xmath72 ) to the cooling time ( @xmath73 ) of a cloud with radius @xmath74 , density @xmath75 , temperature @xmath76 , and velocity @xmath77 . the cloud crushing time of such a cloud is @xmath78 , where the density and velocity of the wind is @xmath79 and @xmath80 respectively . the kelvin - helmholtz timescale is @xmath81 . the cooling time for a 1 pc cloud in our simulations is of the order of @xmath82 seconds , somewhat shorter than the cloud crushing time and the kelvin - helmholtz timescale , suggesting that the cloudlets may remain sufficiently stable to ablation and survive to later times . in addition , we note that the self - gravity of the clouds may cause them to collapse , becoming more difficult to disrupt @xcite . in order to test the dependence of our results on the numerical resolution of the code , we have performed seven simulations of the radiative fractal cloud interacting with a supersonic wind at increasing resolutions from 0.78 - 0.10 pc per cell width ( see table [ sim_param ] ) . we calculated the flux of mass through a surface at @xmath83 over the first 0.7 million years of each simulation in our resolution study . figure [ fig : mass_flux ] shows this mass flux for each simulation as a function of time ( left panel ) , as well as the total mass flux integrated over the first 0.7 myr of the simulation as a function of resolution ( right panel ) . the mass flux over the first 0.4 myr of the evolution is similar at all resolutions . this is a result of the well resolved hot stream of gas passing through the flux surface . after this point , the cloud material begins to pass though the surface and the mass flux starts to vary with resolution . the initial drop in the mass flux seen in the left hand panel of figure [ fig : mass_flux ] is due to the rarefaction that passes though the flux surface ( see fig . [ fig : dens_sph ] ; top left panel ) . from approximately 0.2 to 0.4 myr , the density of the stream increases as mass is ablated from the rear of the cloud resulting in a similar flux at all resolutions . during this time , the mass flux gradually increases as the @xmath84 mass loaded stream of gas passes through the flux surface . at approximately 0.4 myr , the mass flux begins to vary rapidly , dramatically increasing in each simulation . the large variation in the mass flux is due to the turbulent nature of the gas passing though the flux surface and the dense cloudlets immersed within this gas . the cloudlets pass through the surface at different times resulting in fluctuations in the mass flux . there is a general trend of increasing mass flux with the resolution of the simulation . this trend can be explained by the increase in fragmentation of the cloud with increasing numerical resolution . figure [ fig : res_dens ] shows volume renderings of the projected density at 0.7 myr in models rf064 , rf128 , rf256 and rf512 , which have resolutions of 0.78 , 0.39 , 0.20 , and 0.10 pc per cell width respectively . the increase in the fragmentation of the cloud with resolution can clearly be seen , and will be discussed in more detail in [ fragment ] . at the highest resolution attempted in this study ( 0.1 pc ) , the filaments resemble a `` foam '' of cloudlets , while at low resolution the cloud has been broken - up into only a few large fragments . as a consequence , the cross - section of dense material that passes through the flux surface at any given time after 0.4 myr is larger in the higher resolution simulations . the total integrated mass flux passing through the surface at @xmath85 over the first 0.7 myr also increases with numerical resolution ( fig . [ fig : mass_flux ] ; right panel ) . again this is caused by the larger degree of fragmentation at high resolution , resulting in more ablation of cloud material from the back of the cloud . between our highest and lowest resolution simulations the difference in mass flux is approximately 10% . this discrepancy is likely to increase at higher resolutions , although it is possible that convergence may occur at extremely high resolution simulations ( @xmath86 ) that utilize an adaptive mesh . for the 2 pc per cell width resolution of our global simulations in paper i , this error is increased to approximately 20% . we will discuss the impact of the numerical resolution on the results of paper i in [ sb_winds ] . the most significant effect of increasing the numerical resolution of the simulation is the increase in fragmentation of the cloud . the increase in cloud fragmentation can clearly be seen in figure [ fig : res_dens ] , with the the cloud broken into only a few large fragments in the lowest resolution simulations , but 100 s of fragments at higher resolution . nevertheless , filamentary structure , where the concentration of cloudlets is higher , can still be made out at high resolution . these `` filaments '' are located in similar positions to the filaments in the low resolution simulations . we are able to calculate the properties of each cloudlet by using an algorithm which allows us to pick out and select fragments . note that we impose a minimum mass of @xmath87 for a fragment to be selected . figure [ fig : number_clouds ] shows the number of cloudlets produced as a function of numerical resolution . there is a general trend from the number of cloudlets produced in the interaction to increase as a power law with increasing resolution . even with larger computational resources and an adaptive mesh , this trend is likely to continue ad infinitum . the dependence of the degree of fragmentation on the numerical resolution has been observed by other authors in both two - dimensional @xcite and three - dimensional @xcite simulations of a spherical cloud . this effect can be explained by the growth rate of kelvin - helmholtz instability , which is faster at smaller wavelengths . as the resolution is increased , this instability is increasingly resolved and more fragmentation of the cloud is observed . this is illustrated in figure [ fig : kelvin - helmholtz ] , where the kelvin - helmholtz instability can clearly be seen to increasingly fragment the cloud at higher numerical resolution . ) as function of the cloudlet mass . ( online : 0.78 pc / cell [ navy ] , 0.39 pc / cell [ cyan ] , 0.20 pc / cell [ gold ] , and 0.10 pc / cell [ red]).,title="fig : " ] ) as function of the cloudlet mass . ( online : 0.78 pc / cell [ navy ] , 0.39 pc / cell [ cyan ] , 0.20 pc / cell [ gold ] , and 0.10 pc / cell [ red]).,title="fig : " ] ) as function of the cloudlet mass . ( online : 0.78 pc / cell [ navy ] , 0.39 pc / cell [ cyan ] , 0.20 pc / cell [ gold ] , and 0.10 pc / cell [ red]).,title="fig : " ] + figure [ fig : cloud_prop ] gives velocity ( left ) and mass ( centre ) histograms of the cloudlets , as well as the total mass of the cloudlets as function of the cloudlet mass ( right ) , at 0.7 myr for the 4 resolutions shown in figure [ fig : res_dens ] . the massive fragments ( @xmath88 ) present at all resolutions are the remnants of the cloud core . since the number of cloudlets increases with resolution , the highest resolution simulations are comprised of numerous low mass cloudlets . in general , there is a trend towards increasing smaller mass fragments at all resolutions . however , in all cases , the majority of mass of the cloudlet system is found in the massive fragments ( @xmath89 40 % ) rather than the smaller fragments . the velocity of the cloudlets does not vary significantly with resolution . despite the large number of lower mass fragments present in the high resolution simulations , they still fall within the velocity range of @xmath90 . the bulk of the cloudlets at all resolutions have a velocity in the range of @xmath91 . this is likely to increase as the evolution progresses and the cloudlets are further accelerated by the wind . unfortunately , we are unable to fully resolve the interaction of a radiative cloud with a supersonic wind at this time . this limits our ability to draw any reliable conclusions regarding the small scale evolution of the wind / cloud interaction . a significant increase in numerical resolution and/or the use of an adaptive mesh would be required in order for convergence to possibly occur . nevertheless , there is a clear trend in the large - scale evolution of the wind / cloud interaction at all resolutions considered in this study , allowing us to draw some physical conclusions . for example , the soft x - ray luminosity is sufficiently resolved and is almost constant with numerical resolution ( see [ xray ] ) . the effect of radiation in keeping the cloud cool , suppressing the transverse expansion and minimizing the effect of the kelvin - helmholtz instability , is also not effected by the resolution of the simulation . in all cases , the cloud is not immediately destroyed and mixed into the hot wind as seen in adiabatic models , but is instead broken - up into numerous small cloudlets . the major effect of increasing the resolution is the increased fragmentation of the cloud . however , we still see the same qualitative structure at high resolution , with the cloud breaking - up to form a filamentary structure that becomes finer and finer as more detail is resolved . at optical wavelengths ( such as h@xmath0 ) , starburst - driven winds appear as spectacular filamentary systems extending several kpc along the minor axis of the host galaxy , e.g. m82 @xcite , ngc 3079 @xcite , ngc 1569 @xcite . while it has long been proposed that this filamentary material was expelled from the central region of the galaxy @xcite , until now the mechanism behind the formation of the filaments has not been completely understood . in paper i , we proposed that the filaments were formed via clouds of disk gas that are broken - up and accelerated into the outflow by the ram - pressure of the wind . in order for this mechanism to be viable , the cloud fragments need to survive and remain sufficiently cool to emit at h@xmath0 temperatures . the simulations presented in this paper allow us to address this important issue . as in paper i , we define the h@xmath0 emitting gas to be cloud material with temperatures in the range of @xmath92 . we note that as photoionization is not included in our model , the h@xmath0 emission discussed in this paper arises solely from shock ionization . however , photoionization is known to play a role in the ionization of the filaments in many winds . for example , the filaments in m82 s wind are known to be photoionized at low distances above and below the galactic plane , with shock ionization becoming dominant at large distances . an investigation into the effects of photoionization on the cloudlets is warranted , but is beyond the scope of this study . figure [ fig : halpha ] shows a three - dimensional volume rendering of the density of the h@xmath0 emitting gas at 0.5 myr in model rf384 ( online : movie of evolution of the h@xmath0 emitting gas in rf384 over the first 1.37 myr ) . it can be immediately seen in figure [ fig : halpha ] that the h@xmath0 emitting material corresponds to the dense cloud material . thus , the survival mechanism for a cloud proposed in [ survival ] can be invoked to explain the filaments observed in starburst - driven winds . as discussed in [ evolution ] , the cloud is broken - up via the kelvin - helmholtz instability , with the fragments subsequently entrained into the outflow forming a filamentary structure . figure [ fig : vel ] ( left panel ) gives the emission weighted histogram of the z - velocity along the filament at 0.7 myr in model rf384 . the majority of the h@xmath0 emission has velocities in the range of @xmath93 , with the velocity increasing with distance along the z - axis . this is consistent with the h@xmath0 gas reaching higher velocities at larger distances above the galaxy plane , which was observed in the global simulations in paper i. note that this result represents only the velocity dispersion at the base of a _ single _ filament early in its evolution , with the velocity likely to increase as the cloudlets , which form the filament , are further accelerated in the direction of the flow . as in our global simulations , we again find that it is the ram - pressure of the wind that accelerates the clouds . the main difference between the filaments formed in paper i and the filaments found here is the number of cloudlets that comprise the filament ; a direct result of the higher numerical resolution of the simulations in this work . if we were able to increase the resolution of the global simulations , we would find similar fine structure in the filaments that are formed . figure [ fig : ha_mass_flux ] ( left ) shows the mass flux of the h@xmath0 emitting gas through a surface at @xmath83 at each resolution considered in our study . as expected , there is no h@xmath0 emitting gas passing though the surface until approximately 0.3 myr , as the dense cloud material has yet to encounter the flux surface . as in figure [ fig : mass_flux ] , the difference in the mass flux at each resolution is a result of the increasing fragmentation of the h@xmath0 emitting clouds at high resolution . the right hand panel of figure [ fig : ha_mass_flux ] shows the total integrated mass flux , over the first 0.7 myr , of the h@xmath0 emitting gas as a function of resolution . we see an increase in the total mass passing through the flux surface , again caused by the increase in fragmentation as the kelvin - helmholtz instability is further resolved . even at the high resolution of these simulations the h@xmath0 mass flux is yet to converge . in [ survival ] we discussed how a cloud could survive the interaction with a hot supersonic wind . the ability of a cloud to radiate heat is crucial for the clouds survival , allowing it to remain stable to ablation and emit at h@xmath0 temperatures . without this ability , the cloud quickly heats above @xmath94 , expands and becomes susceptible to the kelvin - helmholtz instability . while a radiative cloud is still disrupted , the small sized fragments that are broken off the main cloud have cooling times faster than the cloud crushing time and the kelvin - helmholtz growth rate , and thus , may possibly survive . in the adiabatic case , the fragments get heated and destroyed , with the cloud material quickly becoming mixed into the hot wind . if the fragments survive , they are drawn - out into strands and form a filament downstream of the original cloud position , reminiscent of the filaments seen in starburst - driven winds . in paper i we proposed four mechanisms that could give rise to the soft x - ray emission that is observed to be spatially correlated to the h@xmath0 emitting filaments . this correlation has been observed by _ chandra _ in many starburst - driven winds ( e.g * ? ? ? * ; * ? ? ? * ; * ? ? ? here we summarize the proposed mechanisms : 1 . the mass - loaded wind . this is the largest contributor to the soft x - ray emission in the global simulations . as mass is ablated from the clouds it is mixed into the surrounding turbulent gas , creating a region of hot ( @xmath95 ) rapidly cooling gas that emits strongly at soft x - ray energies . 2 . the intermediate temperature interface between the hot wind and cool filaments . gas at the boundary between the hot and cool gas mixes to produce a thin region of intermediate density and temperature . like the mass - loaded wind , this mixed gas is a strong emitter of soft x - rays . 3 . bow shocks . gas is heated to x - ray temperatures as a bow shock is formed upstream of dense clouds accelerated into the flow . 4 . colliding bow shocks . when two bow shocks interact , the gas is further shock heated to x - ray temperatures . the first two processes involve the mixing of hot and cold gas and could be the result of numerical diffusion in the simulations and therefore not physical . our resolution study allows has to examine the effect of increasing resolution on the soft x - ray emission and determine the realism of the above possible emission processes . as in paper i , we infer the x - ray luminosity in the soft ( 0.5 - 2.0 kev ) energy band using broadband cooling fractions obtained from the mappings iiir code @xcite . figure [ fig : xray ] shows the soft x - ray emissivity in models rf128 ( left ) and rf512 ( right ) at 0.3 , 0.5 , and 0.7 myr epochs . the strongest x - ray emitter in both models is the bow shock that immediately forms upstream of the cloud as it interacts with the wind . regions where bow shocks are interacting result in the highest x - ray emissivities . apart from a few marginally bright tails coming off some of the cloudlets , particularly in model rf128 ( see fig . [ fig : xray ] ; bottom left panels ) , we see no evidence that mass - loading of the wind by ablation from the cloud is a significant contributor to the soft x - ray luminosity . we also see very little evidence that the intermediate temperature interface plays a significant role . there are a few bright regions upstream of cloudlets that have been exposed to the wind . however , it is likely that this enhanced emission is related to the x - ray emitting bow shocks that have formed around each cloudlet . the main difference between the simulations shown in figure [ fig : xray ] at high ( right panel ) and low ( left panel ) resolution is in the structure of the main bow shock and the emission from the cloudlets . while we observe some structure in the low resolution simulations , as the resolution is increased we see clear regions where colliding bow shocks lead to a significant increase in the x - ray emissivity . this is more evident at later times when the main cloud has fragmented and there are many x - ray emitting bow shocks upstream of the resulting cloudlets ( fig . [ fig : xray ] ; bottom right panels ) . it might be expected than that the increase in fragmentation at higher resolutions would lead to an increase in the x - ray luminosity as more bow shocks are formed . however , the majority of the cloudlets formed are sheltered from the impacting wind by the main cloud and do not form a bow shock . thus , they are not seen at soft x - ray energies and do not contribute to the x - ray luminosity ( fig . [ fig : xray ] ; right panels ) . indeed , the soft x - ray luminosity hardly varies with resolution . figure [ fig : xray_lum ] shows the soft x - ray luminosity as a function of resolution for our radiative fractal cloud at 0.7 myr . at all resolutions , the luminosity is of the order of @xmath96 . this amount varies only negligibly from low to high resolution . this is a result of the strongest x - ray source being the main bow shock upstream of the original gas cloud , which is present at all resolutions . as discussed above , we see little evidence that the mass - loaded component of the wind , nor the intermediate temperature interface between the hot and cold gas plays a large role in the soft x - ray emission . while high resolution ( @xmath97 ) global simulations are needed in order to confirm this result , we conclude that bow shocks and their interaction are the main source of the soft x - ray emission in starburst - driven winds . the importance of radiative cooling in the formation and survival of the filaments was discussed in [ survival ] . for our proposal of the formation of a filament , via the break - up and acceleration of cool disk gas into the wind , to be a viable mechanism , cooling must be present in the outflow . observations of the ovi emission line could be used to detect this cooling in the filaments . @xcite report the detection of o vi emission in the dwarf starburst galaxy ngc 1705 , which they associate with cooling in the outflow from gas at temperatures of @xmath98 . they propose turbulent mixing layers ( e.g. * ? ? ? * ) as a possible origin for this emission . in order to determine were the ovi emission may arise in our simulations , we have produced a `` map '' of the predicted ovi emission in models rf384 , rs384 , af384 , and as384 at 0.7 myr ( fig . [ fig : ovi ] ) . we assume that the ovi emission in our simulations falls within the temperature range of @xmath99 . when cooling is neglected ( bottom two panels ) , ovi emission is only observed around the surviving cloud core . this emission is the result of the high degree of mixing of the hot and cold gas seen in the adiabatic simulations . in the radiative models ( top two panels ) , ovi emission is observed throughout the flow and is closely aligned to the filamentary gas . this emission is caused by the mixing of hot and cold gas in the vicinity of each cool cloudlet that comprises the filament . the distribution of the ovi emission is most significant in the radiative fractal model ( rf384 ; top panel ) , where the structure of the filament can still be clearly seen . the emission weighted histogram of the z - velocity for this model at 0.7 myr is shown in the right hand panel of figure [ fig : vel ] . the velocity dispersion of the ovi gas is similar to that of the h@xmath0 emitting filaments , falling in the range of @xmath100 . while our investigation of the ovi emission is only preliminary , further detection of ovi kinematics similar to those proposed in this work would lend support the premise of cooling in the filaments of starburst winds . as discussed in [ method ] , the simulations performed in this study and paper i do not include thermal conduction or magnetic fields . while an investigation into the role these phenomena play in the wind / cloud interaction is beyond the scope of this work , they may influence the evolution and survival of the filaments in starburst winds . as such , we briefly discuss any differences their inclusion may have had on our results . the effects of thermal conduction on the wind / cloud interaction has been investigated over the last decade by a number of authors ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the immediate effects on the evolution of a radiative cloud has been described in detail by @xcite who modeled the interaction of a cloud of radius @xmath101 and temperature @xmath76 with a hot wind of temperature @xmath102 . here we summarize the evolution of a radiative cloud in their two - dimensional simulation that includes the effects of thermal conduction : 1 . in the early phases of the cloud s evolution , thermal conduction results in a converging shock forming around the cloud . the cloud shrinks in size until the central pressure in the cloud s core is high enough to halt the collapse . 2 . the cloud begins to re - expand and over time becomes elongated in the downstream direction . as the cloud evolves , it is compressed in a non - uniform manner due to the shape of the cloud and the complexity of the flow . 4 . after 1 myr , the cloud forms a filamentary structure along the symmetry axis of the simulation . the key difference to the evolution of a cloud in a simulation that neglects thermal conduction is that the cloud does not suffer hydrodynamical instabilities and fragment . this is because thermal conduction smooths the density and velocity gradients across the surface of the cloud , inhibiting the growth of instabilities ( also see * ? ? ? however , @xcite note that there is also significant mass loss from the cloud through evaporation in their thermal conduction models . @xcite also considered the effects of thermal conduction on a cloud s evolution by investigating the survival of a cloud of radius @xmath74 being overrun by mach 30 and 50 shocks . they performed both three - dimensional simulations that ignored radiative cooling and thermal conduction and two - dimensional simulations where they were included . similar to @xcite , they found that thermal conduction inhibited the growth of hydrodynamical instabilities . in their simulations , the structure of a cloud consists of a cool dense core surrounded by a `` corona '' of gas , dominated by thermal conduction , that gradually evaporates as the cloud evolves . the cloud does not fragment and there is significant mass loss through evaporation . in the case of the slower mach 30 shock , the dense core is unaffected by heat conduction and still fragments into multiple cloudlets . they also investigated the x - ray emission that would arise from their simulations @xcite . they found that for the slower shock , the soft x - ray emission arose in the thermally conducting corona , while the cold core would likely emit at optical wavelengths . in the case of the faster mach 50 shock , they suggest that there would be no optical component . in this paper , we find that the soft x - ray emission arises primarily from the main bow shock upstream of the original cloud . the optically emitting filamentary gas originates from the cool cloudlets that are the remnants of this cloud . as the simulations of @xcite are the closest in initial conditions to our study , we use them as a basis for our supposition on the effect of thermal conduction on our results . most significantly , thermal conduction would likely act to inhibit the growth of the kelvin - helmholtz instability , which is responsible for the high degree of cloud fragmentation seen in our simulations . while the initial evolution of the cloud would be similar to that discussed in [ evolution ] , the break - up of the cloud would likely be totally or partially suppressed . the initial structure of the cloud ( e.g. fractal or spherical ) would play a role in determining the degree of any fragmentation that occurs . a more cohesive h@xmath0 emitting filamentary structure would be formed . in their study , @xcite reported the bow shock upstream of the initial cloud to be the main source of the soft x - ray emission in their non thermal conducting simulations . when thermal conduction was considered they found an increase in the soft x - ray emission in the region between the cloud surface and the bow shock . it is likely that our simulations would show a similar result . note that this emission is still complimentary to the filamentary h@xmath0 emission , as seen by observations of starburst winds . we conclude that thermal conduction may help the survival time of a radiative cloud overrun by a supersonic wind by stabilizing it against the destructive effect of the kelvin - helmholtz instability . while there will be less mixing of cloud material into the wind by hydrodynamical effects , there will still be significant mass loss from the cloud via evaporation . however , as discussed below , the presence of a magnetic field may counteract any benefits to cloud stability gained through thermal conduction . magnetic fields have been shown by a number of authors to have a varying effect on the wind / cloud interaction . a variety of different circumstances have been taken into consideration , such as the strength and orientation of the magnetic field . early models were largely two - dimensional and ignored the effects of radiative cooling ( e.g. * ? ? ? * ) , while later models have been three - dimensional ( e.g. * ? ? ? * ; * ? ? ? * ) , and/or included radiative cooling and thermal conduction ( e.g. * ? ? ? * ; * ? ? ? these studies have shown that the orientation of the magnetic field ( i.e. parallel , perpendicular , or oblique to the flow ) has a significant effect on the evolution of the cloud . for example , @xcite found that in the case of a strong oblique magnetic field , the cloud was actually pushed out of the central y=0 plane as it evolved . it has also been suggested that magnetic fields can help suppress the hydrodynamical instabilities that shred the cloud in a non - magnetized simulation @xcite , with less fragmentation occurring with increasing field strength . on the other hand , @xcite show that in a three - dimensional model , the growth of hydrodynamical instabilities can actually be accelerated by the presence of a magnetic field . it is widely thought that the presence of a magnetic field will act to suppress thermal conduction ( e.g. * ? ? ? * ; * ? ? ? @xcite suggest that the coefficient of thermal conductivity @xmath103 may be significantly reduced below the spitzer level in the presence of a magnetic field . to investigate how a reduced degree of conductivity may effect their results , they performed a number of simulations where @xmath103 was below the standard spitzer level @xmath104 . they found that at sub - spitzer levels , thermal conduction is not as efficient at suppressing the hydrodynamical instabilities that fragment the cloud . more recently , @xcite ran a series of two - dimensional axisymmetric simulations that included radiative cooling , thermal conduction , and magnetic fields . they showed that while thermal conduction was not completely suppressed in the presence of a magnetic field , regardless of the orientation of the field , the cloud would be broken up and destroyed by hydrodynamical instabilities . however , they also found that this effect was lessened in the case of a strong magnetic field . while clearly the interplay between thermal conduction and magnetic fields will influence the evolution and survival of the filaments in starburst winds , without detailed three - dimensional simulations ( with initial conditions appropriate to the problem ) that consider radiative cooling , thermal conduction and the complex nature of magnetic fields , it is difficult to hypothesize what effect the magnetic field would have on a filament s formation and survival . nevertheless , a filamentary structure is still likely to be formed by the wind / cloud interaction , with the most significant difference being in the number of cloudlets that comprise the filament . we do not anticipate that there would be a large effect to the h@xmath0 , soft x - ray , and ovi emission processes discussed in [ sb_winds ] . we have performed a series of three - dimensional simulations of the interaction of a supersonic wind with a radiative cloud . we consider two different cloud geometries ( i.e. fractal and spherical ) , which enable us to investigate the impact of the initial shape and structure of the cloud on its subsequent evolution . this work was motivated by the simulations of the formation of a starburst - driven galactic wind in a inhomogeneous interstellar medium , reported in paper i. the aim of this work is to investigate the possible survival mechanism of a cloud accelerated by a hot freely expanding wind . we also set out to determine the effect of the numerical resolution of the evolution of the cloud and the implied soft x - ray emission associated with the interaction . the results of this study are as follows : 1 . both the initial geometry and the density distribution of the cloud significantly affect its evolution . a cloud which has a more inhomogeneous distribution of density fragments more than a cloud with a more uniform structure ( e.g. a sphere ) . the wind more rapidly breaks the cloud apart in regions where it encounters the least density . 2 . a radiative cloud survives longer than an identical adiabatic cloud . this is a result of the lower degree of heating in the radiative cloud , which suppresses the transverse expansion seen in the adiabatic case . the radiative cloud experiences a lower degree of acceleration and has a higher relative mach number to the flow , diminishing the destructive effect of the kelvin - helmholtz instability . 3 . the number of fragments formed by the break - up of the cloud increases as a power law with increasing numerical resolution . this is a direct result of further resolving the kelvin - helmholtz instability , which grows more quickly at shorter wavelengths . the number of fragments formed increases down to the resolution of the simulation and will not converge . 4 . the calculated mass flux increases with numerical resolution . this is due to the turbulent nature of the stream and the increasing fragmentation of the cloud . high ( @xmath105 ) resolution and an adaptive mesh would be required for convergence to possibly occur . 5 . a radiative cloud fragments into numerous cool , small dense cloudlets . these cloudlets are entrained into the turbulent flow , forming an overall filamentary structure , with regions where the concentration of cloudlets is higher . the velocity of the cloudlets at 0.7 myr falls in the range of @xmath90 irrespective of resolution . the filamentary structure that is formed and the range of velocities found are in good agreement with optical observations of starburst - driven winds . thus , we confirm our conclusion from paper i , that h@xmath0 emitting filaments can be formed from clouds accelerated into a supersonic wind by the ram - pressure of the wind . there is little variation in the estimated soft x - ray luminosity of the radiative fractal cloud at all numerical resolutions considered , indicating that the x - ray emission is well resolved . soft x - ray emission arises primarily from the main bow shock , produced in the initial interaction , and from bow shocks produced upstream of fragments that are directly exposed to the wind . regions where these bow shock interact are strong x - ray emitters . we see little evidence that the mixing of hot and cold gas ( e.g. mass - loading and the boundary between the cool cloud material and the hot wind ) , contribute significantly to the x - ray emission . 9 . the ovi emission arises is the same vicinity as the h@xmath0 emission and has comparable emission weighted velocities , suggesting that the detection of ovi in an outflow may be indicative of cooling in the filaments . the ability for a cloud to radiate heat is crucial for it to survive immersed inside a hot , turbulent , supersonic wind . while effects such as thermal conduction and magnetic fields will have an effect on the clouds survival , without radiative cooling the cloud is quickly destroyed by the kelvin - helmholtz instability , with the cloud s material completely mixed into the surrounding stream . thus , for a model of the wind / cloud interaction to be realistic , radiative cooling certainly can not be neglected . through the results of this work and paper i , we have shown that an optically emitting filament is easily formed via the interaction of a cool , dense cloud and a hot , tenuous supersonic wind under the conditions typically found in a starburst wind . we also find soft x - ray emission that has a natural spatial relationship to the filamentary gas . a relationship also seen in chandra observations of these winds . clearly , the multiphase nature of the interstellar medium is crucial for the formation of the filaments in starburst winds and can help explain much of the optical and soft x - ray emission detected in these complex objects .

here we study properties of a homogeneous dipolar bose - einstein condensate in a weak anisotropic random potential with lorentzian correlation at zero temperature . to this end we solve perturbatively the gross - pitaevskii equation to second order in the random potential strength and obtain analytic results for the disorder ensemble averages of both the condensate and the superfluid depletion , the equation of state , and the sound velocity . for a pure contact interaction and a vanishing correlation length , we reproduce the seminal results of huang and meng , which were originally derived within a bogoliubov theory around a disorder - averaged background field . for dipolar interaction and isotropic lorentzian - correlated disorder , we obtain results which are qualitatively similar to the case of an isotropic gaussian - correlated disorder . in the case of an anisotropic disorder , the physical observables show characteristic anisotropies which arise from the formation of fragmented dipolar condensates in the local minima of the disorder potential .

1212.4807

since the realization of bose - einstein condensates ( becs ) in 1995 @xcite , there was a significant interest about effects of the disordered potentials on the properties of ultracold quantum gases @xcite . the reason for this is not only because of the unavoidable irregularities in the trapping potential induced by wire imperfections @xcite , but also due to the fact that disorder can be generated and controlled using laser speckles @xcite . it is well known that cold atoms are a promising tool for simulating other physical systems @xcite in the sense of feynman s quantum simulator @xcite . this applies also to the phenomenon of anderson localization , which was originally used to microscopically describe the absence of diffusion in terms of disorder @xcite . it has a clear bec analogue @xcite , which has been directly observed @xcite . also , localization inside becs due to disorder created by atomic impurities on a lattice was studied theoretically @xcite and recently observed experimentally @xcite . for a theoretical analysis of global dirty boson properties , different methods have been used to describe various limits , ranging from the bogoliubov theory @xcite , numerical approaches @xcite , to the parisi replica method @xcite . it turns out that long - range correlations within both the condensate and the superfluid remain , despite the presence of disorder . however , both quantities are depleted due to the localization of fragmented condensates in the local disorder potential minima . for a strong enough disorder in a homogeneous system , the depletion increases to such an extent that even a critical disorder strength exists , above which a bose - glass phase appears , consisting only of localized mini - condensates @xcite . effects of disorder were also studied for harmonically trapped becs @xcite and becs in optical lattices @xcite , while the temperature behavior of dirty boson properties was examined in refs . @xcite . realization of atomic dipolar becs @xcite with long - range anisotropic interaction has generated large interest in the theory of dipolar quantum gases @xcite . increase in the strength of dipolar interaction is possible by substituting atoms with magnetic dipoles by heteronuclear molecules , which have a strong electric dipolar moment in rovibrational ground state @xcite , or by inducing radiative coupling by placing dipoles into a resonator @xcite . dipolar condensates were studied in the case of isotropic disorder @xcite , which yields characteristic anisotropies for both the superfluid density and the sound velocity at zero temperature due to the anisotropy of the dipolar interaction . although a 3d isotropic laser speckle potential has recently been proposed in ref . @xcite , the typical disorder realized in experiments is cylindrically symmetric and , to the best of our knowledge , it has so far been examined only numerically for contact interaction @xcite . therefore , motivated by the experiments with dipolar becs in anisotropic disorder potentials , we develop in this paper a mean - field theory and analytically study the impact of a weak anisotropic disorder on physical properties of a polarized dipolar bec at zero temperature . to this end we proceed as follows . following the approach developed in ref . @xcite , in sec . [ sec : mf ] we calculate the lowest - order corrections of bec properties due to the presence of disorder within a mean - field theory . for the sake of generality we consider an arbitrary two - particle interaction and a general disorder correlation function . in sec . [ sec : diplor ] we specialize the developed formalism to dipolar interaction and a lorentzian - correlated disorder in fourier space . this yields for both the superfluid density and the sound velocity characteristic anisotropies , which should be measurable in an experiment . in sec . [ sec : con ] we present our conclusions and outlook for further related research . finally , the appendix gives analytical results for the condensate depletion and the disorder correction to the chemical potential , while the supplemental material @xcite gives complete analytical results for the disorder correction of both the superfluid density and the sound velocity for the general case of a bec with contact and dipole - dipole interaction in an anisotropic , lorentzian - correlated disorder . bogoliubov quasiparticles and disorder induced fluctuations decouple in the lowest order @xcite , suggesting that disorder corrections can be calculated at zero temperature by neglecting quantum fluctuations and using a mean - field macroscopic wave function @xmath0 governed by the time - independent gross - pitaevskii ( gp ) equation : @xmath1 here @xmath2 stands for the particle mass , @xmath3 denotes the chemical potential , @xmath4 represents an arbitrary two - body potential , while @xmath5 describes an external disorder potential . denoting the disorder ensemble average as @xmath6 , a homogeneous disordered system can be described , without any loss of generality , by a vanishing mean value @xmath7 and an arbitrary correlation function @xmath8 in this section we present in detail a perturbative theory , which was developed earlier in ref . @xcite , and calculate disorder corrections to the order parameter , the condensate depletion , the chemical potential , the superfluid depletion , and the sound velocity . the range of validity of this perturbation theory is limited by standard requirements for a mean - field approach : dilute , weakly - interacting bec at low temperatures , when quantum fluctuations can be neglected . the perturbation expansion is performed with respect to the disorder strength , thus the disorder potential has to be sufficiently small compared to the chemical potential , i.e. @xmath9 . we start with the observation that the gp eq . ( [ gp ] ) represents a stochastic nonlinear partial differential equation , where the statistics of the condensate wave function @xmath0 is governed by the statistics of the disorder potential @xmath5 @xcite . since @xmath10 describes the macroscopic occupation of the ground state , we assume it to be real without loss of generality . in addition to the statistical properties of the random potential we will also assume that the macroscopic value of some physical quantity @xmath11 , obtained by coarse - graining of a microscopic quantity @xmath12 over a large volume @xmath13 , gives the same result as the disorder ensemble average , namely : @xmath14 here the length of the coarse - graining @xmath15 is assumed to be larger than both the correlation length @xmath16 of the disorder potential @xmath5 , and the healing length @xmath17 , which represents the characteristic distance at which the condensate wave - function responds to some perturbation in the external potential : @xmath18 in the definition of the healing length @xmath19 represents the density of the fluid and @xmath20 denotes the strength of the short - range interaction part of the two - particle interaction potential @xmath21 , expressed in terms of the s - wave scattering length @xmath22 . we consider the case of a sufficiently small random potential @xmath23 , when the perturbative decomposition of the wave function of the system is justified : @xmath24 where @xmath25 corresponds to the correction of the wave function of order @xmath26 in the disorder . solving the gp equation ( [ gp ] ) in the zeroth order of @xmath5 gives @xmath27 whereas the first order correction is straight - forwardly calculated and its fourier transform reads @xmath28 we note that its disorder ensemble average vanishes . therefore , we also have to determine the second - order result , which turns out to be @xmath29\psi_1(\mathbf{k}')\psi_1(\mathbf{k}-\mathbf{k } ' ) } { \frac{\hbar^2 k^2}{2m}+2\psi_0 ^ 2v(\mathbf{k})}\ , .\ ] ] the results obtained in refs . @xcite can be considered as special cases of the above general approach . for instance , we point out that sec . ii of ref . @xcite contains a discrete version of eqs . ( [ expansion])([eq : pert1 ] ) for the case of pure contact interaction , which is here generalized to an arbitrary two - body interaction . note that the second - order correction ( [ eq : pert2 ] ) is slightly different since we take the chemical potential to be constant , whereas in ref . @xcite the density is taken to be constant . in the following subsections we use the above outlined systematic perturbative approach @xcite and calculate several physical properties of the dirty bec and their respective disorder corrections . in analogy to quantum field theory , the one - particle density matrix is defined as @xmath30 @xcite . the macroscopic fluid density is the diagonal part of the one - particle density matrix according to @xmath31 , whereas the condensate density is usually defined as the off - diagonal long - range order ( odlro ) parameter via @xcite @xmath32 performing the coarse - graining of the one - particle density matrix @xmath30 over the fixed volume @xmath13 before taking the limit does not change the result @xmath33 the integration commutes with the disorder ensemble average and using eq . ( [ cg ] ) , we obtain @xmath34 the last equality follows from the fact that the average of an already averaged expression can be omitted . therefore , the depletion of the condensate due to disorder , which is defined as @xmath35 , is simply identified with the variance of the wavefunction . physically , this condensate depletion is due to the formation of fragmented condensates in the respective local minima of the random potential . defining a separate bose - glass order parameter by considering the odlro parameter of the two - particle density matrix @xcite @xmath36 shows that the density of the fragmented condensates @xmath37 defined in eq . ( [ q ] ) coincides with the condensate depletion @xmath38 . to this end the disorder ensemble average is obtained along the same lines as eqs . ( [ eq : n0def])([n0 ] ) . thus , we conclude that the localization phenomenon for quenched disorder follows already from a mean - field description of the dirty boson problem . therefore , our mean - field approach represents a simplified derivation of the disorder - induced condensate depletion in comparison with the bogoliubov theory of refs . note that disorder effects on bogoliubov quasiparticles have recently been analyzed in refs . @xcite . the perturbative expansion ( [ expansion ] ) now yields for the particle density @xmath39 and , correspondingly , for the condensate density @xmath40 with this the condensate depletion results to be @xmath41 using eq . ( [ eq : pert1 ] ) we arrive at the following expression : @xmath42 ^ 2}+\ldots\,.\end{aligned}\ ] ] note that this represents a result for the condensate depletion in second order of the disorder potential for an arbitrary two - particle interaction potential and an arbitrary disorder correlation function . specializing to the delta - correlated disorder @xmath43 and the contact interaction @xmath44 , eq . ( [ eq : deplgen ] ) reduces to @xmath45 which is the seminal result originally obtained by huang and meng @xcite within the bogoliubov theory of dirty bosons . solving the equation @xmath46 for the chemical potential @xmath47 yields its dependence on the average fluid density @xmath48 . we have introduced the notation @xmath47 , denoting the `` bare '' chemical potential , because it diverges for uncorrelated disorder regardless of the density @xmath19 , as can be seen from inserting expressions ( [ eq : pert0])([eq : pert2 ] ) into the second - order correction ( [ eq : dens ] ) : @xmath49 ^ 2}+\ldots\,.\ ] ] this unphysical ultraviolet divergence can be removed by renormalizing the chemical potential @xcite . if the density of the system vanishes , i.e. if there are no particles in the system , the energy needed for a particle to be added also has to vanish @xmath50 . therefore , we define the renormalized chemical potential according to @xmath51 with this we obtain in second order of the disorder strength the renormalized chemical potential : @xmath52 ^ 2}+\ldots\ , , \nonumber\end{aligned}\ ] ] which does not contain an ultraviolet divergence . for calculating the sound velocity later on we will also need the expression for the compressibility of the fluid , or its inverse given by @xmath53 . note that from eq . ( [ ren ] ) it follows that the obtained result does not depend on whether we use @xmath3 or @xmath47 . thus , from the perturbative expansion ( [ eq : eqstgen ] ) we read off : @xmath54 ^ 3}+\ldots\,.\ ] ] without disorder and at @xmath55 , the whole system is in a superfluid state , moving with an arbitrary wavevector @xmath56 , which corresponds to the superfluid velocity @xmath57 . by introducing disorder that moves with the velocity @xmath58 , some part of the fluid will be moving together with it . the normal , i.e. non - superfluid , component of the fluid @xmath59 is defined as the part that moves together with the disorder , while the superfluid component @xmath60 is defined as the fraction of the fluid that moves with the superfluid wavevector @xmath56 . therefore , the macroscopic current density @xmath61 can be separated in this two - fluid picture as follows : @xmath62 the averaged current density @xmath61 can be obtained by analyzing the underlying time - dependent gp equation for the system : @xmath63\psi_s(\mathbf{r},t ) = i\hbar\frac{\partial\psi_s(\mathbf{r},t)}{\partial t}\ , , \ ] ] where the condensate wave function @xmath64 is a product of some as yet unknown function @xmath65 and a plane wave with wavevector @xmath56 that corresponds to the clean - case solution : @xmath66 substituting the ansatz ( [ supsol ] ) into eq . ( [ supgp ] ) , changing variables via @xmath67 and introducing @xmath68 leads to @xmath69 \psi_s(\mathbf{x})=0\ , .\ ] ] although @xmath65 should in general depend on @xmath70 , it can be shown via mathematical induction on the perturbative solution that all orders of @xmath71 turn out to be time - independent @xcite . note that @xmath65 does not depend explicitly on the wavevectors @xmath56 and @xmath72 , but only on their difference @xmath73 . here , we are only interested in small values of @xmath73 and , therefore , perform the expansion @xmath74 , with @xmath75 . an explicit equation for @xmath76 can be obtained by performing the derivative of eq . ( [ gpx ] ) with respect to @xmath73 , yielding @xmath77\mathbf{p}(\mathbf{x } ) + \int d^3 x'\ , v(\mathbf{x}-\mathbf{x'})\big\{[\mathbf{p}^*(\mathbf{x'})+\mathbf{p}(\mathbf{x'})]\psi(\mathbf{x'})\psi(\mathbf{x})+\psi(\mathbf{x'})^2\mathbf{p}(\mathbf{x } ) \big\}=0\ , . \label{eqp}\ ] ] if we take into account eq . ( [ supsol ] ) , the standard definition of the current density , @xmath78 transforms into @xmath79 which then can be further reduced to @xmath80 in the last line we have neglected higher than linear orders in @xmath72 and @xmath56 . for small disorder strengths , we expand eq . ( [ jlexp ] ) with respect to @xmath81 up to second order . to this end we take into account the homogeneity of our problem , that leads to @xmath82 , and note that in zeroth order @xmath65 does not depend on @xmath73 , thus leading to @xmath83 . with this we obtain @xmath84 where also @xmath76 is expanded in the disorder strength according to @xmath85 . solving the imaginary part of eq . ( [ eqp ] ) in first order in @xmath81 yields the fourier transform of @xmath86 : @xmath87 thus , together with the solution for @xmath88 given by eq . ( [ eq : pert1 ] ) and a comparison with eq . ( [ curr ] ) , we obtain from eq . ( [ jlexpu ] ) the normal fluid density in the form @xmath89 ^ 2}+\ldots\,.\label{eq : sdgen}\ ] ] note that in general the non - superfluid component is represented by a tensor @xcite . in the case of a cylindrically symmetric system , we can choose the symmetry axis as the @xmath90-axis and denote the polar and the azimuth angle by @xmath91 and @xmath92 , so integrating eq . ( [ eq : sdgen ] ) in spherical coordinates with respect to @xmath92 yields the angle dependence @xmath93 if both @xmath94 and @xmath95 are @xmath91-independent , i.e. if we have spherical symmetry , integrating eq . ( [ eq : cyl ] ) with respect to @xmath91 leads to a solution in second order of the disorder potential @xcite : @xmath96 this result shows that the superfluid depletion will be larger by a factor of @xmath97 than the condensate depletion . thus , the localized fragmented part of the fluid hinders the superfluid to move . in the mean - field approach , we can also define the sound velocity by perturbing the time - independent solution with a small time - dependent variation . it is expected that sound waves with wavelengths of the order of the correlation length would scatter and interfere due to disorder hills and valleys , making the sound velocity impossible to define precisely . locally , the sound waves would have the same speed as in the clean case . for sound waves with wavelengths much larger than the disorder correlation length , the sound velocity can be calculated using the hydrodynamical approach @xcite . hydrodynamic equations are valid in the macroscopic regime and can only be used for slowly varying quantities that do not depend on the specific microscopic realization . spatial averaging over distances much larger than the correlation length and much smaller than the wavelength solves the problem . assuming that it gives the same result as the disorder ensemble average , we obtain the hydrodynamic equations for the macroscopic , i.e. disorder averaged , quantities in the form @xmath98&=&0\,,\\ \hspace*{-5mm}m\frac{\partial \mathbf{v}_s(\mathbf{x},t)}{\partial t}+\nabla \left[\frac{m\mathbf{v}_s(\mathbf{x},t)^2}{2}+\mu(n(\mathbf{x},t))\right]&=&0\,,\end{aligned}\ ] ] where @xmath19 denotes the macroscopic density , and the disorder velocity @xmath72 is taken to be zero . if we write densities and the superfluid velocity as sums of homogeneous equilibrium values and small variations , @xmath99 as well as neglect second - order terms in the variations , we get the following linearized system of equations : @xmath100&=&0\,,\label{eq : hid1}\\ \frac{\partial \delta\mathbf{v}_s(\mathbf{x},t)}{\partial t}=-\frac{1}{m}\nabla\mu \left[n+\delta n(\mathbf{x},t)\right]&=&\nonumber\\ & & \hspace*{-40mm}-\frac{1}{m}\frac{\partial\mu}{\partial n}\nabla\delta n(\mathbf{x},t)\ , .\label{eq : hid2}\end{aligned}\ ] ] taking the time derivative of eq . ( [ eq : hid1 ] ) and substituting the expression for the superfluid velocity variation from eq . ( [ eq : hid2 ] ) we obtain the generalized wave equation @xmath101=0.\ ] ] from the above equation we deduce that the sound velocity in the direction of some unit vector @xmath102 is given by @xmath103 where the tensorial property of the superfluid density has been taken into account . in order to further evaluate the sound velocity ( [ eq : cgen ] ) for small disorder , the perturbative results for both the inverse compressibility ( [ eq : dmdngen ] ) and the superfluid density following from ( [ eq : sdgen ] ) have to be taken into account . in this section we will specialize the previously developed perturbative formalism and consider bec systems in the presence of two different anisotropies , namely an anisotropic dipolar interaction between the analyzed particles and an anisotropic disorder potential . the latter is widely studied and physically motivated , for instance , by the anisotropy of the laser - speckle potential @xcite . in order to obtain analytical results , we model the disorder correlation function by a cylindrically - symmetric lorentzian in fourier space @xmath104 the lengths @xmath105 and @xmath106 denote the perpendicular and the parallel correlation length , respectively , and their experimentally realistic values are typically in a broad range from a few to several hundreds healing lengths @xmath107 . the function ( [ eq : corf ] ) is not physically realistic , but the corresponding results qualitatively coincide with the case of an isotropic gaussian - correlated disorder , which was numerically calculated in ref . therefore , we expect that all phenomena , that appear here , would also appear qualitatively for a true laser - speckle correlation function in a setup where it decays monotonously with distance . assuming that the van der waals forces between the atoms can be approximated at low energies by an effective contact interaction , the interaction potential in the presence of an external field , that aligns the dipoles in a direction @xmath108 , takes the form @xcite @xmath109\,,\ ] ] where @xmath110 represents the angle between vectors @xmath108 and @xmath111 , and @xmath112 denotes the dipole - dipole interaction strength . in the case of magnetic dipoles @xmath113 , with @xmath114 being the magnetic permeability and the magnetic dipole moment @xmath2 , whereas for electric dipoles we have @xmath115 , with the vacuum permeability @xmath116 and the electric dipole moment @xmath117 . introducing the ratio of the dipole - dipole and the contact interaction @xmath118 and taking the fourier transform of the potential , we obtain @xcite @xmath119\right\}\,.\ ] ] the interaction ratio @xmath120 takes values as small as @xmath121 for @xmath122rb , while for @xmath123cr it is around @xmath124 . for a bec of heteronuclear molecules , its value would be much higher , namely of the order of @xmath125 . the huang and meng result @xcite for the condensate depletion ( [ eq : hm ] ) is linear in @xmath126 , and therefore we will compare the relative change of physical quantities due to disorder to the relative change of the condensate density . to this end we will define a dimensionless disorder correction for each relevant quantity : condensate density , chemical potential , superfluid density , and sound velocity . corrections defined in this way are expressed in terms of only three parameters : the relative dipole - dipole interaction strength @xmath120 , and the correlation lengths in units of the healing lengths , i.e. @xmath127 . we consider systems with an overall cylindrical symmetry , where the disorder symmetry axis is parallel to the direction of the dipoles . otherwise the angle between them would be a fourth parameter that would have to be taken into account . the general case can be calculated analytically for all observables of interest , but the results are too cumbersome to be displayed here @xcite , and we present them in the appendix as well as in the supplemental material @xcite . in this section we derive analytical results for the two special cases , namely the pure contact interaction with anisotropic disorder , and the dipolar interaction with isotropic disorder . we also present and discuss numerical results for the special cases , as well as for the general case with contact and dipole - dipole interaction as well as anisotropic disorder . we now use the disorder correlation function and the interaction potential defined by eqs . ( [ eq : corf ] ) and ( [ eq : poten ] ) in order to calculate the disorder correction of the condensate density . to this end we introduce the dimensionless condensate depletion as follows : @xmath128 taking into account eq . ( [ eq : hm ] ) , making a substitution @xmath129 , denoting the direction of the cylindrical symmetry of the disorder by @xmath130 and introducing direction - dependent anisotropy functions @xmath131 and @xmath132 by @xmath133}\,,\label{eq : v}\end{aligned}\ ] ] eq . ( [ eq : deplgen ] ) yields the dimensionless value of the condensate depletion in second order in the form @xmath134 assuming that the direction of dipoles is parallel to the disorder symmetry , i.e. @xmath135 , the whole system will also become cylindrically symmetric . writing eq . ( [ eq : deplcor ] ) in spherical coordinates @xmath136 , integrating with respect to @xmath137 and @xmath92 and changing the variable @xmath138 , leads to : @xmath139 with functions @xmath131 and @xmath132 from eqs . ( [ eq : r ] ) and ( [ eq : v ] ) having the new form @xmath140 the two special cases , with pure contact interaction @xmath141 and with isotropic disorder @xmath142 can be solved explicitly using euler substitutions @xmath143 and @xmath144 , respectively , in eq . ( [ eq : deplcyl ] ) , which leads to an integral of a rational function with respect to @xmath145 . the analytic results for the two special cases are : @xmath146\,,\\ & & \hspace*{-5mm}-\delta_{n_0}\big|_{z_\rho = z_z = z}= \frac{z ( 1-\lambda ) } { \left(-1+z^2 \delta ^2\right ) [ 1-\lambda + z \delta ( 1+\lambda ) ] } \nonumber\\ & & \hspace*{5mm}+\frac{(-1+\lambda ) } { \delta ( -1+z \delta ) ( 1+z \delta ) ^2}\ , t\left(\frac{z \delta-1}{z\delta+1 } \lambda \right)\ , , \label{eq : depletion2}\end{aligned}\ ] ] where , for brevity , we introduced @xmath147 and @xmath148 by @xmath149 and @xmath150 , and @xmath151 is a new function , well defined for positive values of @xmath145 and analytically continuable for @xmath152 . in fig . [ fig : n0 ] we have displayed the results for the condensate depletion . figure [ fig : n0](a ) corresponds to the special case of pure contact interaction ( @xmath153 ) , described by eq . ( [ eq : depletion ] ) , and fig . [ fig : n0](c ) corresponds to the special case of isotropic disorder ( @xmath154 ) , given by eq . ( [ eq : depletion2 ] ) . note that for @xmath153 and @xmath155 the dimensionless condensate depletion is @xmath156 , which coincides with the huang and meng result . for increasing correlation lengths the depletion decreases , and vanishes for infinite correlation length , as expected , since then the disorder is flat . the tiny asymmetry in fig . [ fig : n0](a ) comes from the fact that @xmath157 describes two spatial dimensions and , therefore , has a more pronounced effect on the depletion than @xmath158 , which represents only one spatial dimension . as we see in fig . [ fig : n0](c ) , increasing the relative dipole - dipole interaction leads to a larger depletion and , eventually , when it reaches the same order as the contact interaction , the bec collapses , which corresponds to the divergence of the depletion for @xmath159 . the general case with both contact and dipole - dipole interaction as well as anisotropic disorder is shown in figs . [ fig : n0](b ) and [ fig : n0](d ) for @xmath160 and @xmath161 . compared to the isotropic disorder case , the depletion decays faster with increasing the correlation length for small value of @xmath162 . in the opposite case , when the radial correlation length is larger than the longitudinal one , the depletion decays much slower with increasing correlation lengths . this can be explained by the fact that , for a fixed value of @xmath157 , the value of @xmath158 is given by @xmath163 , which effectively corresponds to a larger disorder correlation length in fig . [ fig : n0](b ) and leads to a faster decay , while the effective correlation length in fig . [ fig : n0](d ) is smaller and , hence , the decay is slower . therefore , we conclude that the ratio of correlation lengths @xmath162 has a significant impact on the condensate depletion and , thus , can be used for its control . note that the condensate depletion can be measured in matter interference experiments , where the fragmented part of the fluid contributes with a random phase and , therefore , reduces correspondingly the contrast of the interference pattern . we now proceed with the perturbative calculation of the chemical potential and the inverse compressibility using eqs . ( [ eq : eqstgen ] ) and ( [ eq : dmdngen ] ) . their dimensionless disorder corrections are defined as @xmath164 and can be calculated in the similar way as the condensate depletion . analytical results for the general case are given in the appendix and in the supplemental material @xcite , while disorder corrections to the chemical potential for the special cases of a pure contact interaction ( @xmath153 ) and isotropic disorder ( @xmath165 ) are given by : @xmath166\}}{z \left(-1+z^2 \delta ^2\right ) [ 1-\lambda + z \delta ( 1+\lambda ) ] } + \frac{2 ( -1+\lambda ) } { z^2 \delta}\ , t(-\lambda ) \nonumber\\ & & + \frac{2 ( -1+\lambda ) } { z^2 \delta ( -1+z \delta ) ( 1+z \delta ) ^2}\ , t\left(\frac{z \delta-1}{z\delta+1 } \lambda \right)\,.\label{eq : eqstsol2}\end{aligned}\ ] ] the analytically calculated disorder corrections to the chemical potential are shown in fig . [ fig : mu ] . the two special cases ( [ eq : eqstsol ] ) and ( [ eq : eqstsol2 ] ) correspond to figs . [ fig : mu](a ) and [ fig : mu](c ) , respectively , while figs . [ fig : mu](b ) and [ fig : mu](d ) correspond to the general case with both contact and dipole - dipole interaction as well as anisotropic disorder . the disorder correction increases with increasing disorder strength @xmath126 , regardless of the strength of the dipolar interaction and disorder correlation lengths . this is due to the repulsive interparticle interaction , which has a higher potential energy when the fluid is less uniform . the correction of the chemical potential in the case of the pure contact interaction , shown in fig . [ fig : mu](a ) , has a similar dependence on the correlation lengths as the condensate depletion in fig . [ fig : n0](a ) , while , according to fig . [ fig : mu](c ) , the dipole - dipole interaction does not have a significant effect . this is due to the fact that the dipolar interaction contributes partially as attractive and partially as repulsive , thus leading only to a small net effect . note that the chemical potential in the clean case is anisotropic , as can be seen from eq . ( [ eq : eqstgen ] ) and the directional dependence of the limit @xmath167 in eq . ( [ eq : poten ] ) . this peculiar behaviour is discussed in more detail in ref . @xcite . for the general case of anisotropic disorder , we see from figs . [ fig : mu](b)[fig : mu](d ) that increasing anisotropy @xmath168 leads to a slower decay of the disorder correction with increasing correlation lengths for the same reasons as for the condensate depletion . the corresponding results for disorder corrections of the inverse compressibility are : @xmath169\,,\\ \delta_{\frac{\partial\mu}{\partial n}}\big|_{z_\rho = z_z = z}&= & \frac{(-1+\lambda ) \left\{1-\lambda + z^2 \delta ^2 [ 2 - 2 \lambda + 3 z \delta ( 1+\lambda ) ] \right\}}{z \left(-1+z^2 \delta ^2\right)^2 [ 1-\lambda + z \delta ( 1+\lambda ) ] ^2}\,.\end{aligned}\ ] ] they represent intermediate results for calculating later on the sound velocity in subsection [ sec : sound ] , using eq . ( [ eq : cgen ] ) . now we turn to the calculation of the dimensionless disorder correction to the superfluid density tensor , which is defined by @xmath170 it can be separated into a perpendicular and a parallel component after integration with respect to the azimuthal angle @xmath92 . using eq . ( [ eq : cyl ] ) , we get @xmath171 where @xmath172 is already calculated in eqs . ( [ eq : depletion ] ) and ( [ eq : depletion2 ] ) , and @xmath173 is a new integral of the form @xmath174 the solution of this integral for the general case is given in the supplemental material @xcite , while for the two studied special cases the solutions are : @xmath175\,,\\ & & \hspace*{-10mm}i_{\rm sd}\big|_{z_\rho = z_z = z}=\frac{(-1+\lambda ) ^3}{4 z \delta ^2 \lambda [ 1-\lambda + z \delta ( 1+\lambda ) ] } -\frac{(-1+\lambda ) ^3 } { 4 z^2 \delta ^3 \lambda } \ , t(-\lambda ) + \frac{(-1+\lambda ) ^3 } { 4 z^2 \delta ^3 ( \lambda + z \delta \lambda ) } \ , t\left(\frac{z \delta-1}{z\delta+1 } \lambda \right)\,.\end{aligned}\ ] ] in an arbitrary direction of an unit vector @xmath102 the superfluid density can be calculated by describing the tensorial superfluid density according to @xmath176 . in the case of cylindrical symmetry this reduces to @xmath177 thus , obtaining the disorder corrections for @xmath178 and @xmath179 is sufficient for recovering the superfluid depletion in any direction . from eq . ( [ eq : sdepcyl ] ) we directly read off @xmath180 note that the negative sign in front of @xmath181 and @xmath172 suggests that the changes of the superfluid densities and the condensate density are negative or , equivalently , that the depletion is positive . for isotropic systems @xmath182 and @xmath183 both are equal to @xmath184 , as can be seen from eq . ( [ eq : isot ] ) . due to an anisotropy , however , there is a range of correlation lengths and relative dipolar interaction strengths , where the superfluid depletion is smaller than the condensate depletion . some particles from the fragmented fluid contribute to superfluidity , suggesting that they are not localized indefinitely , but have some finite localization time . this localization time was exemplarily calculated in ref . @xcite within the hartree - fock theory of dirty bosons with delta - correlated disorder . the superfluid depletion in the case of a pure contact interaction and anisotropic disorder shows a similar behavior as the condensate depletion , as can be seen in figs . [ fig : scperp](a ) and [ fig : scz](a ) . in the presence of disorder as well as both contact and dipole - dipole interaction , [ fig : scperp](b)-[fig : scperp](d ) show that the depletion of the perpendicular component is similar to the condensate depletion , but the depletion of the parallel component decreases as the relative interaction strength increases , as is depicted in figs . [ fig : scz](b)[fig : scz](d ) . the red lines in figs . [ fig : scperp ] and [ fig : scz ] show where the corresponding superfluid depletion component becomes equal to the condensate depletion . this is illustrated in more detail in fig . [ fig : sccrit ] , which presents the values of the interaction ratio @xmath120 for which the superfluid depletions become equal to the condensate depletion . although defined only for systems without a trap , the above calculated superfluid density could be extended to the trapped case in the simplest way by assuming that it depends only on the local density . if we turn on a slowly moving disorder for a short time @xmath185 , such that @xmath186 is much smaller than the size of the trap , before switching off the trap , this would change the momentum distribution which could be afterwards reconstructed from a time - of - flight measurement . in this way our predictions for the superfluid density in such a system might become observable in experiment . the corresponding disorder correction of the sound velocity in the direction of the unit vector @xmath102 can be calculated using eq . ( [ eq : cgen ] ) : @xmath187\,,\nonumber\end{aligned}\ ] ] if we define the dimensionless disorder correction as @xmath188 from the previous equation we get @xmath189 where @xmath190 denotes the directional dependence of the potential @xmath13 on @xmath102 , according to eq . ( [ eq : poten ] ) . the anisotropy of disorder comes into play in a simple way . from eqs . ( [ eq : cgen ] ) and ( [ eq : sup_separation ] ) it follows that the sound velocity can also be separated into a parallel and a perpendicular component @xmath191 with @xmath192 and the corresponding dimensionless disorder correction @xmath193 figure [ fig : soundcont ] shows disorder corrections to the perpendicular and the parallel component of the sound velocity for a pure contact interaction . the red lines correspond to the values of correlation lengths for which the disorder corrections vanish . for the general case , the anisotropy factor due to the dipolar interaction @xmath194 @xcite is plotted in fig . [ fig : sound](a ) . by introducing weak disorder , the sound velocity changes via two competing effects : the decrease of the compressibility , i.e. the increase in the inverse compressibility , from eq . ( [ eq : dmdngen ] ) , which tends to increase the sound velocity , and the decrease of the superfluid density corresponding to a negative value of @xmath181 , which tends to decrease the sound velocity . the corresponding results are shown in fig . [ fig : sound](b)[fig : sound](d ) . for small disorder correlation lengths the decrease in compressibility is dominant . these corrections can be experimentally measured , for instance , by determining the phonon dispersion relation by using bragg spectroscopy @xcite . we have analyzed in detail how the anisotropy of both the dipolar interaction and the presence of disorder affects the directional dependence of different physical observables of dirty bose - einstein condensates . using the mean - field approach at zero temperature , we have calculated the condensate depletion due to disorder , as well as the corresponding corrections to the equation of state , the superfluid density , and the sound velocity . in particular , we have discussed the consequences for the superfluid density , which becomes a tensorial quantity as a linear response to the moving disorder . whereas ref . @xcite analyzed a dipolar bec in isotropic disorder potential , we have shown here that the anisotropic disorder provides a separate origin for the tensorial nature of the superfluid density . we have found that a large enough disorder anisotropy can even make both the parallel and perpendicular superfluid density component larger than the corresponding condensate density , which happens in the case of dipolar interaction and isotropic disorder only for the parallel component @xcite . these initial results necessitate further studies , as they contribute to the overall physical picture in which the localization of bosons in the respective minima of the disorder potential occurs at a characteristic time scale @xcite . this localization time remains to be analyzed in more detail in a forthcoming publication . we also plan to study effects of disorder on nonlinear oscillation modes and faraday waves in bec @xcite . we would like to thank bakhodir abdullaev , christian krumnow , aristeu r. p. lima , and vladimir lukovi ' c for fruitful discussions . this work was supported in part by the ministry of education , science , and technological development of the republic of serbia under projects no . on171017 and nai - dbec , by daad - german academic and exchange service under project nai - dbec , and by the european commission under eu fp7 projects prace-2ip , prace-3ip , hp - see , and egi - inspire . both a. b. and a. p. gratefully acknowledge a fellowship from the hanse - wissenschaftskolleg . this appendix contains expressions for the dimensionless condensate depletion @xmath172 due to disorder and the dimensionless disorder correction to the chemical potential @xmath195 in the general case , when both contact and dipole - dipole interaction are present and characterized by the ratio @xmath196 . we assume that disorder is cylindrically - symmetric and lorentzian - correlated , characterized by the dimensionless correlation lengths @xmath157 and @xmath158 . these expressions , as well as the expressions for the integral @xmath197 and the dimensionless disorder correction to the inverse compressibility @xmath198 , which define the dimensionless disorder corrections to the superfluid density and the sound velocity , are given in the supplemental material @xcite , as a _ notebook . the expressions for disorder corrections are given in terms of auxiliary functions : @xmath199\ , , \end{aligned}\ ] ] and their values @xmath200 , @xmath201 , where arguments @xmath202 , @xmath203 ( @xmath204 ) are given by : @xmath205}{2 z_z^2 \left[1+(-1+\epsilon ) z_{\rho } ^2\right]}\ , , \quad x_2=\frac{-6 \epsilon -(-1+\epsilon ) ^2 z_z^2+\left(1+\epsilon -2 \epsilon ^2\right ) z_{\rho } ^2}{2 ( 1 + 2 \epsilon ) \left[1+(-1+\epsilon ) z_{\rho } ^2\right]}\ , , \\ & & y_1=\frac{z_z^2 \left[12 \epsilon + ( -1+\epsilon ) ^2 z_z^2\right ] z_{\rho } ^4 + 2 \left[-6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_z^2\right ] z_{\rho } ^6+(1 + 2 \epsilon ) ^2 z_{\rho } ^8}{4 z_z^4 \left[1+(-1+\epsilon ) z_{\rho } ^2\right]{}^2}\ , , \nonumber \\ & & y_2=\frac{(-1+\epsilon ) ^2 \left(\sqrt{1-\epsilon } -\sqrt{1 + 2 \epsilon } \right)^4 \left\{(-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2+(1 + 2 \epsilon ) ^2 z_{\rho } ^4 + 2 z_z^2 \left[6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right]\right\}}{4 \left(-1 - 2 \epsilon + \sqrt{1+\epsilon -2 \epsilon ^2}\right)^4 \left[1+(-1+\epsilon ) z_{\rho } ^2\right]{}^2}\ , .\nonumber\end{aligned}\ ] ] the dimensionless condensate depletion @xmath172 due to disorder , defined by eq . ( [ eq : deltan0def ] ) , is given by : @xmath206 z_{\rho } ^2 + 2 \left[9 \epsilon + ( 1 + 4\epsilon -5 \epsilon^2 ) z_z^2\right ] z_{\rho } ^4+\left[-1 + 2 ( -5+\epsilon ) \epsilon \right ] z_{\rho } ^6\right\}}{2 z_z \left[1+(-1+\epsilon ) z_{\rho } ^2\right]{}^2 \left\{(-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2+(1 + 2 \epsilon ) ^2 z_{\rho } ^4 + 2 z_z^2 \left[6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right]\right\}}\\ & & \hspace*{8 mm } + a_2\ , \big [ \left(\sqrt{1-\epsilon } -\sqrt{1 + 2 \epsilon } \right ) \left(\left(-12 \epsilon + 3 ( -1+\epsilon ) \epsilon z_z^2-(-1+\epsilon ) ^3 z_z^4\right ) z_{\rho } ^2+\left(3 \epsilon ( 1 + 2 \epsilon ) -\left(2 - 3 \epsilon + \epsilon ^3\right ) z_z^2\right ) z_{\rho } ^4 \right . \nonumber\\ & & \hspace*{12 mm } \left . + ( -1+\epsilon ) ^2 ( 1 + 2 \epsilon ) z_{\rho } ^6 + 12 \epsilon z_z^2\right)\big]\big/ \big[2 \left(-1 - 2 \epsilon + \sqrt{1+\epsilon -2 \epsilon ^2}\right ) \left(1+(-1+\epsilon ) z_{\rho } ^2\right){}^2 \left((-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2\right.\nonumber\\ & & \hspace*{12 mm } \left . + ( 1 + 2 \epsilon ) ^2 z_{\rho } ^4 + 2 z_z^2 \left(6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right)\big]\nonumber\\ & & \hspace*{8 mm } + b_1\ , \big[z_z \left(z_z^2 \left(-48 \epsilon + 24 ( -1+\epsilon ) \epsilon z_z^2-(-1+\epsilon ) ^3 z_z^4\right ) z_{\rho } ^2 + 3 \left(8 \epsilon + 8 \epsilon ( 2+\epsilon ) z_z^2+(-1+\epsilon ) ^2 ( -1 + 4 \epsilon ) z_z^4\right ) z_{\rho } ^4\right.\nonumber\\ & & \hspace*{12 mm } \left . + 3 \left(-8 \epsilon ( 1 + 2 \epsilon ) + \left(1 - 9 \epsilon + 8 \epsilon ^3\right ) z_z^2\right ) z_{\rho } ^6+(-1 + 4 \epsilon ( 3 - 2 ( -3+\epsilon ) \epsilon ) ) z_{\rho } ^8 + 24 \epsilon z_z^4\right)\big]\big/ \big[\left(1+(-1+\epsilon ) z_{\rho } ^2\right ) \nonumber\\ & & \hspace*{12 mm } \times \left(z_z^2 \left(12 \epsilon + ( -1+\epsilon ) ^2 z_z^2\right ) z_{\rho } + 2 \left(-6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_z^2\right ) z_{\rho } ^3+(1 + 2 \epsilon ) ^2 z_{\rho } ^5\right){}^2\big]\nonumber\\ & & \hspace*{8 mm } + b_2\ , \big[(-1 - 2 \epsilon + \sqrt{1+\epsilon -2 \epsilon ^2 } ) \big((-1+\epsilon ) ^4 z_z^6 z_{\rho } ^2 + 12 ( 1 - 7 \epsilon ) \epsilon z_{\rho } ^4 + 3 \epsilon ( 1 - 8 ( -2+\epsilon ) \epsilon ) z_{\rho } ^6-\left(1+\epsilon -2 \epsilon ^2\right)^2 z_{\rho } ^8\nonumber\\ & & \hspace*{12 mm } \left . + 3 z_z^2 z_{\rho } ^2 \left(8 \epsilon ( -1 + 4 \epsilon ) + 2 ( -1+\epsilon ) \epsilon ( 1 + 8 \epsilon ) z_{\rho } ^2+(-1+\epsilon ) ^2 ( 1 + 2 \epsilon ) z_{\rho } ^4\right)\right . \nonumber\\ & & \hspace*{12 mm } + 3 ( -1+\epsilon ) z_z^4 \left(-4 \epsilon + ( -1+\epsilon ) z_{\rho } ^2 \left(\epsilon + \left(-1+\epsilon ^2\right ) z_{\rho } ^2\right)\right)\big)\big]\big/ \big[(-1+\epsilon ) \left(\sqrt{1-\epsilon } -\sqrt{1 + 2 \epsilon } \right ) \left(1+(-1+\epsilon ) z_{\rho } ^2\right ) \nonumber\\ & & \hspace*{12 mm } \times \left((-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2+(1 + 2 \epsilon ) ^2 z_{\rho } ^4 + 2 z_z^2 \left(6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right){}^2\big ] \nonumber\\ & & \hspace*{8 mm } + \big[\sqrt{1 + 2 \epsilon } ( -1 + 4 \epsilon ) z_{\rho } ^4-\sqrt{1 + 2 \epsilon } z_z^4 \left(1 + 5 \epsilon + 3 ( -1+\epsilon ) \epsilon z_{\rho } ^2\right)+\sqrt{1 + 2 \epsilon } z_z^2 z_{\rho } ^2 \left(2+\epsilon -3 \epsilon ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\nonumber\\ & & \hspace*{12 mm } + z_z z_{\rho } ^2 \left(-6 \epsilon + \left(1+\epsilon -2 \epsilon ^2\right ) z_{\rho } ^2\right)+z_z^3 \left(6 \epsilon + ( -2+\epsilon ( -2 + 13 \epsilon ) ) z_{\rho } ^2\right)+ \left(1+\epsilon -2 \epsilon ^2\right ) z_z^5\big]\big/\nonumber\\ & & \hspace*{8 mm } \big[\left(-1+(1 + 2 \epsilon ) z_z^2\right ) \left(1+(-1+\epsilon ) z_{\rho } ^2\right ) \left((-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2+(1 + 2 \epsilon ) ^2 z_{\rho } ^4 + 2 z_z^2 \left(6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right)\big]\ , .\nonumber\end{aligned}\ ] ] the disorder correction to the chemical potential , defined by eq . ( [ eq : eqstgen ] ) , is given by : @xmath207\big/ \big[z_z \left(1+(-1+\epsilon ) z_{\rho } ^2\right){}^2 \left((-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2\right.\nonumber\\ & & \hspace*{12 mm } \left . + ( 1 + 2 \epsilon ) ^2 z_{\rho } ^4 + 2 z_z^2 \left(6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right)\big]\nonumber\\ & & \hspace*{8 mm } -a_2\ , \big[(-1+\epsilon ) \left(\sqrt{1-\epsilon } -\sqrt{1 + 2 \epsilon } \right ) \left(-30 \epsilon z_{\rho } ^2+(2 + 2 ( 13 - 5 \epsilon ) \epsilon ) z_{\rho } ^4+(-1+\epsilon ) ( 1 + 2 \epsilon ) ^2 z_{\rho } ^6\right.\nonumber\\ & & \hspace*{12 mm } \left . + ( -1+\epsilon ) ^2 z_z^4 \left(2+(-1+\epsilon ) z_{\rho } ^2\right)+2 z_z^2 \left(15 \epsilon + ( -1+\epsilon ) z_{\rho } ^2 \left(2 + 13 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right)\right)\big]\big/\nonumber\\ & & \hspace*{12 mm } \big[(-1 - 2 \epsilon + \sqrt{1+\epsilon -2 \epsilon ^2 } ) \left(1+(-1+\epsilon ) z_{\rho } ^2\right){}^2 \left((-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2+(1 + 2 \epsilon ) ^2 z_{\rho } ^4\right.\nonumber\\ & & \hspace*{12 mm } \left . + 2 z_z^2 \left(6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right)\big]\nonumber\\ & & \hspace*{8 mm } -b_1\ , \big[2 z_z ( 36 \epsilon ( -1 + 3 \epsilon ) z_{\rho } ^4+(2+\epsilon ( 27 + 8 \epsilon ( -12 + 5 \epsilon ) ) ) z_{\rho } ^6-\left(1+\epsilon -2 \epsilon ^2\right)^2 z_{\rho } ^8+(-1+\epsilon ) ^3 z_z^6 \left(2+(-1+\epsilon ) z_{\rho } ^2\right ) \nonumber\\ & & \hspace*{12 mm } + 3 z_z^2 z_{\rho } ^2 \left(24 ( 1 - 2 \epsilon ) \epsilon + ( -1+\epsilon ) z_{\rho } ^2 \left(2 + 2 ( 9 - 16 \epsilon ) \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right)\nonumber\\ & & \hspace*{12 mm } + 3 ( -1+\epsilon ) z_z^4 \left(12 \epsilon + ( -1+\epsilon ) z_{\rho } ^2 \left(2 + 9 \epsilon + \left(-1+\epsilon ^2\right ) z_{\rho } ^2\right)\right))\big]\big/ \big[z_{\rho } ^2 \left(1+(-1+\epsilon ) z_{\rho } ^2\right ) \left((-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2\right.\nonumber\\ & & \hspace*{12 mm } \left . + ( 1 + 2 \epsilon ) ^2 z_{\rho } ^4 + 2 z_z^2 \left(6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right){}^2\big]\nonumber\\ & & \hspace*{8 mm } + b_2\ , \big[2 ( -1 - 2 \epsilon + \sqrt{1+\epsilon -2 \epsilon ^2 } ) ( -144 \epsilon ^2 z_{\rho } ^2 + 6 \epsilon \left(7 + 22 \epsilon -20 \epsilon ^2\right ) z_{\rho } ^4 - 2 ( -1+\epsilon ) ( 1 + 2 \epsilon ) ( -1 + 8 ( -2+\epsilon ) \epsilon ) z_{\rho } ^6\nonumber\\ & & \hspace*{12 mm } + 3 ( -1+\epsilon ) ^2 z_z^4 \left(14 \epsilon + ( -1+\epsilon ) z_{\rho } ^2 \left(2 + 12 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right)\nonumber\\ & & \hspace*{12 mm } + 3 z_z^2 ( 48 \epsilon ^2+(-1+\epsilon ) z_{\rho } ^2 ( 4 \epsilon ( 7 + 11 \epsilon ) + 2 ( -1+\epsilon ) ( 1 + 4 \epsilon ( 3 + 2 \epsilon ) ) z_{\rho } ^2+\left(1+\epsilon -2 \epsilon ^2\right)^2 z_{\rho } ^4))\nonumber\\ & & \hspace*{12 mm } + ( -1+\epsilon ) ^2 ( 1 + 2 \epsilon ) ^3 z_{\rho } ^8+(-1+\epsilon ) ^4 z_z^6 \left(2+(-1+\epsilon ) z_{\rho } ^2\right))\big]\big/ \big[(-1+\epsilon ) \left(\sqrt{1-\epsilon } -\sqrt{1 + 2 \epsilon } \right ) \nonumber\\ & & \hspace*{12 mm } \times \left(1+(-1+\epsilon ) z_{\rho } ^2\right ) \left((-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2+(1 + 2 \epsilon ) ^2 z_{\rho } ^4 + 2 z_z^2 \left(6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right){}^2\big]\nonumber\\ & & \hspace*{8 mm } + \big[2 ( ( -1+\epsilon ) ( 1 + 2 \epsilon ) ^{3/2 } z_z^4+(-1+\epsilon ) ^2 ( 1 + 2 \epsilon ) z_z^5+\sqrt{1 + 2 \epsilon } z_{\rho } ^2 \left(6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\nonumber\\ & & \hspace*{12 mm } + z_z^3 \left(9 \epsilon ( 1+\epsilon ) + ( -1+\epsilon ) ( 2+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)+\sqrt{1 + 2 \epsilon } z_z^2 \left(-6 \epsilon + ( 2+(2 - 13 \epsilon ) \epsilon ) z_{\rho } ^2\right)\nonumber\\ & & \hspace*{12 mm } + z_z z_{\rho } ^2 \left(-9 \epsilon + \left(1 + 3 \epsilon -4 \epsilon ^3\right ) z_{\rho } ^2\right))\big]\big/ \big[\left(-1+(1 + 2 \epsilon ) z_z^2\right ) \left(1+(-1+\epsilon ) z_{\rho } ^2\right ) \nonumber\\ & & \hspace*{12 mm } \times \left((-1+\epsilon ) ^2 z_z^4 - 12 \epsilon z_{\rho } ^2+(1 + 2 \epsilon ) ^2 z_{\rho } ^4 + 2 z_z^2 \left(6 \epsilon + ( -1+\epsilon ) ( 1 + 2 \epsilon ) z_{\rho } ^2\right)\right)\big]\nonumber\\ & & \hspace*{8 mm } + \frac{8}{z_z+z_{\rho } } \ , t\left(\frac{-z_z+z_{\rho } } { z_z+z_{\rho } } \right)\ , .\nonumber\end{aligned}\ ] ]

the problem of online learning and optimization of unknown markov jump affine models is considered . an online learning policy , referred to as markovian simultaneous perturbations stochastic approximation ( mspsa ) , is proposed for two different optimization objectives : ( i ) the quadratic cost minimization of the regulation problem and ( ii ) the revenue ( profit ) maximization problem . it is shown that the regret of mspsa grows at the order of the square root of the learning horizon . furthermore , by the use of van trees inequality , it is shown that the regret of any policy grows no slower than that of mspsa , making mspsa an order optimal learning policy . in addition , it is also shown that the mspsa policy converges to the optimal control input almost surely as well as in the mean square sense . simulation results are presented to illustrate the regret growth rate of mspsa and to show that mspsa can offer significant gain over the greedy certainty equivalent approach . online learning , stochastic approximation , stochastic cramer - rao bounds , continuum - armed bandit , sequential decision making .

1605.02213

consider the problem of _ online learning and optimization _ of affine memoryless models with unknown parameters that follow a markov jump process . by online learning and optimization we mean that the control input of the unknown model is chosen sequentially to minimize the expected total cost or to maximize the expected cumulative reward procured over a time horizon @xmath0 . in this context , the online learning problem is one of exploration and exploitation ; the need of exploring the space of unknown parameters must be balanced by the need of exploiting the knowledge acquired through learning . for online learning problems with deterministic unknown parameters , a commonly used performance measure is the so - called _ regret _ , defined by the difference between the cumulative cost / reward of an online learning policy and that of a decision maker who knows the model completely and sets the input optimally . the regret grows monotonically with the time horizon @xmath0 , and the rate of growth measures the efficiency of online learning policies . the online learning problem considered in this paper is particularly relevant in dynamic pricing problems when the consumers demand is unknown and possibly varying stochastically . the goal of dynamic pricing is to set the price sequentially , using the observations from the previous sales , to match a certain contracted demand . besides applications in dynamic pricing , results are also relevant to the learning and control problem of markov jump linear systems with unknown parameters . in this paper , we study the online learning and optimization problem of markov jump affine models under two different objectives : ( i ) target matching with a quadratic cost and ( ii ) revenue ( profit ) maximization . our goal is to establish fundamental limits on the rate of regret growth for markov jump affine models and develop an online learning policy that achieves the lowest possible regret growth . without markov jump as part of the model , when there is a single state , the problem considered here is the classical problem of control in experiment design studied by anderson and taylor . anderson and taylor proposed a certainty equivalence rule where the input is determined by using the maximum likelihood estimates of system parameters as if they were the true parameters . despite its intuitive appeal , the anderson - taylor rule was shown to be suboptimal for the quadratic regulation problem by lai and robbins in and also for the revenue maximization problem by den boer and zwart in . in fact , there is a non - zero probability that the anderson - taylor rule produces an input which converges to a suboptimal value for both cases ; therefore , this rule results in _ incomplete learning _ and a linear growth of regret . for the scalar model in which the quadratic cost of the regulation problem is to be minimized , lai and robbins showed that a robbins - monro stochastic approximation approach achieves the optimal regret order of @xmath1 . later , lai and wei showed that this regret order is also achievable for a more general linear dynamic system by an adaptive regulator that uses least square estimates of a reparametrized model and ensures convergence via occasional uses of white - noise probing inputs . the result was further generalized by lai @xcite to multivariate linear dynamic systems with a square invertible system matrix . the result presented in this paper can be viewed as a generalization of this line of work to allow both time - varying linear models and time - invariant models with a non - invertible system matrix . the problem considered in this paper also falls into the category of continuum - armed bandit problem where the control input is chosen from a subset of @xmath2 with the goal of minimizing expected cost ( or maximizing expected reward ) that is an unknown continuous function of the input . this problem was introduced by agrawal @xcite who studied the scalar problem and proposed a policy that combines certainty equivalence control with kernel estimator - based learning . agrawal showed that this policy has a regret growth rate of @xmath3 for a uniformly lipschitz expected cost function . later , kleinberg @xcite proved that the optimal growth rate of regret for this problem can not be smaller than @xmath4 and proposed a policy that achieves @xmath5 . kleinberg @xcite also considered the multivariate problem , @xmath6 , and showed that an adaptation of zinkevich s greedy projection algorithm achieves the regret growth rate of @xmath3 if the cost function is smooth and convex on a closed bounded convex input set . within the continuum - armed bandit formulation , the work of cope @xcite is particularly relevant because of its use of stochastic approximation to achieve the order - optimal regret growth of @xmath7 for a different class of cost functions . cope s results ( both the regret lower bound and the kiefer - wolfowitz technique ) , unfortunately , can not be applied here because of the time - varying markov jump affine models treated here . also relevant is the work of rusmevichientong and tsitsiklis on the so - called linearly parameterized bandit problem where the objective is to minimize a linear cost with input selected from the unit sphere . a learning policy developed in is shown to achieve the lower bound of @xmath7 using decoupled exploration and exploitation phases . even though the model considered in this paper is similar to the one in in terms of the observed output being a linear function of the input , in our paper , the unknown model parameters follow a markov jump process and the specific cost functions studied are quadratic ; thus the problem objective is different . there is a considerable amount of work on dynamic pricing problem with the objective of revenue maximization under a demand model uncertainty in different areas such as operations research , statistics , mathematics , and computer science . in , a multi - armed bandit approach with a regret growth rate of @xmath8 was proposed for a nonparametric formulation of the problem . see also where the same problem under a general parametric demand model is considered and a modified version of myopic maximum likelihood based policy is shown to achieve the regret order of @xmath9 , and where a similar result is obtained for a class of parametric demand models . in both and , authors proved that the lower bound for regret growth rate is @xmath7 . besides more general classes of demand models , affine model similar to the one in this paper has been also studied extensively ; . in both and , it is shown that approximate dynamic programming solutions may outperform greedy method numerically . a special case of our formulation of revenue maximization problem without any markov jump characteristics ( with time - invariant model parameters ) is previously investigated by keskin and zeevi . keskin and zeevi proposed a semi - myopic policy that uses orthogonal pricing idea to explore and learn the system . they showed that the lowest possible regret order is @xmath7 for any policy , and their semi - myopic policy achieves this bound up to a logarithmic factor ; @xmath10 . even though the system model is assumed to be time - invariant in most of the literature , there is a considerable amount of work especially in dynamic pricing that deals with time - varying demand models due to unpredictable environmental factors affecting demand ; see for a demand model that evolves according to a discrete state space markov chain in a revenue management with finite inventory problem , and for a dynamic programming formulation of a profit maximization problem with an unknown demand parameter following an autoregressive process . see also for a revenue maximization problem with an affine demand model where the model parameters are time - varying , yet the cumulative change in the model parameters over the time horizon @xmath0 is bounded . since keskin and zeevi measure the regret of a policy by the difference between the cumulative cost of the policy and that of a clairvoyant who knows all the future temporal changes exactly and chooses the optimal action , their characterization of regret is too pessimistic for the markov jump model considered here . some other examples of related work on online learning with time - varying models apart from dynamic pricing are and . in , besbes , gur , and zeevi studied the online learning problem of more general time - varying cost functions where the cumulative temporal changes is restricted to a budget similar to . however , their characterization of regret is also similar to and thus , incomparable with the one in this paper . yin , ion , and krishnamurthy also considered the problem of estimating a randomly evolving optimum of a cost function which follows a markov jump process . their analysis deals with the convergence of the estimate obtained via stochastic approximation to the limit ( stationary ) solution , whereas in this paper , we are concerned about estimating the optimum of the cost function at each time instant given the previous state of the markov chain . moreover , different than our paper , their analysis relies on the availability of the noisy observations of the cost function gradient and they do not characterize regret . the main contribution of this paper is the generalization of online learning of time - invariant affine models to that of markov jump affine models . extending spall s stochastic approximation method @xcite to the optimization problem of an objective function that evolves according to a markov jump process , we propose an online learning policy , referred to as markovian simultaneous perturbations stochastic approximation ( mspsa ) . we show that mspsa achieves the optimal regret order of @xmath11 for two different objective functions studied for the affine model : ( i ) quadratic regulation and ( ii ) the revenue maximization . furthermore , we also show that the control input of mspsa policy converges to the optimal solution both with probability one and in mean square as @xmath12 . therefore , the proposed policy eventually learns the optimal solution . a key implication of our results is that , in comparing with lai s result on the learning problem of a time - invariant affine model with the quadratic regulation objective @xcite , modulating a linear model by a markov jump process introduces substantial learning complexity ; hence , the regret order increases from @xmath1 to @xmath11 . as a special case , we also show that , even in the absence of markovian jump , when the system matrix is full column rank but not invertible , the best regret order is also @xmath11 . it worths noting that adding just one row to a square and invertible matrix can change the worst case regret from @xmath1 to @xmath11 . in the second part of this paper , we study the profit maximization problem with markov jump demand . to this end , we generalize the lower bound obtained by keskin and zeevi for time - invariant demand to markov jump demand , and we show that this bound is achievable by the mspsa policy for a more general case with markov jump characteristics . the results presented here are obtained using several techniques developed in different contexts . the mspsa policy is a generalization of spall s stochastic approximation method to the optimization problem of an objective function following a markov jump process . to show the optimality of mspsa , we use the van trees inequality to lower bound the estimation error , which is the technique used in . lastly , a result on the convergence of non - negative almost supermartingales @xcite is used to obtain the convergence result for mspsa policy . the model considered here is an affine model , modulated by an exogenous finite state time - homogeneous markov chain @xmath13 where @xmath14 is the state space and @xmath15 $ ] the transition probability matrix . we assume that the state space @xmath16 and the transition matrix @xmath17 are unknown . each state @xmath18 of the markov chain is associated with an affine model whose parameters are denoted by @xmath19 where @xmath20 has full column rank and @xmath21 . all system parameters @xmath22 are assumed deterministic and unknown . at time @xmath23 , the input - output relation of the system is given by rcl [ eq : linear ] y_t & = & a_s_tx_t + b_s_t + w_t , where @xmath24 is the control input , @xmath25 the observable output , @xmath26 the state of the system , and @xmath27 is a random vector that captures the system noise . it is assumed that the random noise @xmath28 is drawn from a possibly state dependent distribution @xmath29 with zero mean ( without loss of generality ) and unknown finite variance @xmath30 . furthermore , for any @xmath31 , @xmath28 and @xmath32 are conditionally independent given the states @xmath33 and @xmath34 . the objective of _ online learning and optimization _ is to find a control input sequence @xmath35 that minimizes the expected cumulative cost incurred at each stage . because @xmath36 is determined before @xmath33 is realized , the stage cost @xmath37 at @xmath23 is a function of @xmath36 and @xmath38 , and the expected cumulative cost is @xmath39 where @xmath0 is the learning and optimization horizon . note that the above quantity is a function of the deterministic parameters @xmath40 and the distributions @xmath17 and @xmath41 . for a decision maker who wants to minimize its expected cumulative cost , the difficulty in finding the optimal control input sequence is that the system parameter @xmath40 and the transition matrix @xmath17 are unknown . if the system parameters @xmath42 were known , then the decision maker would have used this information along with its observation history to determine the optimal decision rule that minimizes the expected cumulative cost . in that case , the problem could be formulated as a dynamic program and solved via backward induction . we refer to this optimal solution under known model as the optimal input and denote it by @xmath43 ( which is made precise in the following sections ) . we assume that a convex compact set @xmath44 containing the optimal input is known to the decision maker . before choosing the control input of period @xmath23 , the only information the decision maker has is a vector @xmath45 containing its decision and observation history up to time @xmath46 , which includes input vector @xmath47 , output vector @xmath48 , state vector @xmath49 , and the set @xmath50 . consequently , a policy @xmath51 of a decision maker is defined as a sequence of decision rules , @xmath52 , such that , at time @xmath46 , @xmath53 maps the information history vector @xmath45 to the system input @xmath54 at time @xmath23 . we denote the input determined by policy @xmath51 as @xmath55 . to measure the performance of an online learning policy , we use the regret measure as a proxy . in particular , the ( cumulative ) regret @xmath56 of a learning policy @xmath51 is measured by the difference between the expected cumulative cost of the decision maker , who follows the policy @xmath51 , and that of a decision maker who knows the system parameters @xmath42 and sets the system input optimally , rcl [ eq : defregret ] _ t^(,p ) & = & ( _ t=1^t ( s_t-1,x_t^ ) - _ t=1^t ( s_t-1,x_t^ * ) ) . since the regret defined above is a function of system parameters , we characterize the performance of @xmath51 by the worst case regret @xmath57 in terms of the worst - case analysis , it is assumed that the worst - case system parameter @xmath40 is chosen from a compact set @xmath58 for any fixed values of the state space size @xmath59 and system dimensions @xmath60 and @xmath61 . since @xmath62 is compact , for any @xmath63 and for all @xmath18 , the largest singular value of @xmath64 is bounded by a positive constant @xmath65 , and the parameter @xmath66 is bounded by a positive constant @xmath67 , @xmath68 . it is also assumed that the variance of @xmath28 is bounded , @xmath69 for some positive constant @xmath70 where @xmath71 denotes the @xmath72th entry of @xmath28 . the optimal input for the worst - case parameters @xmath42 certainly has to be contained in the set @xmath50 , and @xmath64 has to be full column rank for every @xmath18 for the worst - case parameter @xmath40 . note that @xmath73 grows monotonically with @xmath0 . we are interested in the learning rule that has the slowest regret growth . in the following sections , we focus on two different stage costs , hence two different objective functions . the first is a quadratic cost that arises naturally from the regulation problem . in particular , the stage cost at time @xmath23 is given by rcl [ eq : cost ] ^(s_t-1,x_t ) & & ( ||y^ * - y_t ||_2 ^ 2 | s_t-1 , x_t ) , where @xmath74 is a constant target value for output . for the quadratic regulation problem , we assume that the forth order moment of @xmath28 is bounded , @xmath75 , in addition to the previous boundedness assumptions . the second stage cost we consider is the minus revenue that arises from revenue maximization problem ( or profit maximization problem since profit can be expressed as revenue minus total cost ) . specifically , the stage cost of period @xmath23 is given by rcl [ eq : cost_revenue ] ^(s_t-1,x_t ) & & ( - x_t y_t | s_t-1 , x_t ) . here , the revenue is calculated as the inner product of the input and the output vector where the entries of the input and the output vector corresponds to the price and the demand of each product , respectively . therefore , the input and the output dimensions match , @xmath76 @xmath77 , for the revenue maximization problem . for this objective , it is assumed that the matrix @xmath78 is negative definite for all @xmath79 which is a reasonable assumption in dynamic pricing problems , see . in this section , we study the online learning problem with the quadratic cost . we first derive an expression for regret using the optimal solution under known model referred to as the optimal input . then , we introduce an online learning approach and establish its order optimality via the analysis of its regret growth rate and the analysis of the minimum regret growth rate achievable by any policy . we also show that the input of the online learning policy converges to the optimal input both almost surely and in mean square . in order to calculate the regret for any policy , we begin by deriving the optimal solution of a decision maker who knows the system @xmath42 in addition to @xmath45 and aims to minimize the expected cumulative cost , rcl [ eq : cum_risk ] _ \{x_t}_t=1^t ( _ t=1^t ^(s_t-1,x_t ) ) . the problem under known model becomes a dynamic program due to the known @xmath42 . since the markov process is exogenous , independent of the decision policy , the optimization problem decouples to choosing the system input @xmath36 separately for each decision stage with stage cost given in ( [ eq : cost ] ) which is equivalent to rcl [ eq : risk ] ^(s_t-1,x_t ) & = & _ j p_s_t-1,j ( y^*-a_jx_t - b_j_2 ^ 2 + ( _ w^(j ) ) ) + by ( [ eq : linear ] ) . the optimal input @xmath80 minimizing the stage cost is then given by rcl [ eq : optprice ] x_s_t-1^ * & = & ( _ j p_s_t-1,ja_j^ a_j)^-1(_j p_s_t-1,ja_j^(y^*-b_j ) ) . thus , the optimal input @xmath81 at any time @xmath23 depends only on the system parameter @xmath40 , the transition matrix @xmath17 , and the previous state @xmath38 . in the sequel , we use @xmath82 to represent @xmath80 , dropping the explicit parameter dependency on @xmath42 in the notation . hence , the stage regret at @xmath23 , which is the expected difference of the stage cost obtained by policy @xmath51 and the stage cost of the optimal input @xmath83 , can be written as rcl r_t^(,p ) & = & ( ^(s_t-1,x_t^ ) - ^(s_t-1,x_s_t-1^ * ) ) + & = & ( a_s_t(x_t^-x_s_t-1^*)_2 ^ 2 ) , which is obtained using the first order optimality condition ( foc ) for @xmath82 . the t - period regret given in ( [ eq : defregret ] ) can then be expressed as rcl [ eq : regret ] _ t^(,p ) & = & ( _ t=1^t a_s_t(x_t^-x_s_t-1^*)_2 ^ 2 ) . here , we present an online learning policy to the quadratic regulation problem that achieves the slowest regret growth rate possible . referred to as mspsa , the policy is an extension of the simultaneous perturbation stochastic approximation ( spsa ) algorithm proposed by spall @xcite to markov jump models considered here . spall s spsa is a stochastic approximation algorithm that updates the estimate of the optimal input by a stochastic approximation of the objective gradient . the key step is to generate two consecutive observations corresponding to two inputs , that are set to be the current optimal - input estimate perturbed by some random vector in opposite directions , and use them to construct the gradient estimate . in applying this idea to the optimization problem of a markov jump system , a complication arises due to the uncertainty associated with the system state at the time when the system input is determined ; consecutive observations that are used to determine the gradient estimate may correspond to different system states . let @xmath84 be an arbitrary vector @xmath85 @xmath86 @xmath87 @xmath88 where @xmath89 with some positive constant @xmath90 and a non - negative integer @xmath91 , and @xmath92^{\tra}$ ] with @xmath93 s drawn from an independent and identical distribution that is symmetrical around zero , and satisfies @xmath94 and @xmath95 for some positive constants @xmath96 and @xmath97 . @xmath98 @xmath99 @xmath100 @xmath101 @xmath86 update : rcl [ eq : mspsaupdate_regulation ] _ i , t_i+1 & & ( _ i , t_i - a_t_i()|_t_i ) _ where @xmath102 denotes the euclidean projection operator onto @xmath50 , @xmath103^{\tra}$ ] , and @xmath104 with some positive constant @xmath105 and a non - negative integer @xmath106 the key idea of mspsa is to keep track of each state @xmath107 and the estimate of the optimal input associated with each state @xmath107 . when state @xmath72 is realized , the estimate of the optimal input associated with state @xmath72 is perturbed by some random vector and this randomly perturbed estimate is used as input for the next stage . the estimate of the optimal input associated with state @xmath72 is updated only when we obtain two observations of the system output corresponding to two inputs that are generated by perturbing the current estimate in opposite directions by the same amount right after observing state @xmath72 . details of this implementation is given in [ mspsa_for_qr ] . whenever a new state @xmath107 is observed that has not been observed before , mspsa policy assigns an arbitrary predetermined vector @xmath84 as the initial estimate of the optimal input @xmath108 ( line 3 - 7 of [ mspsa_for_qr ] ) . at the beginning of each stage @xmath23 , mspsa checks the previous state @xmath38 ( line 2 ) , and whether any observation is taken using the most recent optimal - input estimate @xmath109 ( line 8) where @xmath110 is the number of times the optimal - input estimate @xmath109 is updated up to time t ( since two observations , that are taken right after observing state @xmath38 , are used for each update of @xmath109 , @xmath110 is approximately half of the number of times state @xmath38 is observed up to @xmath23 . ) . if an observation has not taken using the most recent estimate yet , the input for that stage is set to be a randomly perturbed @xmath109 ( line 10 ) . otherwise mspsa sets the input by perturbing the estimate @xmath109 in the opposite direction by the same amount as the previous one ( line 14 ) . then , it updates the optimal - input estimate by a stochastic approximation ( line 17 ) obtained using the stage costs calculated from both observations ( line 11 and 15 ) and projects it onto @xmath50 . the constant @xmath90 of the perturbation gain sequence @xmath111 should be chosen larger in the high noise setting for an accurate gradient estimate . the choice of the sequence @xmath112 used for the update step determines the step size . the non - negative integers @xmath106 of @xmath112 and @xmath91 of @xmath111 can be set to zero as default , but if the update of the optimal - input estimate fluctuates between the borders of @xmath50 at the beginning of the mspsa policy , setting @xmath106 greater than zero can prevent this fluctuation . we now analyze the regret performance of mspsa . let @xmath113 denote the minimum eigenvalue operator , and @xmath114 be the mean squared error ( mse ) between the optimal input @xmath115 and its estimate @xmath116 given state @xmath72 and @xmath117 , where @xmath117 , as defined in previous section , is the number of times the estimate @xmath116 has been updated up to time @xmath23 by mspsa . the following lemma provides a bound for the decreasing rate of @xmath118 , and hence the convergence rate of the estimate to its true value in terms of the number of times the estimate is updated and thus in terms of the number of times the state @xmath72 has occurred up to @xmath23 ( which is equal either to @xmath119 or to @xmath120 ) . it shows that the mse converges to zero with a rate equal or faster than the inverse of the square root of the number of times state @xmath72 has occurred . [ lemma : regret ] for any @xmath107 , if @xmath121 then there exists a constant @xmath122 satisfying @xmath123 for any @xmath124 . see appendix . to satisfy the condition of lemma [ lemma : regret ] , the decision maker , who follows mspsa , needs to have some information about a lower bound on the minimum eigenvalue of @xmath125 , knowing a non - trivial lower bound @xmath126 for the singular values of the system matrices . this assumption may not be restrictive in practice since the decision maker can set @xmath105 sufficiently large . let the worst - case cumulative input - mse @xmath127 be the worst - case cumulative mse between the input @xmath55 of policy @xmath51 and the optimal input @xmath83 , @xmath128 using the result of lemma @xmath129 , we provide a bound for the growth rate of the worst - case cumulative input - mse and the worst - case regret of mspsa . theorem [ thm : upper ] shows that the mspsa policy achieves the regret growth rate of @xmath9 . [ thm : upper ] if @xmath121 for every @xmath107 , then there exist some positive constants @xmath130 and @xmath131 such that @xmath132 and @xmath133 the input of mspsa @xmath134 is equal to either @xmath135 or @xmath136 given @xmath137 and @xmath117 . by lemma [ lemma : regret ] , observe that rcl [ eq : r_i_t_i ] r_i , t_i & = & ( _ i , t_i c_t_i _ t_i -x_i^*_2 ^ 2|i , t_i ) + & = & e_i , t_i + c_t_i^2 ( _ t_i^_t_i ) c_i/ where @xmath138 . let @xmath139 be the number of times the estimate of the optimal input associated with state @xmath72 has been updated until period @xmath0 . because mspsa uses two observations per update , we can express the worst - case cumulative input - mse for mspsa as @xmath140 by ( [ eq : r_i_t_i ] ) , we bound @xmath141 where @xmath142 . since @xmath139 is smaller than the number of times state @xmath72 is observed , which is a fraction of @xmath0 for any @xmath17 , @xmath143 where @xmath144 . consequently , by ( [ eq : regret ] ) and using the upper bound @xmath65 for the singular values of @xmath145 , @xmath146 where @xmath147 . according to theorem [ thm : upper ] , the average regret converges to zero with a rate equal or faster than @xmath148 . hence , it proves that the average performance of mspsa policy approaches to that of the optimal solution under known model as @xmath12 . however , this does not imply the convergence of the mspsa policy to the optimal input . the following theorem provides both almost sure and mean square convergence of mspsa policy to the optimal input . [ thm : convergence ] for the quadratic regulation problem , rcl [ eq : asconv ] ( _ t x_t^-x^*_s_t-1 ^ 2_2 = 0 ) & = & 1 , and rcl [ eq : inputconv ] _ t ( x_t^-x^*_s_t-1_2 ^ 2 ) & = & 0 . see appendix . theorem [ thm : convergence ] shows that , the input generated by mspsa converges to its optimal value as @xmath12 for any choice of @xmath149 . hence , the condition given in theorem [ thm : upper ] is not necessary for the convergence of mspsa policy . the intuition is as follows ; for any observed recurrent state @xmath107 , the input of mspsa converges to its optimal value at time periods when the previous state is @xmath72 as @xmath12 , and any observed transient state @xmath107 will occcur only a finite number of times . therefore , the input of mspsa converges to its optimal value as @xmath12 . we now show that mspsa in fact provides the slowest possible regret growth . to this end , we provide a lower bound of regret growth for all decision policies . for any policy @xmath51 , the estimate of the optimal input and the actual input of the policy may not be the same ; for example , the input of mspsa is a randomly perturbed estimate of the optimal input and not the estimate itself . hence , let s denote an optimal - input estimate obtained using the past observations corresponding to inputs of policy @xmath51 at time @xmath23 by @xmath150 and the input at time @xmath23 by @xmath55 . we define the worst - case cumulative estimation - mse as @xmath151 in particular , the following theorem states that the product of the growth rate of the worst - case cumulative input - mse @xmath127 and the worst - case cumulative estimation - mse @xmath152 of any sequence @xmath153 can not be lower than @xmath0 for any policy @xmath51 . [ thm : lower ] for any value of @xmath154 , there exists a constant @xmath155 such that , for any policy @xmath51 , rcl [ eq : tradeoffbound ] _ t^ |_t^ & & c t. see appendix . theorem [ thm : lower ] shows the trade - off between exploration ( minimizing the estimation error ) and exploitation ( minimizing the input error ) . if the goal is to minimize the cumulative estimation - mse rather than the regret , than it is possible to find a policy for which @xmath152 grows slower than @xmath156 in which case the cumulative input - mse @xmath127 has to grow faster than @xmath156 . in fact , if @xmath111 is set to be constant rather than a decreasing sequence of @xmath117 , by following the proof of theorem [ thm : upper ] , it is easy to show that mspsa s cumulative estimation - mse grows no faster than @xmath157 whereas its regret would grow linearly with @xmath0 . however , the slowest growth rate of @xmath127 can not be slower than that of @xmath152 for the optimal choice of the estimate sequence @xmath153 ( in other words , one can always take the estimate equal to the input , @xmath158 , in which case @xmath159 ) . therefore , the growth rate of the worst - case cumulative input - mse @xmath127 , and , consequently , the growth rate of the worst - case regret can not be lower than @xmath156 for any policy @xmath51 . therefore , the regret growth rate of mspsa is the optimal one and achieves @xmath7 as stated in theorem [ thm : lower_regret ] . [ thm : lower_regret ] for any value of @xmath154 , there exist some constants @xmath160 such that , for any policy @xmath51 , rcl [ eq : inputmsebound ] |_t^ & & c , and rcl [ eq : regretbound ] |_t^ & & c . we choose the estimate @xmath161 equal to the input @xmath162 . then , by theorem [ thm : lower ] , we have @xmath163 . as a result , @xmath164 . let @xmath165 be the parameter satisfying @xmath166 . in theorem [ thm : lower ] , we fixed @xmath167 for any @xmath168 . hence , by ( [ eq : regret ] ) , @xmath169 where @xmath170 by the extreme value theorem . then , the worst case regret @xmath171 where @xmath172 . to prove theorem [ thm : lower ] , we consider a hypothetical case in which the decision maker receives additional observations at each period @xmath23 . it is assumed that the additional observations provided to the decision maker are the observation values corresponding to input @xmath173 from the states that did nt occur at @xmath23 . since such observations ca nt increase the growth rate of regret of the optimal policy , we establish a lower bound for this case by showing that it becomes equivalent to a single state case with @xmath174 and using the multivariate van trees inequality in a similar way as in . if @xmath175 and @xmath174 , the proofs of theorem [ thm : lower ] and theorem [ thm : lower_regret ] lead to the result regarding the single state case given in corollary [ corollary : lower ] . [ corollary : lower ] for @xmath175 and for any value of @xmath60 and @xmath61 satisfying @xmath176 , there exist some constants @xmath177 such that , for any policy @xmath51 , inequalities ( [ eq : tradeoffbound ] ) , ( [ eq : inputmsebound ] ) , and ( [ eq : regretbound ] ) hold . as mentioned in related work , it has been shown that for @xmath175 and @xmath77 case , the regret growth rate is @xmath178 @xcite . we show that the characteristics of regret growth changes from @xmath178 to @xmath11 for markov jump system . additionally , corollary [ corollary : lower ] states that , even in the absence of markov jump , when system matrix @xmath179 is not invertible , the best regret growth rate also jumps from @xmath178 to @xmath11 . these results can be interpreted as a consequence of the fact that the minimum of the cost function for markov jump system or for single state system with @xmath174 given in ( [ eq : risk ] ) is not a root of the cost function as in the case of @xmath175 with @xmath77 , and decision maker ca nt understand how close it is to the minimum just by looking at its observations . the single state setting of the revenue maximization problem has been previously studied by keskin and zeevi . in this paper , we consider the more general setting where the affine demand parameters can change depending on the state of nature , more precisely the setting where demand parameters follows a markov jump process . as for the quadratic regulation objective , to obtain a regret expression for revenue maximization objective , we first determine the optimal solution of a decision maker who knows the system @xmath42 . then , we present mspsa policy for revenue maximization problem and establish its optimality in regret performance and its convergence to the optimal solution . by following the same argument as before , under known model , the optimal solution of a decision maker aimed at minimizing the expected cumulative cost given in ( [ eq : cum_risk ] ) , which is equal to minus expected t - period revenue , is to choose the system input minimizing the respective stage cost given in ( [ eq : cost_revenue ] ) . by using ( [ eq : linear ] ) , this stage cost can be written as rcl [ eq : risk_revenue ] ^(s_t-1,x_t ) & = & -_j p_s_t-1,j x_t^ ( a_j x_t+b_j ) . the optimal input @xmath80 , which depends only on @xmath42 and the previous state @xmath38 , is then given by @xmath180 by dropping the explicit dependency of @xmath80 on @xmath42 in the notation . using the foc for the optimal input @xmath82 , we obtain the stage regret of a policy @xmath51 , rcl r_t^(,p ) & = & ( ^(s_t-1,x_t^)-^(s_t-1,x_s_t-1^ * ) ) + & = & - ( ( x_t^-x_s_t-1^ * ) ^ a_s_t ( x_t^-x_s_t-1^ * ) ) . since @xmath145 is negative definite , the stage regret is always non - negative . consequently , the t - period regret is given by rcl [ eq : cumulative_revenue ] _ t^(,p ) & = & - ( _ t=1^t ( x_t^-x_s_t-1^ * ) ^ a_s_t ( x_t^-x_s_t-1^ * ) ) . + here , we present the mspsa policy for revenue maximization objective . the only difference between the two problems considered is their respective stage costs ( objectives ) . therefore , the only change in mspsa policy is how the stage costs are calculated to approximate the objective gradient which corresponds to line 11 and 15 of [ mspsa_for_qr ] . in the corresponding steps of mspsa policy for revenue maximization , that is given in [ mspsa_for_rm ] in details , the stage costs are calculated as minus the observed revenue at that stage . let @xmath84 be an arbitrary vector @xmath85 @xmath86 @xmath87 @xmath88 where @xmath89 with some positive constant @xmath90 and a non - negative integer @xmath91 , and @xmath92^{\tra}$ ] with @xmath93 s drawn from an independent and identical distribution that is symmetrical around zero , and satisfies @xmath94 and @xmath95 for some positive constants @xmath96 and @xmath97 . @xmath181 @xmath99 @xmath100 @xmath182 @xmath86 update : rcl _ i , t_i+1 & & ( _ i , t_i - a_t_i()|_t_i ) _ where @xmath102 denotes the euclidean projection operator onto @xmath50 , @xmath103^{\tra}$ ] , and @xmath104 with some positive constant @xmath105 and a non - negative integer @xmath106 . to obtain the regret growth rate for mspsa policy for revenue maximization , we first derive an upper bound on how fast the estimate @xmath116 of the optimal input @xmath108 converges to its true value as we did for the regulation problem . lemma [ lemma : regret_revenue ] shows that the conditional mse @xmath118 between the optimal input and its estimate converges to zero with a rate no smaller than the inverse of the square root of the number of times the estimate @xmath116 is updated by mspsa . [ lemma : regret_revenue ] for any @xmath107 , if @xmath183 then there exists a constant @xmath122 satisfying @xmath123 for any @xmath124 . see appendix . the condition of lemma [ lemma : regret_revenue ] is slightly different than that of lemma [ lemma : regret ] . the bound on the step size constant @xmath105 depends on the minimum eigenvalue of @xmath184 . this difference is due to the choice of a different stage cost . however , the information of a non - trivial lower bound @xmath126 on the singular values of the system matrices is still sufficient to satisfy this condition . using the result of lemma [ lemma : regret_revenue ] , we prove that mspsa achieves the regret growth rate of @xmath9 for revenue maximization objective as given in theorem [ thm : upper_revenue ] , and the input of mspsa converges to the optimal input both almost surely and in mean square as given in theorem [ thm : convergence_revenue ] . [ thm : upper_revenue ] if @xmath185 for every @xmath107 , then there exist some positive constants @xmath130 and @xmath131 such that @xmath132 and @xmath186 same as the proof of theorem [ thm : upper ] up to the step that @xmath143 is obtained . then , by the regret given in ( [ eq : cumulative_revenue ] ) for revenue maximization objective and the fact that @xmath187 is positive definite with eigenvalues upper bounded by @xmath65 , @xmath188 where @xmath189 . [ thm : convergence_revenue ] for revenue maximization problem , @xmath190 and @xmath191 in the proof of lemma [ lemma : regret_revenue ] , we showed that ( [ eq : boundforasconv ] ) holds for any state @xmath107 . therefore , the proof follows the proof of theorem [ thm : convergence ] . previously , for single state setting of this problem , keskin and zeevi have shown that for any policy @xmath51 the worst case regret growth for this problem can not be smaller than @xmath7 and they have shown that @xmath10 is achievable by a semi - myopic policy that they referred to as multivariate constrained iterated least squares ( mcils ) policy . here , we showed that it is possible to achieve the lower bound @xmath7 given in by mspsa policy for more general problem with markov jumped demand . next , we generalize keskin and zeevi s lower bound result to markov jump case by showing that for any policy @xmath51 and for any state space size @xmath192 , the growth rate of worst - case regret is bounded by @xmath7 , and hence mspsa achieves the optimal rate of @xmath11 . [ thm : lower_revenue ] for any value of @xmath192 , there exist some constants @xmath193 such that , for any policy @xmath51 , rcl [ eq : tradeoffbound_revenue ] _ t^ |_t^ & & c^2 t , rcl [ eq : inputmsebound_revenue ] |_t^ & & c , and rcl [ eq : regretbound_revenue ] |_t^ & & c. see appendix . we present simulation results to illustrate the growth rate of regret and the optimal - input estimate convergence of mspsa policy both for quadratic regulation and revenue maximization problems . note that , by these simulation examples , we can only exhibit the performance of `` typical '' parameters and not the worst - case performance as studied in the theoretical characterization of regret . for a benchmark comparison , we consider the greedy least square estimate ( lse ) method proposed by anderson and taylor . at each period , the greedy lse determines the input by using the least square estimates of system parameters as if they were the true parameters and projects it onto the set @xmath50 . in order to calculate the initial lses of the system parameters , the first samples corresponding to the inputs generated by perturbing the initial input @xmath194 in each direction are taken until the lse of @xmath40 is computationally tractable . although , in general , greedy lse performs well numerically , it was shown that it can lead to incomplete learning and may not converge with positive probability which causes linear growth in regret . to illustrate the performance of mspsa policy for quadratic regulation problem , we consider the problem studied in , the problem of an electricity retailer who wants to set hourly electricity prices for the next day to meet its predetermined quantity @xmath195 in demand for each hour . therefore , the system dimension was set to be @xmath196 where each dimension corresponds to an hour of the day . different than where the demand is time - invariant , it is assumed that the demand of its customers changes depending on the state of the day , weather conditions , which follows a markov jump process . for this example , we considered two states and set the transition probability from any state to the same state to be 0.6 and to the other state to be 0.4 . to calculate the average performance , @xmath197 monte carlo runs were used . [ multivariate_target_matching ] shows the average performance of mspsa and greedy lse for this quadratic regulation example . in [ sqrtregret_2state ] , we plot the regret of both policies with respect to square root of the time horizon . we observe that the t - period regret of mspsa grows linearly with @xmath156 , which is consistent with the theoretical upper bound . on the other hand , the regret of greedy lse seems to grow faster than linear . therefore , we observe that mspsa outperforms greedy lse and the difference between the performance of two policies is getting bigger as @xmath0 increases . [ priceconv_2state ] illustrates how averaged normed squared error between the optimal input and its estimate changes with time in a log - log plot . from theorem [ thm : convergence ] , we know that the optimal - input estimate and thus the input itself converges in mean square for mspsa . in [ priceconv_2state ] , we observe that the convergence of mspsa is consistent with this result . furthermore , the logarithm of the estimation error seems to decrease almost linearly with the logarithm of the time horizon . in other words , mse seems to converge with a rate equal to @xmath198 . this is reasonable because in lemma [ lemma : regret ] , we show that , for each @xmath107 , the estimation error decreases with a rate equal or faster than the inverse of the square root of the number of times state @xmath72 is observed . on the other hand , convergence trend for greedy lse seems to be much slower and it performs poorly compared with mspsa s performance . here , we present an example for revenue maximization problem with system size @xmath199 and 3 different states where the transition probability from any state to the same state was set to be @xmath200 and to any other state to be @xmath201 . we used @xmath197 monte carlo runs to calculate the average performance of both policies . the average performance of mspsa and greedy lse for this example with revenue maximization objective is given in [ multivariate_rev_max ] . the regret growth and the convergence of averaged normed squared error between the optimal input and its estimate are illustrated in [ sqrtregret_3state ] and [ priceconv_3state ] , respectively . in both plots , we observe a trend similar to the previous example even though the regret characterizations and the optimal inputs are different due to different objectives . [ sqrtregret_3state ] shows that mspsa s regret grows linearly with @xmath156 whereas greedy lse s regret grows almost exponentially with @xmath156 . therefore , we observe that mspsa eventually outperforms greedy lse as @xmath0 increases even though greedy lse performs better at the beginning of the time horizon . [ priceconv_3state ] shows that , after a slight increase at the beginning of the time horizon , the logarithm of the estimation error of mspsa decreases linearly with the logarithm of the time horizon and becomes sufficiently small ; whereas the estimation error of greedy lse stays almost constant except the spike that is probably due to the poor initial lses . overall , we can say that mspsa outperforms greedy lse in both numerical examples . we present in this paper an online learning and optimization approach for jump markov affine models with unknown parameters for two different objectives : ( i ) quadratic regulation and ( ii ) revenue maximization . for both objectives , we establish that mspsa achieves the optimal rate of regret growth @xmath11 . our results highlight a change of the minimum order of regret growth from @xmath1 of the classical time - invariant affine models to @xmath11 of the jump markov affine models for the quadratic regulation objective . on the other hand , we show that introducing a markov jump process to the system does not change the optimal rate of regret growth for the revenue maximization objective . we also establish the convergence of mspsa policy to the optimal solution . our simulation results verify that proposed method mspsa can outperform the greedy lse method . let @xmath202 . by mspsa update step given in ( [ eq : mspsaupdate_regulation ] ) and the fact that projection onto @xmath50 maps a point closer to @xmath108 , we have , rcl [ eq : boundt ] _ i , t_i+1 & & _ i , t_i - a_t_i()|_t_i - x_i^*_2 ^ 2 + & = & _ i , t_i -2 ( d_i , t_i^+-d_i , t_i^-)(_i , t_i - x_i^*)^|_t_i + & & + ( d_i , t_i^+-d_i , t_i^-)^2|_t_i^|_t_i . our goal is to bound @xmath203 by simplifying ( [ eq : boundt ] ) . by ( [ eq : risk ] ) , we obtain , rcl + & = & 4c_t_i _ t_i^ _ j p_i , j a_j^(a_j_i , t_i+b_j - y^ * ) + & = & 4c_t_i _ t_i^ ( _ j p_i , j a_j^ a_j)(_i , t_i - x_i^ * ) where last equality is obtained using the foc for @xmath108 . let @xmath204 . using the independence of @xmath93 s , we get , rcl + & = & -8_j p_i , j a_j(_i , t_i - x_i^*)_2 ^ 2 + & & -8_min , i_i , t_i . [ eq : secondterm ] since @xmath50 is compact , @xmath205 where constant @xmath206 . for any @xmath207 , because @xmath208 and singular values of @xmath78 are bounded , @xmath209 where constant @xmath210 . by holder s inequality , we have @xmath211 , @xmath212 , and @xmath213 . then , after simplification , we obtain , @xmath214 where @xmath215 , and rcl + & & 8c_t_i^2 ( _ t_i^ _ j p_i , j a_j^(a_j_i , t_i+b_j - y^*))^2 + & = & 8c_t_i^2 ( _ t_i^ ( _ j p_i , j a_j^ a_j)(_i , t_i - x_i^*))^2 , where last equality is obtained using the foc of @xmath108 . consequently , rcl [ eq : lastterm ] ( ( ) ^2|_t_i^|_t_i|i , t_i,_i , t_i ) & & c_2_i , t_i + , + where @xmath216 and @xmath217 . thus , by expressions ( [ eq : boundt ] ) , ( [ eq : secondterm ] ) , and ( [ eq : lastterm ] ) ; rcl + & & ( 1-a_t_i8_min , i + a_t_i^2c_2)_i , t_i + a_t_i^2 . [ eq : boundforasconv ] consequently , @xmath218 using this result recursively and since @xmath219 for all @xmath220 , we have , rcl e_i , t_i+1 & & ( _ j=1^t_i(1 - 8a_j_min , i+a_j^2c_2 ) ) e_i,1 + & & + _ j=1^t_i(_l = j+1^t_i(1 - 8a_l_min , i+a_l^2c_2 ) ) a_j^2 + & & e^_j=1^t_i ( -8a_j_min , i+a_j^2c_2 ) e_i,1 + & & + _ j=1^t_i e^_l = j+1^t_i(-8a_l_min , i+a_l^2c_2 ) a_j^2 . since @xmath221 and @xmath222 , rcl e_i , t_i+1 & & e^-(t_i+1+n_i)+(1+n_i)+2_i^2 c_2e_i,1 + & & + _ j=1^t_i e^(1+n_i)^-1 + 2_i^2 c_2 a_j^2 + & & + _ j=1^t_i + & & , where @xmath223 , @xmath224 , and @xmath225 . in the proof of lemma [ lemma : regret ] , for any state @xmath107 , we showed that the inequality ( [ eq : boundforasconv ] ) holds where @xmath202 . by theorem 1 of robbins and siegmund @xcite , we know that @xmath226 exists and @xmath227 almost surely ( a.s . ) . since @xmath228 , we obtain that @xmath229 let @xmath230 be the indicator function . given @xmath137 and @xmath117 , @xmath134 is equal to either @xmath135 or @xmath136 . hence , @xmath231 . if state @xmath72 is recurrent , @xmath232 because @xmath117 is greater or equal to half of the number of times state @xmath72 is occurred up to @xmath23 . therefore , for a recurrent state @xmath107 , rcl + & = & ( _ t 1_i(s_t-1)x_t^-x^*_i^2_2 = 0 | _ t t_i = ) + & & ( _ t ( 2_i , t_i+2(_i)^2n_1 ^ 2 t_i^-1/2 ) = 0 | _ t t_i = ) + & = & ( _ t_i _ i , t_i=0)=1 . so , for a recurrent state @xmath72 and for any @xmath233 , we have , rcl [ eq : limrecurrent ] _ t ( 1_i(s_t-1)x_t^-x^*_i^2_2 > t t ) & = & 0 . + if a state @xmath107 is transient , then for any @xmath233 , we have , rcl + & & _ t ( s_t-1=i t t ) = 0 , [ eq : limtransient ] where last equality is due to borel - cantelli lemma and the fact that @xmath234 for a trasient state @xmath72 . by definition , expression ( [ eq : asconv ] ) holds , if and only if , for every @xmath235 , @xmath236 . any state @xmath107 is either recurrent or transient . hence , by ( [ eq : limrecurrent ] ) and ( [ eq : limtransient ] ) , we obtain that , for any @xmath235 , rcl + & & _ t _ i=1^k ( 1_i(s_t-1)x_t^-x^*_i^2_2 > t t ) + & = & 0 . since ( [ eq : asconv ] ) holds and @xmath237 where @xmath238 , by lebesgue s dominated convergence theorem , @xmath239 let the transition probability from any state to any other state be @xmath240 . without loss of generality , take @xmath241 . let @xmath28 be i.i.d . with distribution @xmath242 which is independent of the state . because additional observations ca nt increase the growth rate of regret for an optimal policy , we assume that the decision maker receives the observation values corresponding to the input @xmath55 from all other states that did nt occur at time @xmath23 as additional observations at time @xmath23 . hence , at each @xmath23 , the decision maker gets observations from the affine functions of all states for input @xmath55 . let s define @xmath179 , @xmath243 , and @xmath28 as @xmath244 where @xmath245 denotes the system noise of observation from state @xmath72 . now , for any policy @xmath51 , we can express the observation vector at @xmath23 as @xmath246 observe that foc for the optimal input @xmath82 at time @xmath23 obtained from minimizing ( [ eq : risk ] ) is the same for any state @xmath247 for our fixed choice of @xmath17 . hence , we drop the dependence on the previous state @xmath38 along with @xmath17 and denote it as @xmath248 , @xmath249 , to express the dependence on @xmath40 . with the new notation , foc can be expressed as @xmath250 consequently , the optimal price given in ( [ eq : optprice ] ) becomes @xmath251 let s express @xmath252 as @xmath253^{\tra}$ ] where @xmath254 is the @xmath72th entry of @xmath255 and @xmath256 is the @xmath72th row vector of @xmath64 . we fix a compact rectangle @xmath58 such that , for any @xmath63 , @xmath248 is contained in @xmath50 and @xmath64 is full column rank for all @xmath18 . can be shown by the continuity of @xmath248 on a compact rectangle @xmath257 which satisfies @xmath64 to be full column rank for all @xmath18 and for any @xmath258 , and contains a fixed point @xmath259 in its interior for which @xmath260 is in the interior of @xmath50 . the existance of @xmath257 can be shown by using the continuity of the determinant of @xmath261 for each @xmath18 at the fixed point @xmath259 . ] since @xmath17 and @xmath41 are already fixed , our goal is to obtain the performance of the worst - case system parameter @xmath40 that is chosen from the set @xmath62 . applying implicit function theorem on @xmath262 gives rcl [ eq : diffopt ] = -()^-1 = - ( a^a)^-1 , and by calculus , we have , rcl [ eq : difffoc ] ( ) ^ & = & ( ax^*()+b ) _ n^ + _ n + a 1 + x^ * ( ) . let @xmath263 . density of the output vector up to time @xmath23 given the parameter vector @xmath40 and input vector @xmath264 can be written as @xmath265 by writing the joint distribution as a product of conditionals and by the conditional independence of the input for any policy @xmath51 from the parameter @xmath40 given the information history vector @xmath45 , we get , @xmath266 by using the mixed product property @xmath267 and the independence of @xmath268 , we obtain the fisher information for @xmath269 as rcl i_t^ ( ) & = & ( ^ | x^t , ) + & = & i_m ( _ i=1^t 1 + x_i^ 1 , ( x_i^)^ ) . [ eq : fisher ] now , we choose a prior distribution @xmath270 as an absolutely continuous density on @xmath62 taking positive values in the interior of @xmath62 and zero on its boundary . we choose @xmath179 and @xmath243 to be independently distributed with distributions @xmath271 and @xmath272 , respectively , so that @xmath273 . take @xmath274 . now , we use the multivariate van trees inequality in a similar way in . this inequality can be expressed as rcl [ eq : vantreesineq ] ( ^_t - x^*()_2 ^ 2 ) & & + where the expectation operators are also taken over the prior distribution @xmath270 and @xmath275 is some constant given @xmath270 , which can be seen as the fisher information for the distribution @xmath270 . by ( [ eq : fisher ] ) , we have , rcl ( c()i^_t-1()c()^ ) & = & _ i=1^t-1x_i^-x^*()_2 ^ 2 + & & c_0 _ i=1^t-1x_i^-x^*()_2 ^ 2 [ eq : denom ] where @xmath276 . let s define @xmath277 . since @xmath278 is symmetric positive definite , by ( [ eq : diffopt ] ) and ( [ eq : difffoc ] ) , we obtain rcl ( c()^ ) & = & ( -c()^ ( a^a)^-1 ) + & = & - b^ ( ax^*()+b ) ( ( a^ a)^-1 ) + & = & - b^ p b ( ( a^a)^-1 ) . + by singular value decomposition ( svd ) of @xmath179 , observe that @xmath279 where @xmath280 is an orthogonal matrix , and @xmath281 denotes a diagonal matrix with diagonal entries @xmath282 . hence , @xmath17 is symmetric positive semidefinite . also observe that @xmath283 . then , we can bound the numerator term , rcl [ eq : num ] ( ( ( ( ) ^)))^2 & & ( ( b^p b))^2 . observe that @xmath284 is symmetric positive semidefinite and nonzero for @xmath154 ( or @xmath175 and @xmath174 ) since @xmath285 . hence , there exists some direction @xmath286 such that @xmath287 , and , consequently , there exists some distribution @xmath272 such that @xmath288 . more specifically , if @xmath289 for some choice of @xmath272 , we can change that choice of @xmath272 to shift the mean of @xmath243 slightly in the direction of @xmath290 , and have @xmath288 . by independence of @xmath243 and @xmath17 , rcl ( b^ p b ) & = & ( b)^ |p ( b ) + ( ( b-(b))^ p ( b-(b ) ) ) + & & ( b)^ |p ( b ) . [ eq : numineq ] hence , by expressions ( [ eq : vantreesineq ] ) , ( [ eq : denom ] ) , ( [ eq : num ] ) , and ( [ eq : numineq ] ) ; rcl _ t=2^t ( ^_t - x^*()_2 ^ 2 ) & & _ t=2^t + & & _ t=2^t , + [ eq : vantrees ] where @xmath291 and @xmath292 . since @xmath293 , by ( [ eq : vantrees ] ) , we have , @xmath294 let @xmath295 denote the @xmath296th entry of @xmath248 , and , by extreme value theorem , @xmath297 and @xmath298 are attained . since @xmath248 is not a constant over @xmath62 ( otherwise @xmath299 would be zero for all @xmath63 , and left hand side of ( [ eq : num ] ) would be zero for any @xmath270 which is a contradiction ) , @xmath300 . for any policy @xmath51 , @xmath301 . hence , we have , @xmath302 where @xmath303 . we will follow the steps in lemma [ lemma : regret ] and simplify inequality ( [ eq : boundt ] ) . by ( [ eq : risk_revenue ] ) , we obtain , rcl + & = & -2c_t_i _ t_i^ _ j p_i , j ( ( a_j+a_j^ ) _ i , t_i+b_j ) + & = & - 2 c_t_i _ t_i^ ( _ j p_i , j(a_j + a_j^ ) ) ( _ i , t_i - x_i^ * ) , where last equality is obtained using the foc for @xmath108 . let @xmath304 . using the independence of @xmath93 s , we obtain , rcl + & = & 4(_i , t_i - x_i^*)^ ( _ j p_i , j ( a_j + a_j^ ) ) ( _ i , t_i - x_i^ * ) + & & -8_min , i_i , t_i . as in lemma [ lemma : regret ] , @xmath305 . hence , for any @xmath79 , @xmath306 . since @xmath307 , @xmath308 where constant @xmath309 consequently , rcl + & & 2c_t_i^2 ( _ t_i^ ( _ j p_i , j(a_j+a_j^))(_i , t_i - x_i^*))^2 + 2c_1 , and thus , we obtain ( [ eq : lastterm ] ) where @xmath310 and @xmath217 . therefore , ( [ eq : boundforasconv ] ) holds and the rest of the proof is the same as in lemma [ lemma : regret ] . the inequality given in ( [ eq : tradeoffbound_revenue ] ) is used to obtain ( [ eq : inputmsebound_revenue ] ) and ( [ eq : regretbound_revenue ] ) as in theorem [ thm : lower_regret ] . the proof of inequality ( [ eq : tradeoffbound_revenue ] ) follows the proof of theorem [ thm : lower ] with some slight modifications to bound the numerator term of the van trees inequality due to revenue maximization objective . the foc for @xmath248 becomes @xmath311 and the optimal price @xmath312 . for this problem , the compact rectangle @xmath62 is such that , for any @xmath63 , @xmath248 is contained in @xmath50 and @xmath64 is negative definite for all @xmath18 . by using the continuity of the maximum eigenvalue of @xmath64 rather than the determinant . ] the implicit function theorem on @xmath262 gives @xmath313 where @xmath314 we take @xmath272 such that @xmath315 . consequently , the numerator term of the van trees inequality can be bounded as

we present a new measurement of the deuterium - to - hydrogen ratio ( d / h ) in the lyman limit absorption system at towards q19371009 . tytler , fan & burles ( 1996 ; hereafter tfb ) made the first extragalactic detection of deuterium in this absorption system , which remains the best location for a high accuracy measurement of primordial d / h . their detailed analysis of keck spectra gave a low value of d / h = ( statistical & systematic errors ) . now we present a new method to measure d / h in qso absorption systems . we avoid many of the assumptions adopted by tfb ; we allow extra parameters to treat the continuum uncertainties , include a variety of new absorption models which allow for undetected velocity structure , and use the improved measurement of the total hydrogen column density by burles & tytler ( 1997a ) . we find that all models , including contamination , give an upper limit d / h ( 95 % confidence ) . both this and previous analyses find contamination to be unlikely in this absorption system , a analysis in models without contamination gives d / h ( 67% confidence ) , which is higher but consistent with the earlier results of tfb , and a second measurement of d / h towards q1009 + 2956 ( tytler & burles 1997 ) . with calculations of standard big bang nucleosynthesis ( sbbn ) and the assumption that this measurement of d / h is representative of the primordial value , we find a high baryon to photon ratio , . this is consistent with primordial abundance determinations ofhe in h ii regions ( izotov et al . 1997 ) andli in the atmospheres of warm metal - poor population ii stars ( ) . we find a high value for the present - day baryon density , , which is consistent with other inventories of baryonic matter , from low to high redshift : clusters of galaxies , the lyman alpha forest & the cosmic microwave background .

astro-ph9712108

standard big bang nucleosynthesis ( sbbn ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) predicts that a measurement of the primordial ratio d / h will give the most sensitive constraint on the cosmological baryon to photon ratio , @xmath11 . using the photon density measured from the temperature of the cosmic microwave background ( cmb ) , we can then obtain the most sensitive constraint on @xmath12 , the cosmological baryon density in units of the critical density , where @xmath13 mpc@xmath14 . both @xmath11 and @xmath15 are fundamental cosmological parameters , and measurements of d / h in qso absorption systems can determine both to 10% accuracy . when we compare measurements of the primordial abundances of different light elements , h , d and @xmath7he in particular , we test the sbbn theory predictions . measurements of d / h in different astrophysical sites also constrains the history of star formation in these sites , once the primordial value of d / h is established , because d is totally destroyed ( astrated ) as gas is cycled through and ejected from stars ( @xcite ) . adams ( 1976 ) first suggested that deuterium could be measured in intergalactic gas clouds through lyman absorption lines in the spectra of distant qsos . deuterium lyman absorption lines are 82 on the short wavelength side of the corresponding hydrogen lyman lines , but adams notes that d can be detected in only very select absorption systems with simple velocity structure and high neutral hydrogen column density , n(h i ) @xmath16 2 ( lyman limit systems ) . for high redshifts , @xmath17 , the lyman series is redshifted into the optical , and the lines are accessible to large ground based telescopes . the lyman absorption lines of d and h , and the lyman continuum absorption of h i , constrain the column densities of d i and h i , and provide a measurement of d i / h i. the timescale for h and d ionization equilibrium is very short ( @xmath18 yr ) in the photoionized intergalactic medium ( @xmath19 ) , and we assume the ratio d i / h i is identical to d / h throughout this paper ( @xcite ) . the lyman limit system at @xmath20 towards q19371009 is an ideal site to infer a primordial value for d / h . the system is very metal - poor , less than 1/100 solar . deuterium is destroyed as gas is cycled through stars , but metals are produced in the cycle and the system s low metallicity limits the amount of deuterium which could have been destroyed . the gas in the system also has very low internal velocities , which can limit the amount of kinetic energy input through any high energy phenomena , such as supernovae or gamma ray sources . in tfb , we presented high resolution spectra of q19371009 , and made the first measurement of d / h in a qso absorption system . this system has a high total neutral hydrogen column density , log n(h i ) @xmath21 2 ( burles & tytler 1997a ) , and the corresponding absorption is optically thick throughout the entire lyman series . the deuterium feature is well determined by its profile in ly@xmath22 , and the column density of d is well constrained . in this paper , we advance the methods of tfb to place more robust constraints on d / h in this absorption systems . in section 2 , we describe our new method of constraining d / h , and in section 3 , we discuss the new results . in section 4 , we calculate the abundances for a number of metals and discuss the ionization state of the system . in section 5 , we compare our d / h value to the abundance of d in other sites , and other light elements , and we show that our d / h is probably primordial . we model the absorption spectrum along the qso line of sight as a finite number of discrete absorbing components . each component is modeled as a voigt profile ( spitzer 1979 ) , given by three parameters : column density ( n ) , redshift ( @xmath23 ) , and velocity dispersion ( @xmath24 ) . at every wavelength , we sum the optical depth contribution of all absorption line profiles in the model . the optical depth is converted to a normalized flux , and convolved with the instrumental response . the model spectrum can then be directly compared to the observed spectrum . we use the observed spectrum described in tfb . the tfb analysis was sophisticated since it was the first to simultaneously fit many metal and h lines , and to separate thermal and turbulent contributions to line widths . however , it also used many assumptions which we now remove . in tfb , we modeled the absorption system with two d / h components , labeled blue " and red " . the strong feature also required a third h i component ( which contained only 1% of the hydrogen column ) to obtain an adequate fit to the spectrum of the lyman series . the positions of the two main components were tied " to the velocity positions determined from the narrow metal lines which were asymmetric , and well described by these two components . we fit the model to narrow regions of the spectrum containing only the lyman lines of interest , and did not include absorption from other lines along the line of sight . d / h was assumed to be equal in both components . the velocity dispersion of the red d line was determined from the dispersions of the corresponding h and metal lines in the red component . we found the value of d / h which gave the best fit to the data and the formal @xmath2 errors from the d and h column densities , log(d / h ) = 4.64 @xmath25 0.06 . we investigated systematic errors due to incorrect continuum placement in the regions of and the lyman limit , and estimated the systematic error @xmath26 log(d / h ) = 0.06 . the tfb analysis found a consistent model and a well - defined value for the most likely value of d / h . but its limitations lay in the assumptions and the calculation of confidence levels and the total errors on d / h . in this paper , we present major improvements in our analytic techniques . we measure d / h with a variety of models and free parameters . we add new free parameters to describe the unabsorbed quasar spectrum , which we use to normalize the absorption spectrum . previously , we treated continuum uncertainties as systematic errors . the number and velocities of the d / h components are now also free parameters . we no longer require that the d and h lines have the same velocities as the metal lines . the d and h line profiles are constrained only by the spectra covering the lyman series lines . to include all the uncertainties from the modeling into the final uncertainty of d / h , we perform a @xmath4 analysis as a function of d / h . formerly , we calculated the errors from a quadrature sum of the total column density uncertainties of d and h. by calculating @xmath4 as a function of d / h , we can include the uncertainties from all the free parameters in our models , including the velocity dispersions , velocity positions , continuum levels , and column densities . the @xmath4 function gives the relative likelihood of all values of d / h versus the best fit value . the final confidence region can be directly calculated from the @xmath4 function , and can be considered more comprehensive " than the uncertainties which we reported before . the most critical addition to the present analysis is a measurement of the total hydrogen column density , . from our detailed study of the lyman continuum optical depth ( burles & tytler 1997a ) , we measured = 17.86 @xmath25 0.02 , and we use this constraint for all models . is much better constrained from the lyman continuum absorption than from the lyman series line profiles , which gave = 17.94 @xmath27 ( tfb ) , where the first error is statistical and the second is systematic . the new provides a very strong constraint for the models , allows us to add more free parameters to the models to test a variety of models and minimize the model fit with respect to d / h . here we discuss the new models we have adopted to measure d / h . the absorbers in each model can be placed into one of two groups , depending on their association with the d / h absorption system ( dhas ) . the first group contains h i absorbers which will not show d i ( also labeled unassociated " ) , and the second group contains the high column h i absorbers which show d i and are associated with the dhas . we refer to absorbers in the second group as the `` main components '' . for each unassociated absorber , we add three free parameters to the model : one each for n , @xmath24 , and @xmath23 . each main component shows two lines : one of h i and the other d i. we also add one parameter for each line to allow a separation of @xmath24 into t and @xmath28 , because in general @xmath24(h i ) can differ from @xmath24(d i ) , but we tie three of the parameters in the two lines : @xmath29 , t(h i ) = t(d i ) , and @xmath28(h i ) = @xmath28(d i ) . in this paper , we do not use the metal lines to constrain d / h . the velocity dispersions of the main components are given by @xmath30 , where @xmath31 is temperature in units of @xmath32 k , @xmath33 is the atomic mass , and @xmath24 is in units of . for zero turbulent velocity dispersion , we recover the thermal relation for the h i and d i lines , @xmath34 . in all models we require that @xmath35 5000 k in the main components . without this constraint , models which fit the data may include unobserved components with very low t. the background ionizing radiation is hard enough that the absorbing gas is unlikely to be cooler than a few 10@xmath7 k. the results depend on assumptions we make when constructing the models . these assumptions include the number of d / h components , the total number of free parameters , and the spectral regions which we fit . the total number of free parameters is the sum of parameters of all the absorption components and the free parameters included in the continuum . for this analysis , we have chosen 7 models , and we measure d / h for each . we can then compare the best fit values of d / h for each model , and assess the effect of the different assumptions used in each model . all 7 models include the spectral regions listed in table 1 , the number of d / h components chosen for each model , and the 64 h i absorbers listed in table 2 . the models are of three types . models 1 & 2 are the simplest and include 2 and 3 d / h components , respectively , with no free parameters in the continuum . the continuum is held fixed at the initial placement of the unabsorbed continuum level , which is shown as the solid lines at unity in figure 2 . the other five models allow for free parameters in the continuum in each region . the number of allowed continuum free parameters in each region is shown in table 2 . the final two models have three d / h components , but also allow for hydrogen contamination . in model 6 , an extra hydrogen absorber is introduced at redshift @xmath36 , which places h i ly@xmath22 at the position of d - ly@xmath22 , and h i ly@xmath37 at d - ly@xmath37 . in model 7 , the extra hydrogen absorber is introduced at @xmath38 , which places its at the position of d - ly@xmath37 , in the blue wing of . the model parameters are summarized in table 3 . we use the levenberg - marquardt method to minimize @xmath4 ( press et al . the details of the algorithm and computational techniques are presented in burles ( 1997 ) . the algorithm iterates until it converges on the best fit model . the @xmath4 of the final fit is assumed to be the minimum @xmath4 , @xmath39 . a given model will have @xmath40 free parameters , and the algorithm calculates a @xmath41 covariance matrix for each iteration . the 1@xmath42 uncertainties can be directly calculated from the diagonal elements in the covariance matrix of the final iteration . in all the models , we are interested in one parameter specifically , d / h . it is not straight forward to calculate the total uncertainty in d / h from the final covariance matrix directly , even if the errors are normally distributed . due to the intrinsic blending of the h i and d i lines , the formal errors in column density are correlated . instead , we choose to make d / h a free parameter in the models , and calculate as a function of the parameter d / h . we construct a list of d / h values , and model the spectrum for each value in the list . by designating d / h in each model , we reduce the number of free parameters , @xmath40 , by one for each d / h component in the model . now each main component ( components which have d absorption ) has only four free parameters : n(h i ) , @xmath31 , @xmath28 , and @xmath23 . n(d i ) is no longer free , but is given by n(d i ) = n(h i ) @xmath43 ( d / h ) . we assume that d / h is the same in all main components . by constructing an array of as of function of d / h , we can calculate the most likely value of d / h and the confidence levels surrounding this value . we have effectively taken a cross section of the @xmath40-dimensional @xmath4 function to calculate the uncertainties of the single parameter , d / h . the one - dimensional function , @xmath44 , yields both the most likely value of d / h , and confidence levels about the most likely value . figure 1 shows the spectral regions stacked in velocity space . figure 2 presents each spectral region separately on the vacuum heliocentric wavelength scale . the unabsorbed qso continuum is now allowed to vary to achieve the best fit . the qso continuum is modeled by a low - order legendre polynomial , which accounts for smooth variations in the continuum on scales of 100 ( dashed lines in fig . 2 ) . in the data reduction process , we use legendre polynomials to initially normalize the spectrum to the regions of the spectrum showing no absorption ( solid line at unity in fig . 2 ) . in the forest , the continuum is not well defined due to large regions of continuous absorption . we account for this uncertainty in the continuum in the forest by allowing the coefficients of the legendre polynomials to be free parameters , without bounds , in our models . therefore , the continuum is no longer fixed to the level on which we can only speculate . in statistical language , we fit the continuum level with a set of nuisance parameters " , which we allow to vary freely to improve the fit , because they are not of primary interest . we must choose regions of the spectrum to compare to the models . if we model the entire keck spectrum , the model would include over 1000 lines and over 40,000 pixels . for practical purposes , we can not model the entire spectrum simultaneously , the time required to complete the fitting procedure scales as the number of lines times the number of pixels . also , we want our model to be sensitive to d / h , and not other unrelated features in the spectrum . therefore , we select regions of the spectrum which include at least one lyman series line of the dhas . in principle , we want to include all lyman series of the dhas , but some lyman series lines are blended with other strong , unassociated absorption lines , and are not included in the selected regions . in figures 1 & 2 , we show the regions of the spectrum used in the model fitting . table 1 lists the regions used in our analysis , including the number of pixels and the order of the legendre polynomial used to model the unabsorbed continuum in each region . the damping wings of the feature absorb over a large wavelength range , therefore the span of the region is much greater than the other individual line regions . for the higher - order lyman lines , the regions of interest begin to overlap . the overlapping regions are combined into a single region containing multiple lyman lines and this region is labeled ly - limit " . of the 9 regions , the region ( figure 2a ) has the highest snr per pixel , approximately 75 per 4 pixel . thus the entire fit is heavily weighted by ly@xmath22 , and the fit is most tightly constrained by the data is in this region . deuterium can be seen at 5557 , and is composed of 3 components in the model presented here . the fitted continuum contains 5 free parameters , and is reasonably consistent with our initial estimate of the continuum . only at the red edge does it begin to diverge significantly . and the overlap between the estimated and the best fit continua lies directly at the feature . the overlap in this region suggests that the new continuum determination will have little effect in the region . the region ( figure 2b ) is much smaller than ly@xmath22 . the flux returns to the continuum much closer to the center of the ly@xmath37 line , and there is no reason to include a larger section around ly@xmath37 . again , the best fit continuum lies close to the original continuum estimate over the entire region . the model fit to is not as good as ly@xmath22 , there is under - absorption in two places near d- . the blue wing of shows under - absorption , but there is a significant increase in the noise over some of these pixels ( @xmath45 ) , which is most likely due to bad columns being rejected in the original ccd images . the under - absorption on the blueward side of d - is likely due to a hydrogen line ( which is fit in model 7 ) that is blended with the deuterium feature . we label hydrogen lines which overlap and blend with the deuterium features contaminating " hydrogen . contamination will be discussed thoroughly in section 3.1 , and introduced into our models 6 & 7 . the models must produce a good fit to all the lyman lines simultaneously . we show each of the regions separately in fig . 2 to allow a close inspection of the data and model , and to display how well the model reproduces the observed spectrum . notice that there are additional absorbers overlapping the main components . these extra absorbers add free parameters to the fit , and in general , make the model less restrictive . that is , the parameters determining the main components can not be as tightly constrained when overlapping absorbers are included . in figure 2i , we show the spectral region with the highest order lyman lines , ly-12 to ly-19 . the best fit continuum shows a significant difference from the initial estimate . the unabsorbed continuum in this region was difficult to estimate due to the lack of pixels with little or no absorption . the continuum was originally estimated as a constant flux level passing near the few pixels with the highest flux . the large amount of absorption in this region did not allow for a better estimate of the unabsorbed continuum initially . the difference between the best fit continuum ( dotted line ) and the initial continuum ( solid line ) is likely due to the simple approximation of the continuum level during the reduction . the shape of the best fit continuum is more likely and is a function of the instrumental sensitivity in this region . there are several regions of the spectrum shown in figure 2 of tfb which have more flux than expected by their model fit . in the models presented here , this unabsorbed flux is accounted for with the extra hydrogen components and a different continuum . figure 3 shows the major results of the fitting procedure with the seven models . table 3 summarizes the d / h measurements for each model . we performed the @xmath4 minimization procedure for 100 values of d / h over the range @xmath46 log(d / h ) @xmath47 for each of the seven models . for example , figure 3a shows the results for the 2 component model ( model 1 ) . the solid dark gray line represents one of the components and the dashed line represents the other . these graphs show the behavior of the main component parameters as a function of d / h . the parameters sometimes display discontinuous behavior , which is due to the non - uniqueness of voigt profile fitting . the @xmath4 minimization has found two equally likely solutions , and this exhibits itself with a discontinuous change in the parameter values . although the fitted parameters may be non - unique , remains a smooth function of d / h . the exact parameters which give the best fit are not well - determined , but the , and the d / h are well defined . in this analysis , we can not attempt to model the dhas exactly , but we can still find the relative likelihood for all values of d / h . we can not model the individual components exactly because of the intrinsic blending of the absorption lines . but we do find that parameters which include all the components can be measured precisely . in figure 3 , the thick solid line in the column density plot represents the total neutral hydrogen column , . this is merely the sum of the column densities of all components in a given model at a specific value of d / h . we find that all seven models give a which is a smooth function of d / h , even though the individual column densities in the models are not . is the quantity subject to the constraint provided by burles & tytler ( 1997a ) , = 17.86 @xmath25 0.02 . for a model which gives an arbitrarily good fit to the spectral regions , the best fit will depend only on the agreement with the constraint . we can measure d / h directly from figure 3 . with a @xmath4 function of one variable , the confidence levels are easily determined . the 95% confidence levels correspond to @xmath48 , where @xmath49 . for each model in figure 3 , we find the minimum of the @xmath4 function and list this value in table 3 . the 95% confidence levels on d / h are also listed in table 3 , and are shown as vertical dashed lines in figure 3 . the fitting procedure is not perfect , and we can see in figure 3 that there are deviations from perfectly smooth @xmath4 functions . the fitting procedure can converge before reaching the absolute minimum @xmath4 , and we estimate that the difference between the absolute minimum and the calculated @xmath4 can be as large as 0.5 . we take this deficiency into account by setting the 95% confidence levels at @xmath50 . all of the regions of 95% confidence in figure 3 overlap with the central value of constraint . the best fit model of d / h must not only provide the best fit to the spectral regions , but must also fit the constraint of . for all models , the values of decrease smoothly with increasing d / h . n(d i ) is tightly constrained by d - and d- , so as d / h increases , n(h i ) must decrease , which gives smoothly varying functions of for each model . in all models , the 95% confidence regions on d / h represent the model fits where the constraint is well satisfied . all seven models give consistent ranges for the 95% confidence regions , as seen in figure 4 . as expected , the models with more free parameters have larger confidence regions . except for model 6 , which includes hydrogen contamination at ly@xmath22 , all models are consistent with @xmath51 or @xmath52 at 67% confidence . the shape of the functions indicate errors which are normally distributed in log ( d / h ) , and the 67% confidence levels should be one - half of the 95% levels . in models 6 & 7 , we investigate the effects of hydrogen contamination of the d - lyman lines . model 6 includes an additional hydrogen absorber at redshift , @xmath53 , which places its lines near the deuterium lines . figure 3f shows the results of the fitting procedure . for values of log d / h @xmath54 , the results are identical to model 4 , which did not include contamination . but for lower values of d / h , the contaminating hydrogen absorbs a significant amount of flux at the d - and d - lines , and gives a good fit by keeping nearly constant and lowering n(d i ) . although contamination allows for lower d / h as expected , the introduction of an additional absorber with three free parameters does not improve the fit significantly . this result is in sharp contrast to another d / h system , towards q1009 + 2956 , where contamination significantly improves @xmath4 ( burles & tytler 1997b ) . we conclude that contamination is not significant in this d / h system for two reasons : ( 1 ) additional parameters for contamination do not improve the @xmath4 fit , ( 2 ) the likelihood of significant contamination drawn from distributions of the forest ( kirkman & tytler 1997 ; lu et al . 1997 ) is small ( c.f . tytler & burles 1997 , tfb , @xcite , steigman 1994 ) . model 7 includes contamination at the position of d- , at @xmath55 . this redshift is much lower than the dhas , and the hydrogen absorber only affects d- . the extra absorber is initially placed at the center of d- , and is allowed to freely move to achieve the best fit to the data . the fit improves by @xmath56 , as the hydrogen absorber fills in the under - absorption seen in figure 2b . the best fit for model 7 is shown in figure 5 , and the improvement is easily seen upon comparison with figure 2b . the model shows under absorption in three pixels near 4689 (-60 km / s blueward of h- ) , but these pixels fall in a region of spectra with increased noise . in conclusion , we find model 7 is consistent with the results from the other models which did not include this contamination and the measurement of d / h is robust with any extra hydrogen absorption at alone . the presence of contamination will give an overestimate of d / h . therefore , the upper limit on d / h is robust and not affected by the presence of contamination . all seven models are consistent with the upper limit , d / h @xmath3 . the metal absorption lines associated with the dhas were analyzed by tfb . we use the two component fit ( @xmath57 ) of tfb to measure the column densities of the ions shown in table 4 . if no feature is detected at the expected positions of the ionic transitions , then we place a 2@xmath42 upper limit on the total column density . more absorption components could be used to describe the metal lines in this dhas , but the qualitative results will remain unchanged . we used the program cloudy ( ferland 1993 ) , with an ionizing background spectrum calculated by haardt & madau ( 1996 ) , to calculate the ionization state of the gas in each component . table 5 shows the metallicities , temperatures , ionization parameter , and total hydrogen density in each component calculated from the cloudy simulations . the equilibrium temperatures calculated from the cloudy simulations can be directly compared to the temperatures determined from the velocity dispersions of the h , c and si lines . in the blue component , we find a good agreement between the temperatures shown in table 4 . but in the red component the cloudy equilibrium temperature is smaller than the temperature determined from the relative line widths , including @xmath28 . this may indicate another component near the velocity position of the red component . the cloudy simulations did not give an acceptable fit to all ions of si in the red component . a single phase of photoionization could not account for the three column densities observed , and the calculated [ si / h]@xmath58 came from si ii and si iii alone . we excluded si iv for two reasons : ( 1 ) si iv(1393 ) is blended with a c iv(1550 ) at @xmath59 , and ( 2 ) it is likely that there is another component of higher ionization that contributes to si iv . we have not attempted to model the metal lines with more than two components . although the column densities of the individual components would differ , the derived metallicities of the components would not greatly differ from the results shown in table 5 . all components would have a metallicity below 1/100 solar , independent of the model used . we can now test the tfb assumption that the h and d lines were at the velocities of the c and si metal lines . in general , this assumption adds a systematic error to the measurement if the metal lines are not aligned with the h i and d i lines . this dhas is rather unique , because all the metal lines , with different ionizations , show similar velocity structure . in figure 3 , the velocity positions of the blue and red components used by tfb lie at @xmath60 respectively . in our models with three or four components , two of the components lie near the velocity positions of the metal lines , with considerable variation , depending on d / h . but in the two component models , the component separation for the most likely d / h is 19 , rather then 15 , and the red hydrogen component is at a higher velocity position for all values of d / h . this shows that a systematic offset in the red component was introduced in the analysis in tfb by assuming that the h and d lines fell precisely at the metal line positions in the two component model . it is difficult to translate this velocity offset into a d / h change because , as seen in figure 3 , many other coupled parameters are involved . we have measured the deuterium abundance in the dhas at @xmath61 towards q19371009 . within the statistical uncertainties , we conclude that this is the primordial value of d / h produced by big bang nucleosynthesis ( reeves et al . 1973 ; @xcite ) . this absorption system is young ( @xmath62 gyr ) and very metal poor ( [ x / h ] @xmath63 ) . these characteristics place strong constraints on stellar astration of deuterium independent of initial mass functions and star formation rates ( jedamzik & fuller 1997 ; @xcite ) . to astrate significant amounts of deuterium , high mass stars would overproduce cno and si , and low mass stars would not have time to complete their evolution . with no post - bbn processes to create or destroy deuterium , our measured value of d / h must represent the primordial value . due to its importance in the standard cosmological model , d / h has been measured in many astrophysical sites . in the local interstellar medium , piskunov et al . ( 1997 ) find a mean abundance ratio , @xmath64 , in the lines of sight toward nearby stars . chengalur et al . ( 1997 ) have recently detected the d i hyperfine transition in emission toward the galactic anticenter , and find d / h @xmath65 . abundance measurements in meteorites , the lunar soil , and the atmospheres of the jovian planets are consistent with a pre - solar value , @xmath66 ( @xcite ; @xcite ; @xcite ) . in the next section , we use the local measurements of d / h with our measurement at high redshift to discuss the evolution of d / h . tytler & burles ( 1997 ) have made another measurement of d / h in a high - redshift lyman limit system towards q1009 + 2956 , at @xmath67 . preliminary analysis of this object gives d / h = 3.0 @xmath68 , and simulations of hydrogen contamination suggested the most likely value of d / h = 2.5 @xmath69 ( statistical and systematic error ) . the measurement towards q1009 + 2956 agrees very well with the result obtained in this paper towards q19371009 . an analysis of q1009 + 2956 with the methods presented here will be presented in burles & tytler ( 1997 ) . there have been other reports of deuterium detections in high - redshift qso absorption systems : towards q0014 + 8118 ( @xcite ; @xcite ; @xcite ; @xcite ) , towards q12020725 ( @xcite ) , towards q0420388 ( @xcite ) . these other systems do not yield measures of d / h due to either lower quality spectral data , or greater complexities in the velocity structure of the absorption system . the next best candidate system to measure d / h , @xmath70 towards q0014 + 8118 , was shown to have an interloping hydrogen cloud within 10 of the expected position of deuterium ( tytler et al . webb et al . ( 1997 ) recently deduced a d / h value at @xmath71 towards the low redshift qso 1718 + 4807 using a spectrum obtained with the hubble space telescope ( hst ) . unlike q19371009 and q1009 + 2656 , only one h i line was observed , so the velocity structure of the h i is not well known . assuming a single component fit to the , they find d / h = @xmath72 , but this value will remain suggestive unless confirmed with approved hst observations of the high - order lyman lines . many groups have discussed the significance of deuterium in galactic chemical evolution , and have used simple models to calculate the abundance ratio , d / h , as a function of time and metallicity ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . in the past , deuterium evolution was required in order to extrapolate the ism and pre - solar nebula measurements of d / h back in time , and back to zero metallicity , to constrain the primordial d / h value . but now the problem can be inverted , we can use the values of primordial d / h , pre - solar d / h , and the ism d / h to constrain models of chemical evolution , deuterium astration as a function of time and metallicity , and the fraction of baryons which have been cycled through stars in the solar neighborhood . the ratio of the ism and primordial values gives an astration factor in the local ism , @xmath73 so we can make the simple statement that one - half of the gas in the local ism has been cycled through stars . this fraction can be used to place new limits on the stellar initial mass function and the amount of inflow and/or outflow in the local ism . when one considers published models of chemical evolution in the ism , the amount of deuterium astration agrees well with models incorporating standard star formation and some infall of primordial material ( @xcite ; @xcite ; @xcite ) . production of the light elements from standard big bang nucleosynthesis ( sbbn ) depends on a single parameter , @xmath11 , the baryon to photon ratio during sbbn . many groups have calculated the abundance yields of the light elements as a function of @xmath11 ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . the abundance yield of deuterium is a single - valued function of @xmath11 , and using output from the kawano bbn code ( @xcite ) , we find @xmath74 where the first error is the statistical uncertainty in the d / h measurement , and the second error represents the statistical uncertainty in the nuclear cross - sections ( @xcite ) . it is interesting to note that the magnitude of the measurement errors are comparable to the errors in the sbbn calculations . if we use the present - day photon density determined from the cobe firas measurements of the cosmic microwave background ( cmb , @xcite ) , we can directly calculate the present - day baryon density @xmath75 where the error includes both uncertainties in @xmath11 . this value for the baryon density agrees with recent estimates of @xmath76 from observations of the cmb power spectrum , the intergalactic medium at @xmath77 , and rich galaxy clusters at @xmath78 the power spectrum of anisotropies in the cmb determines @xmath15 at the epoch of decoupling , @xmath79 ( hu & white 1996 ) . the locations and heights of the doppler peaks in the cmb power spectrum provide constraints on @xmath15 . the current observational state of the cmb power spectrum does not have the precision to determine @xmath15 , although there are hints of the first doppler peak and statistical analyses suggest @xmath80 consistent with sbbn ( @xcite ) . the future cmb experiments , including the satellites , map and planck , promise to determine @xmath15 to 2% , which will provide an independent check on our measurement and a consistency test for sbbn . at redshifts @xmath81 , the igm is ionized , and can be observed through absorption in high - redshift qso spectra . recent comparisons of high - resolution spectroscopy of the igm and large cosmological simulations require a high baryon density to account for the observed forest , @xmath82 ( @xcite ; @xcite ; @xcite ) . other estimates include the baryons in protogalaxies seen in damped absorption with @xmath83 ( @xcite ) , and the baryons in the diffuse igm which are not seen in absorption , @xmath84 ( @xcite ) . the inventory of baryons at @xmath85 demands a high cosmological baryon density consistent with our measurements of d / h . at low redshift , only a small fraction ( @xmath86% ) of the baryons reside in stars and visible gas , with estimates ranging from @xmath87 ( @xcite ) . but a fair sample of today s baryons are likely to be in gas heated through supernova explosions or accretion into the large potential wells of galaxy clusters . a recent application of oort s method to 14 fields containing rich galaxy clusters yields a baryon density in the form of hot gas , @xmath88 ( @xcite ) , this result is valid assuming that the baryons and emitted light trace the total matter density and that clusters provide a fair sample of the baryon fraction . if most of the baryons today are in the form of hot gas , they must be collisionally ionized ( @xcite ) , and could have already been detected in a survey for o vi at @xmath89 ( burles & tytler 1996 ) . measurements of the total cosmological matter density @xmath90 imply that baryons can not constitute all the matter in the universe . a number of methods have been used to estimate @xmath90 ( c.f . dekel et al . peculiar velocities of galaxies can be treated as cosmic flows induced by large - scale mass distributions . in particular , cosmic flows in large - scale voids places a lower limit on the total mean matter density , @xmath91 ( dekel 1997 ) . this limit on @xmath90 places an upper limit on the baryon fraction , @xmath92 , which agrees with the fraction inferred from x - ray observations of galaxy clusters ( @xcite ; @xcite ; @xcite ) and groups ( @xcite ) . in sbbn , the determination of @xmath11 implies the primordial abundance ratios for the other light elements . we now use our d / h measurement to infer the primordial abundances of the other light elements . in the following , we use the output from the kawano code with @xmath93 s , and n@xmath94 = 3.0 . all errors are 67% confidence and include the uncertainties in the nuclear cross - sections and neutron lifetime . our measurement infers a primordial @xmath7he mass fraction y@xmath95 @xmath96 the comparison between d and @xmath7he provides a crucial test of sbbn ( @xcite ; @xcite ) , but recent measurements of the @xmath7he abundance in extragalactic metal - poor h ii regions have given discrepant results . izotov et al . ( 1997 ) compiled a homogeneous sample of 45 h ii regions taken mainly from the second byurakan survey ( c.f . izotov et al . 1993 ) , and found y@xmath97 ( izotov et al . 1997 ) , which is consistent with the value of y@xmath95 we infer . on the other hand , olive et al . ( 1997 ) have compiled a large sample of h ii regions from studies by olive & steigman ( 1995 ) , izotov et al . ( 1994 ) , and izotov et al . ( 1997 ) and obtain y@xmath98 , which is not consistent with izotov et al . ( 1997 ) nor our inferred value . the differences in the two @xmath7he studies must be properly understood before a direct comparison can be made between d and @xmath7he . olive et al . ( 1997 ) argue that the difference can be entirely accounted for in the atomic data sets and the lowest metallicity region , izw18 . skillman & kennicutt ( 1993 ) observed izw18 and found y@xmath99 . izotov & thuan ( 1997 ) reobserved izw18 , and they infer y@xmath100 . the different results obtained for this h ii region highlights the uncertainties in the current observational determinations of y@xmath95 . the primordial value of d / h infers an abundance for @xmath8li , @xmath101 where a(li ) = 12 + log ( li / h ) . the abundance of @xmath8li observed in the spite plateau " of warm metal - poor halo stars ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) is lower but consistent with the inferred @xmath8li from d / h . bonifacio & molaro have recently analyzed a large sample of suitable halo stars using infrared measurements to give a better indication of effective temperatures . they find a(li ) = 2.238 @xmath25 0.012 @xmath25 0.05 ( statistical and systematic errors , @xcite ) , and no evidence for a dispersion in the plateau abundances , nor correlations of a(li ) with effective temperature or metallicity . our d / h measurements are consistent with the @xmath8li abundance found by bonifacio & molaro ( 1997 ) , but large errors in nuclear reaction rates required in the bbn simulations also allow for non - standard depletion mechanisms in halo stars , which could have lowered the plateau by as much as 0.6 dex . ( @xcite ; @xcite ) . there do not exist any convincing measurements of a primordial @xmath102he abundance , although great efforts have been made to measure @xmath102he in galactic h ii regions ( @xcite ) and planetary nebulae ( @xcite ) using the hyperfine radio line . stars can both produce or destroy @xmath102he , so any attempts to infer the primordial abundance of @xmath102he from galactic measurements are dominated by uncertainties in the chemical evolution of @xmath102he . if we look at the other side of the problem , we can use our inferred primordial value @xmath103 to study the chemical evolution of @xmath102he . with the current state of the observations of the light elements and the associated uncertainties with inferring primordial abundances , we conclude that there is no inconsistency between deuterium and the other light elements with the predictions from sbbn . furthermore , the determination of @xmath11 with predictions from sbbn should be heavily weighted towards measurements of d / h in qso absorption systems ( @xcite ; @xcite ) . we are extremely grateful to w.m . keck foundation which made this work possible , to steve vogt and his team for building hires , to judy cohen , bev oke , and their team for building lris , and to tom bida , randy campbell and wayne wack for assistance at the telescope . we are very grateful to joseph miller and the staff of lick observatory for the construction , maintenance , and assistance with the kast spectrograph , which was made possible by a generous gift from william and marina kast . we thank christian cardall , george fuller , karsten jedamzik , david kirkman , martin lemoine , and jason x. prochaska for many useful conversations . adams , t. f. 1976 , a&a , 50 , 461 audouze , j. , & tinsley , b. 1974 , apj , 192 , 487 balser , d. s. , bania , t. m. , brockway , c. j. , rood , r. t. , & wilson , t. l. 1994 , apj , 430 , 667 balser , d. s. , bania , t. m. , rood , r. t. , & wilson , t. l. 1997 , apj , 483 , 320 bonifacio , p. , & molaro , p. 1997 , mnras , 285 , 847 burles , s. , & tytler , d. 1996 , apj , 460 , 584 burles , s. , & tytler , d. 1997 , aj , 114 , 1330 burles , s. , & tytler , d. 1997 , submitted to apj burles , s. 1997 , in preparation cardall , c. y. & fuller , g. m. 1996 , apj , 472 , 435 carlberg , r. g. , morris , s. l. , yee , h. k. c. , ellingson , e. , abraham , r. , lin , h. , schade , d. , gravel , p. , pritchet , c. j. , smecker - hane , t. , hartwick , f. d. a. , hesser , j. e. , hutchings , j. b. , & oke , j. b. 1997 , submitted to apj letters carswell , r. f. , rauch , m. , weymann , r. j. , cooke , a. j. & webb , j. k. 1994 , mnras , 268 , l1 carswell , r. f. , webb , j. k. , lanzetta , k. m. , baldwin , j. a. , cooke , a. j. , williger , g. m. , rauch , m. , irwin , m. j. , robertson , j. g. , & shaver , p. a. 1996 , mnras , 278 , 506 chengalur , j. n. , braun , r. & burton , w. b. 1997 , a&a , 318 , l35 clayton , d. d. 1985 , apj , 290 , 428 copi , c.j . , schramm , d.n . & turner , m.s . 1995 , science , 267 , 192 dekel , a. , burstein , d. , & white , s. d. m. 1996 , preprint , to appear in critical dialogues in cosmology " , ed . n. turok , world scientific dekel , a. 1997 , in formation of structure in the universe " , eds . a. dekel & j.p . ostriker , cambridge university press deliyannis , c. p. , demarque , p. , kawaler , s. d. 1990 , apjs , 73 , 21 edmunds , m.g . 1994 , mnras , 270 , l37 encrenaz et al . 1996 , a&a , 315 , 397 epstein , r. i. , lattimer , j. m. , & schramm , d. n. 1976 , nature , 268 , 198 ferland , g. 1993 , univ . kentucky dept . physics and astron . internal rept . fields , b. 1996 , apj , 456 , 478 fixsen , d. j. , cheng , e. s. , gales , j. m. , mather , j. c. , shafer , r. a. , & wright , e. l. 1996 , apj , 473 , 576 fuller , g. m. , & cardall , c. y. 1996 , nucl . b , 51 , 71 galli , d. , palla , f. , ferrini , f. & penco , u. 1995 , apj , 443 , 536 gautier , d. , & owen , t. 1983 , nature , 304 , 691 geiss , j. 1993 , in origin and evolution of the elements " , eds . n. prantzos , e. vangioni - flam , & m. casse ( cambridge university press : cambridge ) , 89 giallongo , e. , fontana , a. , & madau , p. 1997 , mnras , 289 , 629 haardt , f. & madau , p. 1997 , apj , 461 , 20 hata , n. , steigman , g. , bludman , s. , & langacker , p. 1997 d , 55 , 540 hu , w. , & white , m. 1996 , apj , 471 , 30 ikeuchi , s. & ostriker , j. p. 1986 , apj , 301 , 522 izotov , yu . i. , lipovetsky , v. a. , guseva , n. g. , kniazev , a. yu . , neizvestny , s. i. , & stepanian , j. a. 1993 , astron . trans . , 3 , 193 izotov , y. i. , thuan , t. x. , & lipovetsky , v. a. 1997 , apj , 435 , 647 izotov , y. i. , thuan , t. x. , & lipovetsky , v. a. 1997 , apjs , 108 , 1 izotov , y. i. & thuan , t. x. 1997 , preprint jedamzik , k. , fuller , g. m. , & mathews , g. 1997 , apj , 483 , 560 jedamzik , k. & fuller , g. m. 1997 , apj , 483 , 560 kawano , l. 1992 , preprint kirkman , d. & tytler , d. 1997 , apj , 484 , 848 krauss , l. m. , kernan , p. j. 1995 , phys . b , 347 , 347 lemoine , m. , vidal - madjar , a. , bertin , p. , ferlet , r. , gry , c. , & lallement , r. 1996 , a&a , 308 , 601 lineweaver , c. h. , barbosa , d. , blanchard , a. , & bartlett , j. g. 1997 , a&a , 322 , 365 lowenstein , m. , & mushotzky , r. f. 1996 , apj , 471 , l83 lu , l. , sargent , w. l. w. , womble , d. s. , & takada - hidai , m. 1997 , apj , 472 , 509 mulchaey , j. s. , davis , d. s. , mushotzky , r. f. , & burstein , d. 1996 , apj , 456 , 80 myers , s. t. , baker , j. e. , readhead , a. c. s. , & leitch , e. m. 1997 , apj , in press olive , k. a. & steigman , g. 1995 , apjs , 97 , 49 olive , k. a. , skillman , e. d. , & steigman , g. 1997 , apj , 483 , 788 osterbrock , d. e. 1989 , astrophysics of gaseous nebulae and active galactic nuclei , ( mill valley , ca : univ . sci . ) persic , m. & salucci , p. 1992 , mnras , 258 , 14 pisonneault , m. h. , deliyannis , c. p. , demarque , p. 1992 , apjs , 78 , 179 piskunov , n. , wood , b. e. , linsky , j. l. , dempsey , r. c. , ayres , t. r. 1997 apj , 474 , 315 prantzos , n. 1996 , a&a , 310 , 106 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , numerical recipes in c , the art of scientific computing , 2nd ed . ( cambridge , uk : cambridge university press ) rauch , m. , miralda - escude , j. , sargent , w. l. w. , barlow , t. a. , weinberg , d. h. , hernquist , l. , katz , n. , cen , r. , & ostriker , j. p. 1997 , apj , 489 , 7 rebolo , r. , beckman , j. e. , & molaro , p. 1988 , a&a , 192 , 192 reeves , h. , audouze , j. , fowler , w. a. , & schramm , d. n. 1973 apj , 179 , 909 reimers , d. , koehler , s. , wisotzki , l. , groote , d. , rodriguez - pascual , p. , wamsteker , w. 1997 , a&a , 329 , 270 rugers , m. , & hogan , c. 1996 , apj , 459 , l1 rugers , m. , & hogan , c. 1996 , aj , 111 , 2135 ryan , s. g. , beers , t. c. , deliyannis , c. p. & thorburn , j. a. 1996 , apj , 458 , 543 scully , s. , casse , m. , olive , k. a. , & vangioni - flam , e. 1997 , apj , 476 , 521 schramm , d. n. & turner , m. s. 1997 , submitted to rev . skillman , e. d. & kennicutt , r. c. jr . 1993 , apj , 411 , 655 smith , m. s. , kawano , l. h. & malaney , r. a. 1993 , apjs , 85 , 219 songaila . a. , cowie , l. l. , hogan c. j. & rugers , m. 1994 , nature , 368 , 599 spite , f. , & spite , m. 1982 , a&a , 115 , 357 spite , m. , maillard , j. p. , & spite , f. 1984 , a&a , 141 , 56 spitzer , l. , jr . 1978 , physical processes in the interstellar medium ( new york , ny : wiley ) steigman , g. & tosi , m. 1992 , apj , 401 , 150 steigman , g. 1994 , mnras , 269 , 53 steigman , g. & tosi , m. 1995 , apj , 453 , 173 thorburn , j. a. 1994 , apj , 421 , 318 tytler , d. , fan , x - m . , & burles , s. 1996 , nature , 381 , 207 tytler , d. , & burles , s. 1997 , in origin of matter and evolution of galaxies " , eds . t. kajino , y. yoshii & s. kubono ( world scientific publ . co. : singapore ) , 37 tytler , d. , burles , s. , & kirkman , d. 1996 , submitted to apj vangioni - flam , e. , olive , k. a. , prantzos , n. 1994 , apj , 427 , 418 vauclair , s. , & charbonnel , c. 1995 , a&a , 295 , 715 wagoner , r. v. , fowler , w. a. , & hoyle , f. 1967 , apj , 148 , 3 walker , t. p. , steigman , g. , schramm , d. n. , olive , k. a. & kang , h. s. 1991 apj , 376 , 51 wampler , e.j . , williger , g.m . , baldwin , j.a . , carswell , r.f . , hazard , c. & mcmahon , r.g . 1996 , a&a , 316 , 33 webb , j. k. , carswell , r. f. , lanzetta , k. m. , ferlet , r. , lemoine , m. , vidal - madjar , a. , & bowen , d. v. 1997 , nature , 388 , 250 weinberg , d. h. , miralda - escude , j. , hernquist , l. , & katz , n. 1997 , apj , 490 , 564 white , s. d. m. , navarro , j. f. , evrard , a. e. , & frenk , c. s. 1993 , nature 366 , 429 white , d. a. , & fabian , a. c. 1995 , mnras , 273 , 72 wolfe , a. m. , lanzetta , k. m. , foltz , c. b. , & chaffee , f. h. 1995 , apj , 454 , 698 zhang , y. , anninos , p. , norman , m. l. , & avery , m. 1997 , apj , 485 , 496 figure 1 : velocity plots of the keck hires spectrum containing lyman series lines in the d / h absorption system towards q1937 - 1009 . zero velocity corresponds to a redshift , @xmath104 , which is identical to figure 1 in tfb . the lyman limit region shows ly-12 through ly-19 in separate frames to emphasize the alignment of the entire lyman series in velocity position . the data is the solid histogram , each bin corresponds to one pixel . the gray line tracing the data is the best fit model with 3 components , model 4 . the three light tick marks represent the position of the deuterium lyman lines , the three solid ticks near @xmath105 mark the corresponding hydrogen lyman lines . the fourth solid tick marks an additional hydrogen absorber , which does not show deuterium . the solid horizontal line at zero flux shows the span of the region of interest around each lyman line . for ly@xmath22 , the region of interest spans the entire velocity range shown in figure 1 , while the other lyman lines have smaller regions . we use all of the spectrum covering the lyman series from ly-12 to the lyman limit . both the spectrum and model fit have been normalized to the unabsorbed quasar continuum , which was included in the fitting procedure . the 1@xmath42 error in the data values are shown by the dotted line near zero flux . figure 2a : spectrum of the region of the dhas . the data and model fit are displayed as in fig . 1 . represents the best fit of model 4 . the dashed line shows the is a 5th order legendre polynomial which gave the best fit to the unabsorbed qso continuum . the tick marks show the positions of all absorbers which lie in each region . the taller tick marks show the 3 main hydrogen absorbers . the corresponding deuterium lines are located approximately 1 blueward of the hydrogen lines , and are easily distinguishable with the same triple pattern . the solid line at unity represents our original estimate of the unabsorbed quasar continuum over the region . figure 3a : results of the fitting procedure for model 1 . the top panel shows the as a function of log ( d / h ) . the remaining panels show n , @xmath24 , and @xmath23 for the main h i components which gave the best fit . the parameters are represented as thin lines with different line styles for each component . the bottom panel shows the relative velocity positions of the two components , @xmath106 corresponds to @xmath104 . ccccc & 5546.08 & 5568.57 & 301 & 5 & 4687.50 & 4692.00 & 70 & 3 & 4445.60 & 4451.56 & 98 & 3 & 4340.40 & 4346.00 & 94 & 3 & 4284.70 & 4293.10 & 145 & 3 ly-6 & 4253.30 & 4257.06 & 63 & 3 ly-7 & 4232.15 & 4239.10 & 120 & 3 ly-9 & 4208.35 & 4213.10 & 82 & 3 ly - limit & 4178.30 & 4196.56 & 325 & 3 12.73 & 30.3 & 3.58001 & 14.27 & 40.8 & 2.65975 & 14.34 & 33.1 & 2.48516 12.59 & 32.0 & 3.57881 & 14.76 & 31.8 & 2.65795 & 14.16 & 37.5 & 2.48387 12.88 & 18.2 & 3.57864 & 13.44 & 16.7 & 2.57458 & 12.64 & 19.8 & 2.48218 13.54 & 21.0 & 3.57846 & 14.59 & 35.4 & 2.57384 & 13.01 & 19.3 & 2.46537 13.39 & 17.1 & 3.57781 & 14.08 & 62.5 & 2.57216 & 12.51 & 11.0 & 2.46489 13.12 & 27.2 & 3.57749 & 13.84 & 26.4 & 2.57127 & 14.57 & 47.3 & 2.46359 13.62 & 20.7 & 3.57673 & 13.58 & 32.3 & 2.53110 & 12.70 & 18.4 & 2.46265 12.28 & 22.8 & 3.57584 & 13.43 & 24.0 & 2.53043 & 12.93 & 19.1 & 2.46216 13.24 & 63.4 & 3.57503 & 13.70 & 62.5 & 2.52962 & 13.84 & 70.2 & 2.45247 12.92 & 20.7 & 3.57381 & 12.31 & 9.3 & 2.52853 & 13.75 & 38.2 & 2.45109 15.40 & 31.1 & 3.57295 & 13.62 & 22.1 & 2.52785 & 14.17 & 34.1 & 2.44969 13.20 & 44.0 & 3.56964 & 13.39 & 43.7 & 2.52595 & 12.52 & 5.9 & 2.44885 12.39 & 22.7 & 3.56837 & 13.59 & 33.7 & 2.52506 & 14.00 & 28.7 & 2.44785 12.43 & 16.3 & 3.56760 & 14.04 & 28.6 & 2.50091 & 13.86 & 22.0 & 2.44761 13.63 & 34.5 & 3.56699 & 13.60 & 35.5 & 2.50053 & 13.13 & 9.3 & 2.44647 14.00 & 26.8 & 3.56640 & 12.80 & 10.7 & 2.49969 & 14.06 & 31.7 & 2.44450 14.31 & 32.9 & 3.56495 & 13.50 & 21.1 & 2.49948 & 13.53 & 20.9 & 2.44366 13.05 & 20.6 & 3.56427 & 13.18 & 30.9 & 2.49919 & 13.11 & 22.2 & 2.44332 13.42 & 33.3 & 3.56333 & 13.24 & 37.0 & 2.48705 & 14.00 & 26.4 & 2.44115 12.30 & 6.8 & 2.66147 & 13.18 & 21.4 & 2.48646 & 12.96 & 7.5 & 2.44042 12.96 & 20.5 & 2.66101 & 13.32 & 38.8 & 2.48615 & 13.60 & 28.5 & 2.43971 & & & & & & 13.83 & 8.9 & 2.43868 cccccccc 1 & 2 & @xmath107 & @xmath108 & @xmath109 & 424.9 & 1097 2 & 3 & @xmath107 & @xmath110 & @xmath111 & 420.6 & 1093 3 & 2 & @xmath107 & @xmath108 & @xmath112 & 394.3 & 1069 4 & 3 & @xmath113 & @xmath114 & @xmath115 & 387.2 & 1065 5 & 4 & @xmath107 & @xmath110 & @xmath116 & 383.5 & 1061 6 & 3 & @xmath117 & @xmath118 & @xmath119 & 382.0 & 1062 7 & 3 & @xmath120 & @xmath114 & @xmath116 & 379.1 & 1062 [ c / h ] & @xmath121 & @xmath122 [ n / h ] & @xmath123 & @xmath63 [ o / h ] & @xmath124 & @xmath124 [ si / h ] & @xmath125 & @xmath126 [ fe / h ] & @xmath127 & @xmath128 log u & @xmath129 & @xmath121 log h i / h & @xmath130 & @xmath131 log @xmath132 ( @xmath133 ) & @xmath134 & @xmath135 l(kpc ) & 1.5 & 0.6 t@xmath136 & 1.62 @xmath25 0.09 & 2.36 @xmath25 0.09 t@xmath137 & 1.72 & 1.67 b@xmath138 & 4.8 @xmath25 0.8 & 8.4 @xmath25 0.4

in this paper we overview the theory of conics and roulettes in four non - euclidean planes . we collect the literature about these classical concepts , from the eighteenth century to the present , including papers available only on arxiv . the comparison of the four non - euclidean planes , in terms of the known results on conics and roulettes , reflects only the very subjective view of the author . _ keywords : _ cayley - klein geometries ; conics ; euler - savary equations ; normed plane ; roulettes _ 2010 msc : _ 46b20 , 51m05 , 52a21 , 53a17

1610.00473

our purpose is to compare the basic properties of certain non - euclidean planes through some known results on two types of curves , which is the motivation for the long introduction before the essential part of this paper . these planes are the so - called hyperbolic , spherical , minkowski and lorentzian planes . we note that even though there is an analytic approach for the examined curves in most of the planes , which we investigate here , the geometric method of construction leads to the definitions that can be interpreted in every case . we do not think that everybody regards our approach of these non - euclidean planes good . we think even less that the collected literature in this paper is complete . however , the large number of rediscovered results in this evergreen topic shows that our paper may not be useless . in the paper , we first present a unified model of the plane geometries , then compare the theories of the examined constructive curves . clearly , there is no room here to mention all results discovered in the last three hundred years . nevertheless , in our approach some important characteristics are shown which behave analogously in all these geometries , and we concentrate on them . in our investigation , it will be very important to know which definitions work simultaneously in all cases , how we can classify the conics , and how basic kinematics works in these planes . we add a large number of references to the introduction of the theory of the curves . we require an analytic approach to the investigated geometries , and thus , we can not mention here non - euclidean planes with only a synthetic model , for example non - paschian planes based on a synthetic construction as in @xcite . the common root of the four geometries used in the paper is a @xmath0-dimensional vector space endowed with a general scalar multiplication of the vectors . this multiplication is a function with two variables in @xmath1 , which , in the case that @xmath1 is a real vector space , we call a _ product_. first we review a way to axiomatize it . this method gives a natural matching of these geometries , the lorentzian and minkowski plane are affine planes , respectively and the hyperbolic and spherical planes are hypersurfaces of certain affine spaces , respectively . the following two subsections contains the comparing of the planes in this direction . a generalization of the inner product and the inner product spaces ( briefly i.p spaces ) was raised by g. lumer in @xcite . the _ semi - inner - product ( s.i.p ) _ on a complex vector space @xmath1 is a complex function @xmath2:v\times v\longrightarrow \mathbb{c}$ ] with the following properties : s1 : : : @xmath3=[x , z]+[y , z]$ ] , s2 : : : @xmath4=\lambda[x , y]$ ] @xmath5 , s3 : : : @xmath6>0 $ ] @xmath7 , s4 : : : @xmath8|^2\leq [ x , x][y , y]$ ] , a vector space @xmath1 with a s.i.p . is an _ space_. g. lumer proved that an s.i.p space is a normed vector space with norm @xmath9}$ ] and , on the other hand , that every normed vector space can be represented as an s.i.p . space . in @xcite j. r. giles showed that the following homogeneity property holds : s5 : : : @xmath10=\bar{\lambda}[x , y]$ ] for all complex @xmath11 . this can be imposed , and all normed vector spaces can be represented as s.i.p . spaces with this property . giles also introduced the concept of _ continuous s.i.p . space _ as an s.i.p . space having the additional property s6 : : : for any unit vectors @xmath12 , @xmath13\}\rightarrow{\mathrm { re}}\{[y , x]\}$ ] for all real @xmath14 . the space is uniformly continuous if the above limit is reached uniformly for all points @xmath15 of the unit sphere @xmath16 . a characterization of the continuous s.i.p . space is based on the differentiability property of the space ( see in @xcite ) . a normed space is gteaux differentiable if for all elements @xmath15 of its unit sphere and real values @xmath17 , the limit @xmath18 exists . a normed vector space is uniformly frchet differentiable if this limit is reached uniformly for the pair @xmath15 of points from the unit sphere . giles proved in @xcite that an s.i.p . space is a continuous ( uniformly continuous ) s.i.p . space if and only if the norm is gteaux ( uniformly frchet ) differentiable . from the geometric point of view we know that if @xmath19 is a @xmath20-symmetric , bounded , convex body in the euclidean @xmath21-space @xmath22 ( with fixed origin o ) , then it defines a norm whose unit ball is @xmath19 itself ( see @xcite ) . such a space is called _ minkowski space _ or normed linear space . basic results on such spaces are collected in the surveys @xcite , @xcite , and @xcite . in fact , the norm is a continuous function which is considered ( in geometric terminology , as in @xcite ) as a gauge function . combining this with the result of lumer and giles we get that a normed linear space can be represented as an s.i.p space . the metric of such a space ( called minkowski metric ) , i.e. the distance of two points induced by this norm , is invariant with respect to translations . another concept of minkowski space was also raised by h. minkowski and used in theoretical physics and differential geometry , based on the concept of indefinite inner product . ( see , e.g. , @xcite . ) the _ indefinite inner product _ ( i.i.p . ) on a complex vector space @xmath1 is a complex function @xmath2:v\times v\longrightarrow \mathbb{c}$ ] with the following properties : i1 : : : @xmath3=[x , z]+[y , z]$ ] , i2 : : : @xmath4=\lambda[x , y]$ ] @xmath5 , i3 : : : @xmath2=\overline{[y , x]}$ ] @xmath23 , i4 : : : @xmath2=0 $ ] @xmath24 then @xmath25 . a vector space @xmath1 with an i.i.p . is called an _ i.i.p . space_. we recall , that a subspace of an i.i.p . space is positive ( non - negative ) if all of its nonzero vectors have positive ( non - negative ) scalar squares . the classification of subspaces of an i.i.p . space with respect to the positivity property is also an interesting question . first we pass to the class of subspaces which are peculiar to i.i.p . spaces , and which have no analogous in the spaces with a definite inner product . a subspace @xmath26 of @xmath1 is called _ neutral ( or isotropic ) _ if @xmath27=0 $ ] for all @xmath28 . in view of the identity @xmath29=\frac{1}{4}\{[x+y , x+y]+i[x+iy , x+iy]-[x - y , x - y]-i[x - iy , x - iy]\},\ ] ] a subspace @xmath26 of an i.i.p . space is neutral if and only if @xmath30=0 $ ] for all @xmath31 . observe also that a neutral subspace is non - positive and non - negative at the same time , and that it is necessarily degenerate . therefore it can be proved that an non - negative ( resp . non - positive ) subspace is the direct sum of a positive ( resp . negative ) subspace and a neutral subspace . we note that the decomposition of a non - negative subspace @xmath32 into a direct sum of a positive and a neutral component is , in general , not unique . however , the dimension of the positive summand is uniquely determined . the standard mathematical model of space - time is a four dimensional i.i.p . space with signature @xmath33 , also called minkowski space in the literature . thus we have a well known homonyms with the notion of minkowski space ! in our paper we call _ lorentzian space _ the minkowski space defined by an indefinite scalar product with signature @xmath34 . let * s1 * , * s2 * , * s3 * , * s4 * , be the four defining properties of an s.i.p . , and * s5 * be the homogeneity property of the second argument imposed by giles , respectively . ( as to the names : * s1 * is the additivity property of the first argument , * s2 * is the homogeneity property of the first argument , * s3 * means the positivity of the function , * s4 * is the cauchy - schwartz inequality . ) on the other hand , * i1*=*s1 * , * i2*=*s2 * , * i3 * is the antisymmetry property and * i4 * is the nondegeneracy property of the product . it is easy to see that * s1 * , * s2 * , * s3 * , * s5 * imply * i4 * , and if @xmath26 is a positive ( negative ) subspace of an i.i.p . space , then * s4 * holds on @xmath26 . in the following definition we combine the concepts of s.i.p . and i.i.p .. the _ semi - indefinite inner product ( s.i.i.p . ) _ on a complex vector space @xmath1 is a complex function @xmath2:v\times v\longrightarrow \mathbb{c}$ ] with the following properties : 1 : : @xmath3=[x , z]+[y , z]$ ] ( additivity in the first argument ) , 2 : : @xmath4=\lambda[x , y]$ ] @xmath5 ( homogeneity in the first argument ) , 3 : : @xmath10=\overline{\lambda}[x , y]$ ] @xmath5 ( homogeneity in the second argument ) , 4 : : @xmath6\in \mathbb{r}$ ] @xmath35 ( the corresponding quadratic form is real - valued ) , 5 : : if either @xmath2=0 $ ] @xmath24 or @xmath36=0 $ ] for all @xmath24 , then @xmath25 ( nondegeneracy ) , 6 : : @xmath8|^2\leq [ x , x][y , y]$ ] holds on non - positive and non - negative subspaces of v , respectively . ( the cauchy - schwartz inequality is valid on positive and negative subspaces , respectively ) . a vector space @xmath1 with a s.i.i.p . is called an s.i.i.p . space . we conclude that an s.i.i.p . space is a homogeneous s.i.p . space if and only if the property * s3 * holds , too . an s.i.i.p . space is an i.i.p . space if and only if the s.i.i.p . is an antisymmetric product . in this latter case @xmath6=\overline{[x , x]}$ ] implies * 4 * , and the function is also hermitian linear in its second argument . in fact , we have : @xmath37=\overline{[\lambda y+\mu z , x]}=\overline{\lambda } \overline{[y , x]}+\overline{\mu } \overline { [ z , x]}=\overline{\lambda}[x , y]+\overline{\mu}[x , z]$ ] . it is clear that both of the classical `` minkowski spaces '' can be represented either by an s.i.p or by an i.i.p . , so automatically they can also be represented as an s.i.i.p . it is possible that the s.i.i.p . space @xmath1 is a direct sum of its two subspaces where one of them is positive and the other one is negative . then there are two more structures on @xmath1 , an s.i.p . structure ( by lemma 2 ) and a natural third one , which was called by minkowskian structure . let @xmath38)$ ] be an s.i.i.p . let @xmath39 be positive and negative subspaces , where @xmath40 is a direct complement of @xmath16 with respect to @xmath1 . define a product on @xmath1 by the equality @xmath41^+=[s_1+t_1,s_2+t_2]^+=[s_1,s_2]+[t_1,t_2],\ ] ] where @xmath42 and @xmath43 , respectively . then we say that the pair @xmath38^+)$ ] is a _ generalized minkowski space with minkowski product @xmath44^+$]_. we also say that @xmath1 is a _ real generalized minkowski space _ if it is a real vector space and the s.i.i.p . is a real valued function . the minkowski product defined by the above equality satisfies properties * 1 * -*5 * of the s.i.i.p .. but in general , property * 6 * does not hold . ( see an example in @xcite . ) if now we consider the theory of s.i.p in the sense of lumer - giles , we have a natural concept of orthogonality . the vector _ @xmath45 is orthogonal to the vector @xmath46 _ if @xmath2=0 $ ] . since s.i.p . is neither antisymmetric in the complex case nor symmetric in the real one , this definition of orthogonality is not symmetric in general . let @xmath38)$ ] be an s.i.i.p . space , where @xmath1 is a complex ( real ) vector space . the orthogonality of such a space can be defined an analogous way to the definition of the orthogonality of an i.i.p . or s.i.p . space . the vector _ @xmath47 is orthogonal to the vector @xmath48 _ if @xmath49=0 $ ] . if @xmath32 is a subspace of @xmath1 , define the orthogonal companion of @xmath32 in @xmath1 by @xmath50=0 \mbox { for all } u\in u\}.\ ] ] we note that , as in the i.i.p . case , the orthogonal companion is always a subspace of @xmath1 . the orthogonal companion of a non - neutral vector @xmath48 is a subspace having a direct complement of the linear hull of @xmath48 in @xmath1 . the orthogonal companion of a neutral vector @xmath47 is a degenerate subspace of dimension @xmath51 containing @xmath47 . let @xmath1 be a generalized minkowski space . then we call a vector _ space - like , light - like , or time - like _ if its scalar square is positive , zero , or negative , respectively . let @xmath52 and @xmath53 denote the sets of the space - like , light - like , and time - like vectors , respectively . in a finite dimensional , real generalized minkowski space with @xmath54 these sets of vectors can be characterized in a geometric way . such a space is called by a _ generalized space - time model_. in this case @xmath53 is a union of its two parts , namely @xmath55 where @xmath56 @xmath57 it has special interest , the imaginary unit sphere of a finite dimensional , real , generalized space - time model . ( see def.8 in @xcite . ) it was given a metric on it , and thus got a structure similar to the hyperboloid model of the hyperbolic space embedded in a space - time model . in the case when the space @xmath16 is an euclidean space this hypersurface is a model of the @xmath21-dimensional hyperbolic space thus it is such - like generalization of it . in @xcite was proved the following statement : @xmath53 is an open double cone with boundary @xmath58 , and the positive part @xmath59 ( resp . negative part @xmath60 ) of @xmath53 is convex . we note that if @xmath61 or the space is complex , then the set of time - like vectors can not be divided into two convex components . so we have to consider that our space is a generalized space - time model . the set @xmath62^+=-1\}$ ] is called the _ imaginary unit sphere _ of the generalized space - time model . with respect to the embedding real normed linear space @xmath38 ^ -)$ ] ( see lemma 2 ) @xmath63 is a generalized two sheets hyperboloid corresponding to the two pieces of @xmath53 , respectively . usually we deal only with one sheet of the hyperboloid , or identify the two sheets projectively . in this case the space - time component @xmath64 of @xmath47 determines uniquely the time - like one , namely @xmath65 . let @xmath66 be arbitrary . let @xmath67 denote the set @xmath68 , where @xmath69 is the orthogonal complement of @xmath47 with respect to the s.i.i.p . , thus a subspace . the set @xmath67 corresponding to the point @xmath70 is a positive , ( n-1)-dimensional affine subspace of the generalized minkowski space @xmath38^+)$ ] . each of the affine spaces @xmath67 of @xmath63 can be considered as a semi - metric space , where the semi - metric arises from the minkowski product restricted to this positive subspace of @xmath1 . we recall that the minkowski product does not satisfy the cauchy - schwartz inequality . thus the corresponding distance function does not satisfy the triangle inequality . such a distance function is called in the literature semi - metric ( see @xcite ) . thus , if the set @xmath63 is sufficiently smooth , then a metric can be adopted for it , which arises from the restriction of the minkowski product to the tangent spaces of @xmath63 . in this subsection we give a natural connection between the metrics of the spherical and hyperbolic planes . it is based on their common trigonometry . in fact , as early as 1766 lambert in @xcite observed that if there are at least two distinct lines through a given point that do nt intersect a given line , then the area of a triangle with angles a , b , c would be @xmath71 for some constant @xmath72 . he knew also the fact that the area of a triangle on a real sphere of radius @xmath72 is @xmath73 . comparing the two formulas he noted the following : `` one could almost conclude that the new geometry would be true on a sphere of imaginary radius '' . it turns out that if we substitute the distance @xmath74 by the imaginary distance @xmath75 in any trigonometric formula of the spherical geometry we get a corresponding trigonometric formula valid in hyperbolic geometry . while in the spherical plane all elements of euclidean trigonometry were reviewed in the nineteenth century ( see @xcite ) the analogous statements in hyperbolic geometry did not were investigated systematically . some recent papers try to make up this deficiency ( see e.g. @xcite , @xcite , @xcite ) . we also can find interesting results in this direction in another recent research . i would like to mention those papers of n.j . wilderberg in which he considered trigonometry in universal geometries . as he said in @xcite _ `` the natural connection with the geometry of lorentz , einstein and minkowski comes from a projective point of view , with trigonometric laws that extend to ` points at infinity ' , here called ` null points ' , and beyond to ` ideal points ' associated to a hyperboloid of one sheet.''_. we note that wildberger s method _ `` ... works over a general field not of characteristic two , and the main laws can be viewed as deformations of those from planar rational trigonometry ... ''_. we propose for study the book @xcite in which first arose the thought of `` universal geometry '' and the paper @xcite containing a detailed list of wildberger s work in this direction . at the end of the previous paragraph we gave a general approach to the imaginary unit sphere of a generalized space - time model . in the easiest situation it leads to the discover of the analytic connection between the spherical plane and the hyperbolic plane . let denote by @xmath76 is the euclidean inner product of the euclidean space and denote by @xmath44 $ ] the indefinite inner product of the lorentzian space of dimension @xmath0 . we can compare the distances of the spherical and hyperbolic geometry because the first distance realizes as a distance of points of the sphere of the euclidean space of radius @xmath77 , and the latter one as a distance of points in the imaginary sphere of radius @xmath78 of the lorentzian space , respectively . the angle measure of the two rays ( @xmath79 and @xmath80 ) in these planes is nothing else as the dihedral angle measure between the planes with unit normals @xmath81 and @xmath82 through the origin intersecting the required sphere in the given rays . from this we get the following results : @xmath83}{r^2 } \\ \hline \cos ( xpy)\angle & \langle \zeta,\xi \rangle & -[\zeta,\xi]\\ \hline \end{array}\ ] ] our first diagram ( fig . [ diagram1 ] ) shows that our general linear algebraic terminology how leads to the four non - euclidean planes . using the notation of the first subsection of the introduction * s.i.i.p * is an abbreviation of a semi - indefinite inner product space which is a generalized space - time model of dimension @xmath0 . we also use a coordinate frame of the embedding space , the coordinates can be considered with respect to this system . the used product either an indefinite inner ( scalar ) product or a semi inner product with the respective abbreviations * i.i.p * or * s.i.p*. an equality over an arrow means the restriction of the space to the set of those vectors which are satisfying it , with the same product . for example the * s.i.i.p * restricted by the equality @xmath84 will be an * s.i.p * showing that the getting plane is a minkowski ( normed ) plane . @xmath85 on the second figure ( fig . [ diagram2 ] ) we can see the so - called cayley - klein geometries . three geometries ( spherical , hyperbolic and lorentzian ) from the investigated ones belong to this class of geometries . if we consider the real projective plane and fix a group of isometries as a special subgroup of the group of projective collinearities then we can associate a projective metric to this subgroup . this metric determines a plane geometry . by klein s program these geometries classified in an algebraic way . the classes depend on the type of the measure of the length and also on the type of the measure of the angle . there are three possibilities for both of the measures , they could be elliptic , parabolic or hyperbolic , respectively . on this way we can get nine possibilities which can be realized , these are the so - called cayley - klein geometries . the cayley - klein geometries have a synthetic description ( see in @xcite ) and also a linear algebraic one ( see the paper of wilderberg @xcite and a recent paper of juhsz @xcite ) . [ cols="^,^,^,^",options="header " , ] + where the dual pairs with parameters @xmath86 and @xmath87 are the same row on the table . * _ the case of two conjugate complex roots : _ + then the last translation can not be performed to eliminate the monom @xmath88 but we can eliminate the monom @xmath89 by an appropriate transformation described in @xcite . shifting to inhomogeneous coordinates and simplifying the coefficients we obtain : the so - called _ semi - hyperbola _ has the normalform @xmath90 where @xmath91 and its dual pair is projectively equivalent with another semi - hyperbola with @xmath92 and @xmath87 . * overviewing the above cases only one remains , when the conic has no symmetry axis at all and @xmath93 . ignoring further explanations we claim the following : if the conic has the normalform @xmath94 where @xmath95 then it is called _ osculating parabola_. its dual is also an osculating parabola by a convenient reflection . the following theorem summarize the above information : @xcite there are @xmath96 types of hyperbolic conics listed in the above discussion . most of them are self - dual and the other ones can be paired into dual - pairs . in a lorentzian plane also there are two possibilities ( analytic and metric ones , respectively ) to define a conic . in the paper of birkhoff and morris @xcite the definition based on the two foci property and hence each of the types of conics have a singular definition . in such a way we can distinguish curves not only on their metric properties but on the base of their causal characters , e.g. we can say about relativistic time - like ellipse or relativistic time - like hyperbola , respectively . more precisely , if @xmath97 means the so - called time interval between two events @xmath98 and @xmath99 then for fixed positive @xmath74 we consider the relativistic _ time - like ellipse _ , as the locus of all points @xmath98 which satisfy @xmath100 with respect to two fixed point @xmath101 and @xmath102 so that their segment is time - like so that @xmath103 is real . these points are the foci of the ellipse . similarly defined the two branch of a _ relativistic hyperbola _ by the equalities @xmath104 respectively . it has been proved , that each geometrical conic of the form @xmath105 is the union of pieces consisting of confocal relativistic conics . ( for more details see the paper @xcite . ) it proved also that the relativistic conics are geometrically tangent to the null ( isotropic ) lines through the foci @xmath101 and @xmath102 . similar definitions can be found in the shonoda s paper @xcite ( and also in the recent arxive @xcite ) . in this paper the apollonius definition of conics were used to generate algebraic curves in the minkowski space - time plane ( in the lorentzian plane ) . it turn out to be different from classical conic sections . it has been extended and classified the sort of `` m - conics '' . also discussed the cases of the singularity points of these m - conics , coming from the transition from time - like world to space - like world through the light - like one . in @xcite , @xcite and @xcite the authors used linear algebra to the definition . in fact , from a geometric point of view , the conics can be represented by a quadric as the zero set of a quadratic form ( or the zero set of its symmetric bilinear form ) . fixing a regular symmetric bilinear form as an indefinite inner product @xmath76 any quadric can be regarded as the zero set of a symmetric bilinear function @xmath106 , where @xmath107 is a selfadjoint transformation with respect to the product . since in an inner product space the self - adjoint transformations has a classification ( see @xcite ) it has also a classification of conics defined by in this way . there is a possibility to define conics in lorentzian plane by projective geometry , since it is also an affine cayley - klein geometry . the lorentzian plane is a projective plane where the metric is induced by a real line @xmath108 and two real points @xmath109 and @xmath110 incidental with it . a curve in the lorentzian plane is circular if it passes through at least one of the absolute points . if it does not share any point with the absolute line except the absolute points , it is said to be entirely circular . every curve of order @xmath21 intersects the absolute line in @xmath21 points . if @xmath109 is an intersection point of the curve and the absolute line with the intersection multiplicity @xmath77 and @xmath110 is an intersection point of the curve and the absolute line with the intersection multiplicity @xmath111 , then it is said to be a curve with the type of circularity @xmath112 and its degree of circularity is defined as @xmath113 . the classification with respect to this method can be found in the papers @xcite and @xcite and says : the conics are classified into : non - circular conics ( ellipses , hyperbolas , parabolas ) , special hyperbolas ( circularity of type @xmath114 ) , special parabolas ( circularity of type @xmath115 ) and circles ( circularity of type @xmath116 ) . the subsection 10.2 in @xcite contains also a nice description of the above six types of conics . in addition , the last pages of the book contains also the definition of hyperbolic conics as conics which are projective conics of the embedding projective plane containing interior points of the cayley - klein model . this definition needed to state that the apollonian definition of conics ( based on the property of the leading line and a focus ) does not work in hyperbolic plane . ( in my point of view this metrical property gives another types of curves in the hyperbolic plane as the analytic one . ) further results on conics in lorentzian plane can be found in @xcite . while the projective geometry under the minkowski plane is the same as under the euclidean plane ( it is an affine plane , too ) the used metric kills the possibility that we classify projective conics using metric properties . from this reason we have several immediate definitions for conics . the first papers in this direction is the paper of wu , ji and alonso @xcite , and the paper of g.horvth and martini @xcite , respectively . minkowski conics was investigated by also fankhanel in the paper @xcite . in normed planes we have three different possibilities to define ellipses metrically . the first one was investigated in the paper @xcite . in @xcite we can find the following definitions refer to a normed plane @xmath117 . * let @xmath118 , @xmath119 , and @xmath120 . the set @xmath121 is called _ the ellipse defined by its foci _ @xmath45 and @xmath46 . * let @xmath122 be a homothetic copy of the unit disk @xmath19 , and @xmath123 be an arbitrary point from it . the locus of points @xmath124 for which there is a positive @xmath125 such that @xmath126 touches @xmath127 and contains @xmath128 on its boundary is called _ the ellipse defined by its leading circle and its focus _ * x*. * let @xmath107 be a straight line , @xmath129 a point , and @xmath130 a ratio larger than 1 . the locus of points @xmath124 , for which there is a positive @xmath131 such that the boundary of the disk @xmath126 contains * x * and the disk @xmath132 touches the line @xmath107 , is called _ the ellipse defined by its leading line and its focus _ * x*. in any normed plane the following holds : an ellipse , defined by its foci , is always an ellipse defined by its leading circle and a focus , and the converse statement is also true . on the other hand , an ellipse defined by its leading line and a focus is not necessarily an ellipse defined by its foci , and again the converse is true . in fig.[mc30 ] we can see that there is an ellipse following the third definition which is not centrally symmetric . by theorem 2 of @xcite it is not an ellipse by the first definition . in our example the norm is the @xmath133 norm , and the leading line @xmath107 and the focus @xmath129 are in `` symmetric position '' with respect to the circle of this minkowski plane , which is a square . plane , title="fig:",height=188 ] + conversely , consider the ellipse @xmath134 defined by its foci and shown in fig.[mc40 ] . first we can see that if it is also an ellipse defined by its leading line , then the leading line @xmath107 and the new focus @xmath135 have to be in `` symmetric position '' with respect to the line joining the original foci . `` symmetric '' means that this line is parallel to a diagonal of the unit square . in fact , if this is not the case , we get a figure as shown on the left side of fig.[mc40 ] . the squares @xmath136 with centers @xmath137 , @xmath138 , @xmath139 , @xmath140 , respectively , touch @xmath107 . the focus has to lie in the shaded rectangle , as the common point of the boundaries of homothetic copies @xmath141 , @xmath142 and @xmath143 of such squares ( with a homothety ratio smaller than 1 ) . on the other hand , the boundary of the square @xmath144 intersects the shaded rectangle in a segment parallel to that one in which it is intersected by @xmath143 . so it is impossible to give a good position for the focus @xmath145 . we now assume that @xmath107 and @xmath135 have symmetric position ( see the right side of fig.[mc40 ] ) . if this holds and the euclidean distance of @xmath107 and @xmath137 is @xmath146 , and that of @xmath147 and @xmath129 is @xmath77 , then , using the fact that the points @xmath137 , @xmath148 and @xmath138 have to lie on the new ellipse , we have the equalities @xmath149 implying that @xmath150 and showing that @xmath151 . thus the leading line and the focus are both determined . on the other hand , the point @xmath152 is not on the obtained ellipse , since the required ratio for it is @xmath153 . + for hyperbolas there are similar metric definitions . it can be proved that in normed planes , a hyperbola defined by its foci is always a hyperbola defined by its leading circle and a focus . the converse statement is also true . in general , the third definition yields a different class of curves . for the case of parabolas , the first two definitions have no analogue , and so we have only the third case . in a normed plane , let @xmath107 be a straight line , and @xmath129 be a point . the locus of the points @xmath154 for which there is a positive @xmath131 such that the boundary of the disk @xmath126 contains * x * and touches the line @xmath107 , will be called _ the parabola defined by its leading line and its focus _ * x*. it is also true that the metric parabola is a simple curve which does not contain segments if and only if the normed plane under consideration is strictly convex . finally some words about the analytic building up of the projective conics of a normed space . the following way is a possibility to define quadrics in the projective augmentation of any smooth , strictly convex space . we describe this method in the two - dimensional case , where the quadric is clearly a conic . every normed plane can be represented as a semi - inner product space ( s.i.p . ; see @xcite and @xcite ) . if the unit disk is strictly convex , this representation is unique . as proved in @xcite , the orthogonality with respect to the s.i.p . is equivalent to the orthogonality concept of birkhoff ( see , e.g. , @xcite and @xcite ) . koehler proved in @xcite that if the generalized riesz - fischer representation theorem is valid in a normed space , then every bounded linear operator @xmath155 has a generalized adjoint @xmath156 defined by the equality @xmath157 = [ x , a^t(y ) ] \mbox { for all } x , y \in v.\ ] ] it can be proved that if in all strictly convex and smooth spaces the above assumption holds , then in such a space there is a generalized adjoint . we remark that @xmath156 is in general not a linear transformation . we say that the linear mapping is _ self - adjoint _ if @xmath158 . if @xmath155 is self - adjoint , then any element of its class in the projective general linear group of @xmath1 is self - adjoint , too . so we can call such a family of operators _ class of self - adjoint linear operators of the projective space _ @xmath159 . now the concept of conics can be introduced as follows . let @xmath159 be a real projective space with the two - dimensional semi - inner product space @xmath38)$ ] . a ( non - degenerate ) projective conic is the zero set of a ( non - degenerate ) form @xmath160 $ ] , with an invertible self - adjoint operator @xmath155 of @xmath159 . we remark that the form @xmath161 is linear in its first argument , homogeneous in its second one , but is neither symmetric , bilinear nor positive . it is symmetric and bilinear if the semi - inner product is symmetric ; bilinear if the semi - inner product is additive in its second argument ; and positive if @xmath155 is a square operator ( meaning that it is the square of another self - adjoint operator , denoted by @xmath162 ) . the group of self - adjoint operators is basically determined by the unit disks , and it determines the projective conics in analytic sense . thus , in this setting the metric of the plane is also used for smooth , strictly convex normed planes . we finish with two problems : * characterize the self - adjoint operators for smooth , strictly convex normed planes . * describe relations between metric conics and general ones . the first question was also investigated in @xcite in the case when the plane also has a lipshitz - type property . some further observations can be found in @xcite . the investigation of roulettes in geometry is important not only with respect to their nice geometric properties but also their influence on kinematics . hence first of all we give a short overview of this connection . after this review we consider the validity of the euclidean results in our non - euclidean geometries . similarly to the case of conics in the minkowski plane we have to define new apparat to consider roulettes while in the cases of the other three geometries the roulettes have an analogous approach as in the euclidean situation . consider a plane @xmath163 which is moving on the fixed plane @xmath164 . the two simplest possibilities for such movements are given by translation and rotation . in euclidean geometry we can substitute the planes with cartesian coordinate frames @xmath165 and @xmath166 , and when we would like to describe the motion of a point @xmath167 of the moving plane , we need the coordinates @xmath168 of the point @xmath167 in the moving frame , the coordinates @xmath169 of @xmath170 in the fixed coordinate system , and the angle @xmath171 of the positive half of the @xmath117-axis of the fixed frame with the positive half of the @xmath45-axis of the moving frame . we get the coordinates @xmath45 , @xmath46 of the point @xmath167 in the fixed system by @xmath172 here @xmath173 are functions of a quantity @xmath111 which determine the motion . ( for example , @xmath111 can denote the time , or any other metric parameter . ) assume that @xmath174 is not zero on an interval of @xmath111 . then it can be inverted , and @xmath169 can also be considered as a function of @xmath175 . ( this assumption says that our motion can not contain translations in that domain . we call such a motion _ non - translative planar motion_. ) the derivative of the coordinate functions with respect to @xmath175 gives the coordinates of the velocity vector of the point @xmath167 . it is more convenient to use vector equality , and hence we introduce some further notion . let @xmath176 denote the rotation about the origin with signed angle @xmath175 . then the first equation array has the form @xmath177 if @xmath178 denotes the rotation with @xmath179 , we have the following rules : @xmath180 where @xmath181 is the unit matrix . we denote by dot the _ derivative with respect to _ @xmath175 , which means in this section the euclidean arc - length parameter . it is clear that @xmath182 for every value of @xmath175 there is precisely one point @xmath183 of the moving plane for which the velocity vector vanishes . this is @xmath184 this point @xmath183 of the moving plane is a so - called _ instantaneous center ( or instantaneous pole ) _ of the motion , and the set of these points is the _ moving polode _ , or curve @xmath185 of instantaneous poles , of the moving plane . the points of the moving polode can also be obtained in the frame as rest . these points @xmath186 are of the form @xmath187 and form the so - called _ fixed polode _ , or curve @xmath188 of instantaneous centers , in the fixed plane . we examine the motion with respect to the point @xmath186 . if @xmath129 is arbitrary , then @xmath189 , and using the equality @xmath190 , we have @xmath191 . since @xmath192 , we get that @xmath193 hence the velocity vector of the motion at the point @xmath129 is orthogonal to the position vector from @xmath186 to @xmath129 . this implies that the moving system in the given moment is a rotation about the center @xmath186 . observe that the velocity vectors of the two polodes at their common point agree ; in fact , @xmath194 hence the arc - length elements of the two curves agree and we get that in every moment the two curves are touching , and their arc - lengths calculated from a point @xmath195 to the point @xmath171 have the same value . hence the moving polode @xmath185 _ rolls without slipping _ ( or without friction ) on the fixed polode @xmath196 , and this is the only rolling process which corresponds to the given motion of the planes . hence we got the fact _ that every non - translatory planar motion of a rigid mechanical system in the plane can be considered as the rolling process of a curve rigidly connected with the system on a fixed curve in the plane_. this motivates the so - called main theorem of planar kinematics , namely : at every moment , any constrained non - translatory planar motion can be approximated ( up to the first derivative ) by an _ instantaneous rotation_. the center of this rotation is called the _ instantaneous pole_. thus , for each position of the moving plane , we generally have exactly one point with velocity zero ( as a result of that , the instantaneous pole is also called velocity center ) . this theorem leads to an interesting class of curves in the euclidean plane . given a curve @xmath185 attached to a plane @xmath163 which is moving so that the curve rolls , without slipping , along a given curve @xmath188 attached to a fixed plane @xmath197 occupying the same space , then a point @xmath167 attached to @xmath163 describes a curve in @xmath164 called a _ roulette_. based on this rolling process we can rewrite the definition of the motion of rigid systems . observe that every planar motion implies the motion of all points of the moving plane with respect to the fixed one . these orbits are said to be roulettes . thus , for the studied motion we consider two curves , also called _ polodes _ , and a suitable rolling process to determine the motion of a singular point . for this purpose a method is needed to determine the fixed position of the point @xmath167 with respect to the moving polode . a usual method is to give a line through the point @xmath167 which intersects the moving polode in the point @xmath198 and fixes the distance of @xmath167 and @xmath198 and the angle of the line @xmath199 with the tangent line @xmath200 of the moving polode at @xmath198 . hence the choice of @xmath198 on the moving polode is arbitrary . fix @xmath201 and @xmath202 . the points of the roulette @xmath203 of @xmath198 can be obtained by the composition of the following transformations : translate the point @xmath204 into the origin , rotate the image of the point of @xmath205 about the origin by the angle @xmath206 , and translate the obtained point by @xmath207 . hence the equality of the roulette of @xmath198 in the fixed system is @xmath208 since the roulette @xmath209 of the point @xmath167 can be described by the formula @xmath210 , we get @xmath211 this means that if we have two touching arcs @xmath212 and @xmath204 of a plane @xmath164 , and we associate to the second arc a moving plane @xmath213 in which its position is fixed , then the rolling process of @xmath204 on @xmath207 ( locally ) uniquely determines an orbit of every point of @xmath163 . in the euclidean plane equation the above equation shows that in every moment with respect to varying @xmath214 we have an isometry . hence the rolling process of the arcs determines a rigid motion of the plane @xmath163 . this representation is locally unique , since a rigid motion uniquely determines its polodes . hence we have if @xmath215\rightarrow \mathbb{r}^2 $ ] are two simple jordan arcs with common touching point @xmath216 such that @xmath146 is the arc - length parameter of both of them ( considered from the points @xmath217 to the point @xmath207 , @xmath204 , respectively ) , then for every @xmath218 $ ] we have an isometry @xmath219 sending the original position vector @xmath220 into the instantaneously position @xmath221 . if @xmath196 and @xmath185 have , for all @xmath218 $ ] , unique tangents at their points @xmath207 and @xmath204 , respectively , then , for all @xmath222 $ ] , @xmath219 is uniquely determined and can be described by the vector equation @xmath223 here @xmath224 and @xmath225 denote the unit tangent vectors at the point @xmath207 and @xmath204 , respectively , and @xmath226 is the rotation with the angle @xmath227 . for fixed @xmath220 , the graph of the function @xmath228\rightarrow \sigma $ ] is said to be the _ roulette _ of the point @xmath229 for the rigid motion given by the system of isometries @xmath230\}$ ] . the most important results in this theory are the so - called euler - savary equations , which compare the curvatures of the moving polode , fixed polode and the corresponding roulettes . we use the following forms of them ( compare with fig . [ eulersavaryfig ] ) : the _ first euler - savary equation _ is @xmath231 where @xmath232 are the curvature radius of the fixed polode at the instantaneous center and the curvature radius of the roulette at the examined point , respectively , @xmath233 is the length of the velocity vector and @xmath234 is the angular velocity of the motion . the _ second euler - savary equation _ says that @xmath235 where @xmath236 is the curvature radius of the moving polode at the instantaneous center . from these equations we can get a common form of the two equations which is @xmath237 spherical geometry of dimension @xmath238 is the geometry of a @xmath238-sphere embedded into a @xmath0-dimensional euclidean space . the motions of the sphere can be get from the three - dimensional special orthogonal linear group @xmath239 , containing those orthogonal transformations which hold the orientation . algebraically , this subgroup contains those linear transformations of the @xmath0-dimensional euclidean space which determinant has value @xmath240 . euler observed that these transformations always has an eigenvector with eigenvalue @xmath240 showing that a motion of the sphere is a rotation about a line which is called by the rotational axis of the transformation . hence the rigid motions of the sphere not only the sequence of certain instantaneous rotations but the sequence of proper rotations of the sphere . in fact , in euclidean geometry we should combine the rotation of the coordinate system with the translation of its origin , contrary on the sphere the translation part is also a rotation ( at another point as the fixed origin ) hence the needed composition gives the product of three rotations which itself also a rotation ( at a new origin ) . to characterize the motion we have to give in a moment a rotation , hence we have a function of the time , mapping the domain time - interval into @xmath239 , continuously . there are several papers on spherical roulettes because spherical kinematics is an intermediate step between planar and spatial kinematics . the earliest one is found by me , the work written by garnier in 1956 @xcite . considering one and two parameters spherical motions in euclidean space , muller gave the relations for absolute , sliding , relative velocities and pole curves of these motions . in addition to that he expressed the corresponding euler savary formula related to the trajectory curves of these @xmath240-parameter spherical motions @xcite . i would like to mention here also the paper of chiang @xcite which contains the results of the most important theorems using spherical kinematics , especially the spherical form of euler - savary equations . for the verification of the results chiang proposed the books @xcite and @xcite . i would like to mention two nice recent papers . the first on by m.a . gungor , s. ersoy , m. tosun @xcite containing more information on the history , too . the second one written by brunnthaler , husty and schrcker on four - bar mechanism ( @xcite ) . in the spherical plane the concept of roulettes also strongly connected to the bar - joint kinematics , in chiang s paper we can find the complete lists of motions induced by bar - joint frameworks . as chiang said : _ spherical four - bar mechanisms are similar to planar four - bar mechanisms . there are also spherical crank - rockers , drag - links and double - rockers , and even also spherical slider - cranks. however , as there is no translation motion in spherical kinematics , because all spherical motions are rotations , there exists therefore no real spherical slider. what we call a slider here exists simply because a joint on the slider is moving along a great circle arc . in fact this joint is @xmath241 apart from its axis of rotation . it is interesting to note the spherical rotary slider - crank as that shown in fig.[chiangfig1 ] , and there is no planar counterpart . _ using the notation of chiang we can say the following . assume that the polodes contact at @xmath167 . denote by @xmath155 the point of the roulette , by @xmath242 the spherical argument of the spherical polar coordinates of @xmath155 , with the pole - tangent great circle as the spherical polar line and by @xmath243 the quantity corresponding to the diameter of the inflection circle in planar kinematics . ( note that @xmath48 is the pole changing velocity , and @xmath244 is the angular velocity ( magnitude ) of the moving body , respectively . ) finally , if @xmath245 means the center of curvature of the roulette then we have the following ( geometric form ) of the first euler - savary equation ( see in fig [ chiangfig2 ] ) : it is very surprising but i can not find any paper on the internet on hyperbolic roulettes . it seems to be that the spherical approach can works also in the lorentzian space ( using it for its imaginary unit sphere ) . this analytically can solves the case of the hyperbolic plane . since roulettes can be defined by non - translatory motions , the basic problem of the general transition to hyperbolic geometry ( that is we have two types of isometries can work as a translation ) will not be occur . we can predict also the form of the euler - savary equation knowing the spherical one of the preceding paragraph . using the connections between the metric of the two planes mentioned in the introduction and the notation of chiang s corresponding to the motion we think that the required equation is : @xmath247 ( so the polodes contact at @xmath167 , @xmath248 is the point of the roulette @xmath249 is the hyperbolic argument of the hyperbolic polar coordinates of @xmath248 , with the tangent line and @xmath243 is a quantity corresponding to the diameter of the inflection circle in planar kinematics , @xmath48 is the pole changing velocity , and @xmath244 is the angular velocity ( magnitude ) of the moving body . ergin @xcite considering the lorentzian plane instead of the euclidean plane , and introduced the one - parameter planar motion in the lorentzian plane and also gave the relations between both the velocities and accelerations . yce and kuruoglu in @xcite using hyperbolic numbers reproduce the results of ergin and in analogy with complex motions as given by mller @xcite , defined one parameter motions in the lorentzian plane . they calling is hyperbolic plane is not good , because of hyperbolic numbers can be identified with the lorentzian plane and not with the hyperbolic plane . the relations between absolute , relative , sliding velocities ( and accelerations ) and pole curves was discussed , too . in the lorentzian plane euler - savary formula is given in references , @xcite and @xcite . we now refer the theorem 4.1 from ikawa s paper . by @xmath250 , ikawa denoted the orthogonal coordinate representation of a curve @xmath251 . the vector field @xmath252 is called the tangent vector field of the curve @xmath251 . if the tangent vector field @xmath117 of @xmath251 is a spacelike , timelike , or null , then the curve @xmath251 is called spacelike , timelike , or null , respectively . ikawa did not consider null curves and proved the following euler - savary types theorem : on the lorentzian plane @xmath253 , suppose that a curve @xmath254 rolls without slipping along a curve @xmath255 . let @xmath256 be a locus of a point @xmath167 that is relative to @xmath254 . let @xmath198 be a point on @xmath256 and @xmath72 a point of contact of @xmath255 and @xmath254 corresponds to @xmath198 relative to the rolling relation . by @xmath257 , we denote a polar coordinate of @xmath198 with respect to the origin @xmath72 and the base line @xmath258 . then curvatures @xmath259 ; @xmath260 and @xmath261 of @xmath255 ; @xmath254 and @xmath256 , respectively , satises @xmath262 @xmath263 the euler - savary equation clarifies the relation between the curvatures of the fixed and the moving polodes ( and the respective roulettes ) of a rolling process without friction , determining a planar motion of a rigid body ( and vice versa ) . this connection follows from euler s theorem of classical mechanics which states that every planar motion can be considered as the sequence of instantaneous rotations , whose centers give the fixed polode . thus , the euler - savary equation has been investigated also in physics , in various contexts ( lorentz spatial motions , elliptical harmonic motions , homothetic motions in the complex plane , etc . ) . however , one reason that there are no deeper and wider investigations on euler - savary applications in physics might be the fact that there are no rotations in the classical meaning . from a certain point of view , in the paper @xcite filled this gap by proposing a concept of rotation which mathematically describes a planar motion with respect to normed planes . thinking about the concept of motion in a wider context , it can be allowed that the duration of the motion of a body a little bit changes its shape . this `` changing of shape '' is determined by the concrete moving ( and thus by the participating polodes and roulettes ) . in order to define a concept of rotation for a minkowski plane , the authors started with extending the definition of brass by considering borel measures in a larger class of curves , not only in the unit circle , and derived angle measures for normed planes from it . ( the recent survey @xcite contains a lot of information on angle measures in minkowski plane . ) let @xmath264 be a closed jordan curve which is starlike with respect to a point @xmath214 of the interior of the region bounded by @xmath196 . an _ angle measure with respect to such a jordan curve _ is a ( normalized ) borel measure @xmath265 on @xmath196 for which the following properties hold : an angle measure defined in this way provides a translation invariant measure of _ angles _ in the plane , which can be defined as the convex hulls of two rays with the same starting point . given an angle @xmath269 with apex @xmath74 , it can be defined its _ generalized angle measure _ @xmath270 to be the measure @xmath265 of the arc determined on @xmath196 by the image of @xmath269 via the translation @xmath271 . figure [ figgenang ] illustrates this concept . [ generalrotation ] let @xmath272 be a minkowski plane and let @xmath196 be a closed jordan curve which is starlike with respect to a point @xmath214 of the interior of the region bounded by @xmath196 . let @xmath273 be a generalized angle measure as in the previous definition . general rotation _ ( with respect to @xmath273 ) is a transform @xmath274 for which the following three properties hold : * ( a ) * the transform @xmath275 leaves invariant the pencil @xmath276 of rays with origin in @xmath214 . in other words , if @xmath277 is a ray with origin @xmath214 , then @xmath278 is also a ray with origin @xmath214 . notice that a general rotation can be considered as acting in the space of directions of @xmath117 . indeed , the set @xmath276 can be seen as this space . for a class @xmath283 we have the following properties : * regarding composition , @xmath283 is an abelian group . more precisely , we have @xmath284 , where @xmath285 is the sum modulo @xmath286 . * for any @xmath287 , the application @xmath288 is a bijection from @xmath289 to @xmath196 . we highlight an interesting fact : the standard euclidean rotation group can be obtained in any minkowski plane . we just have to consider the group @xmath290 where @xmath196 is the _ lwner ellipse _ , which is defined as the ellipse of maximal volume contained in @xmath291 , and @xmath292 is the measure given by twice the area of its sectors . in @xcite there are two examples of general rotations in the euclidean plane . the first one relies on an area - based measure for an ellipse , which is clearly well defined . in the second used the arc - length measure referring to a nephroid . here we quote only the first one . consider the euclidean plane and the system of ellipses with common focus at the origin @xmath293 and with major axis on the @xmath45-axis of the coordinate system , such that the positive half - line of @xmath45 contains the closest point of the ellipse ( see fig . [ heliofig ] ) . in that polar coordinate system ( which is called the heliocentric coordinate system for the ellipse ) , for which the ray @xmath294 is the positive half axis @xmath45 , we can write the radial function @xmath295 of the ellipse @xmath296 by the formula @xmath297 where @xmath214 is the semi - latus rectum of the ellipse and @xmath125 is the eccentricity of it , respectively . let @xmath298 be the area of the sector enclosed by @xmath299 , @xmath300 , and @xmath296 be the arc between these lines . hence @xmath301 with respect to @xmath292 and @xmath296 from above , for every real number @xmath302 there is a generalized rotation of the euclidean plane about @xmath293 with this angle @xmath111 . by kepler s second law about planetary motions , the angle @xmath111 of a generalized rotation is proportional to the time of the motion of the planet . hence the generalized rotation with angle @xmath111 maps the current position @xmath303 of the planet to that point @xmath304 of the orbit where the planet arrives after time @xmath111 . the principle of measuring the angle proportional to the area of the sector intersected by the angle domain from the basic disk @xmath305 works in all minkowski planes and for all basic curves @xmath296 . note that in the euclidean plane with the unit circle as basic curve , this choice of @xmath292 gives the usual angle measure , and that we get the usual rotations as generalized rotations by choosing @xmath167 to be the origin @xmath293 . an advantage of this choice is affine invariance , but there is also a big disadvantage . namely , the length of the arc @xmath296 containing the domain of the angle can not be calculated easily from this angle measure . ( as a known example , we note that the calculation of the arc - length of an ellipse leads to a complete elliptic integral of second kind , which has no closed - form solution in terms of elementary functions . ) if we consider two curves @xmath196 and @xmath185 , then we have to use a suitable lower subscript for the curvature function . we also have the concept of _ curvature radius _ @xmath306 which is , as well - known , the reciprocal value of the curvature at the given point @xmath307 . with these notions we are able to formulate if the unit circle of the minkowski plane is two times continuously differentiable , then the following equality holds : @xmath308 here @xmath306 is the curvature radius of the fixed polode at its point @xmath309 , @xmath310 is the curvature radius of the moving polode at its point @xmath311 , and @xmath312 is the length of the common velocity vector of the fixed and moving polodes at the moment @xmath146 and at the instantaneous pole @xmath313 . the instantaneous center @xmath19 and the curvature center @xmath314 of the roulette at its point @xmath315 satisfy the equality @xmath316 where the second intersection point of the path normal line at @xmath167 with the inflection curve is the point @xmath317 . brunnthaler , k. , schrcker h. p. , husty , m. , _ synthesis of spherical four - bar mechanisms using spherical kinematic mapping . _ in j. lenarcic and b. roth , editors , advances in robot kinematics , pages 377 - 384 . springer , 2006 .

experiments with two - dimensional ( 2d ) plasma crystals are usually carried out in rf plasma sheaths , where the interparticle interactions are modified due to the presence of plasma wakes . the wake - mediated interactions result in the coupling between wave modes in 2d crystals , which can trigger the mode - coupling instability and cause melting . the theory predicts a number of distinct fingerprints to be observed upon the instability onset , such as the emergence of a new hybrid mode , a critical angular dependence , a mixed polarization , and distinct thresholds . in this paper we summarize these key features and provide their detailed discussion , analyze the critical dependence on experimental parameters , and highlight the outstanding issues .

1105.6223

stability of two - dimensional ( 2d ) plasma crystals is a fundamental problem of complex ( dusty ) plasmas . such crystals which are monolayers of hexagonally ordered monodisperse microparticles can be ( routinely ) created in a rf plasma @xcite : particles get negatively charged in a plasma and therefore the electric force exerted on them in the sheath above a horizontal electrode can compensate for gravity , thus providing a stable levitation . there are several mechanisms operating in complex plasmas that can result in the melting of 2d crystals . these mechanisms can generally be divided into two categories _ generic _ and _ plasma - specific_. generic mechanisms of melting are those operating in any ( classical ) system with a given pair interaction between particles , provided the interaction can be described by a hamiltonian . the melting in 2d systems can either be a two - step process ( which involves consecutive unbinding of weakly interacting dislocation and disclination pairs , respectively , with the intermediate hexatic phase ) @xcite or a one - step process ( where the hexatic phase is preempted by the formation of dislocation chains ) @xcite . these generic melting mechanisms can operate in very different 2d systems including complex plasmas @xcite . plasma - specific melting mechanisms can only operate in complex plasmas . such mechanisms are associated with the energy exchange between microparticles and ambient plasma and can be considered as a result of the system openness . for instance , 2d plasma crystals can be strongly perturbed by single particles moving above or below the monolayer @xcite , or they can melt due to fluctuations of particle charges @xcite . the most universal among the plasma - specific mechanisms is that associated with the _ wake - mediated _ interaction between microparticles : in the presence of strong plasma flow the screening cloud around each charged particle becomes highly asymmetric @xcite . these clouds are usually referred to as `` plasma wakes '' @xcite and play the role of a `` third body '' in the interparticle interaction , making it nonreciprocal @xcite . under certain conditions , this makes the system non - hamiltonian and provides effective conversion of the energy of flowing ions into the kinetic energy of microparticles @xcite . the wake - induced mechanism of melting of crystalline monolayers was discovered theoretically a decade ago by ivlev and morfill @xcite . based on a simple model of a particle chain , it was shown that the longitudinal in - plane and transverse out - of - plane dust - lattice ( dl ) wave modes are no longer independent they are coupled due to the wake - mediated interactions . when the modes intersect they become modified by the coupling and form a _ hybrid _ mode in a narrow vicinity of the crossing . this can trigger the _ mode - coupling instability _ which causes the melting . this melting mechanism had received strong confirmation later on , when the instability threshold predicted by the theory was compared @xcite with the experimental observations by u. konopka and numerical simulations by g. joyce . however , the direct comparison of theory and experiment became possible only recently , after an experimental method of measuring the out - of - plane mode was developed @xcite and therefore the essential fingerprint of the mode - coupling instability the hybrid mode became observable . in recent experiment by coudel _ _ @xcite an implementation of this method unambiguously demonstrated that the melting indeed occurs due to the resonance coupling between the longitudinal in - plane and transverse out - of - plane modes . the variation of the wave modes with the experimental conditions , including the emergence of the hybrid branch , revealed exceptionally good agreement with the theory of mode - coupling instability generalized for 2d case by zhdanov _ _ @xcite . the theory of mode - coupling instability @xcite predicts a number of distinct fingerprints to be observed upon the instability onset : along with the emergence of hybrid mode mentioned above , these are a critical angular dependence the hybrid mode first appears only for wave vectors oriented along one of the principal lattice axes , a mixed polarization the two dl modes that form the hybrid mode are no longer purely longitudinal and transverse close to the merging point , and distinct thresholds the instability sets in only when ( i ) the particle number density in the monolayer is high enough or / and vertical confinement eigenfrequency is low enough ( so that the two dl modes can cross and form the hybrid mode ) and ( ii ) the gas pressure is low enough ( so that the growth rate of the hybrid mode exceeds the damping rate ) . all these fingerprints have been mentioned in our previous theoretical publications @xcite , some of them were also illustrated in the followup experimental paper @xcite . however , their discussion was apparently too concise to address all these important properties in necessary detail . the need for in - depth discussion of the wake - mediated mode coupling became evident now , when new publications have appeared where some essential properties of the instability were misinterpreted ( see , e.g. , recent experimental paper @xcite and the subsequent erratum @xcite by liu _ et al . _ ) . therefore , in this paper we summarize the key features of the mode coupling and provide a detailed discussion , analyze the critical dependence on experimental parameters , and highlight the outstanding issues . in 2d plasma crystals , two in - plane wave modes can be sustained ( here we naturally leave aside polydisperse mixtures ) . both modes have an acoustic dispersion , one of them is longitudinal and the other is transverse . since the strength of the vertical confinement in such systems is finite , there is a third fundamental wave mode which has an optical dispersion and is associated with the out - of - plane oscillations @xcite . theory predicts that all three modes depend critically on the parameters of the plasma wakes . yet the role of the wakes is not only the modification of the modes themselves . what is far more important is that the modes are _ coupled _ to each other due to the wake - mediated interactions between particles . it is worth noting that this _ coupling is linear _ and therefore does not depend on the wave amplitude . below we identify the conditions when the wake - mediated coupling becomes crucial for waves in 2d crystals . in general , linear dispersion relations @xmath0 are determined by eigenvalues ( eigenmodes ) of a dynamical matrix @xmath1 . the latter is derived by considering small perturbations of individual particles ( with respect to a stable configuration , e.g. , a hexagonal monolayer for 2d crystals ) of the form @xmath2 . elements of @xmath1 are determined by properties of interparticle interactions and are functions of @xmath3 . if the interactions are reciprocal then @xmath1 is hermitian , and therefore the eigenvalues are always real ( i.e. , the modes are stable ) and the eigenvectors are orthogonal . the situation changes if the interactions are nonreciprocal the eigenvalues in this case can become complex . before specifying a particular functional form of interparticle interactions in 2d crystals , let us generally assume the non - reciprocity and discuss common properties of the dl waves in this case . the dynamical matrix has the following form : @xmath4 the elements @xmath5 , @xmath6 , and @xmath7 determine the dispersion of two in - plane ( horizontal ) modes and @xmath8 characterizes the out - of - plane ( vertical ) mode . the elements @xmath9 emerge due to non - reciprocity and make @xmath1 non - hermitian . the matrix is calculated for the reference frame shown in fig . [ fig1](a ) assuming that vertically the particles are confined in a ( parabolic ) potential well characterized by the eigenfrequency @xmath10 . to derive the dispersion relations @xmath0 , we add a friction force @xmath11 in equations of motion ( with @xmath12 being the damping rate ) which yields : @xmath13=0,\ ] ] where @xmath14 is the unit matrix . thus , @xmath15 are the eigenvalues of @xmath1 , i.e. , the dl wave modes . for reciprocal interactions @xmath16 and then eqs ( [ matrix ] ) and ( [ dispersion ] ) yield @xmath17 there are three independent and real dl modes : two _ acoustic _ in - plane modes @xmath18 ( where @xmath19 and @xmath20 indicate the longitudinal and transverse polarizations and correspond to the plus and minus signs , respectively ) and an _ optical _ out - of - plane mode @xmath21 . when solved for @xmath22 , each mode yields a couple of conjugate branches . when interactions are nonreciprocal the modes are modified they are described by eq . ( [ dispersion1 ] ) with nonzero right - hand side ( r.h.s . ) which is proportional to @xmath23 . the dl modes become coupled with each other and complex . is symmetric and hence can be reduced to the diagonal form also for @xmath24 , but with complex eigenvalues . ] now let us specify interparticle interactions . for the direct interaction between charged grains we naturally adopt the yukawa potential which is characterized by ( negative ) particle charge @xmath25 and the screening length @xmath26 . for the wake we employ the commonly used simple model @xcite of a point - like ( positive ) charge @xmath27 located at the distance @xmath28 downstream each particle and also interacting with the neighboring particles via the yukawa potential . the elements of the dynamical matrix ( [ matrix ] ) for such interactions are calculated in appendix [ appendix1 ] , eqs ( [ elements1 ] ) and ( [ elements2 ] ) . for brevity , theoretical results in this paper are written in the dimensionless form the frequency is normalized by the dl frequency scale , @xmath29 and the wave vector is normalized by the lattice constant @xmath30 : @xmath31 and @xmath32 . after the normalization , the dl modes depend on three dimensionless parameters ( along with the normalized wave vector ) : the screening parameter @xmath33 as well as the relative wake charge @xmath34 and distance @xmath35 . in experiments , @xmath36 is typically about unity whereas both wake parameters are usually small . elements @xmath37 [ eq . ( [ elements2 ] ) ] are proportional to the product @xmath38 and hence the coupling between the dl modes is very small [ @xmath39 . it becomes important only if different dl modes intersect ( or they are very close to each other ) ; otherwise the coupling can be neglected and then the modes can be treated independently , as described by eq . ( [ dispersion1 ] ) . let us elaborate on this point . behavior of two coupled modes , @xmath40 , in a close proximity of their intersection can generally be described by the following equation : @xmath41\left[\omega-\omega_0-{\bf u}_2\cdot({\bf k}-{\bf k}_0 ) \right]=\epsilon.\ ] ] here @xmath42 are the group velocities of the two modes calculated at a point @xmath43 of the intersection line @xmath44 ( the latter is determined by the shape of the crossing modes ) and @xmath45 characterizes the coupling . from eq . ( [ dispersion2 ] ) one readily concludes that the effect of the coupling critically depends on the sign of @xmath45 . this is illustrated in fig . [ fig2 ] : when @xmath46 the crossing modes merge and form a complex hybrid mode in a narrow range @xmath47 , with re @xmath48 and i m @xmath49 . thus , one of the branches of the hybrid mode has a positive imaginary part and hence can be unstable . when @xmath50 the coupling is accompanied by reconnection ( formation of a `` forbidden band '' ) and the modes remain stable ( real ) . furthermore , the sign of the product @xmath51 in the unstable case determines the type of the instability , which is absolute for @xmath52 ( the case illustrated in fig . [ fig2 ] ) and convective otherwise . now we can consider the dl modes described by eqs ( [ matrix ] ) and ( [ dispersion ] ) , with the dispersion and coupling elements from eqs ( [ elements1 ] ) and ( [ elements2 ] ) . the detailed analysis shows that the intersection of the transverse in - plane mode @xmath53 with the longitudinal in - plane , @xmath54 , as well as with the out - of - plane mode @xmath55 is characterized by positive coupling and therefore is accompanied by reconnection illustrated in fig . [ fig2]c , so that the modes remain stable . moreover , the coupling for these two pairs of modes exactly disappears ( fig . [ fig2]a ) if @xmath3 is parallel to one of the principal lattice axes ( i.e. , when @xmath56 or @xmath57 in fig . [ fig1]a ) . in contrast , for the remaining pair of @xmath55 and @xmath54 the coupling is _ always negative_. as illustrated in fig . [ fig2]b , this results in the formation of the _ hybrid _ mode with the imaginary part i m @xmath58 ( by the order of magnitude ) . if damping is low enough , this triggers the _ mode - coupling instability_. thus , only the intersection between the out - of - plane and longitudinal in - plane modes is critical for stability of 2d plasma crystals . it is instructive to present a brief summary of the mode - coupling theory which was originally developed for a 1d string assuming the nearest - neighbor ( nn ) interactions @xcite . as it follows from the subsequent analysis ( see sec . [ fingerprints ] ) , this approach yields very simple expressions which can be conveniently implemented for practical purposes ( e.g. , to calculate the instability thresholds , see sec . [ threshold ] ) . in the framework of 1d nn model @xcite , the coupling between horizontal ( `` longitudinal in - plane '' ) and vertical ( `` out - of - plane '' ) dl modes , @xmath59 \omega_{{\rm v}}^2(k)=\omega_{\rm conf}^2 - 4(1-\tilde q)\left(\kappa^{-2}+\kappa^{-3}\right){\rm e}^{-\kappa}\sin^2\frac12k , \end{array}\ ] ] is described by the following ( exact ) equation : @xmath60 the range of @xmath61 is limited by the first brillouin zone @xmath62 ( we use the same normalization as above ) , and the 1d coupling coefficient is @xmath63 all characteristics of the hybrid mode ( e.g. , width , growth rate , etc . ) can be easily derived from eqs ( [ modes_1d])-([coupling_coefficient_1d ] ) , see appendix [ appendix1a ] . these results are obtained in the limit of `` small '' @xmath64 ( for general case , see ref . @xcite ) . in sec . [ threshold ] they will be compared with the exact 2d results derived in this limit . to conclude this section , we should make two specific remarks which might be useful when considering dl waves and their coupling : \(i ) the dynamical matrix @xmath1 is solely characterized by interparticle interactions , i.e. , does not depend on the damping rate @xmath12 . thus , the _ wave modes _ @xmath65 which are the eigenvalues of @xmath1 do not depend on @xmath12 either . as regards the _ dispersion relations _ @xmath0 , their dependence on @xmath12 is determined by the solutions of the quadratic equation @xmath66 . therefore the effect of friction on waves is straightforward : for weakly damped waves ( when @xmath67 , which is typical for experiments ) , one readily obtains @xmath68 . from the practical point of view this implies that one can analyze undamped dispersion relations ( i.e. , put @xmath69 ) and afterwards simply add @xmath70 to the resulting imaginary part , in a close proximity of @xmath71 and therefore is unimportant for the analysis . ] which simultaneously identifies the instability threshold for the hybrid mode ( see sec . [ threshold2 ] ) . \(ii ) for the analysis of wave modes in a crystal it is _ sufficient _ to consider the wave vectors from within the first brillouin zone @xcite which is shown in fig . this zone is nothing but the wigner - seitz cell of the reciprocal lattice formed by the basis vectors @xmath72 and @xmath73 . hence , the wave vectors @xmath3 and @xmath74 which are different by a linear combination of the basis vectors ( @xmath75 ) are equivalent for wave modes , i.e. , @xmath76 . one has to remember this fact when analyzing dispersion relations at large @xmath77 . for instance , recently liu _ et al . _ @xcite identified one of the `` hot spots '' seen in their fluctuation spectra ( in their fig . 4c ) as a new hybrid mode . this spot is centered at @xmath78 , @xmath56 and therefore is located outside the first brillouin zone ; after the mapping with , say , @xmath79 , its position is @xmath80 , @xmath81 ( equivalent to @xmath57 , see fig . [ fig1]b ; note also the change in the polarization , from the transverse to longitudinal ) . this naturally coincides with the position of the `` regular '' hybrid mode at the border of the first zone , whereas the `` original '' spot is merely its image outside it ( as illustrated in figs [ fig3]c and [ fig4]b ) . in this section we present detailed discussion and analysis of distinct fingerprints characterizing the wake - mediated coupling of dl modes in 2d crystals . based on the results of sec . [ types ] we naturally focus only on the coupling between the out - of - plane and longitudinal in - plane modes , which causes the formation of the hybrid mode and can trigger the mode - coupling instability . to illustrate key features of the mode coupling , the theoretical calculations were combined with the results of molecular dynamics ( md ) simulations ( see description in appendix [ appendix2 ] ) . the dl wave modes were obtained from the simulations by plotting the so - called `` fluctuation spectra '' intensity distributions of thermal velocity fluctuations in the @xmath82 space . figure [ fig3 ] represents a characteristic example of fluctuation spectra measured at the onset of the mode - coupling instability . it was obtained from md simulations performed for the conditions close to those of experiment by coudel _ _ @xcite ( experiment ii in table [ tab1 ] ) . in this case the out - of - plane and longitudinal in - plane modes are just barely touching . such `` shallow '' intersection is nevertheless sufficient to form a distinct hybrid mode and to trigger the instability . figures [ fig3]a and b depict the dl modes in the @xmath83 plane ( positive branches ) . they are obtained for two directions of @xmath3 at @xmath56 and @xmath84 , respectively ( corresponding to the principal lattice axes , see fig . [ fig1]a ) . figure [ fig3]c shows the intensity distribution in the @xmath3-plane ( integrated over frequency ) , which provides a `` top view '' on the fluctuation spectra ( dotted line shows the direction of @xmath84 ) . the hybrid mode can be identified in figs [ fig3]b and c as `` hot spots '' located near the border of the first brillouin zone , where the intersection occurs ( note also images of the hybrid mode outside the first zone ) . the theoretical curves calculated from eqs ( [ matrix ] ) , ( [ dispersion ] ) , ( [ elements1 ] ) , and ( [ elements2 ] ) for the same set of parameters demonstrate excellent agreement with simulations . the reason why the fluctuation intensity is anomalously high at the position of the hybrid mode is the heating induced by the wake - mediated coupling . figure [ fig3 ] represents the marginally unstable regime , when i m @xmath85 . in this regime ( also observed in experiment @xcite ) the effective growth rate of the instability is low , which allows us to obtain the fluctuation spectra before the crystal is eventually destroyed . it is worth noting , however , that the hybrid mode would usually appear as a `` hot spot '' even if the instability is suppressed by friction and i m @xmath86 . this is because the heating due to mode coupling is present also in stable regime . it becomes negligible only when i m @xmath87 , and then the fluctuation intensity coincides with the intensity of dl modes outside the hybrid zone ( where it is determined by the background neutrals or temperature of thermostat in simulations ) . at the initial stage of the instability ( and , of course , when the instability is suppressed ) the effect of nonlinearity on the dl modes is not significant . in this case the contour plot of the growth rate i m @xmath88 reasonably reproduces the distribution of fluctuation intensity in the vicinity of the hybrid mode . this is illustrated in fig . [ fig4 ] where we plotted i m @xmath88 predicted by theory and the ( frequency - integrated ) fluctuation spectra from simulations . at a later stage , however , a nonlinear coupling between different modes becomes essential . this results in a variety of new phenomena , such as the energy cascades from the `` hot spots '' in the @xmath3 space and the generation of secondary harmonics at the double frequency of the hybrid mode ( which are clearly seen in fig . 3 of ref . @xcite ) . nevertheless , since the principal aim of this paper is to focus on distinct features of the _ linear mode coupling _ , we leave aside the discussion of numerous nonlinear coupling effects . for the plots shown in fig . [ fig4 ] we used the same set of parameters as for fig . [ fig3 ] , but for simulations we assumed periodic boundary conditions . the latter not only allowed us to directly compare the theory and simulations , but also to probe the effect of weak deviation from periodicity caused by the horizontal confinement . we see that the only noticeable difference between figs [ fig3]c and [ fig4]b is the intensity of the hybrid mode . this reflects the fact that the growth rate of the instability critically depends on the interparticle distance @xmath30 ( resulting in exponentially growing difference in the heating at the initial stage ) . therefore in the case of parabolic confinement ( where @xmath30 in the center is somewhat smaller than the average value ) the intensity of the hybrid mode is higher than in the homogeneous case shown in figs [ fig4]b . otherwise , the positions and the overall shape of the `` hot spots '' in these two plots practically coincide , suggesting that weak inhomogeneity which is always present in experiments due to the horizontal confinement does not really affect the periodic structure of the brillouin zones . finally , to illustrate the difference between `` shallow '' and `` deep '' intersections of the out - of - plane and longitudinal in - plane modes , in fig . [ fig5 ] we demonstrate the fluctuation spectra obtained for the eigenfrequency of vertical confinement @xmath89 which is slightly lower than that in fig . this naturally shifts the out - of - plane mode down and thus makes the resulting hybrid mode much more pronounced : figure [ fig5 ] shows that for deep intersections the separate `` islands '' of the hybrid mode merge into a joint broad `` ring '' ( within the first brillouin zone ) and the `` hot '' region occupies substantially larger part of the @xmath3-plane . the magnitude of i m @xmath90 for deep intersections is usually much larger than for the shallow intersections and therefore the threshold conditions to trigger the mode - coupling instability in the former case are substantially relaxed ( see discussion in sec . [ threshold ] ) . note that in this case , in order to keep the crystal in the marginally unstable regime ( which allows us to recover the fluctuation spectra ) we increased the damping rate to @xmath91 s@xmath92 . the dependence of dl modes on the direction of @xmath3 is evident from the discussion above . for the hybridization and the onset of the mode - coupling instability this dependence becomes crucial and therefore it deserves a separate discussion . figures [ fig3 ] and [ fig4 ] show that the first crossing of the out - of - plane and longitudinal in - plane modes always occurs at @xmath93 , at the border of the first brillouin zone . the area of the hybrid mode in the @xmath3 plane rapidly grows as the intersection gets deeper , and eventually the hybrid mode exists for any @xmath94 , as in fig . the crossing depth is controlled by changing the particle density or / and the eigenfrequency of the vertical confinement ( see sec . [ threshold ] ) . therefore , when the dl fluctuation spectra are measured in experiments , under conditions close to the onset of hybridization ( and , if not suppressed of the mode - coupling instability ) , special attention should be paid to the orientation of @xmath95 with respect to the crystalline lattice . under these circumstances the direction should be chosen according to the purpose of the experiment : if detection of the mode coupling is the goal , then @xmath84 is the best choice ; if unperturbed modes are to be measured , then @xmath56 should be used . the experimental technique must therefore provide angular resolution , which is easily done for the in - plane modes , but is not always possible for the out - of - plane mode ( since it is difficult to achieve angular positioning of the plasma crystal in an experiment ) . for instance , if the out - of - plane mode is measured with a fixed side - view camera @xcite then the angle @xmath94 is set during the experiment and can not be adjusted at the analysis stage , but if it is measured with a top - view camera ( from the variation of individual particle intensity ) @xcite then @xmath94-resolved wave spectra can be obtained . at @xmath84 and parameters of fig . arrows indicate traces of mixed polarization . far from the hybrid mode ( `` hot spots '' ) the modes become purely out - of - plane ( a ) and in - plane ( b ) . ] as we pointed out in sec . [ theory ] , the dl wave modes are independent only when the wake - mediated coupling can be neglected : far from the intersection or if the modes do not cross at all , the out - of - plane mode is purely transverse , one of the in - plane modes is longitudinal and the other is transverse . however , near the hybrid mode , where the coupling between the first two modes is strong , the theory predicts the emergence of _ mixed polarization_@xcite : as the modes approach the junction point , their polarizations become _ oblique _ ( in the plane formed by @xmath3 and the vertical axis ) , but the eigenvectors remain _ mutually orthogonal_. at the point where they merge the polarization exhibits a discontinuity the eigenvectors of the two modes form a single ( hybrid ) eigenvector with the elliptic polarization . traces of the mixed polarization for the out - of - plane mode have been observed in our recent experiment @xcite ( supplemental fig . s2 therein ) as well as in the experiment by liu _ et al . _ @xcite ( shown in their fig . 4b and erroneously interpreted as a `` new mode '' ) . in fig . [ fig6 ] we illustrate the mixed polarization observed in simulations ( performed for the conditions of our experiment @xcite ) . the magnitude of this effect ( @xmath96 ) is quite small for typical experimental conditions , so that its detection requires very careful analysis and high signal - to - noise ratio . the mode - coupling theory predicts two distinct thresholds : ( i ) _ confinement threshold _ , which determines when the out - of - plane and longitudinal in - plane modes intersect ; it depends on the combination of the vertical confinement frequency and the lattice constant ( particle number density ) . this threshold identifies the sufficient condition for the hybrid mode to form , but only the necessary condition for the instability to set in . the sufficient condition for the instability is determined by ( ii ) _ damping threshold _ this identifies the critical value of the damping rate ( controlled by pressure ) below which the instability is not suppressed . let us discuss thresholds ( i ) and ( ii ) separately ( we recall that all frequencies are normalized by @xmath97 ) . the first intersection of the out - of - plane and longitudinal in - plane modes takes place at the border of the brillouin zone @xcite , for @xmath3 at @xmath84 . then , using the expressions for @xmath55 and @xmath54 we readily obtain the condition for the hybrid mode to form ( confinement threshold ) , @xmath98 where @xmath99 . generally , the critical confinement frequency @xmath100 is a rather complicated function of @xmath36 , @xmath64 , and @xmath101 ; it is given in appendix [ appendix3 ] , eq . ( [ app_confinement1 ] ) . however , eq . ( [ elements1 ] ) suggests that to the accuracy @xmath102 the r.h.s . of eq . ( [ threshold_confinement1 ] ) is independent of @xmath64 and is proportional to @xmath103 . thus , in this limit of `` small '' @xmath64 the combination @xmath104 is a function of @xmath36 only ( i.e. , of the particle density ) which is given by eq . ( [ app_confinement2 ] ) . figure [ fig7]a shows this universal dependence in the useful range @xmath105 ( upper solid line ) . one should always keep in mind that @xmath106 can not be arbitrarily small @xcite : the out - of - plane mode @xmath107 decreases monotonically with @xmath77 and attains a minimum at the border of the first brillouin zone . if @xmath108 this formally implies instability . physically , this is because the vertical confinement in this case becomes too weak to keep the particles in the monolayer and the bifurcation to a multilayer crystal occurs ( i.e. , a monolayer crystal is no longer a ground state ) . thus , irrespective of the value of damping rate , a monolayer plasma crystal can only exist if the vertical confinement frequency exceeds the minimum value defined by is practically independent of @xmath94 at the border of the brillouin zone @xcite ( the variation is @xmath109 ) , so that in order to facilitate the calculations we substitute @xmath3 at @xmath110 . ] @xmath111 this value is given by eqs ( [ app_confinement3 ] ) and ( [ app_confinement4 ] ) for arbitrary and small values of @xmath64 , respectively . in the latter case we obtain a universal dependence @xmath112 on @xmath36 which is also plotted in fig . [ fig7]a ( lower solid line ) . the effect of `` finite '' @xmath64 [ i.e. , beyond the accuracy @xmath102 ] is illustrated in fig . [ fig8 ] for @xmath114 ( results are practically independent of @xmath36 in the range considered here ) . we plot the _ relative deviations _ of @xmath115 and @xmath116 from the values given by eqs ( [ app_confinement2 ] ) and ( [ app_confinement4 ] ) , respectively . the deviations increase as @xmath117 for @xmath118 and then rapidly saturate at the level @xmath119 . [ the latter is because both @xmath115 and @xmath116 do not depend on the wake parameters for @xmath120 , whereas at @xmath121 they scale as @xmath122 . ] note that the curves have almost identical shape at different @xmath101 , which indicates that the shown correction approximately scales as @xmath123 . given typical experimental errors ( see table [ tab1 ] and fig . [ fig10 ] ) , we conclude that the corrections shown in fig . [ fig8 ] can generally be neglected , i.e. , the accuracy of universal dependencies obtained in the limit of small @xmath64 is sufficient for practical purposes . in some cases , however , the effect of `` finite '' @xmath64 can nevertheless play an important role ( see sec . [ large_delta ] ) . finally , let us compare the `` exact '' 2d results ( in the limit of small @xmath64 ) with the results of approximate 1d nn mode - coupling theory summarized in sec . [ 1d_theory ] . for the 1d case , the critical and minimum confinement frequencies are derived from eq . ( [ modes_1d ] ) using the conditions @xmath124 and @xmath125 , respectively . for the comparison with the 2d results , we multiply the 1d results by constant `` form factors '' reflecting mismatch of 1d and 2d environments , which yields @xmath126 \widetilde\omega_{\rm conf}^{\rm(min ) } \simeq 2f_{\rm conf}^{\rm(min)}\sqrt{\left(\kappa^{-2}+\kappa^{-3}\right ) { \rm e}^{-\kappa}}. \end{array}\ ] ] the form factors @xmath127 and @xmath128 are determined from a fit to the 2d results , eqs ( [ app_confinement2 ] ) and ( [ app_confinement4 ] ) . the dependencies given by eq . ( [ threshold_confinement_1d ] ) are shown in fig . [ fig7]a by dashed lines . thus , we see that the 1d nn theory provides excellent qualitative description of the confinement thresholds even for small @xmath36 ; by multiplying these results with _ constant _ factors about unity we also get very good quantitative agreement with 2d results . as pointed out in sec . [ types ] , for @xmath84 ( as well as for @xmath129 ) the transverse in - plane mode becomes exactly decoupled . in this case the out - of - plane and longitudinal in - plane modes are governed by the following equation [ similar to that for the 1d nn model , eq . ( [ coupling_1d ] ) ] : @xmath130 in the vicinity of the intersection @xmath131 the coupled modes are described by eq . ( [ dispersion2 ] ) with @xmath132 . the group velocities @xmath133 and @xmath134 are collinear with @xmath3 due to the azimuthal symmetry at @xmath84 , so that the hybrid mode is given by @xmath135 thus , the width of the hybrid mode is @xmath136 , and the maximum of i m @xmath137 is attained at the crossing point @xmath43 . from this we immediately obtain the sufficient condition for the instability ( damping threshold ) , @xmath138 where @xmath139 and @xmath43 are determined from the crossing condition , @xmath140 in order to relate eq . ( [ threshold_pressure1 ] ) to the necessary instability condition [ confinement threshold , eq . ( [ threshold_confinement1 ] ) ] we introduce the _ confinement control parameter _ , @xmath141 which varies in the range from zero to @xmath142 . it is obtained from eq . ( [ threshold_pressure1a ] ) , @xmath143 where the r.h.s . is calculated as the difference at @xmath144 and @xmath145 . thus , eqs ( [ threshold_pressure1 ] ) and ( [ threshold_pressure2 ] ) determine the dependence of the critical damping rate @xmath146 on the control parameter @xmath147 via the parametric dependence on @xmath148 . for arbitrary @xmath64 , this dependence is derived in appendix [ appendix3 ] , eqs ( [ app_pressure1 ] ) and ( [ app_pressure2 ] ) . similar to the confinement threshold discussed in the previous subsection , for the damping threshold one can also obtain a universal dependence in the limit of small @xmath64 . it is calculated in eqs ( [ app_pressure3 ] ) and ( [ app_pressure4 ] ) , where the r.h.s . are functions of @xmath36 only ( along with the parameter @xmath148 ) and the dependence on the wake parameters is explicitly given by the l.h.s . by introducing the universal normalization for the critical damping rate and the confinement control parameter , @xmath149 we conclude that @xmath150 is a universal function of @xmath36 and @xmath151 . figure [ fig7]b shows @xmath150 versus @xmath151 for four characteristic values of @xmath36 ( solid lines ) . for `` shallow '' intersections , when @xmath106 is slightly below the critical value @xmath152 the critical damping rate increases as @xmath153 , which can be readily derived from eqs ( [ app_pressure3 ] ) and ( [ app_pressure4 ] ) . thus , for given @xmath36 and @xmath151 ( with @xmath154 ) monolayer plasma crystals are unstable if @xmath155 and stable otherwise . the critical damping rate is always bound from above , @xmath156 . for a given @xmath36 , the upper bound @xmath157 corresponds to @xmath158 and is marked by bullets in fig . these points shown in the inset as a function of @xmath36 identify the _ absolutely stable _ region : the mode - coupling instability in monolayer crystals is always suppressed when @xmath12 exceeds @xmath159 . [ cols="<,^,^,^,^,^,^,^,^,^,^,^ " , ] + @xmath160 the value is too high for the mode crossing ( see fig . [ fig10 ] ) , amended in erratum @xcite . + @xmath161 for these data , error bars are available : @xmath162 for @xmath36 and @xmath26 , @xmath163 for @xmath164 , @xmath165 for @xmath89 , and @xmath166 for @xmath30 . according to the 1d nn theory [ see appendix [ appendix1a ] , eq . ( [ hybrid_1d ] ) ] the critical damping rate is @xmath167 i.e. , it also scales as a square root of the control parameter . the upper bound for @xmath168 is obtained by substituting eq . ( [ threshold_confinement_1d ] ) for the maximum of @xmath147 , which yields @xmath169 where the corresponding form factor is @xmath170 . equation ( [ threshold_damping_1d ] ) , which is plotted in the inset of fig . [ fig7]b by the dotted line , also provides excellent agreement with the 2d model . hence , this is a convenient formula to identify conditions of absolute stability for a 2d crystal . wake - induced mode coupling has been observed in several experiments @xcite . however , the analysis of experimental results can be fastidious and needs to be done carefully . the fingerprints of the mode - coupling must be identified properly in order to avoid misinterpretations of the results . therefore , in this section we use the experiment of ref . @xcite as a characteristic example to demonstrate the principal features of the wake - induced mode coupling . then , we give an overview of literature currently available on experiments with 2d crystals ( stable and unstable ) @xcite , the summary of experimental parameters is given in table [ tab1 ] . finally , we discuss the stability of crystals in terms of thresholds introduced in the previous section . one of the most clear experimental evidence of mode - coupling in a 2d plasma crystal was reported in ref . @xcite ( experiment ii in table [ tab1 ] ) . a monolayer of microparticles was formed by levitating them in a plasma sheath above the lower rf electrode in a capacitively coupled discharge in a ( modified ) gec chamber . the microspheres were made of melamine - formaldehyde and had a diameter of @xmath171 m . the low argon pressure of @xmath172 pa ensured that the particle motion was only slightly damped . a high quality of the monolayer ( with no detected particles above or below ) was verified by using a side - view camera . details of the experimental procedure are given in refs.@xcite . the experimentally measured fluctuation spectra of two in - plane dl modes are shown in fig . [ fig9 ] . since this experiment was performed at conditions close to the onset of instability , all essential fingerprints of mode coupling are clearly visible : * the mode coupling critically depends on the direction of the wave vector @xmath3 . the comparison of wave spectra calculated for @xmath56 ( fig . [ fig9]a ) and @xmath84 ( fig . [ fig9]b ) reveals that the onset of mode coupling first occurs at @xmath84 this results in the prominent `` hot spots '' in fig . colored red , they are almost two orders of magnitude more intense than the `` normal '' branches of the fluctuation spectra . the hot spot that lies within the first brillouin zone ( left one ) represents the hybrid mode , the other one is its image in the higher brillouin zone . these observations are in agreement with theory and complimentary simulations ( fig . [ fig3 ] ) , which predicts that the mode coupling first takes place at @xmath84 for `` shallow '' mode intersection . * the less intense spot seen in fig . [ fig9]a at @xmath56 and @xmath173 is outside the first brillouin zone ( in fig . [ fig9]c the location is along @xmath174-axis , at the right edge ) . as explained in sec . [ remarks ] , this spot should be mapped into the first brillouin zone , where it represents the gap between two hot spots of the hybrid mode ( in fig . [ fig9]c , along the border of the brillouin zone at @xmath175 ) , and therefore its intensity is almost one order of magnitude smaller than the intensity of the hot spots . * traces of mixed polarization ( out - of - plane mode ) are clearly seen next to the hybrid mode in figs . [ fig9]a and [ fig9]b . this gives a direct experimental proof that the out - of - plane mode becomes oblique close to the intersection point . a discrepancy of @xmath176 between the experimentally measured in - plane projection of the out - of - plane mode and its theoretical dispersion relation can be explained by `` finite-@xmath64 '' correction to the theory shown in fig . [ fig8 ] ( which is about @xmath177 for the chosen values of wake parameters ) . * finally , all `` hot spots '' predicted by the theory and observed in simulations can be simultaneously seen in the experimental wave spectrum in @xmath3-plane , which is shown in fig . [ fig9]c . next , we analyze the stability of 2d plasma crystals in terms of the confinement threshold ( which determines when the hybrid mode emerges ) . available literature data on stable and unstable crystals is summarized in table [ tab1 ] , and the measured vertical confinement frequency @xmath178 is compared to the theoretical threshold @xmath179 in fig . [ fig10 ] . note that for the comparison we change the normalization of @xmath180 : instead of using @xmath97 a natural frequency scale in theory , now we employ the 2d analogue of dust plasma frequency , @xmath181 this normalization allows us to minimize the experimental errors , since the interparticle distance is usually measured with much higher accuracy than the screening length ( see footnote to table [ tab1 ] ) . figure [ fig10 ] shows that all experiments listed in table [ tab1 ] as having no mode crossing are indeed represented by points well above the confinement threshold line . to the contrary , the points for experiments marked by mode crossing are located very close to the threshold line or below it . the only exception is the experiment viii @xcite by liu _ et al . _ : the wake - induced mode coupling was clearly seen in the fluctuation spectra measured in @xcite , but the reported experimental parameters ( yielding @xmath182 , marked by green up triangle in fig . [ fig10 ] ) certainly do not allow the mode crossing , which indicates large measurement errors ( amended in the subsequent erratum @xcite ) . this example demonstrates that , by comparing experimental parameters to the theoretical confinement threshold near the point where the hybrid mode ( hot spot ) emerges in the fluctuation spectra one can perform very sensitive consistency test for the deduced parameters . at this point , let us briefly discuss the relation between the actual charge carried by particles and the effective charge which determines the dispersion properties of dl waves in 2d plasma crystals . the common method to determine @xmath25 and @xmath26 in 2d crystals is to fit the measured fluctuation spectra for the in - plane modes by theoretical curves . the method is based on the _ assumption _ that the interparticle interaction potential can be approximated by the yukawa form this deviation is negligible indeed ( see ref . @xcite ) . ] and provides rather high accuracy . however , the dispersion relations are determined by the combined effect of direct interparticle interactions and the wake - mediated interactions . while the former contribution scales as @xmath183 , the latter depends on a particular form of the wake . for the model of a point - like wake charge used in this paper the wake - mediated interactions scale as @xmath184 , which modifies the dispersion elements of the dynamical matrix ( [ elements1 ] ) . in the limit of small @xmath64 ( which turns out to be a remarkably accurate approximation , see fig . [ fig8 ] ) this yields the common renormalizing factor @xmath185 in eq . ( [ elements1 ] ) , and therefore the dispersion relations are determined by the _ effective _ charge @xmath186 . this fact should always be kept in mind when fitting the fluctuation spectra by theoretical dispersion relations : in theory , frequencies are normalized by @xmath187 and hence the explicit dependence on the factor @xmath185 naturally disappears . this allows us , e.g. , to directly compare the experimentally measured values of @xmath188 with the theoretical confinement threshold without knowing the wake parameters ( see fig . [ fig10 ] ) . of course , the comparison of theory and experiments in terms of the damping threshold requires knowledge of @xmath101 and @xmath64 , since the growth rate of the hybrid mode is directly proportional to the effective `` dipole moment '' of wake @xmath38 . as was shown in sec . [ threshold2 ] , in order to compare the measured value of @xmath189 ( last column of table [ tab1 ] ) with the theoretical damping threshold plotted in fig . [ fig7]b , we need to known the factor @xmath190 [ see eq . ( [ universal_norm ] ) ; additional factor @xmath103 in the denominator here is because @xmath191 in experiments is determined by @xmath192 . in experiments , however , the wake parameters are generally unknown , so that a reasonable assumption should always be made in order to perform the analysis in terms of the damping threshold . as we mentioned in sec . [ threshold2 ] , under certain conditions the `` finite-@xmath64 '' correction shown in fig . [ fig8 ] can become important . in particular , this happens in `` marginal cases '' , when the normalized value of the confinement frequency is very close to the corresponding threshold line ( viz . , when the distance to the line is comparable or less than the correction ) . for instance , such situation might occur for experiments ii , vi , ix , and x shown in fig . [ fig10 ] . however , typical experimental errors are relatively large , so that accounting for the corrections is hardly required in these cases . as regards the comparison of theory with numerical simulations , where all input parameters are known precisely , the `` finite-@xmath64 '' corrections can become crucial . to illustrate this , let us consider simulations of the shallow mode intersection shown in fig . the ( normalized ) confinement frequency in this case is @xmath193 _ larger _ than @xmath194 , i.e. , according to the universal criterion for the confinement threshold the system is still marginally stable . however , by taking into account the correction to the confinement threshold from fig . [ fig8 ] ( which is about @xmath195 in this case ) we conclude that the system is in fact marginally unstable as observed in the simulations . in contrast , for the case of the deep mode intersection shown in fig . [ fig5 ] ( where the confinement frequency is just @xmath196 smaller ) the correction plays practically no role . the theory of mode - coupling instability @xcite provides detailed picture of a _ plasma - specific _ melting scenario operating in 2d plasma crystals . the melting associated with the wake - mediated coupling between the longitudinal in - plane and out - of - plane modes can only be triggered if ( i ) the modes intersect and ( ii ) the neutral gas damping is sufficiently low . even if the instability is suppressed by the damping and the melting does not occur , the coupling always results in the formation of a hybrid mode which is revealed in fluctuation spectra by anomalous `` hot spots '' emerging at distinct positions in the @xmath3-plane . in the vicinity of the hybrid mode , one can observe traces of mixed polarization for the two intersecting wave branches . in this paper we showed that all these features can be considered as distinct fingerprints to identify the onset of the wake - induced mode coupling . this , in turn , allows us to determine certain combinations of the crystal parameters with a fairly good accuracy . for instance , the theory predicts a well - defined confinement threshold ( for the ratio of the vertical confinement frequency to the dust - lattice frequency ) at which the `` hot spots '' emerge . it is noteworthy that the effect of wakes on the dispersion relations is automatically taken into account via the charge renormalization , which yields the effective charge the value which is deduced from the analysis of experimental fluctuation spectra . on the other hand , the damping threshold as well as the evolution of the mode - coupling instability ( if not suppressed ) are determined by parameters of the wake which are still poorly known for experiments . by identifying the onset of the instability and comparing it with the damping threshold derived above , one could obtain effective `` dipole moment '' of the wake . furthermore , one could measure the energy growth rate at the initial instability stage and compare it with the theoretical prediction . systematic studies in this direction would be highly desirable . however , we note that such measurements would only give us some mean characteristics of the wake , e.g. , the first ( `` dipole '' ) moment in the multipole expansion . we believe that future research into the stability of 2d plasma crystals should be focused , in particular , on the implementation of self - consistent wake models @xcite for the calculation of the mode coupling ( which should also account for the spatial variation of the interaction parameters @xcite ) . such studies will allow us to gain deeper insight into the plasma - specific mechanisms of melting operating in 2d systems and provide more reliable basis for future research into generic melting mechanisms . in ref . @xcite we calculated the elements of @xmath1 with the accuracy @xmath102 . although usually @xmath64 is small indeed , such a linear expansion might not be always sufficient for some experimental conditions ( see sec . [ large_delta ] ) . below we present the results for arbitrary @xmath64 . for brevity , we introduce the following auxiliary functions : @xmath197 \xi(x)=(x^{-1}+3x^{-2}+3x^{-3}){\rm e}^{-x},\\[.2 cm ] \lambda(x)=\frac12[\xi(x)-\psi(x)].\\[.3 cm ] \end{array}\ ] ] then the dispersion elements of eq . ( [ matrix ] ) are given by the following sums over integer @xmath198 and @xmath199 with excluded @xmath200 : @xmath201\right\}\\ \hspace{5.8cm}\times\sin^2\frac12{\bf k}\cdot{\bf s},\\[.6 cm ] \alpha_{\rm v}=\sum\limits_{m , n}\left\{\lambda(\kappa s)-\tilde q \left[\lambda(\kappa s_{\delta})- \tilde\delta^2s_{\delta}^{-2}\xi(\kappa s_{\delta})\right]\right\}\\ \hspace{5.8cm}\times\sin^2\frac12{\bf k}\cdot{\bf s},\\[.6 cm ] \beta=\sum\limits_{m , n}\left[\xi(\kappa s)-\tilde qs^2s_{\delta}^{-2}\xi(\kappa s_{\delta})\right]\\ \hspace{3.9cm}\times(s_y^2-s_x^2)s^{-2}\sin^2\frac12{\bf k}\cdot{\bf s},\\[.6 cm ] \gamma=\sum\limits_{m , n}\left[\xi(\kappa s)-\tilde qs^2s_{\delta}^{-2}\xi(\kappa s_{\delta})\right ] s_xs_ys^{-2}\sin^2\frac12{\bf k}\cdot{\bf s}. \end{array}\ ] ] here , the vector @xmath202 with the components @xmath203 and @xmath204 and the absolute value @xmath205 characterizes the relative positions of all neighbors in the hexagonal lattice ; we also introduced @xmath206 . the coupling elements are given by @xmath207 in the nn approximation ( formally applicable for @xmath208 ) , the modes @xmath209 and @xmath210 ( whose coupling causes the hybridization ) obtained from eqs ( [ elements1 ] ) and ( [ elements2 ] ) coincide with eqs . ( [ modes_1d])-([coupling_coefficient_1d ] ) for a 1d string . in the limits @xmath211 or @xmath212 , eqs ( [ elements1 ] ) and ( [ elements2 ] ) reduce to conventional ( wake - free ) results for dl modes in 2d crystals . for @xmath213 and finite @xmath101 one also obtains the wake - free results , but with the particle charge renormalized by the factor @xmath103 . the intersection point between the horizontal and vertical modes , @xmath214 , is readily derived from eq . ( [ modes_1d ] ) : @xmath215 where @xmath100 is given by the first eq . ( [ threshold_confinement_1d ] ) with @xmath216 . the maximum growth rate of the hybrid mode is obtained from eq . ( [ coupling_1d ] ) , @xmath217 where @xmath218 is given by eq . ( [ coupling_coefficient_1d ] ) . this yields the critical damping rate @xmath219 . the charged particles were confined in a vertical parabolic well with an eigenfrequency which was varied in the range of 15 hz @xmath220 25 hz . in the horizontal plane , two options were investigated : parabolic confinement with an eigenfrequency @xmath221 hz and periodic boundary conditions ( with @xmath222 ) . the equation of motion for each particle is @xmath223 where @xmath198 is the particle mass , @xmath12 is the damping rate , @xmath224 the position of the @xmath225th particle , and @xmath226 represents the confinement force . the force of interparticle interaction is @xmath227 where @xmath228 is the distance from the particle @xmath225 to the particle @xmath229 , @xmath230 is the distance to the wake of the particle @xmath229 , @xmath25 is the particle charge , and @xmath27 is the point - like charge of the wake placed at a distance @xmath28 below each particle . the langevin force is defined as @xmath231 where @xmath232 is the temperature of the thermostat ( here , @xmath28 is the kronecker delta or the dirac delta function ) . in order to simulate systems of a size comparable to experimental ( with @xmath233 particles ) without any approximation ( i.e. , without cutoff radius in the interactions ) , the code was written in cuda c and run on nvidia tesla c1060 gpu or c2050 gpu computing cards . the equations of motion were solved using beaman algorithm with predictor - corrector . to derive the confinement threshold , we substitute the dispersion elements from eq . ( [ elements1 ] ) in eq . ( [ threshold_confinement1 ] ) which yields here the fact that @xmath235 is taken into account , so that it is convenient to introduce the parameter @xmath236 $ ] ( which is equal to 1 or 0 for odd or even @xmath198 , respectively ) . furthermore , for brevity we use @xmath237 . for the accuracy @xmath102 , we can neglect in eq . ( [ app_confinement1 ] ) the second term in curly braces and set @xmath238 , which results in the following expression : @xmath239 where the r.h.s . is a function of @xmath36 only . thus , eq . ( [ app_confinement2 ] ) determine the universal dependence of @xmath100 on @xmath36 . for the damping threshold , we substitute the coupling element [ eq . ( [ elements2 ] ) ] in eq . ( [ threshold_pressure1 ] ) and the dispersion elements in eqs ( [ threshold_pressure1a ] ) and ( [ threshold_pressure2 ] ) . after some algebra we get @xmath242 + \tilde q\tilde\delta^2s_{\delta}^{-2}\left(1+p\right)\xi(\kappa s_{\delta})\right\}\sin^2\frac12|{\bf k}_0|s_y } } \label{app_pressure1 } , \\ \pi & = & \sum\limits_{m , n}\left(\eta_m-\sin^2{\textstyle\frac12}|{\bf k}_0|s_y\right)\left\{\left(1+p\right)\left[\xi(\kappa s ) -\tilde q\xi(\kappa s_{\delta})\right]+\tilde q\tilde\delta^2s_{\delta}^{-2}\left(3+p\right ) \xi(\kappa s_{\delta})\right\}. \label{app_pressure2}\end{aligned}\ ] ] @xmath243 \sin^2\frac12|{\bf k}_0|s_y } } , \label{app_pressure3}\\ \frac{\pi}{1-\tilde q } & = & \sum\limits_{m , n}\left(\eta_m-\sin^2{\textstyle\frac12}|{\bf k}_0|s_y\right)\left(1+p\right)\xi(\kappa s),\label{app_pressure4}\end{aligned}\ ] ] where the r.h.s . depend only on @xmath36 and @xmath148 . thus , eqs ( [ app_pressure3 ] ) and ( [ app_pressure4 ] ) determine the universal dependence of the critical damping rate @xmath146 on the confinement control parameter @xmath147 via the parametric dependence on @xmath148 . 44ifxundefined [ 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 \doibase doi : 10.1016/j.physrep.2005.08.007 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.81.1353 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.41.121 [ * * , ( ) ] http://stacks.iop.org/0022-3719/6/i=7/a=010 [ * * , ( ) ] link:\doibase 10.1103/physrevb.19.2457 [ * * , ( ) ] link:\doibase 10.1103/physrevb.19.1855 [ * * , ( ) ] link:\doibase 10.1103/physrevb.28.178 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.58.1200 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physreve.53.r2049 [ * * , ( ) ] link:\doibase 10.1063/1.3205882 [ * * , ( ) ] link:\doibase 10.1063/1.2993180 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.3656 [ * * , ( ) ] link:\doibase 10.1103/physreve.62.1238 [ * * , ( ) ] s. nunomura , s. zhdanov , d. samsonov , and g. morfill , phys . lett . * 94 * , 045001 ( 2005 ) . link:\doibase 10.1103/physrevlett.83.1970 [ * * , ( ) ] g. morfill , a. v. ivlev , and j. r. jokipii , phys . lett . * 83 * , 971 ( 1999 ) . @noop * * , ( ) link:\doibase 10.1063/1.872112 [ * * , ( ) ] link:\doibase 10.1063/1.872717 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1063/1.1609988 [ * * , ( ) ] lampe , m. , g. joyce , g. ganguli , and v. gavrishchaka , 2000 , phys . plasmas * 7 * , 3851 . link:\doibase 10.1103/physreve.64.046406 [ * * , ( ) ] w. j. miloch , j. trulsen , and h. l. pcseli , phys . e * 77 * , 056408 ( 2008 ) . link:\doibase 10.1063/1.2730498 [ * * , ( ) ] r. kompaneets , s. v. vladimirov , a. v. ivlev , and g. morfill , new j. phys . * 10 * , 063018 ( 2008 ) . link:\doibase 10.1103/physrevlett.83.3194 [ * * , ( ) ] link:\doibase 10.1103/physreve.54.r46 [ * * , ( ) ] g. joyce , m. lampe , and g. ganguli , phys . lett . * 88 * , 095006 ( 2002 ) . a. v. ivlev and g. morfill , phys . e * 63 * , 016409 ( 2001 ) . link:\doibase 10.1103/physreve.68.026405 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.103.215001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.104.195001 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevlett.105.085004 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.269901 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.71.2753 [ * * , ( ) ] link:\doibase 10.1103/physreve.68.046403 [ * * , ( ) ] link:\doibase 10.1103/physreve.71.026410 [ * * , ( ) ] link:\doibase 10.1103/physreve.56.r74 [ * * , ( ) ] link:\doibase 10.1063/1.2188406 [ * * , ( ) ] @noop _ _ , edited by ( , , ) @noop _ _ , edited by ( , , ) v. v. yaroshenko , a. v. ivlev , and g. e. morfill , phys . e * 71 * , 046405 ( 2005 ) . r. kompaneets , a. v. ivlev , v. tsytovich , and g. morfill , phys . plasmas * 12 * , 062107 ( 2005 ) .

we study one and two hole properties of the model at half - filling on the square lattice using series expansion methods at . the dispersion curve for one hole excitations is calculated and found to be qualitatively similar to that obtained by other methods , but the bandwidth for small is some 20% larger than given previously . we also obtain the binding energy and dispersion relation for two hole bound states . the lowest bound state as increases is found to be first-wave , and then-wave , in accordance with predictions based upon the kohn - luttinger effect . we also make a similar study for the model .

cond-mat9806367

models of strongly correlated electrons in two dimensions have been much studied in recent years as part of a search for a minimal model of the high - t@xmath6 superconducting cuprates , and other fascinating new antiferromagnetic systems@xcite . the @xmath0 model , defined by the hamiltonian @xmath7 has become a generic model for mobile electrons / holes in doped antiferromagnets . in this expression the @xmath8 , @xmath9 are the usual electron creation and destruction operators , @xmath10 are the electron spin operators and the sum is over nearest neighbors pairs . @xmath11 is a projection operator which ensures that doubly occupied states are excluded . the @xmath0 model originally was derived@xcite as the strong correlation limit of the hubbard model , with @xmath12 , but it is now generally treated as an effective hamiltonian in its own right , with @xmath13 and @xmath14 independent parameters . for the cuprates it is estimated@xcite that @xmath15 , which is well outside the region of validity of the hubbard mapping . an independent `` derivation '' of the @xmath0 model as an effective hamiltonian for the cuprates has been presented by zhang and rice@xcite . at exactly half - filling ( one electron per site ) the kinetic term vanishes because of the single - occupancy restriction and the model reduces to an antiferromagnetic insulator . it is known from a variety of numerical and analytic calculations@xcite that this model , on the square lattice , has nonzero long range antiferromagnetic ( nel ) order , reduced by strong quantum fluctuations . removal of a small number of electrons ( by doping ) will allow mobility of holes and will reduce the antiferromagnetic order . the physics of a small number of holes in a dynamic antiferromagnetic background remains a challenging problem . previous studies of the @xmath0 model at @xmath1 have used a variety of approaches . exact diagonalization methods have been used extensively on clusters up to 32 sites@xcite , but suffer from substantial finite - size corrections and an inability to treat extended excitations . green s function monte carlo calculations have been performed for larger lattices , up to @xmath16@xcite , and show , for example , a significant decrease of the two - hole binding energy with increasing lattice size . a variety of variational approaches and analytic many body methods have also been employed@xcite . the picture that emerges from this body of work is roughly as follows . for one hole the ground state energy is given by @xmath17 where the coefficients are approximate numerical estimates . the dispersion curve for 1-hole states is qualitatively similar from different methods , with minima at @xmath18 and equivalent points . the bandwidth from the largest diagonalizations scales as @xmath19 where the coefficients are again estimates and are not known with great precision . the spectral function for 1-hole states has been calculated@xcite and shows a clear quasiparticle peak , together with a continuum at higher energies . for two holes the binding energy , spectral function and pair susceptibility have been computed . the results support the existence of a bound state with @xmath20 and d - wave symmetry for @xmath21 , with @xmath22 although not all methods agree and this is still controversial . the dispersion curve for two - hole bound states is also of interest @xmath23 to our knowledge the only calculation of this is due to eder@xcite . the aim of the present work is to study the physics of one - hole and two - hole states in the half - filled @xmath0 model at zero temperature via linked - cluster series expansions . this approach has been developed extensively in recent years by our group@xcite and others@xcite . most of the applications to date have been to spin systems , although we have included fermions in lattice gauge calculations . shi and singh@xcite have used this approach to study the hubbard model at half filling , via an expansion about an additional ising term included in the hamiltonian . for completeness we also mention the use of high - temperature expansions to study the @xmath0 model@xcite . the present work of course addresses different questions which can not be probed by high - temperature series . the plan of the paper is as follows . in sec . ii , the series expansion method is briefly reviewed , and the results for one - hole properties of the @xmath0 model are presented . in sec . iii , we study the dispersion relation and the binding energy for two - hole bound states . in sec . iv , we study the one - hole and two - hole properties of the @xmath5 model . v is devoted to a summary and discussion . the series expansion method is based on a linked cluster formalism associated with standard rayleigh - schrdinger perturbation theory . the ground state energy @xmath24 for a lattice of @xmath25 sites can be written as a sum over all connected clusters @xmath26 as @xmath27 where @xmath28 is a geometrical embedding factor and @xmath29 is the `` cumulant energy '' of cluster @xmath30 . the same formula can be used for the energy of any particular cluster @xmath31 @xmath32 where the sum is now over all sub - clusters embeddable in @xmath31 . the cluster energy is calculated perturbatively , then the cumulant energies are obtained iteratively from eq . ( [ eq3 ] ) and substituted into eq . ( [ eq2 ] ) to obtain the final series . the formalism for excited states , needed to compute the dispersion curves , is a little more complex@xcite . we introduce an ising spin anisotropy @xmath33\end{aligned}\ ] ] where @xmath34 , and rewrite the hamiltonian for the @xmath0 model as @xmath35 with @xmath36 with a staggered @xmath37-field of strength @xmath38 to improve convergence@xcite . this term vanishes in the total hamiltonian in the isotropic exchange limit @xmath39 . the ising part of the exchange , @xmath40 , forms the unperturbed hamiltonian while the perturbation @xmath41 includes both the hole hopping terms and transverse spin fluctuations . the original hamiltonian for the @xmath0 model is recovered in the limit @xmath42 ( i.e. @xmath43 ) . we expect the series expansion to give more reliable results in the small @xmath2 region , where the perturbation @xmath41 is less important . we directly compute the series for the ground state energy difference between the system with one ( or two ) hole(s ) and the system with no hole . two types of series can be obtained in this way . by fixing @xmath44 and @xmath38 we can compute series in the single variable @xmath45 up to order @xmath46 ( where @xmath46 is typically of order 11 - 13 ) , @xmath47 we refer to these as @xmath45-series . extrapolation to @xmath42 is carried out by standard pad approximant or integrated differential approximant methods@xcite . in the other approach we keep both @xmath45 and @xmath48 as expansion parameters and obtain double series of the form @xmath49 where the coefficients @xmath50 are computed up to order @xmath51 . extrapolation to @xmath42 then yields a series in @xmath2 , @xmath52 which we refer to as a @xmath13-series . we first consider states with one hole , which may exist either on the up - sublattice ( a ) or the down - sublattice ( b ) , where the up and down sublattice are represented in fig . [ fig_lattice](a ) by open and full circles , respectively . here @xmath53 are the usual primitive translation vectors . the unperturbed wave function for the zero - momentum spin - up state is taken to be @xmath54 where @xmath55 is the classical nel state . now a hole in the up ( down ) sublattice can never transform into a single hole in the down ( up ) sublattice , since the hamiltonian conserves the total spin @xmath56 ; so that we should choose @xmath57 shown in fig . [ fig_lattice](a ) as the primitive translation vectors , rather than @xmath58 . the unit vectors @xmath57 are those of the antiferromagnetic bipartite lattice . the brillouin zones in reciprocal space corresponding to these different choices are shown in fig . [ fig_lattice](b ) . most previous calculations , however , have used @xmath53 as unit vectors ( corresponding to @xmath59 in momentum space ) , instead of @xmath60 , so we will also conform to this convention ( and set @xmath61 to one ) unless it is specified otherwise . for momentum @xmath62 we obtain the energy to order @xmath63 , involving 8739 distinct clusters containing up to 13 sites . for @xmath64 it is necessary to distinguish between clusters having different spatial orientation . we expand the energy to order @xmath65 , involving 23546 clusters with up to 11 sites . the series coefficients for @xmath66 at fixed momentum @xmath67 and different @xmath38 are listed in table [ tab_1hmin ] , where we can see the series converges best around @xmath68 . the extrapolations of these series by using the integrated differential approximant methods are shown in figure [ fig_diffr ] , where we can see that at @xmath42 , the extrapolations from different @xmath38 series agree with each other very well , while the case @xmath69 gives the smallest error bar . in table [ tab_tjk ] , we list the full spectrum series for @xmath66 and @xmath70 . the other coefficients are too lengthy to publish but can be supplied on request . in figure [ fig_ek1h ] we show the energy of a one - hole state for @xmath71 along symmetry directions in the brillouin zone , for various ratios @xmath2 obtained from @xmath45-series . the lowest energy occurs at @xmath72 and equivalent points . this result , and the overall shape of the one - hole dispersion curve are in very good agreement with previous work@xcite . we note that the symmetry and the redundant information evident in fig . [ fig_ek1h ] are the results of choosing a brillouin zone which is two times too big . if we choose @xmath73 as defining the bipartite brillouin zone , the corresponding dispersion relation along symmetry directions in the brillouin zone is shown in fig . [ fig_ek1h2 ] . figure [ fig_w ] shows the one - hole bandwidth , @xmath74 , as a function of @xmath75 , compared with results from other approaches . for @xmath75 around 0.5 our results agree well with other work , but for large @xmath75 our result is significantly larger than that obtained by martinez and horsch@xcite . also we found a peak value of @xmath76 at @xmath77 , rather than @xmath78 at @xmath79 as obtained by martinez and horsch@xcite , and liu and manousakis@xcite . using the two - variable expansion we obtain the one - hole energy @xmath80 at some particular values of @xmath71 : @xmath81 \nonumber \\ e^{\rm 1h}(\pi/2,\pi/2)/j & = & 2.164(2 ) -3.91(4 ) ( t / j)^2 + 9.1(4 ) ( t / j)^4 + o[(t / j)^6 ] \label{eq_e1h } \\ e^{\rm 1h}(\pi,0)/j & = & 2.164(2 ) -2.68(2 ) ( t / j)^2 + 3.06(4 ) ( t / j)^4 + o[(t / j)^6]\nonumber % old res . from y=0 only % e^{\rm 1h } ( 0,0)/j & = & 2.164(3 ) -0.650(6 ) ( t / j)^2 - 0.036(9 ) ( t / j)^4 + o[(t / j)^6 ] \nonumber \\ % e^{\rm 1h}(\pi/2,\pi/2)/j & = & 2.164(3 ) -3.91(4 ) ( t / j)^2 + 9.1(4 ) ( t / j)^4 + o[(t / j)^6 ] \label{eq_e1h } \\ % e^{\rm 1h}(\pi,0)/j & = & 2.164(3 ) -2.70(4 ) ( t / j)^2 + 3.01(4 ) ( t / j)^4 + o[(t / j)^6]\nonumber\end{aligned}\ ] ] these coefficients are not exact as they involve an extrapolation to @xmath42 . the leading coefficient 2.164(3 ) can be compared with 2.1552 from a spin wave calculation@xcite and 2.193(7 ) from a monte carlo simulation@xcite . the bandwidth can be obtained as @xmath82 \label{w_2p}\ ] ] this result is also plotted in figure [ fig_w ] as a dashed line , and it agrees well at small @xmath2 with that obtained from @xmath45-series . the leading coefficient 3.26(5 ) should be compared with previous calculations giving a value of 2@xcite or 2.6@xcite . in the previous section we developed a series expansion approach to calculate one - hole properties of the @xmath0 model , and showed how reliable quantitative results could be obtained . in the present section we extend this approach to consider states with two holes . we start from an unperturbed wave - function consisting of two nearest - neighbour holes with opposite spin , and overall momentum @xmath71 : @xmath83 here again there are two different choices of brillouin zone possible . one may choose to use a diagonal unit vector of length @xmath84 ( where @xmath61 is the lattice spacing ) , which connects the centre points of two adjacent bonds of the original lattice as shown in fig . [ fig_lattice2h](a ) ; this corresponds to the large , outer brillouin zone in fig . [ fig_lattice2h](b ) . this was basically the convention used by eder@xcite . alternatively , and more correctly , one may choose a diagonal unit vector of length @xmath85 , differentiating between the @xmath86 and @xmath87 sublattices . this gives rise again to the brillouin zone of the bipartite lattice , marked as the inner zone in fig . [ fig_lattice2h](b ) . in this second case , there are 4 independent bond states in the unit cell , and one may form linear combinations of them making up the conventional @xmath88- , @xmath89- , @xmath90- and @xmath3-wave states . with the first choice , all these four states are included within one brillouin zone : the points corresponding to zero - momentum @xmath88- , @xmath89- , @xmath90- and @xmath3-wave states are shown- '' and `` @xmath90- '' wave states are degenerate by reflection symmetry . the points marked @xmath89 and @xmath90 in fig . [ fig_lattice2h](b ) are really linear combinations of the proper @xmath89 and @xmath90 states . [ fig_lattice2h](b ) . then one finds that the different states are related by equations such as @xmath91 ( in units where @xmath84 has been rescaled to one ) . this relationship is evident in fig . 19 of ref . @xcite . we have computed series for the two - hole spectrum @xmath92 up to order @xmath65 . this involves 46,440 distinct clusters of up to 12 sites . the @xmath45-series coefficients for @xmath66 and @xmath70 are listed in table [ tab_tjk ] . using the two - variable expansion , after extrapolation to @xmath42 , we obtain the following estimates for the two - hole @xmath88- , @xmath4- and @xmath3-wave pair energy : @xmath93\nonumber \\ e^{2p}/j & \equiv & e^{\rm 2h}(0,\pi)/j = 3.798(2 ) - 3.705(10 ) ( t / j)^2 - 4.1(1 ) ( t / j)^4 + o[(t / j)^6 ] \label{eq_e2h } \\ e^{2s}/j & \equiv & e^{\rm 2h}(\pi,\pi)/j = 3.798(2 ) - 0.52(3 ) ( t / j)^2 - 1.62(5 ) ( t / j)^4 + o[(t / j)^6 ] \nonumber \end{aligned}\ ] ] so the binding energies defined by @xmath94 for @xmath88- , @xmath4- and @xmath3-wave pair states are : @xmath95 \nonumber \\ e_{\rm b}^{2p}/j & = & -0.529(4 ) + 4.1(1 ) ( t / j)^2 - 22.3(9 ) ( t / j)^4 + o[(t / j)^6 ] \label{eq_binding } \\ e_{\rm b}^{2s}/j & = & -0.529(4 ) + 7.3(1 ) ( t / j)^2 - 19.8(9 ) ( t / j)^4 + o[(t / j)^6 ] \nonumber\end{aligned}\ ] ] the binding energy for @xmath96 has previously been given as @xmath97 by spin wave theory@xcite , and @xmath98 by monte carlo@xcite . our result is in relatively good agreement but suggests that the monte carlo calculations may overestimate the binding energy somewhat . we have also computed the dispersion curves for the two - hole bound state from @xmath45-series , and these are shown in figure [ fig_ek2h ] , for various small @xmath2 . the minimum occurs at @xmath99 ( @xmath3-wave pair ) , and the maximum occurs at @xmath100 ( @xmath88-wave pair ) , while the energy at @xmath100 is almost unchanged as @xmath2 increases . the curves are quite flat for small @xmath2 , as expected , and develop more structure as @xmath2 increases . the results are broadly similar to those of eder@xcite . the @xmath88- , @xmath4- and @xmath3-wave pair binding energies , obtained from both the @xmath45-series and the two - variable expansion as a function of @xmath2 , are shown in figure [ fig_binding ] . note that for @xmath45-series , the extrapolation is done on series for @xmath101 , rather than the 1-hole and 2-hole energies individually , because values of these two energies are quite large , while their difference is usually small . the convergence is excellent for small @xmath2 , and we believe our results are very accurate there . the uncertainty increases for larger @xmath13 . we can see from this figure that as @xmath2 increases from zero , the @xmath88-wave binding energy decreases very quickly , and the binding disappears around @xmath102 . the @xmath4-wave binding energy also decreases quickly in the small @xmath2 region , but it turns out to be very flat beyond @xmath103 , and the binding energy is very small . the @xmath3-wave binding energy decreases slightly as @xmath2 increases from zero . our results are consistent with the @xmath3-wave binding disappearing around @xmath104 . beyond that point , the @xmath4-wave state appears to have the lower energy , and remains very weakly bound according to our results . figure [ fig_binding ] also shows some other numerical results for the 2-hole binding energy . the solid circles with error bars are gfmc results for the @xmath16 lattice obtained by boninsegni and manousakis@xcite , and lie somewhat below our result , though broadly similar in trend : the discrepancy may easily be attributed to a finite - size effect . the solid triangles are the results of a recent 32-site finite - lattice calculation@xcite . the open circles are the results of a linear extrapolation in @xmath105 from the finite - lattice data by chernyshev , leung and gooding@xcite ; but they discuss reasons why such a naive extrapolation is not to be relied on especially at large @xmath2 . the results lie somewhat above our series results for the @xmath3-wave binding energy , but again have a broadly similar trend . chernyshev , leung and gooding@xcite find that the lowest bound state on the 32-site lattice is a singlet state with @xmath106 symmetry . the apparent cross - over between the @xmath4-wave and @xmath3-wave bound states came initially as a surprise . but in fact it accords very well with predictions@xcite based on the kohn - luttinger effect . in a repulsive fermi liquid , pairing in high - angular momentum states can occur due to a singularity in the 2-particle scattering amplitude . this effect was previously believed not to occur in two dimensions , because it does not appear up to second order in perturbation theory ; but it was recently shown@xcite that a singularity does arise at higher order . chubukov@xcite showed that in the dilute limit the effect survives down to @xmath107 and gives rise to @xmath4-wave pairing . he also pointed out that the same effect will occur in an arbitrarily dense fermi liquid . both numerical@xcite and analytical@xcite approaches have shown that at low densities @xmath4-wave pairing is dominant for @xmath108 , @xmath3-wave is dominant at @xmath109 , and there is a region of @xmath88-wave dominance for @xmath110 . figure [ fig_binding ] appears to indicate that a similar pattern extends even to the region of half - filling . in fact , the earlier variational calculation in the hamiltonian . ] of eder also shows a crossover between @xmath4-wave and @xmath3-wave states ( fig . 21 of ref . @xcite ) at around @xmath111 . several theoretical approaches@xcite have predicted that magnon exchange should give rise to shallow , long - range bound states even at large @xmath2 values . the behaviour of the @xmath4-wave state looks consistent with this idea , but not the @xmath3-wave state . it is possible that another state ( or states ) in the @xmath3-wave sector could cross over the one we have calculated , and become the lowest energy state at large @xmath2 . such a crossover could typically imply a square - root singularity in the energy eigenvalue , and the series extrapolation would no longer be valid beyond that point . we also study a simplified version of the @xmath0 model , the @xmath5 model , which has an ising rather than a heisenberg spin interaction @xmath112 here we naturally take the first term in @xmath113 as the unperturbed hamiltonian , and the second term as the perturbation , and perform an expansion in @xmath48 . this expansion only converges well up to about @xmath114 . to improve the convergence in the large @xmath48 region , we can include a staggered @xmath37-field of strength @xmath38 as in the @xmath115 model , and perform the following separation , @xmath116 where @xmath117 we then take @xmath40 as the unperturbed hamiltonian and @xmath41 as the perturbation , and perform an expansion in @xmath45 for fixed values of @xmath48 and @xmath38 . here again we need to extrapolate the series to @xmath42 . the series for the one - hole state have been computed up to order 21 ( that is , to @xmath118 for the hamiltonian in eq . ( [ h_tjz ] ) or to order @xmath119 for the hamiltonian in eq . ( [ h_tjz2 ] ) ) involving a list of 23546 clusters with up to 11 sites , while the series for the two - hole bound state have been computed up to order 18 , involving a list of 23546 clusters up to 11 sites . the final series in @xmath48 for the spectrum of the one - hole state and two - hole bound state are listed in table [ tabtjz ] . [ fig_ek1h_tjz ] shows the one - hole spectrum for some particular values of @xmath48 . here we can see that the shape of the dispersion curve is quite different from that for the @xmath0 model : the lowest energy occurs at @xmath100 rather than @xmath18 as in the @xmath0 model , and the bandwidth here is about 10 times smaller than that for the @xmath0 model . the minimum one - hole energy as a function of @xmath120 is shown in fig . [ fig_e1hmin ] . in the interval @xmath121 , the minimum one - hole energy can be very well fitted@xcite as @xmath122 where the linear term comes from the @xmath123 term in the hamiltonian . this is in good agreement with previous estimates@xcite . the two - hole dispersion is shown in fig . [ fig_ek2s_tjz ] . we note that the energy at @xmath100 is exactly @xmath124 , independent of @xmath13 , and the energy along path @xmath125 ( @xmath4-wave pair ) to @xmath99 ( @xmath3-wave pair ) is very flat . the @xmath88- , @xmath4- and @xmath3-wave binding energies defined by @xmath126 are shown in figure [ fig_binding_tjz ] . we can see from this figure that @xmath88-wave binding disappears around @xmath127 . the @xmath3-wave binding energies are slightly greater than for @xmath4-wave when @xmath128 , but when @xmath129 , the @xmath4-wave pair has a greater binding energy than the @xmath3-wave , similarly to that found for the full @xmath0 model . thus the @xmath5 model also shows a crossover between @xmath3-wave and @xmath4-wave bound states , occurring even earlier than in the full @xmath0 model . again , the earlier variational calculations of eder showed very similar results ( fig . 12 of ref . @xcite ) . the solid points in figure [ fig_binding_tjz ] are the results of a reduced basis diagonalization study on clusters of up to 50 sites by riera and dagotto@xcite . the trend of these results agrees extremely well with the series estimates , although their estimates lie systematically a little lower . they do not discuss the symmetry of the bound state@xcite . they find the binding of the 2-hole state disappears at @xmath130 , or @xmath131 , which agrees very well with our results . the @xmath0 model on the square lattice has been studied by many workers over the last decade , using a variety of techniques . these methods are all approximate and have uncertain accuracy . independent confirmation from a variety of approaches increases one s confidence in the results . we have used the linked - cluster series expansion method to study one and two hole properties of the half - filled @xmath0 and @xmath5 models in 2-dimensions at @xmath1 . to our knowledge this is the first time this has been done . our results for 1-hole and 2-hole states can be summarized as follows : * the 1-hole dispersion curve has been computed throughout the brillouin zone , and agrees in shape with previous calculations . * the minimum for 1-hole dispersion occurs at @xmath132 and equivalent points , again in agreement with previous work . * the bandwidth @xmath133 for the 1-hole dispersion curve of the @xmath0 model has been calculated . at small @xmath75 it agrees with the spin - wave calculations of martinez and horsch@xcite and liu and manousakis@xcite , but beyond @xmath134 our results lie substantially higher than theirs . we get a peak value of @xmath135 at @xmath136 , rather than @xmath78 at @xmath137 as given in previous work . * we have computed the dispersion curve for 2-hole bound states . to our knowledge this has not been well calculated previously . * in the limit @xmath138 of the @xmath0 model , we can compare our results for the one - hole and two - hole energy ( @xmath139 and @xmath140 ) , the binding energy ( @xmath141 ) , and the bandwidth for the one - hole state ( @xmath133 ) with previous spin - wave@xcite , monte carlo@xcite and other calculations . + [ cols="<,^,<",options="header " , ] rr|rr|rr|rr & & & & & & & + + ( 0 , 0 , 0 ) & 2.000000000 & ( 8 , 2 , 0 ) & -8.693161815@xmath142 & ( 5 , 1 , 1 ) & -3.650599647@xmath143 & ( 11 , 5 , 1 ) & -2.756493796@xmath142 + ( 2 , 0 , 0 ) & -5.333333333@xmath143 & ( 9 , 2 , 0 ) & 2.176985203@xmath143 & ( 6 , 1 , 1 ) & 2.714260859@xmath143 & ( 6 , 2 , 2 ) & 7.130158730@xmath142 + ( 4 , 0 , 0 ) & 1.529021164@xmath143 & ( 10 , 2 , 0 ) & -1.017391598@xmath142 & ( 7 , 1 , 1 ) & 2.890627842@xmath143 & ( 7 , 2 , 2 ) & -6.731669606@xmath142 + ( 5 , 0 , 0 ) & 9.481481481@xmath142 & ( 11 , 2 , 0 ) & -3.628184110@xmath143 & ( 8 , 1 , 1 ) & -2.000742064@xmath143 & ( 8 , 2 , 2 ) & -1.075223788@xmath143 + ( 6 , 0 , 0 ) & -1.772367448@xmath142 & ( 6 , 4 , 0 ) & 1.188359788@xmath142 & ( 9 , 1 , 1 ) & -1.603987396@xmath143 & ( 9 , 2 , 2 ) & 1.539341160@xmath143 + ( 7 , 0 , 0 ) & -1.684678469@xmath143 & ( 7 , 4 , 0 ) & -4.980599647@xmath144 & ( 10 , 1 , 1 ) & -1.153947213@xmath143 & ( 10 , 2 , 2 ) & 9.103499935@xmath142 + ( 8 , 0 , 0 ) & -6.427987242@xmath142 & ( 8 , 4 , 0 ) & -1.402790612@xmath142 & ( 11 , 1 , 1 ) & 1.409738327@xmath143 & ( 11 , 2 , 2 ) & -2.080643863@xmath143 + ( 9 , 0 , 0 ) & 2.538548842@xmath143 & ( 9 , 4 , 0 ) & 2.637981607@xmath142 & ( 6 , 3 , 1 ) & 9.506878307@xmath142 & ( 9 , 4 , 2 ) & 3.146104767@xmath142 + ( 10 , 0 , 0 ) & 1.029816377@xmath143 & ( 10 , 4 , 0 ) & -1.865761552@xmath142 & ( 7 , 3 , 1 ) & -7.727789536@xmath142 & ( 10 , 4 , 2 ) & -4.054217598@xmath142 + ( 11 , 0 , 0 ) & -2.786820323@xmath143 & ( 11 , 4 , 0 ) & -4.652610741@xmath142 & ( 8 , 3 , 1 ) & -1.457528700@xmath143 & ( 11 , 4 , 2 ) & -7.000652266@xmath142 + ( 3 , 2 , 0 ) & 2.666666667@xmath143 & ( 9 , 6 , 0 ) & 1.048701589@xmath144 & ( 9 , 3 , 1 ) & 1.806575878@xmath143 & ( 9 , 3 , 3 ) & 2.097403178@xmath142 + ( 4 , 2 , 0 ) & -5.333333333@xmath142 & ( 10 , 6 , 0 ) & -6.651097800@xmath145 & ( 10 , 3 , 1 ) & 8.928712117@xmath142 & ( 10 , 3 , 3 ) & -2.874385753@xmath142 + ( 5 , 2 , 0 ) & -1.173562610@xmath143 & ( 11 , 6 , 0 ) & -1.940149264@xmath144 & ( 11 , 3 , 1 ) & -2.245856051@xmath143 & ( 11 , 3 , 3 ) & -4.632188322@xmath142 + ( 6 , 2 , 0 ) & 8.604515263@xmath142 & ( 3 , 1 , 1 ) & 5.333333333@xmath143 & ( 9 , 5 , 1 ) & 1.258441907@xmath142 + ( 7 , 2 , 0 ) & -1.320134011@xmath142 & ( 4 , 1 , 1 ) & -2.962962963@xmath143 & ( 10 , 5 , 1 ) & -1.312853800@xmath142 + + ( 0 , 0 , 0 ) & 3.500000000 & ( 6 , 3 , 3 ) & 1.457965902@xmath144 & ( 6 , 3 , 2 ) & 5.831863610@xmath144 & ( 9 , 4 , 2 ) & @xmath237.128118497@xmath145 + ( 2 , 0 , 0 ) & @xmath237.333333333@xmath143 & ( 8 , 3 , 3 ) & @xmath238.399019298@xmath144 & ( 7 , 3 , 2 ) & @xmath234.705118313@xmath144 & ( 10 , 4 , 2 ) & @xmath236.158730192@xmath142 + ( 4 , 0 , 0 ) & 1.945310847@xmath143 & ( 9 , 3 , 3 ) & 2.106020257@xmath142 & ( 8 , 3 , 2 ) & @xmath235.937350737@xmath142 & ( 11 , 4 , 2 ) & 6.702194733@xmath142 + ( 5 , 0 , 0 ) & 1.570370370@xmath143 & ( 10 , 3 , 3 ) & 5.953632826@xmath144 & ( 9 , 3 , 2 ) & 3.295758313@xmath142 & ( 10 , 5 , 3 ) & 2.643901061@xmath145 + ( 6 , 0 , 0 ) & @xmath231.814429222@xmath143 & ( 11 , 3 , 3 ) & 4.246104998@xmath144 & ( 10 , 3 , 2 ) & 1.653086795@xmath142 & ( 11 , 5 , 3 ) & @xmath238.754870105@xmath146 + ( 7 , 0 , 0 ) & 3.336368382@xmath142 & ( 8 , 4 , 4 ) & 1.843920010@xmath145 & ( 11 , 3 , 2 ) & @xmath231.386696140@xmath142 & ( 6 , 3 , 0 ) & 5.831863610@xmath144 + ( 8 , 0 , 0 ) & @xmath231.977570791@xmath143 & ( 10 , 4 , 4 ) & @xmath233.073373856@xmath144 & ( 8 , 4 , 3 ) & 7.375680039@xmath145 & ( 7 , 3 , 0 ) & @xmath231.162496326@xmath144 + ( 9 , 0 , 0 ) & @xmath231.214962788@xmath143 & ( 11 , 4 , 4 ) & 4.127194476@xmath145 & ( 9 , 4 , 3 ) & @xmath235.394036567@xmath145 & ( 8 , 3 , 0 ) & @xmath231.040888381@xmath143 + ( 10 , 0 , 0 ) & 1.897994391@xmath143 & ( 10 , 5 , 5 ) & 2.643901061@xmath146 & ( 10 , 4 , 3 ) & @xmath232.344210680@xmath142 & ( 9 , 3 , 0 ) & 8.179624689@xmath142 + ( 11 , 0 , 0 ) & 7.710636344@xmath143 & ( 2 , 1 , 0 ) & 6.666666667@xmath143 & ( 11 , 4 , 3 ) & 3.300423401@xmath142 & ( 10 , 3 , 0 ) & 1.171660236@xmath143 + ( 2 , 1 , 1 ) & 3.333333333@xmath143 & ( 3 , 1 , 0 ) & 3.333333333@xmath143 & ( 10 , 5 , 4 ) & 1.057560424@xmath145 & ( 11 , 3 , 0 ) & 1.531781787@xmath142 + ( 3 , 1 , 1 ) & @xmath233.333333333@xmath143 & ( 4 , 1 , 0 ) & @xmath232.673809524@xmath143 & ( 11 , 5 , 4 ) & @xmath237.959003190@xmath146 & ( 8 , 4 , 1 ) & 2.212704012@xmath144 + ( 4 , 1 , 1 ) & @xmath231.192857143@xmath143 & ( 5 , 1 , 0 ) & @xmath234.024977954@xmath143 & ( 4 , 2 , 0 ) & 2.962962963@xmath142 & ( 9 , 4 , 1 ) & @xmath238.862200427@xmath145 + ( 5 , 1 , 1 ) & 2.904905203@xmath143 & ( 6 , 1 , 0 ) & @xmath231.302727805@xmath143 & ( 5 , 2 , 0 ) & @xmath232.629629630@xmath142 & ( 10 , 4 , 1 ) & @xmath239.978608234@xmath142 + ( 6 , 1 , 1 ) & 5.329329186@xmath143 & ( 7 , 1 , 0 ) & 3.499536691@xmath143 & ( 6 , 2 , 0 ) & @xmath232.375711537@xmath143 & ( 11 , 4 , 1 ) & 1.153622881@xmath143 + ( 7 , 1 , 1 ) & @xmath236.271803816@xmath143 & ( 8 , 1 , 0 ) & 4.079418136@xmath143 & ( 7 , 2 , 0 ) & 2.330090787@xmath143 & ( 10 , 5 , 2 ) & 4.230241698@xmath145 + ( 8 , 1 , 1 ) & @xmath237.522337850@xmath143 & ( 9 , 1 , 0 ) & 5.301735104@xmath142 & ( 8 , 2 , 0 ) & 3.039557383@xmath143 & ( 11 , 5 , 2 ) & @xmath231.750974021@xmath145 + ( 9 , 1 , 1 ) & 5.347837206@xmath143 & ( 10 , 1 , 0 ) & @xmath231.026434348 & ( 9 , 2 , 0 ) & @xmath233.529882617@xmath143 & ( 8 , 4 , 0 ) & 1.106352006@xmath144 + ( 10 , 1 , 1 ) & 1.146091910 & ( 11 , 1 , 0 ) & @xmath237.412817539@xmath143 & ( 10 , 2 , 0 ) & @xmath236.813379022@xmath143 & ( 9 , 4 , 0 ) & @xmath237.128118497@xmath145 + ( 11 , 1 , 1 ) & @xmath234.863394606@xmath143 & ( 4 , 2 , 1 ) & 5.925925926@xmath142 & ( 11 , 2 , 0 ) & 9.790814131@xmath143 & ( 10 , 4 , 0 ) & @xmath235.538838348@xmath142 + ( 4 , 2 , 2 ) & 1.481481481@xmath142 & ( 5 , 2 , 1 ) & @xmath232.629629630@xmath142 & ( 6 , 3 , 1 ) & 8.747795414@xmath144 & ( 11 , 4 , 0 ) & 8.043954590@xmath142 + ( 6 , 2 , 2 ) & @xmath235.533170820@xmath142 & ( 6 , 2 , 1 ) & @xmath238.610412317@xmath142 & ( 7 , 3 , 1 ) & @xmath235.867614638@xmath144 & ( 10 , 5 , 1 ) & 5.287802122@xmath145 + ( 7 , 2 , 2 ) & 1.147526740@xmath142 & ( 7 , 2 , 1 ) & 5.769980192@xmath142 & ( 8 , 3 , 1 ) & @xmath231.459267389@xmath143 & ( 11 , 5 , 1 ) & @xmath232.626461032@xmath145 + ( 8 , 2 , 2 ) & 3.484612337@xmath142 & ( 8 , 2 , 1 ) & 2.118965896@xmath143 & ( 9 , 3 , 1 ) & 1.611775400@xmath143 & ( 10 , 5 , 0 ) & 3.172681273@xmath145 + ( 9 , 2 , 2 ) & @xmath238.489781072@xmath142 & ( 9 , 2 , 1 ) & @xmath239.184491129@xmath142 & ( 10 , 3 , 1 ) & 1.623787831@xmath143 & ( 11 , 5 , 0 ) & @xmath239.550737020@xmath146 + ( 10 , 2 , 2 ) & @xmath238.365174384@xmath142 & ( 10 , 2 , 1 ) & @xmath233.638022235@xmath143 & ( 11 , 3 , 1 ) & @xmath233.687197934@xmath143 + ( 11 , 2 , 2 ) & 3.945274143@xmath143 & ( 11 , 2 , 1 ) & @xmath235.611998341@xmath144 & ( 8 , 4 , 2 ) & 1.475136008@xmath144 + rr|rr|rr|rr & & & & & & & + + ( 0 , 0 , 0 ) & 2.000000000 & ( 20 , 0 , 0 ) & 3.269830958@xmath147 & ( 8 , 1 , 1 ) & 1.107677837 & ( 20 , 3 , 1 ) & 1.363566623 + ( 2 , 0 , 0 ) & @xmath232.666666667 & ( 10 , 2 , 0 ) & @xmath239.187011686@xmath144 & ( 10 , 1 , 1 ) & @xmath235.258263841 & ( 12 , 2 , 2 ) & 2.213826787@xmath142 + ( 4 , 0 , 0 ) & 2.607407407 & ( 12 , 2 , 0 ) & 1.237459977@xmath143 & ( 12 , 1 , 1 ) & 2.361854957@xmath148 & ( 14 , 2 , 2 ) & @xmath232.801396443@xmath143 + ( 6 , 0 , 0 ) & @xmath235.667818930 & ( 14 , 2 , 0 ) & @xmath231.042288265 & ( 14 , 1 , 1 ) & @xmath231.041949744@xmath149 & ( 16 , 2 , 2 ) & 2.293697931 + ( 8 , 0 , 0 ) & 1.579276071@xmath148 & ( 16 , 2 , 0 ) & 7.139692772 & ( 16 , 1 , 1 ) & 4.570753372@xmath149 & ( 18 , 2 , 2 ) & @xmath231.550982309@xmath148 + ( 10 , 0 , 0 ) & @xmath234.977461910@xmath148 & ( 18 , 2 , 0 ) & @xmath234.367374242@xmath148 & ( 18 , 1 , 1 ) & @xmath232.003192131@xmath150 & ( 20 , 2 , 2 ) & 9.435217647@xmath148 + ( 12 , 0 , 0 ) & 1.689229038@xmath149 & ( 20 , 2 , 0 ) & 2.492266705@xmath149 & ( 20 , 1 , 1 ) & 8.787977137@xmath150 & ( 20 , 4 , 2 ) & 4.053738087@xmath146 + ( 14 , 0 , 0 ) & @xmath236.023423934@xmath149 & ( 18 , 4 , 0 ) & @xmath231.275445883@xmath151 & ( 14 , 3 , 1 ) & @xmath231.112602736@xmath145 & ( 18 , 3 , 3 ) & @xmath234.160122546@xmath144 + ( 16 , 0 , 0 ) & 2.225264150@xmath150 & ( 20 , 4 , 0 ) & 1.432818794@xmath145 & ( 16 , 3 , 1 ) & 6.891013162@xmath144 & ( 20 , 3 , 3 ) & 8.089745849@xmath142 + ( 18 , 0 , 0 ) & @xmath238.447437043@xmath150 & ( 6 , 1 , 1 ) & @xmath231.896296296@xmath143 & ( 18 , 3 , 1 ) & @xmath231.208165780@xmath143 + + ( 0 , 0 , 0 ) & 3.500000000 & ( 2 , 1 , 0 ) & 2.666666667 & ( 10 , 3 , 1 ) & @xmath233.833568660 & ( 10 , 4 , 0 ) & @xmath234.187749256@xmath143 + ( 2 , 0 , 0 ) & @xmath234.000000000 & ( 4 , 1 , 0 ) & @xmath232.607407407 & ( 12 , 3 , 1 ) & 3.749428458@xmath148 & ( 12 , 4 , 0 ) & @xmath234.625682644 + ( 4 , 0 , 0 ) & 2.311111111 & ( 6 , 1 , 0 ) & @xmath239.124044680@xmath143 & ( 14 , 3 , 1 ) & @xmath231.763568623@xmath149 & ( 14 , 4 , 0 ) & 3.879009243@xmath148 + ( 6 , 0 , 0 ) & @xmath233.578130511@xmath143 & ( 8 , 1 , 0 ) & 2.709462265@xmath148 & ( 16 , 3 , 1 ) & 5.362296757@xmath149 & ( 16 , 4 , 0 ) & @xmath231.755530196@xmath149 + ( 8 , 0 , 0 ) & @xmath231.447791909@xmath148 & ( 10 , 1 , 0 ) & @xmath231.454354060@xmath149 & ( 18 , 3 , 1 ) & @xmath234.338040480@xmath149 & ( 18 , 4 , 0 ) & 5.141129944@xmath149 + ( 10 , 0 , 0 ) & 8.449051521@xmath148 & ( 12 , 1 , 0 ) & 4.973154521@xmath149 & ( 8 , 4 , 2 ) & 3.776348180@xmath143 & ( 10 , 5 , 1 ) & 5.414709373@xmath143 + ( 12 , 0 , 0 ) & @xmath233.205718341@xmath149 & ( 14 , 1 , 0 ) & @xmath238.044807264@xmath149 & ( 10 , 4 , 2 ) & @xmath237.694903208@xmath143 & ( 12 , 5 , 1 ) & @xmath236.844512414@xmath143 + ( 14 , 0 , 0 ) & 7.683946158@xmath149 & ( 16 , 1 , 0 ) & @xmath233.592648436@xmath150 & ( 12 , 4 , 2 ) & @xmath233.834200865 & ( 14 , 5 , 1 ) & @xmath231.368955321@xmath148 + ( 16 , 0 , 0 ) & 1.864995221@xmath149 & ( 18 , 1 , 0 ) & 4.011882196@xmath147 & ( 14 , 4 , 2 ) & 4.301073460@xmath148 & ( 16 , 5 , 1 ) & 1.169642314@xmath149 + ( 18 , 0 , 0 ) & @xmath231.447204343@xmath147 & ( 4 , 2 , 1 ) & 9.481481481@xmath143 & ( 16 , 4 , 2 ) & @xmath232.363456039@xmath149 & ( 18 , 5 , 1 ) & @xmath235.467419040@xmath149 + ( 2 , 1 , 1 ) & 1.333333333 & ( 6 , 2 , 1 ) & @xmath232.072004703 & ( 18 , 4 , 2 ) & 9.094821992@xmath149 & ( 12 , 6 , 2 ) & 5.024245295@xmath143 + ( 4 , 1 , 1 ) & @xmath231.362962963 & ( 8 , 2 , 1 ) & 1.071796284 & ( 10 , 5 , 3 ) & 2.707354687@xmath143 & ( 14 , 6 , 2 ) & @xmath235.607740036@xmath143 + ( 6 , 1 , 1 ) & 3.117601411 & ( 10 , 2 , 1 ) & 1.934131152@xmath148 & ( 12 , 5 , 3 ) & @xmath236.099856791@xmath143 & ( 16 , 6 , 2 ) & @xmath231.688448287@xmath148 + ( 8 , 1 , 1 ) & @xmath231.157205899@xmath148 & ( 12 , 2 , 1 ) & @xmath231.610227509@xmath149 & ( 14 , 5 , 3 ) & @xmath233.591949866 & ( 18 , 6 , 2 ) & 1.530376091@xmath149 + ( 10 , 1 , 1 ) & 3.724437583@xmath148 & ( 14 , 2 , 1 ) & 9.100666397@xmath149 & ( 16 , 5 , 3 ) & 4.680082984@xmath148 & ( 14 , 7 , 3 ) & 4.576926174@xmath143 + ( 12 , 1 , 1 ) & @xmath234.436033620@xmath148 & ( 16 , 2 , 1 ) & @xmath234.053430224@xmath150 & ( 18 , 5 , 3 ) & @xmath232.944841481@xmath149 & ( 16 , 7 , 3 ) & @xmath234.694339733@xmath143 + ( 14 , 1 , 1 ) & @xmath235.334708723@xmath149 & ( 18 , 2 , 1 ) & 1.279544679@xmath147 & ( 12 , 6 , 4 ) & 2.009698118@xmath143 & ( 18 , 7 , 3 ) & @xmath231.886986003@xmath148 + ( 16 , 1 , 1 ) & 5.421019234@xmath150 & ( 6 , 3 , 2 ) & 3.732392710@xmath143 & ( 14 , 6 , 4 ) & @xmath234.925030503@xmath143 & ( 16 , 8 , 4 ) & 4.117367040@xmath143 + ( 18 , 1 , 1 ) & @xmath233.093983511@xmath147 & ( 8 , 3 , 2 ) & @xmath237.739781828@xmath143 & ( 16 , 6 , 4 ) & @xmath233.212472559 & ( 18 , 8 , 4 ) & @xmath234.052076547@xmath143 + ( 4 , 2 , 2 ) & 2.370370370@xmath143 & ( 10 , 3 , 2 ) & @xmath231.053007756 & ( 18 , 6 , 4 ) & 4.903204193@xmath148 & ( 18 , 9 , 5 ) & 3.670192948@xmath143 + ( 6 , 2 , 2 ) & @xmath235.237389771@xmath143 & ( 12 , 3 , 2 ) & 1.967483048@xmath148 & ( 14 , 7 , 5 ) & 1.525642058@xmath143 & ( 10 , 5 , 0 ) & 3.248825624@xmath143 + ( 8 , 2 , 2 ) & 1.010441376 & ( 14 , 3 , 2 ) & @xmath231.305817944@xmath149 & ( 16 , 7 , 5 ) & @xmath234.034853659@xmath143 & ( 12 , 5 , 0 ) & @xmath236.899424873@xmath143 + ( 10 , 2 , 2 ) & @xmath231.494008026 & ( 16 , 3 , 2 ) & 6.744405152@xmath149 & ( 18 , 7 , 5 ) & @xmath232.770916581 & ( 14 , 5 , 0 ) & @xmath236.463536863 + ( 12 , 2 , 2 ) & 1.846909265@xmath143 & ( 18 , 3 , 2 ) & @xmath232.972249754@xmath150 & ( 16 , 8 , 6 ) & 1.176390583@xmath143 & ( 16 , 5 , 0 ) & 6.567477615@xmath148 + ( 14 , 2 , 2 ) & 1.042676508@xmath148 & ( 8 , 4 , 3 ) & 1.888174090@xmath143 & ( 18 , 8 , 6 ) & @xmath233.342928209@xmath143 & ( 18 , 5 , 0 ) & @xmath233.388841937@xmath149 + ( 16 , 2 , 2 ) & @xmath231.111625374@xmath149 & ( 10 , 4 , 3 ) & @xmath232.886375899@xmath143 & ( 18 , 9 , 7 ) & 9.175482371@xmath142 & ( 12 , 6 , 1 ) & 6.698993726@xmath143 + ( 18 , 2 , 2 ) & 1.249271047@xmath150 & ( 12 , 4 , 3 ) & @xmath231.577708121 & ( 6 , 3 , 0 ) & 3.732392710@xmath143 & ( 14 , 6 , 1 ) & @xmath231.079421253 + ( 6 , 3 , 3 ) & 9.330981775@xmath142 & ( 14 , 4 , 3 ) & 1.855760450@xmath148 & ( 8 , 3 , 0 ) & @xmath239.806190570@xmath143 & ( 16 , 6 , 1 ) & @xmath232.116718725@xmath148 + ( 8 , 3 , 3 ) & @xmath232.869718840@xmath143 & ( 16 , 4 , 3 ) & @xmath231.206715261@xmath149 & ( 10 , 3 , 0 ) & @xmath237.606578761@xmath143 & ( 18 , 6 , 1 ) & 2.016971553@xmath149 + ( 10 , 3 , 3 ) & 7.302649849@xmath143 & ( 18 , 4 , 3 ) & 6.228213110@xmath149 & ( 12 , 3 , 0 ) & 1.900360917@xmath148 & ( 14 , 7 , 2 ) & 6.538465963@xmath143 + ( 12 , 3 , 3 ) & 6.474454323@xmath142 & ( 10 , 5 , 4 ) & 1.082941875@xmath143 & ( 14 , 3 , 0 ) & @xmath231.096944607@xmath149 & ( 16 , 7 , 2 ) & @xmath236.381513372@xmath143 + ( 14 , 3 , 3 ) & @xmath231.507976852@xmath148 & ( 12 , 5 , 4 ) & @xmath237.182655253@xmath142 & ( 16 , 3 , 0 ) & 4.525126033@xmath149 & ( 18 , 7 , 2 ) & @xmath232.878600133@xmath148 + ( 16 , 3 , 3 ) & 1.112739551@xmath149 & ( 14 , 5 , 4 ) & @xmath231.662709009 & ( 18 , 3 , 0 ) & @xmath231.604127694@xmath150 & ( 16 , 8 , 3 ) & 6.176050560@xmath143 + ( 18 , 3 , 3 ) & @xmath235.501017323@xmath149 & ( 16 , 5 , 4 ) & 1.713732791@xmath148 & ( 8 , 4 , 1 ) & 5.664522270@xmath143 & ( 18 , 8 , 3 ) & @xmath231.660697208@xmath143 + ( 8 , 4 , 4 ) & 4.720435225@xmath142 & ( 18 , 5 , 4 ) & @xmath231.134996640@xmath149 & ( 10 , 4 , 1 ) & @xmath231.358637239 & ( 18 , 9 , 4 ) & 5.709189031@xmath143 + ( 10 , 4 , 4 ) & @xmath231.753576976@xmath143 & ( 12 , 6 , 5 ) & 6.698993726@xmath142 & ( 12 , 4 , 1 ) & @xmath235.476288758 & ( 12 , 6 , 0 ) & 3.349496863@xmath143 + ( 12 , 4 , 4 ) & 5.665451332@xmath143 & ( 14 , 6 , 5 ) & 2.348849413@xmath142 & ( 14 , 4 , 1 ) & 6.555818772@xmath148 & ( 14 , 6 , 0 ) & @xmath232.248529752@xmath143 + ( 14 , 4 , 4 ) & 7.031489323@xmath143 & ( 16 , 6 , 5 ) & @xmath231.498021733 & ( 16 , 4 , 1 ) & @xmath233.561148091@xmath149 & ( 16 , 6 , 0 ) & @xmath231.304144759@xmath148 + ( 16 , 4 , 4 ) & @xmath232.234394856@xmath148 & ( 18 , 6 , 5 ) & 1.514875202@xmath148 & ( 18 , 4 , 1 ) & 1.327749407@xmath150 & ( 18 , 6 , 0 ) & 1.069591051@xmath149 + ( 18 , 4 , 4 ) & 1.696295810@xmath149 & ( 14 , 7 , 6 ) & 4.358977309@xmath142 & ( 10 , 5 , 2 ) & 4.331767498@xmath143 & ( 14 , 7 , 1 ) & 7.628210290@xmath143 + ( 10 , 5 , 5 ) & 2.707354687@xmath142 & ( 16 , 7 , 6 ) & 6.397044909@xmath142 & ( 12 , 5 , 2 ) & @xmath238.287589771@xmath143 & ( 16 , 7 , 1 ) & @xmath231.648715187@xmath143 + ( 12 , 5 , 5 ) & @xmath231.145505796@xmath143 & ( 18 , 7 , 6 ) & @xmath231.267992012 & ( 14 , 5 , 2 ) & @xmath238.117281873 & ( 18 , 7 , 1 ) & @xmath233.918669310@xmath148 + ( 14 , 5 , 5 ) & 4.544498080@xmath143 & ( 16 , 8 , 7 ) & 2.940976457@xmath142 & ( 16 , 5 , 2 ) & 8.399862105@xmath148 & ( 16 , 8 , 2 ) & 8.234734079@xmath143 + ( 16 , 5 , 5 ) & 1.078497914 & ( 18 , 8 , 7 ) & 7.770975088@xmath142 & ( 18 , 5 , 2 ) & @xmath234.609565870@xmath149 & ( 18 , 8 , 2 ) & 1.073653528@xmath143 + ( 18 , 5 , 5 ) & @xmath232.724985485@xmath148 & ( 18 , 9 , 8) & 2.038996082@xmath142 & ( 12 , 6 , 3 ) & 3.349496863@xmath143 & ( 18 , 9 , 3 ) & 8.563783546@xmath143 + ( 12 , 6 , 6 ) & 1.674748432@xmath142 & ( 4 , 2 , 0 ) & 4.740740741@xmath143 & ( 14 , 6 , 3 ) & @xmath234.223690648@xmath143 & ( 14 , 7 , 0 ) & 4.358977309@xmath143 + ( 14 , 6 , 6 ) & @xmath237.829101098@xmath142 & ( 6 , 2 , 0 ) & @xmath236.512874780@xmath143 & ( 16 , 6 , 3 ) & @xmath239.671323639 & ( 16 , 7 , 0 ) & @xmath234.703472705@xmath143 + ( 16 , 6 , 6 ) & 3.741008913@xmath143 & ( 8 , 2 , 0 ) & @xmath231.556811959 & ( 18 , 6 , 3 ) & 9.548744493@xmath148 & ( 18 , 7 , 0 ) & @xmath231.979896930@xmath148 + ( 18 , 6 , 6 ) & 1.267008305 & ( 10 , 2 , 0 ) & 1.207544510@xmath148 & ( 14 , 7 , 4 ) & 2.615386385@xmath143 & ( 16 , 8 , 1 ) & 1.029341760 + ( 14 , 7 , 7 ) & 1.089744327@xmath142 & ( 12 , 2 , 0 ) & @xmath233.196277031@xmath148 & ( 16 , 7 , 4 ) & @xmath231.332433821@xmath143 & ( 18 , 8 , 1 ) & @xmath234.479185205@xmath143 + ( 16 , 7 , 7 ) & @xmath235.528545914@xmath142 & ( 14 , 2 , 0 ) & @xmath235.271684834@xmath148 & ( 18 , 7 , 4 ) & @xmath231.033273379@xmath148 & ( 18 , 9 , 2 ) & 1.141837806 + ( 18 , 7 , 7 ) & 3.138641071@xmath143 & ( 16 , 2 , 0 ) & 1.018254901@xmath150 & ( 16 , 8 , 5 ) & 2.058683520@xmath143 & ( 16 , 8 , 0 ) & 5.146708800@xmath143 + ( 16 , 8 , 8) & 7.352441142@xmath144 & ( 18 , 2 , 0 ) & @xmath235.459613629@xmath150 & ( 18 , 8 , 5 ) & 6.203359335@xmath142 & ( 18 , 8 , 0 ) & 2.583076501@xmath143 + ( 18 , 8 , 8) & @xmath234.001373177@xmath142 & ( 6 , 3 , 1 ) & 5.598589065@xmath143 & ( 18 , 9 , 6 ) & 1.631196866@xmath143 & ( 18 , 9 , 1 ) & 1.284567532 + ( 18 , 9 , 9 ) & 5.097490206@xmath144 & ( 8 , 3 , 1 ) & @xmath239.918360628@xmath143 & ( 8 , 4 , 0 ) & 2.832261135@xmath143 & ( 18 , 9 , 0 ) & 7.136486288@xmath143 +